i
Vowel Quality and Phonological Pro jection
Marc van Oostendorp
PhD Thesis
Tilburg University
September
Acknowledgements
The following p eople have help ed me prepare and write this dissertation
John Alderete Elena Anagnostop oulou Sjef Barbiers Outi Bat El
Dorothee Beermann Clemens Bennink Adams Bo domo Geert Bo oij Hans
Bro ekhuis Norb ert Corver Martine Dhondt Ruud and Henny Dhondt
Jo e Emonds Dicky Gilb ers Janet Grijzenhout Carlos Gussenhoven Gert
jan Hakkenb erg Marco Haverkort Lars Hellan Ben Hermans Bart Holle
brandse Hannekevan Ho of Angeliek van Hout Ro eland van Hout Harry
van der Hulst Riny Huybregts Ren e Kager Hans Peter Kolb Emiel Krah
mer David Leblanc Winnie Lechner Klarien van der Linde John Mc
Carthy Dominique Nouveau Rolf Noyer Jaap and Hannyvan Oosten
dorp Paola Monachesi Krisztina Polg ardi Alan Prince Curt Rice Henk
van Riemsdijk Iggy Ro ca Sam Rosenthall Grazyna Rowicka Lisa Selkirk
Chris Sijtsma Craig Thiersch MiekeTrommelen Rub en van der Vijver
Janneke Visser Riet Vos Jero en van de Weijer Wim Zonneveld
Iwant to thank them all They have made the past four years for what
it was the most interesting and happiest p erio d in mylife until now ii
Contents
Intro duction
The Headedness of Syllables
The Headedness Hyp othesis HH
Theoretical Background
Syllable Structure
Feature geometry
Sp eci cation and Undersp eci cati on
Skeletal tier
Mo del of the grammar
Data
Organisation of the thesis
Chapter
Chapter
Chapter
Chapter
Chapter
Chapter
Chapter
App endices
Vowel Quality and Rhyme Structure in Dutch
Intro duction
The Dutchvowel system
Tenseness versus Length The Case of Dutch
The argument for length
The arguments against length
A theory based on the feature lax
lax and syllable structure in vowel harmony
Tenseness and branchingness in Dutch iii
CONTENTS iv
Formalization in Optimality theory
Some more arguments for the length of A vowels
Tenseness cannot b e de ned in a satisfactory way
phonetically
Minimality requires branching
A vowels form the domain of tonal contour in Lim
burg Dutch
Conclusion
App endix Historical overview
Dutch structuralism
Pre generative literature
Early generative grammar
Bisegmental analyses in generative phonology
Tilburg Dutch and Standard DutchVowel Length
Details in the Standard Dutchvowel system
Ambisyllabicity
r lengthening
The phonetic nature of the tensing feature
Extrasyllabicity and catalexis
A dialect with real length Tilburg Dutch
The vowel system
Why only lax vowels can b e long
Vowel shortening
Analysis
Long vowels in other Brabant dialects
The limited distribution of long vowels
Conclusion
Derivation of the Dutchvowel system
Conclusion
Derived schwa in Dutch
Intro duction
Prop erties of r schwa
Word initial p osition
Word nal p osition
Vowel quality
Closed syllables
Style registers
Prop erties of e schwa
CONTENTS v
The ep enthetic vowel is schwa
E schwa do es not o ccur at the end of the word
E schwa only o ccurs in the last syllable of the word
Word internal contexts in whiche schwa do es not o ccur
Style registers
Summary and conclusion
Dutch U schwa
Intro duction
Prop erties of u schwa
Syllable weight
U schwa do es not o ccur word initially
Some other segmental e ects
The onset of schwa syllables
The co da of schwa headed syllables
Degenerate and sc hwa headed syllables
Obligatory versus optional ep enthesis
Again on complex onsets
Schwa surrounded by identical consonants
Schwa after ng
U schwa and stress
Adjacency b etween schwa and full vowels
Complementary distribution of u schwa and e schwa
Schwa deletion
Conclusion
Previous analyses of u schwa
Reduction Theory
Ep enthesis Theory
No Syllable Theory
Remaining problems
Final Devoicing
Sup erheavy syllables b efore schwa
Post lexical u schwa
Conclusion
Table of Prop erties
SchwainFrench and Norwegian
Intro duction
French
E schwaistheepenthetic vowel
E schwa do es not o ccur at the end of the word
CONTENTS vi
U schwamust o ccur in an op en syllable
Laxing in the head of a fo ot
U schwa do es not o ccur at the b eginning of the word
Consonant clusters b efore schwa cannot b e p ossible
complex onsets
Schwa is stressless
Schwa cannot o ccur next to a vowel
Schwa deletion
A parameter
Conclusion plus a note on learnability
Norwegian
Schwa is the ep enthetic vowel
Ep enthetic schwadoesnotoccurattheendoftheword
Schwamust o ccur in an op en syllable
Schwa do es not o ccur at the b eginning of the word
Consonant clusters b efore schwa cannot b e p ossible
complex onsets
Alternation with degenerate syllables
Conclusion and another note on learnability
Conclusion
AVowel Glide Alternation in Rotterdam Dutch
Intro duction
The second p erson clitic
Hiatus after high vowels
After coronal stops
Third p erson singular clitic
The diminutivesu x
Sievers s Law
Other issues
Clitics and the diminutive in Standard Dutch
Lexical forms
The underlying form of S is not i
High vowel followed byschwa
st p erson plural clitic
Conclusion
The Pro jection ConstraintFamily
Intro duction
Pro jection and weakness
The fo ot level
CONTENTS vii
N level The
N level
Nuclear level
Features
Constraints con icting with pro jection and weakness
Conclusion
A Constraints and Families of Constraints
A Proso dic well formedness
A Syllable well formedness
A Foot well formedness
A Word well formedness
A Autosegmental Representations
A Feature Co o ccurrence and Licensing
A Parsing
A Constraints against unnecessary structure
A Ad Ho c Constraint
B Arguments for ranking
B Top ology of the Dutch lexicon
B Top ology of the Dutch p ostlexical phonology
C Ranking schemes
C Top ology of the Standard Dutch lexicon
C Top ology of the Standard Dutch p ostlexical phonology
C Top ology of the French phonology
C Top ology of the Rotterdam Dutch phonology
References
Summary in Dutch
Summary in English
Intro duction
The Headedness of Syllables
Almost all phonological theories which recognise the syllable as a con
stituent also p ostulate that this constituent has a head The notions head
and constituent are formalised in di erentways in di erent theories but
the underlying intuition seems to b e the same in all formalisms In this
thesis I explore some consequences of this headedness hyp othesis whichI
think have not su ciently b een worked out
The criteria for headedness are quite diverse but fortunately they con
verge most of the time I will mention the most imp ortant ones here
One criterion is sonority As is well known universally the syllable
consists of a monotonic sonority rise optionally followed by a monotonic
sonority fall see Basb ll Clements for a few recent discussions
Therefore every syllable has exactly one sonority p eak We could call this
p eak segment the head
Another criterion is prominence Some phonological features like stress
or tone seem phonologically not to b e assigned to individual segments but
rather to syllables or other proso dic constituents When these features get
realized they generally do so on one sp eci c segment Again we could
call this segmentthe head of the syllable Clements Hermans a
Odden
Inversely it has sometimes b een argued that the fact that vowel fea
tures can spread skipping consonants in vowel harmony pro cesses whereas
owels is an consonantal features seem to never spread over intervening v
argument for headedness Kaye Lowenstamm and Vergnaud Van Lit
Van der Hulst and Van de Weijer Vowels b eing typical syl
lable heads pro ject to a level for instance the nucleus where consonants
are no longer visible Whereas there is a level at which these vowels are
adjacent to one another there is no similar level for consonants
A third criterion is dispensability Whenwehave a syllable like tap in
a language we can strip o the t and or the p still retaining a complete
syllable ta ap a But in most languages we cannot do the same
The Headedness of Syllables
with the a tp We could call this indisp ensi ble segmentthehead
Unfortunately this criterion is not very strong as it stands b ecause it gives
undesirable results in the case of diphthongs pai and in languages with
syllabic sonorants pan In the rst case b oth a and i would b e
assigned the status of head b ecause the language would presumably allow
b oth pa and pi syllables in the second case b oth a and n would
get the same status b ecause the language would allow b oth pa and pn
On the other hand we will see that some syllable heads in some languages
can only o ccur in closed syllables Also the onset seems to b e an obligatory
sub constituent of syllables in many languages without it b eing a head In
those cases this criterion would say the syllable has no head at all Still the
intuition b ehind this criterion seems clear the head is the segmentthat
determines to a considerable extent the nature of the syllable
A last p ossible criterion amply discussed in Dresher and Van der Hulst
is complexity According to these scholars heads in proso dic struc
ture in general allow more complexity than their dep endents For instance
the typical head of a fo ot is a heavy syllable whereas the typical dep endent
is light Although Dresher and Van der Hulst do not explicitly discuss the
internal structure of syllables we might assume that segments have some
internal complexity measured for instance bythenumb er of marked fea
tures they contain we then exp ect every syllable to contain exactly one
most complex segment whichwewould call the head Some phonologists
haverecently argued for a complexity metric in the sense discussed here
Harris Rice Dogil Interestingly this notion of complex
ity seems to coincide with sonority The most sonorous segmen ts are shown
to b e the most complex ones by the latter two authors but the rst author
gives exactly the opp osite interpretation This criterion is the only one that
do es not seem to b e directly compatible with the framework prop osed in
this thesis where some kinds of complexity cause segments to b e syllable
heads whereas other kinds of complexity cause segments to b e dep endents
in the syllable
Although the Headedness Hyp othesis is accepted bymany phonologists
Icontend that it has more consequences than is usually acknowledged
In other areas of grammatical theory it is often assumed that features of
the head can determine the structure of the pro jection which is built on that
This situation would b ecome even worse if we consider languages suchasBerber Dell
and Elmedlaoui or Bella Co ola Bagemihl in such languages anysegment
can form a syllable of its own so that the criterion would collapse
The most well known current theory that do es not accept the syllable as a constituent
at all is Government Phonology Kaye Lowenstamm and Vergnaud Kaye
The most well known theory that recognizes the syllable but no headedness is mora
theory under some of its guises Hyman Hayes b but cf also Hayes
The Headedness of Syllables
head For instance in the so called minimalist view of syntactic structure
Chomsky a lexical head which p ossesses a strong feature F requires
a functional head to b e present higher up in the pro jection which can license
F cf Muysken and Van Riemsdijk for an older collection of articles
dealing with features and pro jection in syntax Similarily in morphology
it is often assumed that morphological heads can have sub categorization
frames which determine the structure of the word i e the presence vs
absence of other morphemes
In phonological theory the head is not often assumed to play this role
In this thesis I will derive and defend a more sp eci c hyp othesis on the role
of the syllable head in proso dic structure whichsays that the head of a
syllable de nes the internal structure of that syllable
Headedness The structure of a syllable is determined by the
feature structure of its head
In particular I will argue that the feature structure of heads determines
to a large extent the licensing of onset and co da p ositions in the syllable
For instance it is well known that reduced vowels often show a severely
simpli ed syllable structure and do not allow co das or complex onsets I
will argue that there are constraints of the following typ es i if a segment
dominates a feature F it should head a branching constituent of typeTand
ii if a segment heads a branching constituent of type T it should dominate
afeatureF I call these constraints pro jection constraints I will show
that there are such constraints for every level of the metrical hierarchy upto
and including the fo ot level and for several typ es of features most notably
those features referring to vowel height This is the central h yp othesis of
this thesis
Below I will brie y intro duce the meaning of this hyp othesis Subse
quently I will o er a division into chapters of the material which has to b e
discussed
The Headedness Hyp othesis HH Icho ose to adopt the
following template for a maximal syllable cf section on several details
of this structure like the status of the onset graphically representing the
headedness prop erty in the most straightforward way
The Headedness of Syllables
N
N
N
N
b r k
The Headedness Hyp othesis claims that the variable prop erties of the struc
ture in are dep endent on the feature material in the head Sp eci cally
I will prop ose certain conditions relating the presence of phonological fea
tures in the head and the branchingness of X no des of some pro jection
level
It is a direct consequence of the Headedness Hyp othesis that no part of
syllable structure should b e underlying Underlying representations only
contain phonological features and p erhaps asso ciation lines I assume that
skeletal p oints X slots or morae cannot b e part of underlying repre
sentations This means that underlying segmental length long vowels or
geminate consonants can only b e represented bywayofatwo ro ot rep
resentation Selkirk It also has some obvious consequences for the
underlying representation of glides and vowels as will b e discussed in chap
ter
Yet I will argue that also this typ e of representation is not always nec
essary and that phonetic length can in some cases also b e the result of
the presence versus absence of some feature I will show that this is one
instance of the well known tense lax distinction which corresp onds to the
op en versus closed structure of syllables
It is usually claimed that Dutch phonology draws a distinction b e
tween vowels which are long and vowels which are short on the basis
of paradigms such as the following
a kim horizon k m kip chicken k p krimp shrink kr mp
b kiem germ kim keep goalie kip kriemp krimp
Phonetically the di erence is one of tenseness rather than length at least
in the case of the high vowels but this tenseness can b e derived from
the length The phonetic mo dule automatically interprets long vowels as
tense and subsequently shortens tense high vowels under this standard
interpretation
The gap in can b e explained if we accept that i Dutchallows at
most three segments in the rhyme mo dulo coronal stops and ii the vowel
The Headedness of Syllables
i o ccupies two slots in the rhyme whereas the vowel o ccupies only one
slot i e we distinguish b etween the twovowels by representing them as
in where I stands for the segmental material which is supp osed to b e
the same for i and under some analyses Similar distinctions could b e
made for instance b etween a and o and and and
a x
I
b x x
I
A lot of empirical problems arise however if we assume the representations
in All these problems lead me to conclude that Dutch do es not havea
length distinction at all The underlying di erence rather has to b e stated
in terms of lax Furthermore the Dutchrhyme allows two p ositions rather
than three b oth r mp and rip are larger than monosyllabic parsing the
p into a degenerate syllable These assumptions allow me to explain the
paradigm in by the following constraint
Connect N lax
N N dominates lax N branches a Project lax
b Project N lax N branches N dominates lax
This is an instance of what I call a biconditional pro jection constraint
Such constraints will b e generally referred to in this thesis by the name
Connect X Y In many places it will b e useful to split the biconditional
formulation into two simple conditional formulations These will b e referred
to as Project X Y Both parameters X and Y can range over a set of
features or the set of pro jection levels
The constraint Connect lax N can b e shown to b e op erativealsoin
a harmony language like Eastern Javanese in this language all the vowels
in a word are lax i its nal syllable is closed Similar conclusions can b e
reached for some Portuguese Spanish and French dialects and for Tagalog
Chamorro has a constraint relating the op en closed character of a syllable
to the height of the dominating vowel
All the prop erties of Dutch just mentioned get a natural explanation
under the present assumptions Furthermore by it is also explained
why syllables cannot end in a lax vowel Dutch has C V syllables in all
cases where V is not a lax vowel
The use of a parameter like notation in Connect X Y of course sug
gests that there are other p ossibiliti es Ihopetoshowinthefollowing
The Headedness of Syllables
chapters that Dutch has evidence for pro jection constraints at every level
of pro jection up to and including the fo ot i e at the N N N and fo ot
level In most cases we refer to branching constituents for instance the
constraint Project Ft lax in the analysis of French crucially refers to
branching feet this is indicated by the subscript Yet in the chapters
and I will refer in a few cases to a constraint on pro jection to the head
of a fo ot regardless whether this fo ot is branching or not
The range of features which can b e mentioned in pro jection constraints
is also quite large and includes at least the ma jor class features consonantal
and sonorant the features high low and lax which I will call the
aperture features tone and the feature sets V the set of all vo calic features
and F the set of all features
a V fhigh low lax coronal labial dorsal cons son g
v v v
the set of vo calic features
b C fcoronal labial dorsal g the set of consonantal fea
c c c
tures
c F the set of all segmental features
A complete formal discussion of all p ossible and imp ossible pro jection con
straints will b e provided in chapter
It mightbeinteresting that one of the most well known constrain ts of
syllabic theory can now b e reformulated as a pro jection constraint
Onset Project F N IfanN has any feature F F in its
head N should branch
According to this version of Onset the only onsetless syllables that are
allowed are syllables with a completely empty head This is what makes
it di erent from other formulations of this constraint Yet we could argue
that empy headed syllables without an onset will b e ruled out by indep en
dent considerations of representational economy As so on as a syllable has
anything in its head it should also have something in its onset also accord
ing to For ease of reference I will stick to the name Onset instead of
Project X Y throughout this thesis
For now it is imp ortant to note that I assume that the pro jection con
straints can b e divided into two subsets one b ottom up e g Project lax
N and one top down e g Project N lax It is useful to distinguish
between the twotyp es of constraints b ecause in some languages we nd
evidence for the b ottom up version of a constraint but not for the top down
Or by the constraint Contour p
The Headedness of Syllables
version or vice versa If we assume a principle of constraintinteraction
it can b e shown that b ottom up and top down constraints play di erent
roles in the hierarchy For instance in some French dialects mid vow
els in closed syllables get always laxed chapter showing an e ect of
Project N lax Yet the re ex of Project lax N simply cannot b e
observed mid lax vowels can also o ccur in op en syllables This shows that
the former constraint is visibly activeinFrench but the latter is not
Pro jection constraints in this thesis refer either to the sp eci c features
cons son lax or low or to one of the feature sets F the set of all
features or V the set of all vo calic features
It will also b e necessary in a few cases to refer to the logical opp osite of
pro jection constraints These are constraints that sayeither i if a segment
has a feature F it should beinadependent position at level T or ii if a
segment is in a dependent position at level T it should have a featureF All
weakness constraints we encounter in this thesis refer to either high or
consonantal Such constraints I call weakness constraints Examples
of such constraints are the constraints whichsaythathighvowels are the
b est candidates for gliding Weak N high and that consonants prefer
to o ccur in the margin of a syllable Weak cons N Constraints like
these will b e intro duced in chapters and
Inow turn to schwa I will analyse this vowel as an empty root node
with just the ma jor class features sp eci ed on it I think it is p ossible to
show that the many prop erties of schwa can b e derived from its b eing an
empty root node
The main prop erties can b e stated as follows schwadoesnotallow
complex onsets it cannot b e stressed and it only allows sonorant consonants
in the co da I wish to argue that each of these prop erties follows from the
defective feature contentofthisvowel For this reason it can only pro ject
a minimal structure i e a core CV syllable
We already knowthatschwa cannot head a closed syllable b ecause of
I supp ose syllables where schwa seems to b e followed by a sonorant
actually have this sonorant as the head of the syllable phonologically the
schwa is lled in later say in the phonetic comp onent
What I need to derive then is that placeless vowels do not allow adjunc
tion even though like all vowels they require an onset Potential complex
onsets always consist of an obstruent sonorant sequence in Dutch By the
reasoning just outlined these sequences could as well b e syllables of their
own I will argue it is actually even b etter for them to b e complete syllables
than to b e adjunction structures to an empty headed syllable
If these assumptions are worked out in the prop er way they explain at
least the rst two prop erties of schwa immediately Supp ose for instance
The Headedness of Syllables
that a full syllable like drie three dri has the following structure
N
N
N
N
d r i
Supp ose that vo calic place features are necessary in order to license the
higher level N Since schwadoesnothave these features it can only license
the lower level N
N
N
N
m
We can also assume that N cannot branch If it did by it would get a
sp eci cation as lax Because vowels with only ap erture sp eci cations do
not surface in Dutch this would in turn trigger a whole apparatus of rules
lling in also place features for schwa whichwould nally surface as as
in French ach ter to buy il ach te he buys The question now
is why in Dutchwe can nd sonorant consonants after the schwa vader
father adem breathe open id vad r ad m op n I prop ose
that in these cases the schwa is not underlying but rather a sp eci c form
of phonetic sp ell out Standard Dutch uses for syllabic sonorants This
means that at the end of the word sonorant consonants have the p ossibility
of pro jecting a CV syllable The last syllable of vader can b e represented
as
N
N
N
d r
Theoretical Background
This concludes the outline of my attempt to uphold the Headedness Hy
p othesis in the phonology of Dutch Approximately the same problems
arise in the phonology of French and Norwegian I will show these prob
lems can b e solved in this same framework and with a few minor di erences
in the sp eci cation of the grammar of these three languages
Theoretical Background
In this section I will present a discussion of the phonological framework
adopted in this study Unfortunately there is nothing like a uniform Stan
dard theory of phonology to which one can refer at this moment In a study
like this one therefore has to make partly arbitrary c hoices b etween
comp eting phonological theories on topics which are not of direct concern
to the study at hand In this section I present some of the particular the
oretical choices I have made A more complete overview of recent theories
can b e found in works such as Kenstowicz a Ro ca and Gold
smith This section can b e used as a reference while studying the
chapters that follow
Syllable Structure The representation of syllable structure I
X theoretic constituent but one without skeletal p oints X slots use is an
or mor cf section
N
N
N
N
b r k
Versions of X bar theoretic representations have b een prop osed in Levin
Uriagereka Michaels Harris and Hermans
a We will see b elow that there is also some evidence for branching
nuclei in the representation of long vowels and diphthongs chapter
The choice for this particular notation is partly arbitrary InVan Oost
endorp I have established equivalence classes of trees X structures
bracketed grids dep endency graphs and autosegmental representations
Each of these can imitate imp ortant asp ects of the b ehaviour of anyof
Theoretical Background
the others We can therefore cho ose the notation which suits our purp oses
b est
An advantage of X structures is that they graphically representthehead
relations that form the theme of this dissertation in a fairly straightforward
way With a few adaptations an alternative notation like the mora theory
Hyman Hayes b can also represent headedness but it do es
not seem to b e particularily suitable for this goal The same is true for the
classical onset rhyme nucleus theory The notation used in Government
Phonology Kaye Lowenstamm and Vergnaud is not able to
represent the relation b etween syllable head and onset segments in the way
which is needed here In other words I crucially need a syllable no de
for some of my analyses to hold Dep endency trees nally are simply
equivalent to the X bar structures used here I havechosen for the latter
b ecause they are more familiar b ecause the morphological and syntactic
theory I assume throughout this study is Government and Binding theory
Chomsky rather than Dep endency syntax and b ecause
my assumptions on segmental structure di er from those of Dep endency
Phonology
Furthermore recentdevelopments in syntactic X theory seem to help
solving some problems The classical problem for instance why onsets
o ccur on the left and co das on the right can in X theory b e reduced to the
more general problem of why al l X structure syntactic or phonological has
a Sp eci er Head Complement order Kayne and Chomsky
have argued that syntactic X bar structure universally corresp onds to one
constituent order only This order seems to b e exactly the one prop osed
for the syllable
The representation in would not b e an acceptable X bar repre
sentation in the sense of Kayne b ecause the sisters to the main
pro jection line are not maximal pro jections themselves and b ecause the N
no de pro jects twice Both of these latter two options are allowed within
Chomsky s version of bare phrase structure theory the graphical
asp ects of which I will therefore adopt Another parallelism b etween the
X bar structure used here and Chomsky s bare phrase structures is that
N no des This corresp onds exactly to the number there can b e at most two
of segments in the onset
The most signi cant problem with the syllable tree in lies in its
relation to stress The di erence b etween heavy versus light syllables cannot
b e captured easily in this framework This is the reason why many theorists
have adopted a moraic mo del of syllable structure Stress is not the main
topic of this study however and furthermore various prop osals have b een
Cf Chomsky for yet another view of syntactic X structure
Theoretical Background
made in the literature to reconcile non moraic mo dels of syllable structure
with the stress facts One of these prop osals could b e adopted
A large part of this dissertation will b e devoted to an argumentthat
n
the branching or non branching of a given N no de is dep endent on the
presence or absence of phonological features under N and vice versa I will
for instance argue that N can only branch if its head is lax in Dutch The
structure of a syllable therefore is very much dep endent on the head of that
syllable
I will not discuss the structure of the onset in any detail With Levin
McCarthy and Prince Hayes b and many others I as
sume that the onset do es not exist as a constituent Since I also assume
that phonological metrical structure is binary branching Lieb erman and
Prince Kiparsky seems the most plausible structure For
syntax it has b een argued that every phrase contains maximally one sp eci
er and one adjunct i e two elements dominated by N byZwart
and Chomsky Although it is hard to nd criteria distinguishing
between sp eci ers and adjuncts in syllable structure it seems to b e gen
erally accepted that in many languages the maximal numb er of segments
inanonsetistwo mo dulo the coronal fricatives I will give indep endent
evidence for the assumption that b oth onset segments are dep endenton
the head in chapter
Feature geometry I adopt the following feature tree McCarthy
Clements and Hume
cons son
Laryngeal continuant nasal
voice C place
coronal vo calic dorsal labial
lateral
Ap erture
V place high low lax
coronal labial dorsal
Some asp ects of this tree I havechosen simply for the sake of explicitnes s
Theoretical Background
For instance in most chapters I am not concerned with the structure
of the Laryngeal no de Ihave adopted the prop osal byLombardi
concerning this p oint Also the exact p osition of nasal lateral and
esp ecially continuant Selkirk Wetzels Steriade Padgett
is of no direct relevance to anyofmy claims
While I do need to draw a distinction b etween a C place and a V place
no de I am not sure the dep endency relation established here is crucial
A tree structure like the one prop osed in Lahiri and Koreman can
p ossibly work as well I also assume that the class no des lab eled C place
V place Laryngeal vo calic and Ap erture are structural non terminals
they cannot o ccur in a representation unless they dominate some feature
Except for the ma jor class features Clements all features have
only one activevalue in the parts of the phonology that I study I will
assume that these features are monovalent but similar results can of course
b e obtained by theories of undersp eci cati on or of feature value markedness
The activevalue of all features is F
Ihavechosen what seemed to me the most common lab els for eachof
the features except p erhaps for the Ap erture features for which again I
refer to the next chapter I omitted several features to whichInever make
any reference at all such as strident and aspirated
Crucial asp ects of this tree for my analysis are the fact that the ma
jor class features are sp eci ed on the ro ot McCarthy and further
more the structure directly under the ro ot C place vo calic V place and
Ap erture no des Clements a and the identityofvo calic and
consonantal place features Clements and Hume
Finally there is some ongoing debate on the status of the class no des
in The classical interpretation of these no des as real autosegmental
elements present on their own tier has recently b een questioned byHa yes
Van de Weijer and Padgett among others These au
thors have argued for an alternativeinterpretation under which lab els such
as Place or Supralaryngeal refer to sets of features rather than to indep en
dent no des Since these prop osals do not seem to diminish the descriptive
power of the theory in anyway my prop osals here are neutral with resp ect
to this issue of interpretation Yet I do assume that class no des V place
or Ap erture can never b e terminal elements in the representation and this
assumption seems to b e more easily captured under the interpretation of
Hayes Van de Weijer and Padgett than under the classical interpretation
A large part of this thesis deals with Dutch phonology Inchapter I
will amply discuss the feature structure of the Dutchvowels and in chapter
In chapter I will brie y consider the p osition of tone in the hierarchy and adopt a
prop osal under which it is dep endent on the Laryngeal no de
Theoretical Background
I discuss the glides j and w and some palatalized consonants typical
for Rotterdam Dutch Here I simply list the Standard Dutch consonants
with their feature sp eci cations
b t p k d b s f x
consonantal
sonorant
voice
coronal
labial
dorsal
continuant
nasal
r l n m z v h
consonantal
sonorant
voice
coronal
labial
dorsal
continuant
nasal
rhotic
This sp eci cation is not to o distant from what most Dutch phonologists
assume cf Bo oij
I sometimes refer to the glottal stop in the discussion that follows al
though I do not assume that it is an underlying segment of Dutch It will
turn out to b e an imp ortant assumption that the glottal stop and h are
almost empty consonants the former consisting of a ro ot no de without
any dep endents and the latter of a ro ot no de which only dominates the fea
ture continuant I will return to the prop er sp eci cation of these segments
in chapter I am not very sure ab out the feature sp eci cation of r On
the one hand I will showbelow that it cannot have a place sp eci cation
such as coronal or dorsal On the other hand it also do es not b ehave
like the really empty consonants b and h I haveprovisionally solved
this problem by assuming a feature rhotic
The features high lax and low have b een omitted since they are only connected
to vowels
Cf also Mester and It o Steriade ab out the placelessness of r
Theoretical Background
Sp eci cation and Undersp eci cation As I have already in
dicated in the previous section I need to refer only to one value for every
feature except for the ma jor class features I therefore will assume all these
features are monovalent again except for cons and son The latter
features are supp osed to b e sp eci ed throughout the derivation
Undersp eci c ation theory b ecomes less relevant under suchaviewof
features Nevertheless some additional assumptions have b een made First
of all I assume that coronal has the status of the most unmarked place
feature Paradis and Prunet Mohanan McCarthy and
Taub it is the place feature that is most likely to reduce and most
likely to b e inserted in the case of ep enthesis Yet it is still presentinall
coronal consonants and in all frontvowels
Furthermore I have decided to use a version of the theory of Combina
tory Sp eci cation Archangeli and Pulleyblank which seemed most
compatible with OptimalityTheory I assume that every geometrically well
formed feature tree can b e input to the grammar Surface lters will then
lter out unwellformed feature combinations For instance Dutchdoesnot
haveanylow rounded vowels as we will see b elow In my view there can
still b e an input segmentwhichislinked to b oth low and labial Yet for
such a segment to surface something has to happ en either the feature low
or the feature labial or b oth has to b e delinked in order to escap e the
surface lter against low labial combinations
In addition to Combinatory Sp eci cation I adopt a theory of p ossible
output lters or feature co o ccurrence restrictions All lters I need
t into one of the following twoschemes I assume A B
A
a
B
b A B
Constraints with the general form of a forbid the combination of in
compatible features such as low and labial or high and lax
Constraints following the scheme in b state that a certain feature A
should b e accompanied by another feature B If we ll in low for A and
lax for B we get the statement that all lowvowels are lax
The variables A and B can only refer to single features not to feature
bundles This already restricts the numb er of p ossible feature co o ccurrence
restrictions to in the case of features If wewant
to restrict the numb er of lters even more we can assume the so called
If wewould allow for binary features we could furthermore rewrite b as A B
or a as A B B A
Theoretical Background
Grounded Phonology Hypothesis of Archangeli and Pulleyblank ev
ery feature co o ccurrence restriction needs to b e grounded in the phonetics
This means for instance that we assume there can b e a constraint against
the combination of low and labial say b ecause it is relatively di cult to
push the tongue to its lowest p osition and op en the mouth and at the same
time round the lips but no similar constraint against high and labial
For more discussion of this issue I refer to section on page
Skeletal tier For the representation of vo calic and consonantal
length I adopt a more radical version of the so callled two ro ot theory of
Selkirk According to this theory long consonants i e geminates
are represented bytwo ro ot no des on the melo dy rather than bytwoskeletal
p oints or by moraic consonants My approach is slightly more radical than
Selkirk s b ecause I assume the same typ e of representation also for
vowels
Of course nothing in principle will prevent us from adorning the struc
tures used in this thesis with moras or x slots Yet I have found no real
argument for suchamove I think that one of the main arguments in favour
of a two ro ot analysis is that it do es not need any additions to the theory
If we accept the notion of an autosegmental ro ot no de there is nothing in
the theory which can exclude a representation of the following typ e for long
vowels
ro ot ro ot
vocalic
We also predict this representation to contrast with the following
ro ot
vo calic
In fact given the fairly rich notion of a ro ot tier that I employIhave found
no particular reason to explicitly distinguish b etween a melo dic ro ot tier
and a skeletal tier consisting of either x slots or moras in anyofmy analyses
and therefore I havechosen to leave the latter tier out of my representations
Theoretical Background
Mo del of the grammar I assume the following minimal orga
nization of the phonological comp onent for Dutch
Underlying representations
Lexical Phonology
Postlexical Phonology
Phonetic forms
The only division is that into lexical and p ost lexical phonology This
division also seems to b e the only residue of derivational theory that I
need The typ e of argument that will b e made for this division b elowruns
as follows Some phenomena can b e derived on the surface but it can b e
shown that they cannot b e present at some deep er level For instance we
will see in chapter that Standard Dutch has b oth an underlying schwa and
an ep enthetic schwawhich are in complementary distribution and which
have di erent prop erties in particular ep enthetic schwa can form a closed
syllable whereas underlying schwa can not I can only explain this di erence
by assuming that ep enthetic schwa is the result of a p ost lexical pro cess
arising at a level where the condition forcing schwa to b e in an op en syllable
is no longer very strong
Another imp ortant assumption I make is that a pro cess of strayera
sure applies b oth after the lexical and after the p ostlexical comp onent It o
This means that in b oth cases non parsed material is erased and
will not b e input to the p ostlexical comp onent or to the phonetics resp ec
tively This assumption seems to b e in accordance with current practice in
Optimality Theory as it is represented for instance in McCarthy and Prince
a
Optimality Theory The hyp othesis which I defend in this dis
sertation is a hyp othesis on representations there is a relation b etween
features on a segment and the prop erties of the proso dic structure whichis
headed by that segment Derivational issues therefore do not seem to play
a crucial role Y et in some cases the analyst needs to refer to them I have
solved this problem in the following way
Theoretical Background
Ihavechosen to use a framework of constraint based phonology in which
the linguistic generalisations are expressed in terms of output constraints
More in particular I havechosen to adopt the framework of Optimality
Theory as it is presented in Prince and Smolensky a McCarthy and
Prince and related work In this section I will brie y outline
the main prop erties of this theory as used in this thesis
In this framework linguistic theory consists of two functions called Gen
and Eval resp ectively
Gen takes at its input any phonological representation and generates
an in nite set of output candidates forms whichhave b een parsed into
proso dic constituents to whichepenthetic material has b een added Fur
thermore under the standard version of OT adopted here Gen is b ounded
by the following restricti ons only
Containment No elementmay b e removed from the input form The
input is thus contained in every candidate form
Consistency of Exp onence No changes in the exp onence of a phonologically
sp eci ed morpheme are p ermitted
Freedom of Analysis Any amount of structure may b e p osited
The principle of Containment is abandoned in some more recentversions of
OT McCarthy and Prince b in favour of some other mechanism which
makes the input representation available for output constraintevaluations
Yet I think this development is of no particular consequence to the theory
I presenthere
The function Gen thus generates every p ossible analysis for a given input
form A subsequent function Eval takes this set of output candidates as
its input and selects one of them as the actual optimal output It do es this
by comparing all the candidate outputs in parallel to a linearily ordered
ranked set of constraints The higher ranked a constraint the more
forceful it is
Supp ose that a language has the constraints A B C and D ranked in
that order in OT notation A B C D Supp ose furthermore that for
some given input Gen generates the output candidates and An
arbitrary constraintevaluation is now pictured in the following way
Following current practice I assume this representation do es not contain any
proso dic structure cf McCarthy and Prince
Theoretical Background
Candidates A B C D
p
An asterisk in a b ox means that the candidate in the horizontal line violates
the constraintinthevertical column The optimal candidate the actual
p
output is marked bya sign This candidate is optimal b ecause all other
candidates violate some higher ranked constraint more often than this one
do es
Candidate for instance is non optimal b ecause it violates the high
est constraint A while there are other forms which do not violate this
constraint Violation of A is therefore fatal for asindicatedby the ex
clamation mark b ehind the violation sign the asterisk Notice that this
candidate is non optimal even though in all it has the smallest number of
violations
Similarily candidate is non optimal b ecause it violates constraintB
while there are other candiadtes notably and which do not violate B
and also do not have more violations of any more highly ranking constraint
in casu A than
Now the only two candidates which are left in the candidate set are
and Both of these candidates violate constraintC yet is chosen as
the optimal form b ecause it violates this constraint only once whereas
violates the same constrainttwice Multiply violated constraints playa
crucial role in many OT analyses also in the present study An instance is
the constraint Onset This constraintsays syllables should have onsets
it can b e violated byanynumb er of syllables in a word If Onset were
b e constraint C in the tableau in could b e a candidate with one
onsetless syllable whereas would b e a candidate output with twosuch
o ending syllables Candidate is therefore selected
In my view using Optimality Theory in the analysis has the following
advantages In the rst place manyofmy analyses will b e based on the
family of pro jection constraints This notion of a family of constraints is
central to Optimality Theory
Also the b etter part of the analysis of language variation is based on the
interaction of the pro jection constraints with each other e g in chapter
and with other constraints e g in chapter The most imp ortant
For practical reasons I do not conform to all representational conventions of the OT
literature For instance I do not shade irrelevantboxes and I also do not use dotted lines
to indicated unordered constraints
Theoretical Background
among these are the so called Faithfulness constraints which demand that
input and ouput structures are maximally similar Almost all Faithfulness
constraints used here refer to features I assume that there are two general
typ es of Faithfulness constraints
a Parse If is present in the input it should b e present in the
output and parsed into higher order structure e g Parse
coronal Parse labial
b If is not present in the input it should also not b e present
in the output e g coronal labial
We can also see b as a set of surface constraints prohibiting all o c
currences of a certain feature Parse would then give exactly the
required result and Parse causes a feature to never surface at all
high tone Parse high tone in a language so that we can saythat
that do es not use tone for linguistic purp oses
Finally I wanttosay a few words ab out mybackground assumptions of
learnability I assume that something like the Subset Principle of Berwick
holds the initial grammar is the most restrictive p ossible and the
child only minimally changes its grammar on the basis of p ositiv e evidence
In Optimality theoretic terms at the initial stage of the acquisition pro
cess approximately the following statement holds all constraints are ranked
very high are assumed to b e unviolated except for the faithulness con
straints This gives approximately as an e ect that the child generates only
universally unmarked forms and this in turn has the e ect of the Subset
Principle whichmakes language learning easier to comprehend When a
child hears a form it cannot generate she assumes this is the consequence
of some lexical marking on that form and that the violation of the con
straint she observes is due to the higher ranking of one of the relevant
Faithfulness constraints Of course this is not a full worked out mo del of
OT language acquisition For instance it do es not tell us what is assumed
in the case of two p otentially con icting constraints that are b oth member
of the Faithfulness family or that b oth do not b elong to this family and it
also abstracts away from problems of noise in the input but for the cases
at hand it seems to work reasonably well
Initially the child assumes only the word tata or baba Chomsky
to b e well formed b ecause it satis es all relevant constraints for instance
Onset Now the child hears a word starting with a vowel It knows that
it can only realize this form by ranking some faithfulness constraint for
instance Parse vowel above Onset
I also abstract away from the partial reranking b etween lexical and p ostlexical phonology that seems to b e needed
Data
This is admittedly a heavily simpli ed theory of phonological acquisi
tion but it serves its limited purp oses in this thesis rather well
Data
In this section I will brie y outline the sources of my data Most of my
data come from Standard Dutchofwhich I consider myself a nativespeaker
The Standard Dutch data come partly from written sources most often the
phonological literature and in a few cases dictionaries and partly from my
own observations I havealways checked all data against myown intuition
but it seems to me that hardly any piece of Standard Dutch data is con
troversial I have tried to indicate the few exceptions All words and parts
of words were considered to b e part of my data with two classes of excep
tions names and interjectives Names sometimes re ect facts ab out older
stages of the language and interjectives seem to have a p eculiar phonology
of their own For instance Dutch has an interjective transcrib ed as t t t
indicating disagreement or moral ob jection with a rep eated clickwhich
is not otherwise part of the Dutch segmentinventory The phonology of
the class of names seems not so extremely deviant but still I haveassumed
that there are no forms with underlying schwabetween l and m if the
only counterexamples app ear to b e the names Wil lem w l m and Col lem
k l m
ve also used data from Dutch dialects most prominently from Iha
Tilburg Dutch and Rotterdam Dutch For the data of these dialects I
have relied more heavily on written sources for these dialects and on my
own observations I do not havenativespeaker knowledge of either fortu
nately these proved to b e su ciently adequate I have furthermore checked
all my data against at least one nativespeaker It turned out that this typ e
of data is p otentially sub ject to more controversy
For French and Norwegian I used approximately the same metho d as
for the non standard Dutch dialects i e I checked written sources against
native sp eaker intuitions For all other languages I have relied on written
sources only unless indicated otherwise
In my transciption I followed a system which is closely related to IPA
Furthermore I used x and x to denote input and output representa
tions of the grammar resp ectivel y this means that x always denotes a
phonological representation
Organisation of the thesis
Organisation of the thesis
I will now prop ose a division into chapters of the material to b e discussed
I will brie y indicate the topic of every chapter Themainlineofthe
argument is as follows In chapter I give evidence for the constraint
N lax a pro jection constraintwhich is op erativeontheNlevel Connect
and refers to the feature lax and in chapter I discuss the interaction of
this constraint with phenomena of vowel length In chapter and I
discuss schwa the emptyvowel inDutch French and Norwegian I will
show that this vowel shows a defective b ehaviour at almost all pro jection
levels i e N N and the Fo ot In chapter I discuss the b ehaviour of high
vowels and argue that in addition to pro jection constraints we also need
weakness constraints which are the logical opp osites of pro jection Weak
ness constraints say thatifasegment dominates a feature F it wants to b e
in a dep endent p osition High vowels are a go o d example of this they are
the typical vowels that o ccur in the weak part of a diphthong that alter
nate with glides and that in some languages o ccur in unstressed p osition
Chapter nally gives an overview of all the pro jection constraints
that are intro duced in this thesis Ishow that the constraint set is not
some arbitrary unprincipled set of unrelated constraints but instead can
b e sub divided into a small numb er of classes of tightly related constraints
Chapter This chapter discusses the tense long vowels of
Dutch The analysis based on Connect N lax sketched ab oveisworked
out I argue that a length based theory of the Dutchvowel system encoun
ters to o many problems and that for this reason it should b e replaced by
a theory based on the feature lax I will refute the arguments that have
b een adduced in favour of a length based theory and present all arguments
Ihave found to b e in favour of a theory based on lax This chapter is
concluded by an app endix in which the history of the length vs tenseness
debate of the last years is reviewed
Chapter In this chapter I develop a complete analysis of length
in the Dutchvowel system including diphthongs and loan vowels and its
relation to syllable structure on the basis of the Headedness Hyp othesis
A prominent role is assigned to Brabant dialects of Dutch These dialects
have a complete set of long lax vowels in addition to the system found in
Standard Dutch I will take the Tilburg dialect as an instance The Tilburg
dialect has vowels schwa tense a e i o y u lax
y and long lax q q q q q yq It can b e shown that eachofthese
vo wels is phonologically distinct from all the others
Organisation of the thesis
On the one hand Tilburg Dutchgives extra evidence for the tense lax
distinction in Standard Dutch b ecause this distinction is needed in this
dialect in order to describ e the di erence b etween long q ontheone
hand and tense o on the other On the other hand Tilburg Dutch seems
to contradict the hyp othesis that Dutch dialects do not have underlying
length distinctions at all b ecause the di erence b etween and q can
phonetically only b e describ ed as one of length
This chapter also includes sections on the nature of the feature that
is needed to describ e the di erence b etween the two main sets of vowels
tense ATR op en p eripheral etc on diphthongs and on
extrasyllabicityandonr colouring
Chapter Chapter is devoted to Dutch derived schwa I ar
gue that there are three typ es of schwainDutch The most well studied
of these is underlying and while this o ccurrence has manyinteresting prop
erties in this chapter I discuss twotyp es of derived schwa viz schwaas
a reduction vowel and schwaasanepenthetic vowel In some ways these
b ehave exactly as the underlying sch wainDutch but in others they do
not The reduction pro cess can b e seen as the loss of all vo calic qualities in
a given unstressed non nal p osition The ep enthetic schwaisadefault
vowel which is inserted for syllabic and sometimes metrical reasons In this
way we can hop e to establish a uni ed theory of all o ccurrences of Dutch
schwa I also review the prop erties of p ost lexical schwa the schwaof
clitics and in ectional elements
Chapter In this chapter I argue that there is ample evidence
showing that schwa do es not license a complex onset at an underlying level
of representation Also schwa is notoriously invisible for rules of stress
In this chapter I will show that this follows if we assume that schwacan
only pro ject to a minimal syllable as the result of indep endently needed
constraints on syllable structure The defective pro jecting b ehaviour of
schwa itself follows from the fact that it is maximally undersp eci e d and
in particular from the fact that it lacks a place no de It is well known that
segments lacking a place no de e g ep enthetic segments do not participate
fully in the syllable structure
Chapter Chapter deals with classical questions with resp ect
to French and Norwegian schwa There is ample literature on b oth the
French and the Dutchschwa as well as the default vow els of some other
languages but hardly anywork has b een done on language variation in this area
Organisation of the thesis
NowtheFrenchschwa clearly is a puzzle for the theory sketched for
Dutchschwa Phonetically the twovowels are very similar and their b e
haviour also has a lot of prop erties in common like the stressless nes s of
schwa On the other hand the b ehaviour of Frenchschwa in the syllable
is di erent from the one of Dutchschwa For Dutchschwawe could sayit
licenses elements in its nucleus taf l but not in its onset ta but
for French it seems to b e exactly the opp osite cf tabl tab l On
the other hand Dutchschwa and Frenchschwa should have the same pho
netic representation b ecause they are indistinguish able at least in some
of their dialects and they share some imp ortant phonological prop erties
b oth cannot b e stressed and b oth cannot o ccur at the b eginning of a word
for instance
Norwegian is very interesting in the comparison b etween French and
Dutch on the surface this language seems to have a Dutch typ e schwa
allowing a co da tig r but it also has a French typ e schwaallowing a
complex onset epl This can b e explained bytheinteraction of prin
ciples governing schwa with the principles of stress in Norwegian where
primary stressed word initial syl labl es have to b e bimoraic as a result of
Prokosch Law
Chapter Although chapters through are basically con
cerned with vowels in the head p osition of a syllable we can also wonder
what happ ens if a vowel o ccurs in a dep endent p osition such as the syllable
onset In other words gliding is also of relevance for the theory on vo calic
features and syllable structure
Chapter presents as a case study the j i alternation in monomor
phemic forms diminutives and nd p erson clitics in the Rotterdam dialect
of Dutch
The forms raampje and koningkje are normally realized as rampi and
kon ki in the Rotterdam dialect with a long high vowel instead of Stan
dard Dutch j whereas voetje and dingetje are always realized as vutj
and d tj In other words the form j shows up after a coronal stop
and the form i shows up in other contexts Similar but sligh tly di erent
alternations can b e found in the case of the second p erson clitic j i and
in some monomorphemic forms
The analysis I want to prop ose for these facts is based on work done
in previous chapters here I intro duce the notion of a weakness constraint
shortly discussed ab ove
In this way I hop e to show the theories develop ed in these chapters can
beputtowork for these complicated facts A comparison of this vowel glide
alternation with the working of Sievers s Law in Gothic will furthermore
Organisation of the thesis
reveal additional arguments against vowel length in Dutch
Chapter In this chapter I review the formalism used in this
thesis and o er a conclusion It will app ear that the constraints which are
used in this thesis can b e organised into several separate families sets of
related constraints which are derived from a common ancestor by the setting
of several parameters Iwork out these families in some formal detail
In particular I concentrate on the pro jection constraints and weakness
constraints mentioned earlier in this intro duction
It app ears that the pro jection constraints always refer either to the
whole set of vo calic features or to the particular features low and lax
and in some cases to tone and or the feature voice whereas the weakness
constraints all refer to the feature high
App endices OT analyses can get rather complicated esp ecially
when we treat a somewhat larger fragment of a grammar as is the case in
this thesis In App endix A I giveanoverview of all constraints used in the
analyses in this dissertation in app endix B I subsequently list all ranking
arguments for Dutchwhich are referred to in this thesis and in app endix C
Ihave graphically represented the top ology of the grammar of Dutchwhich
is studied in this thesis My exp erience is that it is extremely di cult to
rememb er all individual rankings This is the reason whyIhave added the
latter two app endices
Vowel Quality and Rhyme Structure in
Dutch
Intro duction
In many languages there is a phonetic correlation b etween vowel length
and tenseness If this correlation is complete that is to say if all long
vowels in a language are tense and all short vowels are lax it is attractive
to assume that either length or tenseness is absent from the phonological
representations and derived from the other prop ertyby a rule of phonetic
interpretation Once this p oint is established we still have to face the
question of which prop erty is phonological in a given language The aim of
this chapter is to reevaluate some of the evidence that has b een presented
to settle this latter p oint
Awell known instance of the problem mentioned here arises in the study
of the Germanic languages in particular German and Dutch In those lan
guages the correlation just noted can b e observed with some minor prob
lems to b e discussed b elow Phonologically the prop erty distinguishing the
two sets of vowels seems to b e most relevant for syllable structure long
tense vowels can b e followed by at most one non coronal consonantword
nally whereas short lax vowels can b e followed bytwo Most analysts
have taken this to b e an argument for phonological length from which tense
ness can b e deriv ed in the phonetic comp onent I will argue that this view
is mistaken and that there are strong arguments for considering tenseness
rather than length relevant for the phonological comp onent The sylla
ble structure facts can b e derived from a constraint relating vowel quality
to the structure of the syllable This typ e of relation gets indep endent
supp ort from languages in which the relevant phonological feature shows
up prop erties of vowel harmony under standard autosegmentalist assump
tions length is not something that can spread Finally I try to address
the question whether also other elements of the proso dic hierarchycanbe
in uenced by phonological features like tenseness and whether there are
phonological features other than tense which can a ect proso dic structure
The Dutchvowel system
I pro ceed along the following lines in my argumentation In the next
section I giveanoverview of the Dutchvowel system The main arguments
for and against tenseness theory will b e discussed in section In section
I will reinterpret the main argument for length in suchawaythatit
can b e seen as an argumentinfavour of tenseness In section I discuss
the role of lax in harmony systems In section I then examine whether
there are any arguments in favour of length left and I try to refute these
The pre nal section contains a conclusion I have also added an app endix
section to this chapter in which I brie y summarize and review some
of the th century literature on the Dutchvowel system
The Dutchvowel system
The tenseness versus length debate most probably is the oldest in Dutch
phonology The issue has received some discussion by almost all scholars
working on Dutch such as De Gro ot Van Ginneken Trub et
zkoy Van Wijk Cohen et al Van Haeringen
Heeroma Moulton De Rijk Brink Zonneveld
Bo oij a Trommelen Van der Hulst
b Kager Lahiri and Koreman Smith et al and
Hermans a an overview of the debate will b e provided in an
app endix to this chapter section Similar partly parallel discussions
have also b een conducted on the German vowel set see Ramers
Wiese Vennemann Hall Kager and references
cited there In the following table I ha ve listed all Dutchvowels
i iq yq uq
ii i y u
iii y
iv
v e o
vi i y u
vii
viii q q q
ix
x a
Some of the vowels I will set apart in this chapter to b e treated in one
of the chapters that follow The vowelsinrows i and viii are usually
Throughout this study I assume that the vowel marked y is phonologically
rounded although phoneticallytherounding of this vowel is very slight See
Fudge and Gilbers for discussion
The Dutchvowel system
considered to b e loan segments b ecause they o ccur almost exclusively in
loan words from French and English see section in the next chapter
the schwa in iv is in manyways a very exceptional vowel in the Dutch
system and will b e discussed in the chapters and and the sounds in vi
are diphthongs not simple segments section
The core system of Dutchvowelsisthus formed by the segments in rows
ii iii v vii ix and x
Of these the segments in rows ii v and x are usually referred to as
tense or long and the segments in rows iii and vii as lax or short
I will follow Moulton and use the terms A vowel and B vowel
to refer to tense long vowels or lax short vowels resp ectivel y wherever a
theory neutral term is needed I also assume that the vowels can b e paired
in the following way a o e A B The high
vowels i y u do not have a counterpart in the B set The lowvowel
do es not have a counterpart in the A set The reasons why I mak ethese
pairings will hop efully get clear from the discussion that follows The most
controversial assumption probably is that isa midvowel and not high
An empirical reason for making this assumption is that this vowel always
alternates with e and never with i For instance in plural formation we
nd pairs suchasschip schepen ship s sx p sxep n sxip n along with
dag dagen day s d x da n Spa Noske Also mid vowels
Also the vowels in v are slightly diphthongized phonetically they movetowards
the centre of the vowel triangle but most scholars agree that this is not a phonological
issue See Van de Velde for a recent so ciolinguistic study on the phonetic di er
ences b etween Standard Dutch as sp oken in Flanders and as sp oken in the Netherlands
in this resp ect
The rules of Dutch orthography with resp ect to vowels are somewhat complicated
In closed syllables which notion is supp osed to include word nal syllables ending in
one consonant tense long vowels are written as doubled vowels zeep soap zep
boos angry b os raam window ram The high vowels i y u are written ie dief
thief dif uu bluut philtrum blyt and oe boef scoundrel buf resp ectively
is written eu deuk dent d k
In op en syllables most tense vowels except i u and are written with a single
letter ramen windows ra m ga go a bozer angrier b o z r zie see zi
doe do du keu cue k e is written with a single letter in word internal op en
syllables but with a double letter word nally reden reason re d zee sea ze
Single orthographic e at the end of a word is always read as schwa made maggot
ma d
Lax vowels are sp elled with single letters zit sit z t zet put z t mug mosquito
myx koppel b elt k p l dapper brave d p r As the last two examples indicate
consonants after word internal lax vowels are doubled when they are not in a cluster
This do es not mean that consonants are long but that the precding vowel is lax or
short
u resp ectively Schwais Diphthongs are sp elled ei or ij i ui y au or ou
normally sp elled e These are not all rules of Dutchvowel sp elling but they should su ce to understand the examples in this thesis
Tenseness versus Length The Case of Dutch
get lax and lengthen b efore an r in the same fo ot koor choir k q r
keur choice k qr keer turn k q r Gussenhoven Again wesee
e and b ehaving as a pair High vowels b efore r only lengthen so that
i alternates with iq rather than with q See Levelt and Bo oij
a for more arguments that islowand mid
Tenseness versus Length The Case of Dutch
In this section I will discuss the most imp ortant argumentinfavour of
length as the characteristic distinguishing A vowels from B vowels and on
the other hand the arguments for viewing tenseness as that prop erty This
argumentation will lead to a paradox on the one hand syllable structure
seems to indicate that A vowels are long but on the other hand all other
tests that we can develop seem to indicate that they are not
The argument for length After Zonneveld and with the
exception of Van der Hulst b all authors on Dutch mentioned ab ove
have assumed that length is a phonological prop erty present at least at the
stage of the deriv ation where syllable structure is assigned The strongest
and as far as I can see the only reason for this is that A vowels pattern
with diphthongs as opp osed to B vowels with resp ect to syllable structure
as is shown in the following paradigm
ra ra raam ram raap rap raamp ramp
yard window turnip
rah r ram r m rap r p ramp r mp
ram quickly disaster
rui r y ruim r ym kruip kr yp kruimp kr ymp
moulting time wide creep
The paradigm gap in can b e lled by just a handful of words For un
known reasons often the vowel is i and the form is an irregular past tense
forms of a verb hielp help ed hilp stierf died stirf twaalf twelve
twalf and a few others Usually one abstracts awayfromthesecounterex
amples but I will explain the high vowel cases in section
twaalf thus is the really problematic case Numerals form an The form
exceptional stratum in the phonologies of many languages For instance in
French word nal consonants are deleted b efore a consonant initial word
if and only if they are also deleted phrase nally six id sis and dix
The same is true for prop er names of course which also sometimes are counterex
amples Tjeenk tje k Rienk ri k etc
Tenseness versus Length The Case of Dutch
ten dis are counterexamples to this generalisation however Their nal
s deletes b efore a consonant but not phrase nally
Furthermore a b efore a sonorant obstruent cluster might b e sub ject
to a sp ecial tensing rule We will see in section that this pro cess is
pro ductive in at least one Dutch dialect It might b e that this form is a
last remnant in Standard Dutch of the same pro cess
At the end of the word A vowels and diphthongs can b e followed by
one non coronal consonantorby no consonant at all B vowels on the other
hand can b e followed by one or two non coronal consonants In the middle
of the word B vowels are normally only followed by one tautosyllabic con
sonant A vowels and diphthongs in those p ositions are not followed byany
tautosyllabic consonant at all The picture is somewhat blurred b ecause all
of the well formed structures in can b e followed by one or more voiceless
coronal obstruents
raad rat raamt ramt raapt rapt raampt
advice he plans carries
rat r t rand r nt hapt h pt stampt st mpt
rat edge he bites he stamps
ruit r yt ruimt r ymt kruipt kr ypt kruimpt
window he empties he creeps
Forms like rat rand or raad can b e analysed as falling under the template
in This is not true however for forms like stampt or staand Often the
more complex forms are morphologically derived but some phonologically
complex monomorphemic forms suchasherfst autumn h rfst and ernst
earnest rnst do also o ccur Wemay surmise that the coronal obstruents
have some sp ecial status with resp ect to the proso dic word I will refer to
these coronals as extraprosodic Furthermore I assume that coronal ob
struents are only extraproso dic when they havetobe thus the t in raad
or rand is not extraproso dic since it can b e incorp orated into the proso dic
word in a normal way Coronal obstruents can o ccur in extraproso dic p o
sitions they do not have to o ccur in those p ositions
Now diphthongs are usually assumed to b e bip ositional in Dutch The
facts in can therefore b e explained in an elegantwayifweassumeA
vowels are phonologically long and if in addition we make the following
assumptions
A Length based Theory of the Dutch Syllable
The analysis presented here comes closest to that presented in Zonneveld
This is only one p ossible theory based on length Actually all assumptions presented
here havebeencontested by some length theorist I do not think however that this
a ects the strength of my argumentinanyway
Tenseness versus Length The Case of Dutch
a A vowels are bip ositional B vowels are monop ositional
b Coronal obstruents can b e extraproso dic i e if necessary they
can get licensed outside of the phonological word
c Syllable rhymes contain exactly two p ositions
d At the end of the phonological word there can b e at most one
non coronal extrasyllabic consonant
It can b e observed that sup erheavy C VVCorC VCC syllables in most
languages also in Dutch are limited to word nal p ositions while word
internally we nd at most heavy syllables This has led some researchers to
assume this is one instance of a case where the system is more p ermissive
regarding incomplete structure at word edge than elsewhere In particular
it is assumed that word nal syllables are only heavy not sup erheavy The
word nal consonant is extrasyllabic in one way or another This has to
b e seen under anyb o dy s theory as a di erenttyp e of sp ecial p osition than
the one of coronal obstruents discussed earlier after a word nal syllable
we can nd at most one non coronal consonant but a much larger number
of coronals The di erenttyp es of proso dic p ositions will b e discussed in
section of the next chapter
I will refer to the p ositions at the end of words in whichalltyp es of
consonants can o ccur as extrasyl labic whereas I refer to the p ositions for
coronals only as extraprosodic The p in kruipt is extrasyllabic and the t
ys p eriph is extraproso dic By de nition extraproso dic p ositions are alwa
eral to extrasyllabic p ositions Extraproso dic p ositions are outside of the
proso dic word whereas extrasyllabic p ositions are inside of those
Dutch has a class of words like aalmoes alms al mus which seem
exceptions to this rule On the basis of their stress b ehaviour wehave
to conclude however that these words are proso dic comp ounds Zonneveld
and Trommelen which consist of two separate proso dic words aal
and moes at some level of the representation If aalmoes were a normal
word wewould exp ect a stress pattern aalm o es Ifaalmoes proso dically is
aal moes we exp ect a almoes whichiswhatwe actually nd
The idea that the nal consonant of a sup erheavy syllable is outside the
syllable structure can b e worked out in at least two di erentways Wecan
assume the consonant remains completely outside syllable structure and
gets licensed in a non proso dic way Alternatively we can assume it forms
the onset of a degenerate syllable with a phonetically emptyvowel as its
head The last metho d rst prop osed in generative phonology by Aoun
McCarthy and Giegerich is esp ecially known from
so called Gov ernment Phonology Kaye Lowenstamm and Vergnaud
Charette Recently Zonneveld has prop osed a similar
Tenseness versus Length The Case of Dutch
analysis for Dutch outside of the framework of Government Phonology cf
also Langeweg An advantage of this analysis is that it can explain
why sup erheavy syllables b ehaveinmany resp ects as bisyllabic units In
this theory they are bisyllabic
I will p ostp one the formal choice b etween these two analyses to section
Empirically they servemy purp oses equally well Both haveasa
consequence under tenseness theory that Dutch has only twoorthree
typ es of syllable CV CV C andmayb e CV V for diphthongs The
A B i j
nal consonantofziek ill zien see zink sink and zijn b e z in is
outside of syllable structure or forms a syllable with an empty V The nal
consonantin zin or tik on the other hand is tautosyllabic with the vowel
as we will see b elow
According to this length based theory then the form r is ruled out
b ecause it only contains a monomoraic vowel thereby violating the con
straint in c the forms ra r y with a bimoraic vowel and diphthong
resp ectivel y r m with a monomoraic vowel and a moraic consonant
all are p erfectly well formed syllables and rap kr yp and r mp are all
analysed as a bimoraic syllable followed by a degenerate syllable Forms
suchas ramp r ymp and r lmp are ruled out b ecause they cannot
b e generated in anyway according to the templates in
The length theory has a quite elegant analysis of the syllable structure
facts The reason for this is that length can b e analysed in terms of a
numb er of syllabic p ositions The distinction b etween and a can b e
describ ed in the following way where the x s stand for skeletal p ositions or
ro ot no des and the F for the relevant feature tree
a
x x x
F F
On the other hand it is not easy to see how a phonological feature such
as tense can b e related to the numb er of p ositions in the syllable or in the
word
The arguments against length Iwillnowshow that despite
its usefulness in the theory of syllable structure length theory su ers from
several empirical problems whichwould b e very easy to circumventinathe
ory which uses a feature I will also show that this feature should probably
b e lax rather than tense in the case of Dutch
The rst three arguments have also b een discussed in Van der Hulst b
Tenseness versus Length The Case of Dutch
The Dutch stress system treats A vowels as light
Most theorists agree that Dutch hasaquantity sensitive stress system
in which closed syllables and syllables ending in a diphthong countas
heavy but syllables which end in an A vowel count as light Dutch
stress is tro chaic and seems to b e b ound by a three syllable window
at the right edge of the word One of the facts which p oint at the
quantity sensitivity of Dutch stress is that primary stress can b e on
the antep enultimate syllable only if the p enult has an A vowel not
when it contains a closed syllable Evers and Huybregts Trom
melen and Zonneveld
tremolo id tr emolo agenda id ax nda ax nda
From a typ ological p oint of view this situation is very marked if weas
sume a length theory There do es not seem to b e any non Germanic
system for which the generalisation holds true that CVV syllables are
light and CVC syllables heavy A generally accepted universal says
that if a language distinguishes b etween heavy and light syllables
CVV syllables always countasheavy The universal holds even par
tially for Dutch but only in the realm of diphthongs
This problem received a lot of attention in the prop osals of those de
fending length theory A large numb er of prop osals can b e collected
from the literature Dutchvowels would havetwo ro ot no des which
are attached to the same proso dic p osition and therefore they would
count as light Lahiri and Koreman they would rather haveone
ro ot no de and two proso dic p ositions and Dutchwould exceptionally
calculate syllable weight at the level of ro ot no des Kager or
they would b e analysed as a short vowel followed byaschwa like
element at the level of stress assignment where schwawould b e invis
ible for stress Smith et al Hermans a It seems therefore
p ossible to technically solve the problem in a length theory
This do es not diminish the imp ortance of the fact that all of these
theories need some very sp eci c stipulations while the whole problem
simply do es not exist for tenseness theory b ecause syllables ending
in an A vowel in the latter theory get a CV structure universally the
characteristic structure of a light syllable
See however Bo oij a for a recent analysis of Dutch stress which do es not dep end
on the notion quantity sensitivity
Duanmu has argued that also Chinese imp oses a bimoraic minimum on its
syllables Yet even if this were true it could b e p ointed out that the Chinese proso dic
system is di erent in the sense that there do es not seem to b e a clear cut distinction
between syllable and fo ot Bimoraicity requirements on feet are of course quite common
Tenseness versus Length The Case of Dutch
Dutchdoesnothave CV syllables
Another well known universal has it that all languages have core syl
lables consisting of exactly one consonantandonevowel Trub etzkoy
Jakobson As wehave seen ab ove DutchB vowels are
always in a checked p osition whichistosay that they always have
to b e followed by a consonant Only A vowels can app ear at the end
of a syllable but if these vowels are long Dutch only has CVV CVC
and mayb e CVVCandCVCC syllables
Again tenseness theory do es not have this problem As wewillsee
b elow there is evidence that all A vowels head a C V syllable Under
tenseness theory Dutch therefore do es have core syllables
There are more A vowels than B vowels
Trub etzkoy gives a third markedness criterion that causes
problems for length theory the set of A vowelsinDutch andin
owels Germanic languages in general is bigger than the set of B v
Such a division normally is taken to suggest that the bigger set in
this case the set of A vowels is unmarked Length theory cannot
make this prediction If A vowels are identical to B vowels except
that they are asso ciated to two rather than to one proso dic p osition
we exp ect that the A vowels are more marked than B vowels Length
theory also has a problem explaining the exact p osition of the gap
in the vowel triangle It is a strange coincidence that exactly the
high vowels lack a short counterpart while all long high vowels are
phonetically short
Tenseness theory do es not have this problem provided we assume
that lax if we assume this feature is to b e interpreted as ATR
or RTR is the marked value in Dutch The B vowels can nowbe
seen as formally marked Furthermore the restriction on high vowels
is a natural one The lter against lax high can get a phonetic mo
tivation at least if we assume laxness involves tongue ro ot retraction
Archangeli and Pulleyblank Asymmetries in the realm of high
and lowvowels are not uncommon in ATR vowel systems
The problems for length theory b ecome even harder if we consider
sc hwa foramoment Dutchschwacounts as the asymmetric vowel
of Dutch showing a p eculiar b ehaviour with resp ect to onsets stress
and other phenomena cf chapters and
The most plausible analysis for this segmentisthatitisheavily under
sp eci ed b oth phonologically and phonetically Keating Brow
man and Goldstein Ko opmans Van Beinum Further
more phonetically this vowel is extremely short Yet according to
Tenseness versus Length The Case of Dutch
the syllable structure test it should b e classi ed as phonologically
long
moede tired mud moeder mother mud r adem breath
ad m moederm d rm
Some length theorists Trommelen Bo oij a have indeed
prop osed that schwa should b e classi ed as phonologically long in
spite of the fact that there is no indep endent evidence for suchan
assumption Again the additional problem arises that there is no
short counterpart to schwa Furthermore if we adopt the p osition
that Dutch stress is quantity sensitive we get the undesirable result
that Dutchisaquantity senstive system in which some CVV syllables
count as sup erlight
A theory based on the feature lax do es not encounter anysuch prob
lems If schwaisheavily undersp eci e d it also lacks the feature lax
Therefore we simply exp ect it to pattern with the A vowels rather
than with the B vowels
Morpheme Structure Constraints distinguish b etween tense
and lax vowels
Some constraints on morpheme structure referring to the di erence
betw een A vowels and B vowels cannot b e expressed naturally in
terms of length For instance Dutch has a constraint against tau
tomorphemic j i This constraint can b e seen as an instance of
the OCP It is absolute for the A vowel i but there are a few p er
fectly normal words with j plus short B vowel jiddisch jicht
Similarly there is a constraint banning w plus long y or but
there are p erfectly acceptable words with wy wurm wurg wulp
Both of these constraints seem to b e related to the OCP The j i
lter esp eciall y can b e analysed as a constraint forbidding twoidenti
cal segments in a rowifwe assume that j is an i in a consonantal
In chapter I will argue that the constraint should b e Contour a constraint some
what similar but not equal to the OCP
The argument based on Morpheme Structure Constraints is due to Van Wijk
I will argue b elow that is b est seen as the lax counterpart of e not i there are a few
words starting with je all Biblical names and words derived from biblical names Jezus
jesu et Jesuit Jeremias jeremi eren lament This somewhat weakens the argument
based on ji but do es not a ect the parallel argument based on wy w
Again the mid vowels are slightly more p ermissive than the low ones witness
the new acronym WEU West Europese Unie w which is the only counterexample I found
Tenseness versus Length The Case of Dutch
p osition But it is unclear why all of a sudden j isallowed if the
feature tree of is attached to only one skeletal p osition
The working of this lter is also observable in the pro cess of homor
ganic glide insertion MiekeTrommelen p c a j is inserted in hiaat
hiatus hi j at as well as in hobbyist amateur h bi j st but not
in shiiet shiite si it si jit
Secret languages treat A vowelsasaunit
For the last few years secret languages and language games whichI
will uniformly call secret languages have received more attention
as p ossible sources of evidence for phonological structure than was
usual in earlier generative literature In an overview of several such
systems McCarthy a p oints out that English secret languages
suchasGoat Latin and Alfalfa without exception treat long vowels as
unities In Goat Latin each syllable gets an in x Vb w between the
onset and the rhyme where V isacopy of the head of the original
syllable An example of an English sentence and its translation in
Goat Latin McCarthy s a example d is given immediately
b elow
a This is Goat Latin
b b w s b b w zgowb wowt l b w t b w n
According to McCarthy a it is typically the case in English
secret languages that the long vowel is copied as a whole and never
just the rst half of it in whichcasewewould get gob wowt for
goat At the same time Goat Latin excludes copying of a rhyme
constituentaswe can see in the translation of latin This translation
is l b w t b w n and not l b w t nb w n If Dutchwould
have a secret language like Goat Latin this would again provide us
with a problem for length theory whichwould not necessarily b e a
problem for tenseness theory
Unfortunately the study of Dutch secret languages is heavily under
develop ed Such languages probably exist but I have not b een able to
nd any phonological or non phonological study dealing with them I
ha ve therefore informally conducted a miniature exp erimentinwhich
I explained Goat Latin to some i e nativespeakers of Dutch
all non linguists by telling them they had to nd the simple rule
of a secret language of whichIgave and if necessary rep eated an
example sentence that only included B vowels and schwa
Tenseness versus Length The Case of Dutch
After this I gave them a few other Dutchsentences to translate
into the secret language in order to see whether they actually under
sto o d the basic rules These sentences also included B vowels and
schwa only Finally I gave them a set of sentences whichalsoin
cluded A vowels and diphthongs These sentences are given in
b elow where I rst give the Dutchsentence I read to the informants
subsequently the translation that most of the informants gave and
thirdly variants that were given by a minority it happ ened that for
no form more than di erent translations were given eachtimewith
an indication of the numb er of informants that pro duced the variant
o ccasional apparent mistakes like bubuwuk have b een left out
Dit is erg simp el
D b w t b b w s b b w rx s b w mp b w l
This is very easy
a Ik lees dat boek erg
b b w k leb wes d b w t bub wuk b b w r x
kb w k
graag als het
rab wax b b w ls b w t
rax b wax h b w t
regent buiten
reb we b w nt b yb w yt b w n
n reb we nb w nt byb w yt b w
I like to read that b o ok when it rains outside
b Het is ook erg
h b w t b b w s bob wok b b w r x
b b w t rb w r x
mo oi
mob woj
mo jb wo j
Because it is very b eautiful
c Wij allen houden
w ib w i b b w l b w n h ub w ud b w n
b b w l nb w n hob w ud b n
van lezen
v b w n leb wez b w n
v nb w n leb wez nb w n
We all love reading
Tenseness versus Length The Case of Dutch
In so far as there is variation at all it do es not directly b ear on the
issue under discussion Some informants had problem with schwa
headed syllables presumably b ecause in secret languages the original
syllables are reanalyzed as proso dic words and proso dic words cannot
only haveschwa syllables Also o ccasionally clusters of B vowels plus
a following consonant where copied This has to b e seen as a mistake
against the basic rule given that this copying was not attested in the
example sentence not even where this was p ossible Almost all of
these mistakes were made by the same informant The same mistake
also o ccurred once with an A vowel raxb wax
More imp ortantly aword like lees was translated by all of myinfor
mants as leb wes and never as l b web w s or l b web w s
Crosslinguistical ly underlying long vowels tend to get split in this
typeofword game Vago A complication arises however b e
cause all informants but one treated diphthongs as unities as well wij
we w i was never translated as w b w i Diphthongs however
b ehave like bisegmental unities in other resp ects most signi cantly
they make syllables heavy I will try to solve this paradox in the next
chapter
A less sup er cial empirical research could clearly shed more lighton
this p oint but for now the conclusion seems justi ed that native
sp eakers of Dutch tend to treat A vowels more as a unity than length
theory predicts
There are dialects with a real length contrast Some dialects of
Dutchhave long lax vowels in addition to the short lax and tense ones
of Standard Dutch For instance Tilburg Dutchhasthefollowing
underlying vowel set Boutkan and chapter fi y u e o
qg Tenseness and length thus are q q q q q
indep endent parameters in this dialect Some dialects also havelong
variants of the high tense vowels Hermans Geert Bo oij p c
Yet in all of the dialects I examined the short tense vowels b ehave
exactly likeinDutch This means that for these dialects wehaveto
invoke a sp ecial mechanism to account for the long b ehaviour of the
tense segments If wehave todothisanyway itisnotclearwhy
we could not use the same mechanism for the analysis of Standard
The one informant who did split the diphthongs seems to have b een in uenced by
sp elling given the phonetic outcome of her variant of Goat Latin Another informant
even copied the long diphthong o j in mooi
As a matter of fact if wetakeinto consideration the loan phonemes of rows i and
viii even Standard Dutch has suchlongvowels in its inventory
A theory based on the feature lax
Dutch I will return to Tilburg Dutch in the next chapter
A theory based on the feature lax
Wearethus faced with a paradox On the one hand the syllable structure
facts seem to p oint in the direction of a length based theory On the other
hand there are many other facts which seem to p oint to a theory based on
the feature lax rather than on length Some of the problems noted ab ove
have also b een observed by prop onents of a length based analysis they have
tried to solvetheminvarious ways I refer to Kager Trommelen and
Zonneveld Smith et al for a few fairly recent prop osals Yet
none of these prop osals seem to b e able to deal with all the problems noted
here For this reason I prop ose the following theory based on the feature
lax
A Lax based Theory of the Dutch Syllable
a B vowels have a sp eci cation for lax A vowels are unsp eci ed
for this feature Only diphthongs and truly long vowels are
bip ositional
b Coronal obstruents can b e extraproso dic i e if necessary they
can get licensed outside of the phonological word
c Connect N lax
d At the end of the phonological word there can b e at most one
noncoronal extrasyllabic consonant
The constraint in c is an instance of what I call a projection constraint
N pro jection has a certain feature in this case lax the If the head of an
syllable itself has a certain structure it branches Inversely the head of
a branching rhyme should always carry the feature lax The de nition of
Connect N lax is as follows
def
Connect N lax Project N lax Project lax N N
dominates lax i N branches
English to o has twosetsofvowels which can b e distinguished in terms of b oth
length and tenseness English di ers from Dutch and German in that we probably
need both features in its phonology Halle and Mohanan Lab ov but cf
Giegerich Harris for arguments that this is not necessarily true for all
English dialects Still most arguments against length adduced in this chapter for Dutch
cannot b e transferred to English in English long vowels do count for stress for instance
and wealsohave syllables ending in short vowels
A theory based on the feature lax
The formulation in is stated in terms of the X bar representation used
in this thesis but it seems p ossible to phrase a version of it in other
frameworks of syllabic representation For instance wecouldalsosay
that a syl lable is bimoraic i the head of dominates a feature lax
Connect lax in a moraic theory of syllable structure enriched
with the notion of a head The precise nature of the feature lax will b e
discussed in section in the next chapter
The analysis in is minimally di erent from the length theory pre
sented under It can explain the facts in as follows r is ruled
out b ecause its head vowel contains a sp eci cation lax but the rhyme
do es not branch this causes a violation of Connect N lax ra is an
op en syllable without a lax sp eci cation on the head r m and r y each
have branching rhymes with a lax head and rap r mp and kr yp are
analysed as sequences of a syllable followed by a degenerate syllable just as
under length theory Notice that tenseness theory also explains without
extra costs why the head of a diphthong is lax The phonetic length of the
non high tense vowels can b e derived by p ostulating a phonetic rule which
lengthens non high vowels in op en syllables
This analysis solves the problems mentioned for length theory Dutch
can b e analysed as having a large numb er of CV syllables viz all the syl
lables of which the head is not lax Furthermore these syllables all count
as light as exp ected Grounded constraints can explain the asymmetries
in the vowel system it is quite common in languages of the world that lax
and high or tense lax and low do not go together That the
MSC works against w and not against w can b e accounted for b ecause
and are di erent segments Finally it is not surprising that secret
languages treat A vowels as a unit b ecause that is what they are at every
level of the analysis
Another interesting prop erty of this theory is that it explains why truly
long vowels i e the loan phonemes and diphthongs as opp osed to the
A vowels b ehave as heavy in the stress system Trommelen and Zonneveld
In the theory presented here they are bisegmental hence heavy just
like closed syllables
The reader maywonder whether a should not b e analysed as low lax The
a pair b ehaves in an exceptional way in that there are exceptions to the general
pattern in whichvery often involve either the high vowels or a In some dialects
for instance in Tilburg Dutch the a also seems to serve as a p erfectly normal vowel in
closed syllables Similar anomalies with the a pair have b een observed for German
Kager and French Dell This pair should therefore b e treated as exceptional
in some way
lax and syllable structure in vowel harmony
lax and syllable structure in vowel harmony
N lax The The crucial assumption in the lax based theory is Connect
reader mightwonder whether this do es not count as a language sp eci c
assumption In this section I want to argue however that it is a universal
constraint b elonging also to the grammars of other languages
One di erence b etween a feature like lax or ATR and a prop erty
like length is that under standard autosegmentalist assumptions the for
mer can spread whereas the latter cannot Therefore if we nd a language
in which there is vowel harmony on the relevantproperty while there is
also an interaction b etween the prop erty and the op en vs closed nature of
the syllable wehave an additional argument for a lax based theory
Now it is true that there are many languages with ATR harmonyin
which there do es not seem to b e a relation with syllable structure at
all I b elieve however that there is su cient evidence that the con
straint Connect N lax is a principle of Universal Grammar whichis
also op erative in other languages some of which are not genetically or ge
ographically related to Dutch Some examples are Tagalog Carrier
Eastern Javanese Archangeli and Pulleyblank Andalusian Span
ish Ho op er Lusitian Portuguese Redenb erger and Midi and
Qu eb ec French Durand Dumas D echaine
In all these languages we can distinguish b etween A vowels and B vowels
in a way that is phonetically rather similar to Dutch In eachofthese
languages wehave to assume the di erence is to b e describ ed in terms
of a feature b ecause there is vowel harmony with resp ect to exactly this
prop erty All vowels in a certain domain usually a proso dic word either
b elong to the A set or to the B set In an autosegmental mo del this can only
b e explained if wehave a feature for the standard autosegmental analysis
of harmonyinvolves spreading It is much more di cult if not imp ossible
to see howvowel length i e b eing linked to two p ositions can harmonize
The harmonic b ehaviour of length tenseness in these languages under
autosegmental assumptions unambiguously shows wehaveafeature in these
languages Yet in each of these languages there is also a relation b etween
vowel quality and syllable structure whic h is b est describ ed as Connect N
lax
Interestingly in each of these languages there also is a correlation b e
tween tense and the ap erture features high and low which I argue b elow
also holds for the Dutchvowel system In particular it seems that the
combinations high ATR or high lax and low ATR are often
disallowed or at least avoided in many languages
Eastern Javanese Schlindwein Archangeli and Pulleyblank is p erhaps the clearest example of this state of a airs In this language
lax and syllable structure in vowel harmony
all vowels in the word are lax ATR according to the authors just
mentioned if the nal syllable is closed otherwise all vowels in the word
are tense
Actually the situation in Javanese is somewhat more complex b ecause
the harmony takes place only b etween vowels that agree in the features
high and low What we could assume is that there are two di erent
harmony pro cesses one a ecting the non high vowels and one a ecting the
high vowels or alternatively that there is one pro cess which can only apply
between vowels that agree on the feature high cf the discussion of
Andalusian Spanish b elow
a Distribution of lax high vowels
buri back mVr t student
turu sleep pl p r edge
bali return adVs bathe
b Distribution of lax mid vowels
ombe drink b b t weight
bodo stupid l r n stop
kere b eggar g l b get
rame noisy a l di cult
Archangeli and Pulleyblank explain this distribution as follows
tense is underlyingly absent in Eastern Javanese morphemes If the last
vowel o ccurs in a closed syllable it pro jects a tense feature according
to a rule much like Connect N lax which subsequently spreads through
the word Otherwise all vowels get the value tense by default Inter
estingly thelowvowel a do es not participate in the opp osition It o ccurs
b oth in words with lax vowelsasinwords with tense vowels In the next
section I will argue that a plays an ambiguous role in the Dutchvowel
system as well
We might assume that there is a harmony constraint op erativeinthese
languages which cannot b e seen at work in a language likeDutch Since
vowel harmony is not a topic of this dissertation I will provisionally for
mulate this constraintasfollows
Harmony Avowel V should have the same sp eci cation for RTR
as the vowel W that o ccurs on the right of V if V and W are in
the same proso dic word
N lax in Javanese all vowels except If Harmony dominates Connect
the word nal one prefer to b e harmonic over having the required value
lax and syllable structure in vowel harmony
to satisfy Connect N lax Harmony is not applicable to the word
nal vowel however so that Connect N lax is able to select the optimal
candidate here This will give the e ect just describ ed
A somewhat more complicated example of the same phenomenon is
seen in Tagalog Carrier asserts that a lmost without exception
closed syl lables do not contain long vowels in native stems However some
su xes such as adjectivizing an and in includeavowel whichislongand
makes all vowels in the stem long Finally all short vowels get laxed at
the end of the derivation while all long vowels stay tense The result is a
pattern very muchlike the one in Eastern Javanese
Tagalog adjectivizing su xes
N A
hiqmaqtay fainting hiqmaqtaqyiqn given to fainting
duqgob blood duqguqbaqn covered with blo o d
puqt k mud puqtiq kaqn
In mo dern autosegmental theory it would b e more plausible to say the
feature tense is underlying and length is phonetically derived b ecause
of the spreading pro cess But still there would b e the relation with rhyme
structure the op en syllable e ect whichwould have to b e accounted for
by something like Connect N lax
The same constraint Connect N lax can b e observed at work in Ro
mance dialects Most scholars do not assume Romance languages havea
length distinction In some dialects however we ndlaxvowels in closed
syllables and tense vowels in op en syllables
Take for instance the following paradigms with mid vowels of Languedo c
French copied from Durand
Notice that in this case it should b e tense rather than lax that spreads I will
not go into the question of whether the features presented here are binary or whether
there are features with opp osing interpretations See chapter and app endix A for some discussion
lax and syllable structure in vowel harmony
e
ses se his sec s k dry
mettra metra will put mettre m tr put
s echait seMe will dry sceptique s ptik sceptical
peu p little p eur p r fear
heureux r happy heurter rte hit
aveugler av gle to blind aveugle av gl blind
o
beau b o handsome bord b r side
p oser p ose put p oster p ste p ost
rosier rozje rosebush rose r z rose
Durand describ es this pro cess as lowering of the mid vowels in closed sylla
bles yet we could just as well see it as laxing of these vowels However
French do es not haveany stress rule whichwould justify a length distinc
tion b etween the di erentvowels
Again the restriction to mid vowels p oints to a sp ecial relation of high
and low to the pro cess involved
Ho op er prop oses the rules a and b to account for a similar
phenomenon in Spanish
lej l j ley law
le le le him you
lerdo l rdo lerdo slow
piedra p jedra piedra stone
ekuestre ekw stre ecuestre equestrian
a syll tense syll
b syll tense
These rules are exactly the same as the ones prop osed for Dutch That the
rules are really pro ductiv e can b e seen in morphological pro cesses with s
plural formation for nouns second p erson singular formation for verbs
klase klas s clase clases class
t nto t nt s tonto tontos stupid
korqe korq s corre corres run
Interestingly some Spanish dialects have a pro cess whichisvery muchlike
Eastern Javanese tense harmony Igive the following Andalusian Ho op er
In Languedo c French the vowels are also sub ject to this rule in the environment
b efore C
lax and syllable structure in vowel harmony
Zubizaretta Lieb er paradigm as an example the authors
just mentioned have shown that the s is deleted phonetically in this dialect
orthography singular plural gloss
p edazo p e aSo p S piece
piso piso p s o or
grup o grup o grVp group
Again this typ e of harmony cannot b e explained if vowel length is at play
b ecause length is not a feature that can spread
In Andalusian as in Javanese the relation with vowel heightisagain
very clear According to Zubizaretta there are twotyp es of rules
laxing the nal vowelofaword The rst rule is obligatory in all Andalusian
dialects it makes the vowel very lax and it feeds spreading This rule applies
to non high vowels only The second rule is not obligatory do es not o ccur
in all Andalusian dialects and makes the vowel only slightly lax This rule
applies to all vowels also the high ones
The lowvowel a also plays a sp ecial role in Andalusian harmony be
cause it b ehaves as an opaque vowel it do es not undergo laxing itself and
it blo cks harmony to pass through it Although a formalisation of the
harmony pro cesses would lead us to o far a eld I refer to the work of the
scholars cited earlier it should b e clear bynow that constraints on co o ccur
rence of height and lax harmony enforcing constraints and Connect N
lax interact in Andalusian to pro duce a complicated pattern in which the
role of Connect N lax is nevertheless undeniable
Finally I want to remark that lax is mayb e not the only feature that
has a relation to rhyme structure At least one other ap erture feature
seems to play a similar role as lax viz high This has b een observed for
Chamorro byChung Her formulation of the distribution of high
her ex and some examples her exx followbelow
Non lowvowels surface as mid in stressed closed syllables and as
high elsewhere
m etgut strong p oddu to fall
w
man e i cold c obg i to do
ettigu short mund o gu cow s stomach
w w
g esg is to brush k uras on heart
l ebblu b o ok ucan rain
p saw shing line p eskad ot sherman
imp atcu b ored tired m umu to ght
disp asyu slow goby enu government governor
w
g to dream c etnut wound
lax and syllable structure in vowel harmony
Since tense and high seem to b e intimately connected feature values in
many languages including Dutch cf section and since Chamorro
do es not seem to have a phonetic tense lax distinction indep endentof
height it is not very clear whether phonologically should b e anal
ysed as simply an instance of Connect N lax or as a variation of this
constraint in terms of high rather than tense In any case Chamorro
shows more evidence that a relation b etween vowel features and branch
ingness of rhymes is not just a p eculiarity of Dutch but o ccurs in many
other languages as well A more complete overview of all p ossible relations
between ap erture features and syllable structure will b e provided in chapter
It seems that a feature like lax can havetwo e ects an e ect on
syllable structure and a harmonizing e ect We can draw the following
table of languages
Vowel Harmony No Vowel Harmony
Syll Str Javanese Dutch
No Syll Str Dagaare French
Dutch has an e ect on syllable structure but no harmony theWest African
language Dagaare inversely has ATR harmony but no relation b etween the
feature and syllable structure and Javanese is an example of a language
which has b oth I nd it most di cult to come up with a language that has
a tense lax distinction but neither harmony or a syllable structure e ect
Paris French sup er cially seems a case in p oint but on the other hand
this language shows some marginal syllable structure e ect closed syllable
adjustment of e cf chapter It could b e that the reason whyitis
di cult to nd a language without harmony or closed syllable adjustment
is that in such a language it is di cult to argue for the existence of an
indep endent feature lax
More formally we mightsay that there are three groups of constraints
at work The rst constraint of course is Connect N lax the second con
straintisHarmony and the third group of constraints is Faithfulness ban
ning all deletion and insertion of the feature lax i e Parse lax lax
These constraints together de ne four di erenttyp es of language
N lax Faithfull Javanese a Harmony Connect
b Harmony Faithfull Connect N lax Dagaare
For the sake of simplicityI have assumed that nothing can happ en to a representation
except for the insertion or deletion of lax i e that the constraints against deletion and
insertion of consonants are undominated in all examples presented here
lax and syllable structure in vowel harmony
c Faithfull Harmony Connect N lax
Faithfull Connect N lax Harmony Paris French
N lax Harmony Faithfull d Connect
Connect N lax Faithfull Harmony Dutch
If the faithfulness constraints are undominated no feature lax can ever b e
deleted or inserted in order for a structure to meet either Connect N lax
or Harmony The relative ordering of the latter two constraints is therefore
irrelevant If Connect N lax is undominated similarly the ranking of
the other two constraints is irrelevant since the value for lax is dep endent
on the structure of the syllable in every p osition in the word Only when
Harmony is the highest ranking constraintcanwe determine what the
relative ranking of the other constraint is since this constraint is dep endent
on an initial value on whichwe can harmonize When this initial value is
N lax determined by the structure of the nal syllable hence Connect
we get Javanese When it is determined by an underlying feature weget
Dagaare
Tenseness and branchingness in Dutch Ihavenow moti
vated the constraint Connect N lax on the basis of some languages not
directly related to Dutch Let us see what it can do for us in the analysis
of Dutch itself In order to explain the Trub etzkoyan asymmetry b etween
the two sets I assume that lax is the marked value I furthermore assume
that lax is absent in order to b e maximally explicit but my prop osal
do es not crucially dep end on any particular version of undersp eci c ation
theory
It mightbeofinterest also that Dutch is a language which has nal
devoicing This could b e seen as the result of a constraint disallowing voice
to o ccur at the end of a syllable Nowwe will see at various places b elow
that there is reason to assume a strong connection b etween the features
voice and lax For instance in Dutch monomorhemic VCV sequences lax
vowels are almost always followed byvoiceless fricatives and tense vowels
are almost always followed byvoiced fricatives cf section on page
It therefore seems quite suggestive that b oth features are disallowed to
o ccur at the end of the syllable
Nowif is a lax and these are the only twovowels that have
b een recognized as a tense lax pair by all phonologists wehave logically
sp eaking four p ossible syllables whichareofinterest to us
My prop osals for schwainchapters and will however dep end on the assumption
that lax is absent indeed
lax and syllable structure in vowel harmony
N N
N N
d
d t
lax
lax
N N
N N
d a
d a t
Of these four logical p ossibiliti es only two are actually allowed by Connect N
lax viz a a lax vowel in a closed syllable and d a tense
vowel in an op en syllable b is disallowed b ecause the vowel is lax
but the rhyme is not branching c is disallowed b ecause the rhyme is
branching but the vowel is not lax I once again emphasize that daad is
a p ossible word of Dutch but the nal t falls outside the rst syllable in
my view cf section in the next chapter
Formalization in Optimality theory The observations made
ab ove can b e formalized in Optimality Theory in the following way
N lax Parse consonant Project N V Connect
Parse consonant forces all consonants to b e parsed into syllables Project N
V is a constraint against degenerate syllables Since degenerate syllables
are syllables with branching N they have an onset segment but without
a head such a constraint can b e formulated as a pro jection constraintas
follows
Project N V If an Nbranches N should dominate some vo calic
feature F V
I assume that the onset consonant is parsed i the vowel is parsed all other
analyses will crash due to indep endent constraints The four underlying
The analysis will b e changed slightly b elow section where Parse consonant
will fall to a lower rank and another constraint will take its place to ensure da t
Some more arguments for the length of A vowels
forms d d t da and dat will get the following evaluations the
dots represent syllable b oundaries the angled brackets unparsed material
d
Candidates Connect N lax Parse consonant
p
d
d
d t
Candidates Connect N lax Parse consonant Project N V
d t
p
d t
d t
d t
da
Candidates Connect N lax Parse consonant
da
p
da
dat
Candidates Connect N lax Parse consonant Project N V
dat
dat
da t
p
da t
The two prop erties that count as the most serious arguments in favour
of vo calic length can b e explained in this system as well viz b ehaviour
b efore consonant clusters and b ehaviour at the end of the syllable A vowel
b efore a consonant cluster at the end of the word is in a closed syllable and
therefore by necessity lax a vowel at the end of the syllable is in an op en
syllable and therefore by necessityitisnot lax
Some more arguments for the length of A vowels
In section I mentioned one imp ortant argumentwhich sup er cially
seemed to b e in favour of the length based theory This was the argu
One output candidate that has to b e ruled out as well is d b with an ep enthetic
consonant This will b e done in section
Some more arguments for the length of A vowels
ment based on syllable structure which has however b een refuted in the
previous sections
As far as I know three other arguments have b een mentioned in the
literature in favour of length theory I think that also these arguments are
not necessarily leading us to a length theory however I will now discuss
each of these arguments in a separate subsection
Tenseness cannot b e de ned in a satisfactory way pho
netically The phonetic argument is in fact more an argument against
tenseness than an argumentinfavour of length As I indicated in the intro
duction it is very di cult to giveawater tight p erceptual or articulatory
de nition of the contrast b etween tense and lax vowels A strong tradition
in mo dern phonology demands that phonological features always havea
phonetic basis Therefore more abstract features are less desirable
I will discuss the issue of the phonetic basis of lax in more detail in
section I will show there that there is a straightforward articulatory
de nition of the feature viz Retracted Tongue Ro ot
For now I note that this argument is the weakest of the ones discussed
here b ecause length theory has the same problems in providing a phonetic
basis There is a number of vowels which b ehave as long phonologically but
which are phonetically just as short as the phonologically unambiguously
short vowels These recalcitrantvowels are i y and u V an Haeringen
Heeroma Moulton No oteb o om Trommelen
Hermans Bo oij Therefore the notion of phonological length is
at least as abstract as the notion of laxness
Minimality requires branching A p otential argumentinfa
vour of length theory comes from Word Minimality McCarthy and Prince
quite convincingly argued that we should recognize a principle of
Universal Grammar requiring all lexical words to have a minimum size
This minimum is one branching Fo ot either bimoraic or bisyllabic sub ject
to language sp eci c requirements
Now Dutch do es not have lexical words consisting of a CV syllable where
V is a B vowel It does have a su ciently large sto ckofwords of the CV
shap e with V an A vowel mee with P vree made love V thee tea
N wee queer A to mention just one example for every lexical category
This could b e describ ed as a consequence of Minimality with the parameter
setting bimoraic for Dutch under a length theory
It has b een observed that Dutch syllables can never end in a lax vowel
not even when they o ccur in p olysyllabic monomorphemic forms Wedo
not nd words ending in a lax vowel and we also do not nd lax vowels in
Some more arguments for the length of A vowels
front of another vowel Van der Hulst
t k s taxi id t ksi
ka d cadeau gift kado
h at hiaat hiatus hi j at
kr ol creool creole kre j ol
Since all the unattested forms would still b e bimoraic even bisyllabic
these gaps cannot b e the consequence of Word Minimality Itisthus usually
assumed that Dutch has a minimality requirement on syllables Trommelen
Van der Hulst Kager Syllable rhymes should contain at
least two p ositions in Dutch
Tenseness theory do es not need to p osit such a minimality requirement
but it is still faced with the problem howtoaccount for monomoraic mee
thee vree and wee It seems that wewould have to assume Dutchhasno
minimality requirement
Kiparsky and Kager have argued that in tro chaic stress
systems there is a relationship b etween apparent minimality violations and
word nal stress a tro chaic system has forms with nal stress only if it has
apparent minimality violations What this suggests is that these systems
have some underlying marking which enables seemingly monomoraic feet
Both p olysyllabic words with nal stress and monomoraic words would have
this exception marking
Kiparsky and Kager argue this marking is catalexis tro chaic systems
whichhave nal stress have a catalectic mora or syllable i e a stress
b earing unit which is not lled by melo dic material We can assume this
catalectic element is also presentinwords which are sup er cially coun
terexamples to the proso dic minimality requirement These forms are then
bimoraic and their second mora is catalectic
Dutchhasatrochaic stress system and it clearly has words with nal
stress menu chocol a etc Therefore we could assume that Dutchhas
catalexis in its technical vo cabulary cf Nouveau in whatever way
this formal instrument is expressed I will discuss the representation of
catalexis in section in the next chapter In any case it is not necessary
to assume that the second mora of CV lexical words is lled It might
b e catalectic Therefore we can discard the Proso dic Minimality argument
Minimality is not a problem for a tenseness theory at all
A vowels form the domain of tonal contour in Limburg
Dutch The last argument for length theory I want to discuss is provided
by Hermans a in a discussion on the interaction b et ween syllable
structure and tone in Maasbracht Limburg Dutch a dialect sp oken in the
Some more arguments for the length of A vowels
Dutch province of Limburg as well as in surrounding areas in Germany and
Belgium The facts are pretty complicated and involveseveral factors such
as stress and consonantal feature comp osition not under discussion here I
will therefore not b e able to fully discuss them here and have to con ne
myself to a summary
Limburg Dutch phonetically has two distinct tonal contours a falling
tone analyzed as HL and a dragging tone analyzed as HLH A contrast
between the twocontours can only b e established in the syllable b earing
primary stress and only if that syllable has a well de ned structure Its
rhyme should contain either a non high A vowel or a B vowel followed
by one of the sonorant consonants fm n l g No contrast is ever found
in any of the other cases All the following facts have b een copied from
Hermans a I use a to indicate a falling tone and a to indicate a
dragging tone
a non high A vowel
falling tone dragging tone
steeg alley st ex stex rose st ex
slaag succeed sl ax slaag blow sl ax
Koos name k os koos crust k os
weer again w er weer we w er
b B vowel followed by tautosyllabic sonorant
falling tone dragging tone
kan jug k n kan can k n
klam trap kl m klam hardly kl m
min minus m n min vile m n
c Other categories
pit kernel p t
kat cat k t
rok skirt r k
Piet name pit
koes quiet kus
Mien name min
Hermans a explanation of this paradigm is as follows The rhyme
includes a smaller constituent the nucleus This nucleus dominates either
just an A vowel or a B vow el plus a following tautosyllabic sonorantcon
sonant r l m or n Obstruents can by stipulation never b e part of the
nucleus
The sonorants r and by stipulation are not part of the nucleus Limburg
Dutch also has a set of overlong vowels from which I abstract away In the next chapter
I will discuss this typ e of vowels in another dialect of Dutch
Some more arguments for the length of A vowels
N
N
N
r k
Phonologically the falling tone is an HL contour and the dragging tone
an HH contour The rst high tone is assigned to the rst mora of a syl
lable automatically Both contours are bitonal According to Hermans it
is attractive to assume the nucleus is bip ositional and tones can only b e
assigned within the domain of the nucleus
If b oth nuclear slots are lled either by a long vowel or by a short
vowel followed by a sonorant the contours can get complete realization If
however we let a short vowel b e followed by an obstruent which will b e a
sister rather than a daughter of N only one nuclear p osition is available
This p osition is lled by the high tone whichissharedbythetwocontours
The low tone in the dragging contour is inserted in the phonetic com
p onent to satisfy the OCP The nucleus will b e phonetically lengthened to
giveway to three rather than to two tones
The phonetic comp onent is also needed in the analysis of high A vowels
which do not allow for a contrast even though they are phonologically long
in the sense that they o ccupytwo p ositions in the nucleus According to
Hermans a the high vowels phonologically take the only two p ositions
which are available in the nucleus If they are followed b y a sonorant this
sonorant therefore can only b e syllabi ed outside of the nucleus Since it is
extranuclear it cannot b e a tone b earer In the phonetic comp onent the
high vowels are however shortened by stipulation After this op eration they
are to o short to b ear a tonal contour The nal part of the contour will
therefore disapp ear The sonorant gets a low tone at the end of the deriva
tion In this way the sequence A vowel sonorant is predicted to always
surface with a falling tone
In my view these multiple references to the phonetic comp onenttode
rive the right contours weaken the force of Hermans a argumentfor
a phonological role of length to a considerable degree Tenseness theory
also recognizes the phonetic length of non high A vowels A further obser
vation we could make is that the contours are needed only on syllables with
primary stress That vowels should lengthen under such a condition is not
very surprising and indeed lengthening of non high tense vowels under
primary stress is a pro cess that is well known from the literature
Conclusion
Hermans a arguments give a strong indication that A vowels are
phonetical ly long but this is not a matter of dispute for either length theory
or tenseness theory
Conclusion
In this chapter I have argued for one sp eci c pro jection constraint Connect N
lax This constraint can regulate the relation b etween the op en vs closed
nature of a syllable and the tenseness vs laxness of a vowel This con
straintisvery highly ranked in Dutch so that it regulates the structure
of every indep endent syllable In other languages most notably in regu
lar ATR harmony languages the working of the constraint is obscured by
other factors Yet there are also languages in whichvowel harmony and
N lax interact in suchawaythatwe can see the e ect of b oth Connect
these are Polynesian languages suchasJavanese or Romance dialects such
as Andalusian Spanish
The assumption of Connect N lax also solves a problem of Dutch
phonology and it seems to indicate that the length of Dutch A vowels is
a phonetic derivative of tenseness tense vowels always are syllabi ed into
op en syllables and vowels in op en syllables get lengthened esp ecially when
they are stressed
The analysis of word nal consonants as p otentially b elonging to a de
generate syllable led me to p ositing one other pro jection constraint viz
Project N V the constraintthatevery syllable with an onset i e ev
ery syllable with a branching N no de has to have at least some vo calic
feature material in its head something that degenerate syllables violate by
de nition This constraintwillbeworked out in more detail in the next
chapter
The other constraints that were prop osed were fairly straightforward for
instance faithfulness constraints blo cking unnecessary consonant deletion
The exact p osition in the hierarchy will b e approximated more exactly in
the course of the chapters that follow App endix B gives a top ology of
the constraint hierarchyofDutch as far as it could b e determined by the
research rep orted on in this thesis
App endix Historical overview
In this app endix I will brie y review the phonological literature that has
app eared since the year of the rst article on the phonology of Dutch
written by A W de Gro ot It is not myintention to give a complete and
ob jective historical survey of the literature for an overview of the history
App endix Historical overview
of Dutch linguistics see Bakker and Dibb ets for an overview of the
twentieth century see Anderson In the b o dy of this chapter I present
ideas that go back to the p erio d of early structuralism while hop efully
preserving the insights of more recentwork
It is therefore useful to have at least a sup er cial idea ab out b oth the
structuralist ideas and ab out the insights that were formulated later in
order to reject these This also gives me the p ossibility to discuss some
classical problems from the p ointofviewdevelop ed here I havechosen
to skip most of the literature already discussed in previous sections but it
is di cult to avoid all overlap
Dutch structuralism The Europ ean structuralist tradition had
a signi cant impactonDutch phonological research One of the most im
p ortant founding conferences of this new branch of linguistics was organized
by Dutchscholars in The Hague In the decade after this confer
ence Dutch phonology came to prosp erity The problem of distinguishing
A vowels and B vowels was among the very rst to b e attacked
Traditional linguists likeVan Helten Van Wijk Verdam
who were often working on earlier stages of the language or on pho
netics had referred to the distinction as one of short versus long vowels
Most phoneticians in this century havekept to this distinction as the only
real one This p osition was rejected by all phonological authors of the
early structuralist p erio d however
De Gro ot prop osed to distinguish b etween vowels whichwere
bright helder and those whichwere dull dof Vowel length was
a concomitant qualityofbrightvowels not a distinctiveproperty itself
However De Gro ot s own terminology was based on phonetic introsp ection
that most subsequent authors have found hard to replicate It has b een
rejected as b eing to o much based on intuition In the words of Moulton
this proposal does not advance matters It merely tel ls us that the
two classes exist Aswehave seen it is still di cult to nd a term that is
more insightful
Other phonologists used a terminology that was more directly related
to Sievers s stark geschnitten versus schwach geschnitten strongly
cut versus weakly cut and Trub etzkoy s Silbenschnittkorrelation
Syllable Cut
Es ist klar dass auch die Silb enschnittkorre lati on ein proso dis
cher Anschlussgegensatz ist Sie ist ja eigentlichnichts an
deres als eine Opp osition zwischen dem sogenannten festen
und dem sogenannten losen Anschluss eines vokalischen Sil
agers an einen folgenden Konsonanten Wenn dab ei der bentr
App endix Historical overview
Vokal mit festem Anschluss kurzer als der Vokal mit losem An
schluss ist so ist dies nur eine phonetische Folgeerscheinung
Die Silb enschnitts korrel ation b eruht somit auf einer
privativen Opp osition deren merkmallosen Glied der unge
schnittene voll ablaufende Vokal ohne festen Anschluss an einen
folgenden Konsonanten ist Dadurcherkl aren sich auchdie
Ergebnisse der Aufhebung dieser Korrelation sie wird im Aus
laute o der vor Vokalen aufgehob en wob ei in der Aufhebungsstel
lung naturlic herweise nur die phonetisch langen b ezw halb
langen voll ablaufenden Vokalphoneme stehen so im Englis
chen Holl andischen Deutschen Norwegischen Schwedischen
im Schottisch Gaelischen im Hopi usw
Idevoted a relatively large amount of space to this quotation b ecause it ex
presses the structuralist s p oints in a very concise way First vowel length
is describ ed as an epiphenomenon of the real privative opp osition namely
the one b etween a tight and a lo ose connection b etween consonants and vow
els But most imp ortantly it gives one of the most comp elling arguments
in fa vour of the Silb enschnittkorre lati on instead of the length hyp othesis
we can explain why it is the memb ers of group A that count as unmarked
There are two reasons for this In the rst place these are the vowels
that o ccur at the end of the word as wehave seen In a Trub etzkoyan way
of thinking it often happ ens that a privative opp osition is lifted at the edge
the end of a domain we only nd the unmarked form there Therefore
we can see set A as the unmarked set
Another way not mentioned in the quotation but immediately b elow
it in whichwecanseesetAasunmarked is by counting the number of
memb ers set B is smaller than set A byone ortwo dep ending on whether
we recognise the distinction memb ers This means that there are
afewvowels in set A whichdonothave a corresp onding vowel in set B
This is an indication that set A really is the unmarked set If B would
b e unmarked there would b e vowels which can only exist if they b ear
the marked feature This line of reasoning seems to have b een accepted
by all Dutch structuralists A lot of discussion subsequently wentinto
the exact nature of the phonological feature corresp onding to Trub etzkoy s
correlation As noted De Gro ot s prop osal to distinguish b etween brigh t
and dull vowels has not b een accepted Among the other terms that have
b een prop osed I mention scherp gesneden versus zwak gesneden Van
Wijk and gedekt checked versus ongedekt free Van Ginneken
The connection to Trub etzkoy s line of thinking is obvious even
though it is not clear there has b een a one way relation of in uence only
Yet the phonetic arbitrariness remained a problem For instance Van
App endix Historical overview
Haeringen remarked with resp ect to Van Wijk s division that
there is no a priori way to tell that the syllable b oundary which follows the
zwak gesneden vowels of gieten futen boeken is in anyway di erentfrom
the scherp gesneden vowels of pitten putten pokken The Syllable Cut
did not seem to corresp ond to anything in the phonetic reality
Pre generative literature The next few steps were taken at
the end of the s and the b eginning of the s I call this p erio d
pre generativist not b ecause the scholars in this p erio d could b e seen as
a group with a sp eci c goal in mind but b ecause the ideas suggested in
this p erio d seem to have had the largest impact in subsequent generativist
work
Two di erent lines of reasoning were set up First came Cohen et al
who can b e set in the structuralist tradition as far as the analysis
ofsetAversus set B is concerned b e it that these authors intro duced the
terms tense gespannen versus lax ongespannen in the Dutch literature
in order to describ e the phonetic di erence They set up the following
corresp ondences
lax y
tense i e a o u y
At the same time some scholars b ecame aware of the problems that were
inherent in this approach I already mentioned the criticism against the
earlier structuralist work and indeed the same questions were raised with
resp ect to Cohen et al by Moulton the division into two groups
was not well motivated phonetically and esp eciall y the corresp ondence pairs
were arbitrary in some resp ects Other divisions were p ossible whichwould
do the phonological job of distinguishing pairs of phonemes equally well
e ee y o
i y u a
The distinction b etween e and i phonetically seems to b e more or less
the same as that b etween i and so that it is not very clear what is
exactly the reason for lab eling the one pair as tense lax and describing the
other in some other way
An imp ortant problem for the theory was discussed byVan Haeringen
Heeroma and Moulton the classi cation of the high
vowels fi y ug Phonetically these vowels pattern with the short vowels
rather than with the long ones cf No oteb o om but phonologically
they are more ambiguous
The most imp ortant reasons to distinguish phonologically b etween group
A and group B are distributional In this p erio d the following citeria were recognized
App endix Historical overview
Extensibility At least Van Haeringen claimed that vowels
in group A are extensible they can b e delib erately pronounced in a
longer p erio d of time whereas the vowels in group B are not extensible
Now i y and u could not b e extended in the sp eechofVan Haeringen
and therefore this author argued they were memb ers of group
B The non extensibility of i y and u as well as that of the lax
vowels was immediately questioned by Heeroma and nowadays
it do es no longer count as a criterion for the distinction b etween A
and B altogether As a matter of fact the extensibilityoflaxvowels is
sometimes used as an argument against length theory A sp eaker can
freely extend the lax ofbom b omb b m to b qqqqqm which still
do es not sound like boom tree b om In a theory of grammatical
comp etence the extensibili ty criterion is however very di cult to
evaluate in either direction
Intervo calic fricatives In the context VF V a vowel F a frica
tive F is voiceless if and only if V is a memb er of group B Heeroma
OK OK OK
Compare l av n refresh l ev n live l ov n
OK OK OK
praise l v r hank l iv dear l uv n lu l v n
OK OK
l v n l v n l y v n l v n bl f n bark b s n
OK OK OK
b erries r f n ri s h s n jig and jog k y s n do o dd
jobs bl af n b es n r if n h us n k ys n According to
this criterion the high vowels should clearly b e in group A There are
however some pretty normal words P as n Eastern m z l luck
p yz l puzzle that count as exceptions to the generalization which
therefore is not water tight I have no explanation for the interaction
between voiced fricatives and tenseness in vowels I will return to this
issue in the next chapter
Occurrence b efore consonant clusters Vowels from group B
o ccur b efore word nal consonant clusters C C where C is a non
i j j
coronal The dorsal nasal b ehaves like a consonant cluster in this
OK OK OK
resp ect For instance b afraid kr carrion j
OK
kr e j o d i d u d young d thing b a
OK OK OK
b lk b eam b alk w lp whelp w elp kr mp shrink
kr imp This criterion seems to indicate that i y and u are in group
A A substantivenumber of counterexamples however involves exactly
OK OK
these vowels esp ecially i h ilp help ed w irp throwed etc
This criterion has b een extensively discussed ab ove The reason why
high vowels count as exceptions might b e exactly that these vowels
Henk van Riemsdijk p c notes that there is some variation in the pronunciation
of puzzel some sp eakers say pyz l
App endix Historical overview
do not have a lax variant I will return to this b elow
The criterion is also not water tight however b ecause there are again
OK
some counterexamples like tw alf on the one hand On the other
hand for instance the combination C C is also very rare while
i j
is in group B according to all other criteria The only words where
OK OK OK
it shows up are w lx willow zw lk oil cloth sch lf er
OK
chip z lv er silver
Occurrence at end of a word Vowels of group B cannot o ccur at
the end of the word Van Haeringen questioned this criterion
b ecause schwa phonetically the shortest vowel can o ccur at the end
of the word as well Yet Heeroma argued that this cannot count
as an argument b ecause schwa is an exceptional vowel in many other
resp ects Therefore this remains to b e one of the most imp ortant
are clearly in group A according criteria The high vowels i y and u
OK OK OK
to this criterion z i see d u do n y now
Diminutives the diminutiveshows up as tj after V Swhere
B
V isavowel in group B and S a sonorant consonant d tj
B
m n tj b m tj h l tj k r tj andasCj C some stop
after V S where V is a vowel in group A According to Van Haerin
A A
gen the high vowels b ehave sometimes as if they were in
group A and sometimes as if they were in group B w iltj w il tj
bl ump j bl um tj r imp j str im tj Heeroma ho wever
observes that also this criterion is weak b ecause there are to o many
counterexamples also in the other classes Trommelen notes
that Van Haeringen s arguments mighthavebeentypical for his Rot
terdam Dutch background
The discussion was brought to a conclusion byavery in uential article
by Moulton who prop osed to distinguish b etween a set of long
and a set of short vowels The long vowels were e a o the short
vowels y and the vowels i y u did not participate in the
opp osition This explained the ambiguous b ehaviour of the latter group
Notice however that this typ e of analysis can only b e parsed if long is
a feature not under an autosegmental approach to length
Early generative grammar In spite of Moulton s argu
ments against it early workers on the generative phonology of Dutch De
Rijk Spa Brink used the feature tense This mightbe
traced back to the in uence of Chomsky and Halle who used the
same feature for a similar distinction in English The interpretation of the
feature varied however from author to author
App endix Historical overview
For instance De Rijk argued that we may regarda tense
vowel as a sequence of lax ones The reasons that were given for this
version of length theory were somewhat di erent from the ones hitherto
discussed This pap er is full of original observations most of which are
unaccounted for in most of the following literature on the distribution of
vowels Following SPE the author furthermore allowed himself to givea
highly abstract analysis of these data
One of the observations was that y o ccurs in all contexts except
b efore w and r y on the other hand is very frequent b efore w and r
but hardly o ccurs in any other context according to De Rijk For
instance in word nal p osition it o ccurs only in u you y and nu now
ny He therefore prop osed a diphthongization rule deriving y from
y in all contexts except b efore glides the y b efore r was assumed to
b e the result of a tensing rule applying after diphthongization He tried
to extend this approachtoderivealso i and u from underlying i and
u resp ectivel y but as the author admitted the p oint there turned out not
very strong
Spa remarked that an approach along the lines of SPE was able
to solve a problem of Dutch morphophonemics It could describ e plural
formation in en in a uniform way In earlier accounts examples like the
following were problematic
a d x da n
b vl x vl n
c w p w b n
d b v l b vel n
Spa prop osed to solve this problem by assuming the following un
derlying representations for these forms d vl w bb b v l Plural
formation now simply is the addition of n Subsequently there is a rule
of vowel tensing in op en syllables the rule version of Connect N lax
syll
C tense tense
cons
Finally a rule degeminates the obstruent clusters at the end of the word Of
course this is not really an argumentinfavour of tense nor is it intended
to b e We could just as well assume that the vowel is lengthened in op en
syllables The other analysts of this p erio d also do not give convincing
We mightaddmenu menu m ny individu individual ndividy paraplu um
brella paraply and some other words admittedly all originally loans from Frenchbut
I think accepted by most sp eakers as normal Dutchforms
App endix Historical overview
arguments in favour of any p osition De Rijk remarks it is very
di cult to cho ose b etween long and tense by phonological criteria he
cho oses the latter b ecause u nfortunately we cannot postpone our decision
until the happy times when phonological theory wil l have a rm grasp on its
problems He therefore cho oses to follow Cohen et al
Bisegmental analyses in generative phonology AsfarasI
know the rst to level substantive criticism against the use of the feature
tense within the generative framework is Zonneveld Interestingly
Zonneveld bases himself partly on an attackby Lass on the
use of tense for English in SPE Lass s main argument sounds
familiar bynow
Even the few apparently empirical tests for recognizing the fea
ture fail either through b eing hop elessly sub jective or through
logical incoherence
Zonneveld s own main motivation was that he wanted to put the diphthongs
i y u in one class together with the A vowels The diphthongs b e
have like A vowels with resp ect to many of the criteria mentioned ab ove
they can o ccur at the end of a word but not b efore a consonantclus
ter they cho ose the same allomorph of the diminutive Zonneveld
also mentioned that diphthongs and A vowels alternate in ablaut rijd reed
r it ret buig boog b yx b ox etc that in sp eech errors long vowels
were often substituted for diphthongs and vice versa No oteb o om
and that precisel y the A vowels and the diphthongs formed the left hand
context for a rule of Dutch Zonneveld was interested in Weakening
Zonneveld argued that the simplest assumption would b e that
b oth diphthongs and A vowels are bisegmental the diphthongs would con
sist of two di erentvowels following one another and the A vowels would b e
diphthongs of which the two memb ers happ ened to b e identical Because
otherwise he was following SPE he prop osed that a set of morpheme struc
ture constraints would take care of the di erences b etween B vowels on the
one hand and diphthongs and A vowels on the other These constraints
now could all refer to the di erence b etween cons and cons cons
The account presented ab ove can explain these data without assuming
length Ablaut is nowadays regarded as a typically very early pro cess in the
Lexical Phonology something applying to stems if it is considered to b e
of interest for phonological theory at all Substitution may therefore take
place at a moment where diphthongs are still regarded as monosegmental
just like Goat Latin Weakening may b e sensitive to preceding syllables
ending in a vowel etc
App endix Historical overview
This prop osal was accepted and recast in a non linear framework by
Trommelen Van der Hulst and Bo oij a among others
In these theories the vowels and a are represented as follows
a
x x x
A A
A diphthong is a sequence of two short vowels each linked to its own
x slot The most salient prop erties of the Dutchvowel system can nowbe
explained in an elegantway the o ccurrence b efore consonant clusters
and at the end of the syllable
As for the clusters it is assumed that the rhyme of the syllable contains
at most three skeletal p ositions at the edge of the word NowifA vowels
o ccupy already two p ositions in the rhyme for themselves there is no further
ro om for two consonants at most one consonant can o ccur after an A vowel
B vowels on the other hand o ccupyonlyoneskeletal p osition and therefore
they leaveroomfortwo consonants in the rhyme
With regard to the o ccurrence at the end of the word it is assumed that
Dutch syllables are minimally bimoric i e they should contain at least
two p ositions in the rhyme Bip ositional A vowels satisfy this criterion
immediately monop ositional B vowels on the other hand need another
segmentintherhyme in order to ful ll this requirement
During the rst years the problem of the high vowel receives relatively
little attention This changes at the end of the s Trommelen
gives a list of prop erties of the high vowels Apart from the prop erties
already mentioned in this app endix she mentions a few prop erties which
would lead us to classify these vowels in the set of B vowels
The consonant s app ears either after short vowels or after high vow
els but not after non high vowels in set A rush id r s broche
brooch br s fetisj fetish f t s pistache pistachio pist s douche
shower du s pluche plush ply s quiche id ki s
Dutch has a three syllable window at the end of the word for main
stress Sup erheavy syllables at the end of the word can only b e
skipp ed in a few exceptional cases evenaar equator ooievaar stork
etc However the class of stressles s sup erheavy syllables including
For a description of the distribution of the diminutive which requires more elab ora
tion in this typ e of approachseeTrommelen
App endix Historical overview
a high vowel is considerably larger emir id t almoed talmud
t ureluur redshank etc
Sup erheavy syllables are very rare when not at the right edge of the
word Again a large class of exceptions is formed by sup erheavies
with high vowels toendra tundra junta id bourgeois id pizza
id muesli id
Most of these words are fairly recentloans Trommelen therefore
argues that Dutchhastwotyp es of high vowels the long ones in the
native lexicon and short ones in the lexicon of loan words One of two
things will now happ en according to her prediction either the loan words
will adapt themselves to the grammar and get shortened or the grammar
will adapt itself to the loan words and start tolerating short high vowels
In our system short high vowels are actually high vowels in a closed
syllable If Trommelen s observations are right it seems to me
that they are we can say that two constraints are currently in com
p etition Connect N lax and high lax In previous stages of Dutch
Connect N lax was inviolable vowels in closed syllables had to b e lax
It seems that the constraint against high lax vowels is stronger in Mo dern
Dutch High vowels can o ccur in closed syllables without laxing in order to
meet the demands of high lax
One of the most imp ortant problems facing the length theory is related
to stress as discussed b efore Dutch has a quantity sensitive stress system
but CV V syllables i e syllables containing a tense vowel countaslight
i i
whereas CV V and CVC syllables count as heavy This is an extremely
i j
marked situation cross linguistic all y
Lahiri and Koreman solved this problem by assuming that long
vowels consisted of two ro ot no des whichwere attached to only one weight
unit one mora The nonsense syllables vaa and vam got the following
representations resp ectively
vaa vam
v a a v a m
Syllable length was counted at the melo dic tier whichmakes the de nition
of length somewhat awkward syllable weight at the moraic tier
The problem of syllable weightissolved by Kager in a somewhat
di erentway Like Lahiri and Koreman Kager adopts a
App endix Historical overview
moraic theory of syllable structure In his view B vowels are monomoraic
and A vowels are bimoraic Kager suggests that Dutch syllables are
all sub ject to a bimoraicity requirement Therefore counting the number
of moras is not a p ossible way to compute the weight of a syllable in Dutch
Instead weight is computed in terms of melodic complexity ACVV
i i
syllable is light b ecause it has just one vowel at the melo dic level On the
other hand CV V and CVChavetwo di erent segments at the melo dic
i j
level dep endent on the moras Kager s prop osal therefore can b e
seen as the mirror image of Lahiri and Koreman s prop osal the
former computes syllable length at the level of moras and weightatthe
melo dy while the latter computed weight at the mora tier and length at
the melo dy
An imp ortant problem for the prop osals of Lahiri and Koreman
and Kager lies exactly in this computation at di erent levels In
the rst place one of the main motivations for moraic theory is that seg
ment counting was deemed to b e an imp ossible option for the grammar
McCarthy and Prince In the second place b oth theories haveto
p ostulate a rather complicated theory of syllable length Kager has
to b e able to count to three if he wants to explain the sup erheavy status
of raam window ram Lahiri and Koreman have to p ostulate that
twovo calic ro ot no des maybeattached to one mora The same is not true
for a vo wel and a consonant at the rst mora m is heavy but it is true
for a vowel and consonant at the second mora am is p ossible
A di erent approach to solving the same problem is o ered in Smith
et al and Hermans a Under each of these approaches
while there are twoskeletal p ositions involved with the A vowels the second
p osition is empty at the p oint in the derivation where stress is assigned It
therefore do es not count as making the syllable heavy
Both approaches also involve a reintro duction of the feature tense in a
certain way This is literally true for Hermans Smith et al in
stead follow the Dep endency Phonology vowel geometry of Van der Hulst
and Smith and use a dep endent element I denoting pharyngeal
expansion instead All A vowels are identical to B vowels except that they
have an extra dep endent feature I and of course there are twoA vowels
This typ e of account can essentially b e traced backtoTrager and Smith who
prop ose that there are twotyp es of vowels in English simple vowels and complex
vowels The latter have the form of a simple vowel followed by a glide This glide could
be any element of the set fh j wg For instance the long vowel of bought actually
consists of a sequence h
In order to see the connection b etween this prop osal and the ones discussed in the text
it is necessary to recall that h is a consonant without any supralaryngeal sp eci cation
If it is a glide at all it is the glide variantofschwa Nowtheemptyskeletal p osition of
Smith et al and Hermans is also schwa This is made explicit in Hermans a
App endix Historical overview
which do not have a corresp onding B vowel
Smith et al also recognise an underlying length distinction represented
by underlying x slots This length distinction is used for the representation
of the vowels of fr ele frail fr ql freule dame fryql zone id z qn
and also for the lengthened high vowels which o ccur in some loan words such
as surprise id syrpriqs louvredeurtje salo on do or luqvr d qrtj and
centrifuge spinner s ntrifyq According to Smith et al these vowels
underlyingly havetwo x slots
Furthermore the second x slot of the non high A vowels is lengthened
by general rule after stress assignment has applied The high A vowels do
not spread in this way except p erhaps b efore r b ecause otherwise no
distinction could b e made b etween underlyingly long i y and u and short
i y and u which are lengthened by rule
Hermans a has an analysis whichisvery similar in all rel
evant resp ects to the one prop osed in Smith et al All A vowels havean
underlying feature tense This feature ensures that the vowel is sup
plemented with an emptyskeletal p osition via some technical steps which
need not concern us here This p osition is lled up after stress assignment
in the case of the non high vowels In the case of the high vowels it re
mains empty and gets eventually deleted for lack of phonetic content The
solution of Hermans a for really long vowels is also very similar to
the one of Smith et al The main di erence is that for technical
reasons Hermans do es not want to assume underlying x slots He is mainly
interested in Maasbracht Dutch which has long variants of the high vowels
and of the low lax vowels just like Standard Dutch He prop oses that
these vowels underlyingly havetwo ro ots
The main problem with the Smith Hermans typ e of approach apart
from the empirical problems amply discussed in the b o dy of the chapter
is that it is highly abstract It p ostulates a vo calic length that in some
cases never surfaces and that is also irrelevantformany comp onents of
the grammar that normally do consider weight Basically the length is
only seen by the comp onent that needs to see it viz the syllabi cation
algorithm
All these recent prop osals are based on the insight that the main argu
ment for vowel length is syllable structure They all try to establish that
this length is invisible or irrelevant for other phonological mo dules suchas
stress assignment If however syllable length is irrelevant also for syllable
structure itself no argument for vo calic length except in the truly long
vowels is left
Tilburg Dutch and Standard DutchVowel
Length
In the discussion of A vowels versus B vowels in the previous chapter I
have put aside several vowels In particular these were the truly long
vowels and the diphthongs All of these have to b e treated as cases of real
vo calic length One of the main p oints of this chapter is that the lax based
pro jection theory presented in the previous chapter also has to o er an
interesting p oint of view for those vowels I will showthisonthebasisof
Standard Dutch facts and also on the basis of Tilburg Dutch a dialect in
which the parameters length and tenseness arguably b oth play a role in the
underlying system Undeniably long vowels and diphthongs typ ologically
b ehavemuch more like real long vowels for instance they b ehaveasheavy
in the stress system Their head is also most often lax a fact whichI
argue to follow from Connect N lax
Another goal of this chapter is to b e somewhat more sp eci c on some
asp ects of Dutchword structure that have b een left untouched in the pre
vious chapter For instance I will treat notions suchasambisyllabicity
extrasyllabicity and catalexis and the phonetic content of the feature lax
This chapter is structured in the following way In section I discuss
the relevant asp ects of the Standard Dutch long vowel system In sec
tion I then turn to Tilburg Dutch and discuss long vowels and vowel
shortening in that dialect In section I giveanoverview of the fea
ture co o ccurrence constraints that are needed to derive the whole Dutch
segmentinventory The last section is devoted to some concluding remarks
Details in the Standard Dutchvowel system
In this section I rst discuss the Dutchvowel system as I have established
it until now and sp eci cally the role of the diphthongs in that system In
the second section I brie y discuss my assumptions on the ambisyllabicity
of consonants Thirdly I lo ok at the b ehaviour of tense vowels b efore tau
tomorphemic r The topic of section is the phonetic nature of the
Details in the Standard Dutchvowel system
feature I hitherto called lax In section I conclude with an overview
of di erent forms of exceptional syllable structure at the edges of words
Diphthongs The three Dutchdiphthongs i y u have
b een left out of consideration until now I will assume eachconsistsofalax
vowel followed by a high vowel and that b oth vowels have the same vo calic
place features Schematically they therefore will b e represented here as i
y and u resp ectively and they will receive the following structure
cons cons cons cons cons cons
lax high lax high lax high
coronal labial coronal labial dorsal
The rst ro ot no de gets the sp eci cations of a lax mid vowel and the second
ro ot no de those of a high tense vowel
These diphthongs are problematic in one resp ect Sometimes for in
stance in the Dutchversion of Goat Latin cf section they b ehave
as unities If diphthongs are really long in the sense that they havetwo
ro ot no de p ositions this weakens the argumentinfavour of length theory
My prop osal which is in some asp ects similar to prop osals of Calabrese
for Italian dialects and to the prop osals of McCarthy a to
explain the almost identical b ehaviour of English diphthongs is to treat
diphthongs as underlyingly short lax vowels
If we lookatthevowel system wehave established until now we nda
gap There seem to b e no lax high vowels
tense lax
high i y u
e o y
low a
Some authors assume is actually a high vowel but there is some evidence
that this vowel is in opp osition to e rather than to i In morphological
tensing environments i e the plural of a certain class of nouns where
lax vowels b ecome tense the vowel shifts to e systematically lid leden
memb er s l t led smid smeden smith s sm t smed While it is
true that in this same environment also shifts to e whichwemight
The exact organization of the lower part of the system is to b e discussed in section
Details in the Standard Dutchvowel system
attribute to a repair rule since there is no tense counterpart to crucially
never alternates with i in this environment
In the next chapter we will see that e and are the tense and lax
vowel that are most susceptibl e to reduction to while other vowels are
harder to reduce
The same is true in r colouring to b e discussed in section where
e laxes to q not to q leer leather l qr l qr High i only lengthens
in this environment Finally the rules of shortening in Tilburg Dutchalso
treat as the short variantof e
If this is true there really are no high lax vowels in Dutch This gap
has to b e accounted for somewhere in the theory p erhaps by the lter in
high
lax
Wehave already seen this lter can get a phonetic motivation at least if
we assume that lax is de ned as the retraction of the tongue ro ot since
we need di erent almost opp osite tongue movements for high vowels and
for lax vowels Archangeli and Pulleyblank
Usually it is assumed that is a restriction on underlying representa
er that this is not a p ossible option in a theory in which tions Notice howev
it is assumed that any combination of features forms a p ossible input for
the grammar I think the latter assumption makes an interesting prediction
in this particular case Also vowels with the feature combination high lax
plus any of the p ossible place feature combinations are allowed as input
The combination of high and lax is however not allowed on the surface
according to if read as a surface lter whichisvery strong in Dutch
Agiven language can use one of several strategies to avoid violation of
feature co o ccurrence constraints like One p ossibilityisleaving either
high or lax unparsed and replacing it by a non o ending feature value
high or lax tense resp ectivel y
This typ e of clean up strategy is attested in some Italian dialects Cal
abrese and p ossibly also in some non diphtongizing Dutch dialects
whichhave pin like for instance Aalten Dutch and Maasbracht Dutch or
p q n like for instance The Hague Dutch and Tilburg Dutch instead of
Standard Dutch p in for pijn pain Aalten and Maasbrachthavechosen
to parse high The Hague and Tilburg havechosen to parse lax in order
to satisfy cf Van Eekelen et al
Standard Dutch on the other hand seems to have a preference for pars
ing underlying features It cho oses a di erent strategy also attested in
dialects of Italian according to Calabrese ibid which is inserting a new
Details in the Standard Dutchvowel system
ro ot no de Both ro ot no des are linked to the same place features but only
one gets high and the other one lax
The lax high front labial vowel for instance b ecomes either y or y
The surface order of the twovowels is determined by Connect N lax
Because wehavetwovowels in the rhyme this rhyme technically counts
as heavy Therefore its head must b e the lax vowel and b ecause the head
precedes the complementwe get y Similar reasoning leads us to p osit
i and u for the front unrounded diphthong and the back rounded
diphthong resp ectivel y
I will now pro ceed to formalize these observations in terms of interacting
constraints high lax seems to b e unviolated in Dutch Other constraints
we need are listed b elow
a Parse high Parse lax high lax has to b e parsed
b NoLongVowel NLV Two adjacentvo calic ro ots maynot
be linked to the same material
high
Parse high Parse lax NLV
lax
As can b e seen hardly any ordering b etween these constraints can or need
to b e established at this p oint It is only relevant that NoLongVowel is
rank ed b elow most of them Only the ranking of Connect N lax with
resp ect to this constraint has b een imp ossible for me to establish
The candidate set generated over the input high lax coronal gets
the following evaluation for convenience I representvowels with matrices
rather than with feature trees in this b ox and the di erentParse and Fill
constraints for all features have b een con ated
Another inviolable constraintofDutch as indeed in most if not all languages of
the world is the one against high low combinations Also the constraint Parse low
may b e surmised to have a high ranking b ecause diphthongs havea low part and a high
part as we will see in the next chapter Strictly sp eaking surface diphthongs mayalso
b e derived from underlying high low combinations The rest of the argumentwould
work in the same way
I will show b elow that this constraint probably is not needed in the analysis of Dutch
where it can b e replaced by a constraint against the insertion of the feature low plus a
pro jection constraint
In this chapter I implicitly assume that empty cons ro ots are avoided wherever
this is p ossible an assumption for which I refer to the next chapter
Details in the Standard Dutchvowel system
high
Candidates Parse NLV Con N lax
lax
high
lax
coronal
high
lax
coronal
high
lax
coronal
high lax
coronal coronal
p
lax high
coronal coronal
The constraint NoLongVowel gets violated in the optimal candidate
However there is no alternative which do es not violate at least one of the
higher ranking constraints Splitting the o ensive feature combination in
two ro ots is the only thing wecando
Furthermore the rst part of the diphthong phonetically or p ostlexi
cally gets lowsothatwehave y i and u resp ectively probably in
order to makethetwo parts of the diphthong maximally distinct in qual
ity This desire for a maximally falling diphthong apparently overrides the
constraint against insertion of low We could assume this lowering is pho
netic or p ost lexical b ecause it is not structure preser ving and are
no part of the vowel inventory of Standard Dutch except in diphthongs
This argumentmightintheendprove not to b e to o relevant In that case
the lowering might just as well b e assumed to b e lexical Both following
In particular it is imp ortant that Parse lax and Parse high dominate the constraint
against long vowels Beside high lax some other feature co o ccurrence constraints are
op erative in Dutch One instance is low labial since there are no low rounded vowels
in Standard Dutch Yet none of these other constraints seem to havediphtongization as
an e ect Rather in these cases one of the o ending features gets deleted This means
that wehave evidence for the following constraint ordering
i Parse high Parse lax NoLongVowel Parse low or Parse labial
Interestingly there is evidence that Parse high and Parse lax are ranked much higher
in Dutch than the Parse constraints relating to other features This evidence comes
from reduction We will see in the next chapter that high and lax vowels can hardly
b e reduced to schwa whereas other vowels can much more easily This will b e taken as
evidence that the features lax and high are much harder to get rid of than the other
features i e that Parse high P arse lax Parse low Parse labial etc
Details in the Standard Dutchvowel system
forms will still b e in the candidate set after the part of the hierarchyin
has applied
Project N low Heads of branching nuclei need to b e supp orted
by low
low Do not insert low
Candidates Project N low low
low
p
high
lax
coronal
coronal
lax high
coronal coronal
Project N low is yet another memb er of the family of pro jection con
straints under a certain traditional interpretation of the structure of the
nucleus If we assume that all vowels are dominated by the nucleus N long
vowels and diphthongs by de nition imply a branching nucleus cf chapter
for more discussion
Furthermore Project N low and low together have the same ef
fect as NoLongVowel in the grammar of Dutch together they also pro
hibit the formation of unnecessary long vowels Reference to NoLongVowel
is therefore not necessary in the analysis of Dutch I refer the reader to the
last chapter for fuller discussion of b oth the notion of a branching nucleus
and of the status of NoLongVowel in the theory
If we assume the condition on secret languages like Goat Latin is that all
vowel features have to b e copied the unit likebehaviour of diphthongs in
this system follows Both lax and high have to b e copied The o ending
vowel splits b oth in the original and in the copied p osition
Finally Iwanttomention that sometimes o j a j uj and iw ew are
mentioned in the literature as diphthongs I will treat these as sequences
of an indep endentvow el plus glide following Bo oij among others
That is to say the tense vowel would b e in an op en syllable and the glide
would b e in the onset of a degenerate syllable It is interesting that these
sequences seem sub ject to the OCP and that sequences like ij yj and
ej with a frontvowel followed by j or uw ow with a labial vowel
followed by w do not exist We do nd yw clusters zenuw nerve
This constraintmight b e b etter seen as a con ation of structure preserving con
straints blo cking low labial combinations etc
The status of aw is unclear it do es not seem to b e attested in Standard Dutch
but it is found in dialects such as Rotterdam Dutch or Tilburg Dutch
Details in the Standard Dutchvowel system
schwaduw shadow which is somewhat surprising b ecause y is round as
well I will return to these issues in chapter
Ambisyllabicity In this section I will brie y discuss the syllable
structure of Dutchwords like the following
app el app eal p l
hobbel knob h b l
al lofoon allophone lof on
kudde herd k d
rol lade rolled b eef r l ad
All these words contain a V CV sequence of whichV is lax This means
i j i
that the consonant should b e syllabi ed into the co da of the rst syllable to
satisfy Connect N lax Yet there is also evidence that this consonantis
syllabi ed into the onset of the second syllable One reason to assume this
is that the consonantcanbevoiced which is only allowed in onset p ositions
in Dutch a language with Final Devoicing Another reason is that onsetless
syllables headed byschwa do not exist as we will see in the next chapter
This means that the consonant in question is in two syllabic p ositions
In most analyses at least since Trommelen and Van der Hulst
this is taken to mean that the consonant in question is literally ambisyl
labic i e one ro ot no de linked to two syllables In principle it could b e
that phonologically these consonants are even geminates which are short
ened p ost lexically or phonetically Borowsky It o and Mester I will
arbitrarily adopt the former assumption that these are single consonants
attached to two syllabic p ositions so that kudde gets the following repre
sentation
N N
N N
N N
k d
As wehave seen in section the voicing of fricatives in V FV p osition
i j
is almost predictable from the vowel qualityofV ifthisvowel is tense
i
the fricativeisvoiced griezel horror xriz l and if this vowel is lax the
fricativeisvoiceless mossel mussel m s l The problematic cases again
include a or Pasen Easter mazzel luck cf section b elow
Details in the Standard Dutchvowel system
This di erence might b e attributed to a di erence in syllabic a liation
of the fricatives fricatives linked to one syllable would acquire voice
whereas ambisyllabic fricatives would stayvoiceless We nd a similar
di erence in Northern Italian with resp ect to s Voicing Nesp or and Vogel
a simple intervo calic s voices in that dialect but a geminate i e
ambisyllabic s do es not see Slis and Van Heugten for a rep ort that
voiceless fricatives are generally longer than voiced ones
Van der Hulst has shown that lax vowels cannot o ccur b efore
a OL sequence O an obstruent L a liquid While there are words such
as zebra id zebra and microfoon microphone mikrofon there are no
words like z bra or m krofon
In order to rule out these surface forms wehave to rule out two syllabi
cations One is z b bra with an ambisyllabic b and the other is z b ra
with a syllable b oundary b etween the obstruent and the liquid In order
to rule out the rst syllabi cation we need to assume that only segments
which are under the lowest N node can be ambisyllabic In order to rule
out the second syllabi cation I prop ose to use the Syllable Contact Law
Vennemann Clements
Syllable Contact Law SCL In any sequence C C C must
a b a
exceed C in sonority
b
The SCL exlcudes z b ra syllabi cations on the grounds that the word
contains a less felicitous syllable contact We will need this constraintagain
in chapter where it will b e used to rule out certain clusters b efore
r lengthening I will now pro ceed to discuss a pro cess of Dutch
phonology which is related to the issue of vowel length lengthening and
colouring of tense vowels b efore r Thisr has to b e tautomorphemic to
the vowel and furthermore has to o ccur within the same fo ot Gussenhoven
These vowels get a di erent quality b efore this consonant The
following minimal pairs are copied from Trommelen and Zonneveld
The phonological representations used here have b een somewhat idealised from the
phonetic forms
Gussenhoven has p ointed out that the domain of lengthening colouring seems
to b e the fo ot and that there is also a pro cess of colouring without lengthening at the level
of the phonological word cf Europa Europ e r opa Oeralisch Uralic uralis cf also
Van der Hulst I only deal with the fo ot level pro cess and have no explanation of
the word level pro cess The fact that unfo oted vowels do not lengthen seems to b e due
to the strong correlation b etween vowel length and stress
Details in the Standard Dutchvowel system
giek gig ik gier vulture giqr
beet bite b et beer b ear b qr
huub name hyp huur rent hyqr
kleum shiver kl m kleur colour klyqr
boet mend but boer farmer buqr
stoof stove stof stoor disturb st qr
blaat bleat blat blaar blister bla q r
kegel cone ke l kerel guy k qr l
eega sp ouse e a era id qra
ethica ethics etika Erica name qrika
fusie fusion fyzi furie fury fyqri
According to the phonetic measurements of t Hart the colouring
e ect in high vowels is basicaly lengthening to overlong in mid vowels it
is lengthening plus laxing plus some diphthongization towards In the
lowvowel a the e ect is hardest to observe if it can b e observed at all
Iwant to claim that the e ects of colouring on high mid and lowvowels
can nevertheless get a uniform explanation in terms of the theory devel
op ed in this chapter I will takethebackmidvowel o as a paradigm case
Lengthening of vowels b efore sonorants and esp ecially b efore r is a famil
iar pro cess in many languages see for instance Giegerich on Scottish
English I will attribute the pro cess to an admittedly ad hoc constraint
against CV r a sequence of a short vowel followed by r p ending further
phonological and phonetic analysis as to why short tense vowels would b e
disallowed b efore r
R Colour CV r
Because of the way lexical syllabi cation works in Dutch the syllable nal
vowel will always b e tense Notice that this condition gives some evidence
for a degenerate syllable analysis of sup erheavy syllables b ecause suchan
analysis makes it p ossible to treat vuur re vyqr and furie fury fy qri
in the same way
As far as I can see the analysis can only work if we recognize a lexical
p ostlexical distinction It is essential that heer lord is rst syllabi ed as
he r in the lexicon and that this is later changed into h q r Otherwise we
cannot explain whywe prefer to add an extra ro ot no de instead of simply
laxing the vowel and syllabifying r into the co da so that wewould get
h r a non existing well formed word of Dutch
Some schwa like element is inserted b etween the a and the r in manyDutch
dialects so that blaar is pronounced bla r See also section for some discussion
of schwaepenthesis in these cases
Details in the Standard Dutchvowel system
What we need then is a rst lexical round of syllabi cation in which
he r is created and a constraint which I will call Faithful and which
says that previously built syllable structure maynotbechanged I do not
see any other way to derive this same e ect
The e ect of this constraint should b e that the candidates to b e consid
ered all have lengthened their tense vowels b efore r they have added a
ro ot no de and made the vo calic features dep end on b oth ro ots Furthermore
these vowels have received a feature lax
ro ot ro ot
vo calic
V place lax
O
Wehave to explain why lengthening and laxing go hand in hand in this
case
Only lax vowels can p ossibly b e long in this system if we assume
that the twohalves of a long vowel should b e tautosyllabic This is so
b ecause long vowels automatically make a syllable heavy And according
to Connect N lax only lax vowels can head a heavy syllable
This reasoning is summarized in the following tableau
o r
Candidates R Colour Connect N lax Faithful
o r
o o r
o o r
r
p
r
I already noted that I assume therefore that the analysis given here applies
at the p ostlexical stratum The Faithfulness constraint ruling out r
the result of merely laxing the formerly tense vowel not lengthening it is a
constraint to the e ect that previously built syllable structure may not b e
a ected In this case o and r have b een syllabi ed into di erent sylla
bles at the rst lexical stratum This heterosyllabici ty relation is preferably
not changed Therefore r is preferred over r
It is necessary that RColour is ranked very low in the lexical hierarchy in order for
the underlying sequence or not to surface as r already in the lexicon
Details in the Standard Dutchvowel system
As mentioned ab ove the long lax vowels b efore r also have a tendency
to diphthongize towards I will return to this in section
Wenowhave the following explanation for the mid vowels These vowels
have to lengthen Because in this way they create heavy syllables they also
have to b ecome lax Laxing alone is not su cient b ecause syllable structure
built in previous strata has to b e satis ed
The phonetic e ect of colouring on the lowvowel a is much harder to
discern I claim that this is just a sup er cial phonetic fact The quality
di erence b etween a and seems to b e smaller in Dutch than the dif
ference b etween other tense lax pairs Furthermore long vowels get some
slight phonetic tensing automatically q and a therefore are phonetically
very close This is the reason why it is hard or even imp ossible to observe
a di erence b etween the two We will see b elow that this asymmetry in the
system is even clearer in Tilburg Dutch section
The b ehaviour of the high vowels can b e understo o d in terms of already
established principles These vowels have to lengthen and in order to satisfy
Connect N lax they also have to b ecome lax If we assume the con
straint against high lax is undominated in Dutch b ecause lax high vowels
never surface in the language this laxing is not p ossible however
high
Candidates R Colour Connect N lax
lax
i r
i i r
p
ii r
r i
i i r
The high vowels thus are predicted to only lengthen Which is con rmed
by the ndings of t Hart
The phonetic nature of the tensing feature Wehave es
tablished that the di erence b etween A vowels and B vowels can b est b e
describ ed in terms of a feature Ab ove I informally suggested ATR as
The symbol i is used to indicate a lax version of i
Under a slight reformulation the R Colour constraint against V r could b e made
to predict that also diphthongs are dissallowed b efore r This prediction is indeed
con rmed at least as far as word nal p ositions are concerned for u and y ir
is disallowed in tautomorphemic words altogether indeed Trommelen and Zonneveld
argue that this diphthong colours to q The facts are somewhat complicated
and it seems di cult to work out an analysis why the colouring of diphthong is so
restricted to apply only to ir not to the other twodiphthongs
Details in the Standard Dutchvowel system
an interpretation but until now I informally used the lab el lax to refer to
this feature However quite a few other lab els have b een suggested and I
have not yet provided any argumenttofavour ATR or tense over their
comp etitors
In this section I will brie y discuss some of the lab els that can b e found
in the twentieth century literature I will argue that each of these lab els
has its advantages but its problems to o My conclusion will b e that while
no conclusive argumentation is p ossible at our present state of knowledge
the strongest phonological arguments may indeed b e p ointing towards a
tongue ro ot feature
tense This probably is the lab el that is most commonly
used in the literature Unfortunately it is not very well de ned indeed
the lab el tense has b een used to refer to almost all relevantvo calic prop
erties mentioned b elow De Rijk assumes that we may regarda
tense vowel as a sequenceoflaxones and uses tense to describ e what
wewould have called phonological length Very often the term is also
used to describ e what Lindau and Lab ov among others call
p eripheral This use of the term has b een common in Germanic stud
ies at least from Winteler and Sievers until Giegerich
and Bo oij a The most p ersistentproponent of the hyp othesis that
tense and ATR denote the same phonological ob ject is Halle Jakobson
and Halle Halle and Stevens Halle Halle and Clemen ts
Tense thus can b e seen as a cover term and the same is true for its logical
counterpart lax The terms themselves are of course somewhat impression
istic which makes them very useful for informal discussion as long as we
have not settled the exact nature of the distinction at hand It is just one
step less abstract than the A vowel vs B vowel distinction I used ab ove
Its very imprecision makes the term useless in a maximally explicit theory
of sound structure however In the end tenseness should b e reduced to
one of the following features or to another one yet unknown
p eripheral Peripherality within the vowel triangle is prob
ably the most common de nition of tenseness in linguistic theory Lindau
Lab ov is the only scholar who uses it as an indep endentfea
ture in recentwork Lab ov uses tense as an additional purely abstract
feature to cross classify English vowels
The Dutchvowel triangle for instance can b e roughly drawn as follows
lax vowels are underlined
The very rst to prop ose the lab el p eripheral at all is Lindau
Details in the Standard Dutchvowel system
i y u
e o
a
In a sense the lax vowels form an inner triangle and the tense vowels an
outer triangle If a vowel is back and round its tense counterpart is even
more back and more round If a vowel is low its tense counterpart is even
lower
A lot of problems have b een raised against this interpretation of the fea
ture tense It seems to b e a comp osite feature and most tense lax contrasts
can also b e expressed in terms of the indep endently needed parameters
vowel height and backness vs frontness cf Gilb ers Furthermore
it is the only relative feature in mo dern phonological theory We cannot
tell of a given vowel in isolation whether it is p eripheral or not This pred
icate can only b e attributed to a vowel within avowel triangle Lass
Donegan Furthermore it is not very clear how to use this lab el to
non arbitrarily distinguish e g b etween and y while not at the same
time distinguishin g b etween for instance i and e in exactly the same way
The lab el do es have some explanatory use with resp ect to the constraints
prop osed in this chapter If the articulatory muscles havetomovetoan
extreme p osition and syllables can just use a limited amount of time the
strong syllable cut of tense vowels can b e explained It is still di cult to
see the phonetic reason why lax vowels should o ccur in closed syllables but
none of the features discussed here can provide us with a ready explanation
for that fact
More problematic is the sp ecial b eha viour of high vowels The con
straint against the combination of high and p eripheral is di cult to
understand Esp ecially i and u are real extremes in the vowel triangle
There is no apparent reason why they could not have less extreme variants
in Dutch and many other languages while this p olarityisavailable for
roundness and backness In section we will see that the a pair
is somewhat sp ecial to o In general the strong corresp ondence b etween
vowel height and tenseness cannot b e explained if weinterpret tense as
p eripheral
The reason why Lab ov prop osed p eripherality as a feature can unfortunately
not b e discussed in this dissertation He uses it to describ e vowel shift phenomena
mainly in the history of English and in mo dern American and British English dialects
The vowel shift and most asp ects of English phonology unfortunately fall b eyond the
scop e of this study
Details in the Standard Dutchvowel system
op en The reasoning ab ove leads us to the p osition of Clements
i
a who argues that the distinction in Bantu languages previously de
scrib ed in terms of tense or ATR should b e describ ed byanaperture
feature op en low and high should similarily b e replaced byin
stances of op en so that we get a scale of ap erture features The Dutch
system could under Clements a prop osal b e analysed as follows
op en
op en i y u
op en
op en
y op en
op en
op en
e o op en
op en
op en
op en
op en
op en
op en a
op en
Ihave assumed op en low and op en high Wethenget
to the conclusion that op en corresp onds to ATR quite surprisingly
b ecause ATR seems to b ehavelike high hence likea op en feature
in pro jection constraints not like a op en feature suchas low Another
problem is that there is no inherent reason to numb er the di erentvowel
heights this way Why shouldn t wesaythate g op en op en
op en
Indep endent of the p ossible uses of the feature in the analysis of Bantu
see however Kenstowicz a for discussion of a p ossible reanalysis of
Clements a examples in terms of high and ATR the multiply
o ccurring feature op en seems problematic in the analysis of Dutchpre
sented here
On the other hand Clements a and Odden undeniably
make a forceful argument that tense should b e attached to a separate no de
together with the height features and to the exclusion of all other vo calic
features They show that these features spread together in some Bantu
languages Because we also established a sp ecial relation b etween vowel
height and tenseness in feature co o ccurrence restrictions I conclude that
Although I do not discuss these explicitly in the main text I think similar ob jections
can b e raised against prop osals in the framework of Dep endency Phonology suchasthe
one in Smith et al as I raise here against op en
Details in the Standard Dutchvowel system
there is su cient evidence for an ap erture no de but that the daughters of
this no de are high low and ATR
ATR RTR Several arguments have b een broughtup
against the equation of ATR and tense One is that these features seems
to have slightly di erent phonetic re exes In particular ATR do es not
usually involve the fronting of backvowels It remains a question whether or
not languages mayvary in their phonetic realization of a given phonological
feature Another ob jection is that in real ATR languages as sp oken for
instance in West Africa there is no connection b etween ATR and syllable
structure I think that the fact that a feature can have an e ect in certain
languages which it do es not have in others is not necessaril y an argument
for not using that feature esp ecially not in a theory likeOTinwhich
constraints which are very strong in one language may b e ranked very low
in another language
According to Clements an imp ortant problem with the as
sumption that high ATR and low are sisters is that ATR normally
is de ned in terms of the activity of a sp eci c articulator the tongue ro ot
Painter Lindau As an articulator b ound feature wewould
exp ect it to pattern with other articulator b ound features This means
that its most natural place is under a place no de for which assumption
Clements refers to McCarthy b The features high and lo w on
the other hand are articulator free features they can b e brought ab out by
movements of b oth the jaw and the tongue b o dy and ro ot
Although the feature theory sketched byClements a in whicha
sharp division is made b etween articulator b ound and articulator free
features each with their own no des seems very interesting it remains to
be worked out Furthermore a precise phonetic articulator free de nition
of op en is not provided so that wehave no exact idea of the naturalness
i
of Clements alternative Finally as Clements a himself remarks the
hierarchical nature of op en is not really mirrored in other phonological
features except p erhaps tone So whereas Clements a criticism in
itself is to the p oint and should lead to a more precise idea of the phonetic
Yet another ob jection Stewart Ramers is that whereas tongue ro ot re
traction seems to b e the marked value in Europ ean languages as wehave seen African
languages often seem to treat tongue ro ot advancement as the marked feature Several
answers are p ossible to this ob jection In the rst place it do es not seem necessarily true
that African languages only refer to ATR Archangeli and Pulleyblank analyse
several languages as having the ATR value as the active value amongst them Wolof
and Yoruba In the second place we could assume either that languages can cho ose
between having the or the value of ATR as the marked value or we can assume
that there are two complementary features ATR and RTR
Details in the Standard Dutchvowel system
and phonological nature of the features involved it is not clear whether his
own theory do es not su er from very similar problems
In my view the fact that high low on the one hand and ATR on the
other interact with each other byway of feature co o ccurrence restrictions
has to b e accounted for by reference to the fact that all these features refer
to movements of the tongue b e it that the former two refer to the b o dy of
the tongue and ATR refers to the tongue ro ot By now familiar constraints
like a therefore get a natural explanation in phonetic terms while the
constraints in b are predicted to b e imp ossible in a grounded phonology
high high
low ATR a
ATR low
high ATR
b low high
ATR low
A theory of op en cannot makethistyp e of prediction at all The equiv
alents of the forms in a as well as of those in b seem to refer to
arbitrary sets of ap erture features
op en op en
a op en op en
op en op en
op en op en
b op en op en
op en op en
In the remainder of this work I will use ATR wherever phonetic explicitnes s
is required and lax everywhere else
Extrasyllabicity and catalexis Iwillnow discuss two formal
mechanisms I have used without de ning in the previous chapter catalexis
and extrasyllabicity I will brie y give these twomechanisms a somewhat
more precise account in this section It is very imp ortant to de ne them
without referring to underlying proso dic structure My claim is that no un
derlying proso dic syllable structure is necessary to describ e Dutchvowel
length I do refer to catalexis and extrasyllabicityhowever so these notions
should get a non proso dic interpretation
yp es of Extrasyllabicity Ihave distinguished b etween twot
word nal extrasyllabicity the extraprosidicity of coronal consonants and
the extrasyllabici ty of other segments
At the right p eriphery of a Proso dic Word more than one coronal voice
less obstruent can b e extrasyllabic In real monomorphemic words the
The constraint low ATR will b e shown to b e op erativeinDutch in section
Details in the Standard Dutchvowel system
maximum is toorts torch torts herfst autumn h rfst and in derived
forms vreemdst strangest vremtst but it is not clear whether this up
p er limit is an accidental fact ab out the Dutch lexicon or a consequence
of some grammatical principle I will assume that these coronal segments
are extraproso dic in a broader sense They are outside the Proso dic Word
altogether where they are licensed in some sp ecial way See Paradis and
Prunet McCarthy and Taub and references cited there for
discussion of the sp ecial status of coronals in this resp ect and others I will
not further develop the coronal extraproso dicity here
My real concern is with the other class of extrasyllabic segments which
are not necessarily coronal These are the nal segments of boom tree and
ramp disaster It is absolutely crucial that these segments are extrasyl
labic b ecause this allows us to say that tense vowels o ccur in op en syllables
only It is also crucial that these non coronal extrasyllabics cannot cluster
otherwise wewould get unattested forms like raamp I assume these seg
ments o ccupy a nal p osition within the proso dic word These I will call
extrasyllabic whereas the coronals I call extraproso dic just to havea
convenienttermathandwhich can distinguish b etween the two classes
It seems that segments are always attached at the lowest p ossible p osi
tion In kat cat k t the t lls a rhyme p osition to satisfy Connect N
lax In president t is still within the Proso dic Word b ecause it creates
the e ect of a sup erheavy nal syllable Only in herfst where it cannot
get a regular proso dic p osition it stays outside the Proso dic Word
I already discussed the two theoretical p ossibilitie s wehaveforthede
scription of extrasyllabics One was to let these segments b e unsyllabi ed
altogether the second was to syllabify them into a degenerate syllable I
will cho ose the second option here even though nothing in my analysis
seems to crucially hinge on this assumption My reason for cho osing it
is stress where sup erheavy syllables b ehave as disyllabic A degenerate
syllable is a syllable with just consonant and an empty head The notion
will b e further develop ed in the next chapter For now wemight assume
the following representation where C is the consonant p osition
N
N
N
C
Because degenerate syllables are marked constituents they cannot b e freely
distributed over the word we need a pro jection constraint banning them
Details in the Standard Dutchvowel system
whichwehave already seen at work in the previous chapter
N V A branching N should have a head with vo calic Project
features
In order to make sure that degenerate syllables are in most cases disallowed
wehave to stipulate that Project N V dominates Parse consonant
It is generally b etter to leave a consonant unparsed than to put it into a
degenerate syllable
The problem that has to b e solved now is howwecanmakesurethat
extrasyllabic segments escap e Project N V in absolutely word nal p o
sition
In the next chapters we will see that there is quite a lot of evidence
that phonological word edges in Dutch are almost inalterable Pro cesses of
vowel ep enthesis for instance never insert a schwavowel at the edge of a
word I assume that here wehave another result of an Alignment constraint
on the edges of words McCarthy and Prince b comparing the edges
of lexical sp eci cation and output form for the correct formal de nition of
this constraint see chapters and app endix A
Align The edges of a word may not b e altered
If Align is ranked su ciently high at least ab ove the pro jection con
straint against degenerate syllables it will blo ckword nal deletion of con
sonants but not word internal deletion It will similarily blo ck insertion of
ep enthetic material to add weight to the head of the degenerate syllable
To illustrate this I will give the tableaux for some simple examples such
as maan mo on man and ma mum ma
a maan
Candidates Align Project N V Parse C
ma n
p
ma n
ma ne
ma n
b ma
Candidates Align Project N V
ma t
p ma
Details in the Standard Dutchvowel system
The t in the last tableau is an arbitrarily chosen ep enthetic consonant such
consonants do not surface in Dutch The underlined e in the rst tableau
similarily is an arbitrarily chosen ep enthetic vowel not equal to schwa such
vowels never surface either
Since Align refers exclusively to word edge p osition it cannot savede
generate syllables word internally a It cannot savemultiple stacked
degenerate syllables either b I arbitrarily assume that Project N
V is satis ed by deletion of the consonant it could also b e satis ed byin
sertion of a schwavowel as we will see in chapter where schwaepenthesis
is discussed
a maatka
Candidates Align Project N V Parse C
p
ma t ka
ma t ka
b raamp
Candidates Align Project N V Parse C
ra m p
ra m p
p
ra m p
ra m p
ramp
Candidates Align Contiguity Project N V Parse consonant
p
r m p
r m p
r m p
Iintro duced one new constraint in the last tableau Contiguity
Contiguity Given a string X Y if there is a syllable such
that are dominated by should b e dominated by
Contiguity helpsustocho ose r m p over r m p
Since Align crucially refers to both word edges as we will see in the
chapters that follow one could ob ject that at least we predict degenerate
syllables to o ccur also at the left edge of the word Thus for instance we
would predict non o ccurring words suchas xvar Yet in this case there
is a way to satisfy all constraints mentioned We can insert a schwa in the
There are a few words suchas psychologie or mnemoniem whichmight b e seen
Details in the Standard Dutchvowel system
nucleus word nally this schwawould have to b e inserted at the end of the
word thereby still violating Align Yet at the b eginning of the word the
nucleus is word internal and Align is not violated byschwaepenthesis
as having an extrasyllbic initial consonant but these typ e of clusters are sub ject to
very sp eci c conditions the real onset has to consist of a coronal and furthermore it
may not b e complex For the latter reason I susp ect that these clusters are exceptional
bisegmental onsets not a degenerate syllable followed by a normal syllable A language
where degenerate syllables seem to b e allowed b oth word initally and word nally is
p ossibly Polish Gussmann
Details in the Standard Dutchvowel system
xvar
Candidates Align Contiguity Project N V Parse C
x va r
x va r
x va r
x v a r
p
va r x
In this case there is a way to satisfy all constraints mentioned We can insert
aschwa in the nucleus word nally this schwawould have to b e inserted
at the end of the word thereby still violating Align butword initially the
schwa is inserted after the consonantsothatAlign is not violated
The preference for the edges of Proso dic Words to remain unaltered
has the e ect in Dutch that consonants are put in a degenerate syllable
at the end of a word only This explains the distribution of sup erheavy
syllables in Dutch which o ccur only at the end of the word This topic in
connection to the b ehaviour of syllables headed byschwa will b e discussed
in the next chapter
Catalexis Inow turn to the other phenomenon that is related
to the p eriphery of words catalexis We need this device to explain i
whyCVwords still satisfy Proso dic Minimalityand ii howwords with
exceptional word nal stress are marked in the lexicon
In the original prop osal of Kiparsky catalexis was an underlying
mora or syllable without features attached to it CV words in this way
could b e seen as bimoraic or bisyllabic and nal syllables could b e marked
as b eing sup er heavy thereby attracting stress
Clearly the option of underlying moras or syllables is not compatible
with the assumptions whichIhave made until now Up to this p oint I
have b een able to stick to the null hyp othesis that no proso dic structure
is underlying Notice however that nothing excludes underlying bare ro ot
no des These ro ot no des have the same e ect as underlying moras Supp ose
they are place less obstruents which means they will b e represented only
as a cons sonorant ro ot I assume this will b e interpreted as one of
the empty consonants h or b A similar prop osal has b een made by
Dunlap for Spanish whichalsohasatrochaic system and word nal
stress and for which catalectic analyses have b een prop osed as well see also
Nouveau and for Italian Vagra
On the emptiness of h and b see chapter
Details in the Standard Dutchvowel system
Because they are obstruents they will surface as degenerate syllables
after heavy syllables I give an example tableau for ma mum b elow this
tableau supplements b ab ove it do es not replace it
Details in the Standard Dutchvowel system
ma
Candidates Align Project N V Parse C
p
ma h
ma h
Aword like chocol a can also get a nal h in underlying representation
so that the nal fo ot will b e bisyllabic This h is actually phonetically
realised in the case of words with a nal stressed op en syllable cf No ote
b o om
The question remains how the nal syllable in ma h is treated by the
phonetic comp onent It might b e that a syllable without any place sp eci
cation at all is simply left uninterpreted On the other hand esp ecially in
word nal p osition after A vowels a phonetic h can b e observed
Another question that arises is whether this line of reasoning do es not
destroy our explanation of the di erence b etween tense and lax vowels al
together Whyis m h not p ermitted to surface as a word The answer
is that h can never o ccur in co da p osition in Dutch not word internally
and not at the p eriphery It can only o ccur in onset p osition This fact has
to b e stipulated byany theory in one way or another I prop ose to do it
by the following constraint
Coda h A h may only o ccur in an onset
Coda H Parse lax
Finally Iwant to discuss one prediction I make If these words with nal
stress really end in a catalectic consonant wewould exp ect this consonant
to have other e ects as well Indeed one such e ect can b e found in an
area not directly related to stress viz in ectional morphology
Adjectives take an in ectional ending in some syntactic contexts for
instance in inde nite DPs of common gender This in ectional ending only
surfaces when the word in question ends in a consonant not when it ends
in an unstressed vowel
In a theory of phonological pro jection the constraint Coda h mightbeseenasthe
result of the interaction of twoweakness constraints Weak N C the weakness con
straint requiring weak p ositions in rhymes to have at least some consonantal feature
apart from cons son would b e very strong dominating for instance Parse C
N C would b e muchweaker and b e dominated by for instance Parse whereas Weak
C Since I consider the whole matter not su ciently imp ortant I will not go into it any
further
For some of these cases the lackofovert in ection might also b e due to totally
di erent factors For instance it mightalsobethattheword albino is more noun like
than real adjectives and therefore not able to carry overt in ection Henk van Riemsdijk
A dialect with real length Tilburg Dutch
a een jn tijd
a happy time
b een albino man
an albino man
c een mica tafel
a micaceous table
d een sexy dame
a sexy lady
Words ending in a stressed vowel and monosyllables ending in a vowel
b ehave as if they were ending in a consonant in this resp ect as predicted
They are the only vowel nal words that takea in ection
a een wee j geur
a sickly smell
b een continu w stro om
a continuous ow
It is hard to see whywords ending in stressed vowels would b ehavelike
words ending in consonants if we do not adopt a theory of catalectic con
sonants
A dialect with real length Tilburg Dutch
This section deals with vowel length and vowel shortening in the Tilburg
dialect of Dutch Tilburg Dutch has a much larger set of vowels than
Standard Dutch Two prop erties of the Tilburg vowel set are of sp ecial
interest to us In the rst place it lls in a few gaps in the Dutchvowel
system since it has low rounded vowels In the second place it seems to
have a real length distinction on top of the tenseness distinction familiar
from the standard language Furthermore it has some phonological and
morphological pro cesses that seem to refer to length I will showinthis
section that even in Tilburg long vowels haveavery limited distribution
Ihave collected my data on Tilburg mainly from the dictionary of Van
Rijen and from the dialectological study of Boutkan The
pro cesses describ ed here are not unique to the Tilburg dialect however
They are quite wide spread in the Brabant area which ranges approx
imately from Brussels Belgium in the south to s Hertogenb osch The
p c noun like adjectives ending in consonants suchasgenoeg su cient are not able
to carry in ection either The same is not true for a form suchassexy whichisaclear
adjective The set of data mighthowever b e considered to o small to settle the p oint
A dialect with real length Tilburg Dutch
Netherlands in the north and in its broadest sense includes the provinces
North Brabant in the Netherlands Antwerp Flemish Brabant and the
East Flemish Dender area Belgium See for instance Zonneveld on
Brussels Dutch Keymeulen and Taeldeman on Hofstade Dutch and
Nuyts on Antwerp en Dutch Furthermore traces of the pro cess can
also b e found in Standard Dutch and Standard German
Below I will rst describ e the Tilburg vowel set in some detail Subse
quently I will giveanoverview of the various lengthening and shortening
phenomena attested in the dialect After this I will give an analysis of
Tilburg I will show that vo calic length has a very limited distribution and
that the Tilburg system helps us de ne the Standard Dutchvowel system
in a more principled way In section I brie y discuss the b ehaviour of
long vowels in some other Brabant dialects In section I will discuss
the limited distribution of long vowels in b oth Standard Dutch and the
Tilburg dialect
The key hyp othesis will b e that Tilburg has the same typ e of syllable
structure as Standard Dutch in spite of its somewhat di erentvowel sys
tem Also the stress system and the relation b etween syllable structure
and stress seems to b e the same in the two dialects Ideally we therefore
wanttohave an analysis of Tilburg proso dic structure which is maximally
close to the theory wehave for Standard Dutch There is one minor su
p er cial di erence b etween the two dialects whichIwant to note at the
outset Tilburg Dutc h has a small numb er of functional items suchasw e
what w and d e that d which end in the lax vowel and sometimes
or Historically these forms are derived from words ending in a t
d t w t Synchronically I have no explanation for this fact
The vowel system The Tilburg dialect of Dutch has di erent
vowels The consonant system of Tilburg Dutch is exactly the same
as in the Standard language Weijnen All vowels of Standard
Dutch are found in this dialect but in addition we nd long versions of
the lax vowels and a height distinction in the lax rounded vowels Wecan
divide the Tilburg vowel system in groups the tense vowels the short lax
vowels the long lax vowels the dorsal vowels a and schwa All
of these vowels have distinctive status According to Boutkan the
only vowels for which it is di cult to nd minimal pairs are and yq
WhileIwas completing this thesis I b ecame acquainted with the work of Boutkan
and Kossmann who give an ample overview of Tilburg phonology I refer the interested
reader to that work for a list of minimal pairs showing convincingly that eachofthese
vowels has a phonemic status in the language
The term dorsal vowel is meanttoputthesetwovowels apart from the rest of the
vowel system I will argue b elow that low is not an adequate term to set this pair apart
A dialect with real length Tilburg Dutch
He mentions r o ok smoke V ryqk and reuk smell r k as the only pair
he found Boutkan do es not discuss diphthongs Van Rijen
claims that older sp eakers of Tilburg do not haveanydiphthongs Younger
sp eakers havedevelop ed a j and au presumably under the in uence of
Standard Dutch i and u Yet here I will concentrate on the diphthong
less variant of the Tilburg dialect I will now brie y discuss eachofthe ve
subsets Where I give Tilburg words in non phonetic transcription I follow
the conventions of Van Rijen
Non dorsal tense vowels The set of non dorsal tense vowels
of Tilburg Dutch is exactly the same as that of Standard Dutch mo dulo
some minor phonetic di erences I listed the six vowels in this set in
Belowevery vowelIgive an example word and its gloss
front back
unrounded rounded
i y u
plisi yp rt sjurt
p olice man b ed he lo oks
e o
me s l lot r
with forest meadow draw lots
The set can b e further divided into front unrounded i e front rounded
y and back rounded u o or alternatively high i y u andmid e
o vowels
According to Van Rijen there is also a set of over long high
tense vowels f iq yq uq g Yet these vowels seem to b e in complementary
distribution with the normal high tense vowels The long variants o ccur
b efore tautosyllabic r and the other variants in all other contexts I will
ignore this contrast in the rest of this section see section ab ovefora
discussion of a similar pro cess in Standard Dutch
Non dorsal short lax vowels The next group to consider are
the non dorsal lax vowels In Standard Dutch this subpart of the system has
two gaps as compared to the tense vowel set There are just two rounded
vowels one front y and the other one back Standard Dutch
makes no height distinction in this part of the system These two gaps are
The fact that I call these vowels non dorsal is purely descriptive As a matter of
fact the back labial vowelsofcoursearesupposedtohave a dorsal comp onent
A dialect with real length Tilburg Dutch
lled in as it were in the Tilburg dialect which has six non dorsal short lax
vowels Compare to
front back
unrounded rounded
t d r r ms
eats through Roman Catholic
d b k sx p
that I sto op sheep
If we compare to it might seem that they are exactly parallel
Because the vowels in are high and mid the vowels in could then
also b e classi ed as high and mid I will argue b elow however that this is
not the case y and in Tilburg Dutch as in Standard Dutch are mid
vowels and are low
Technically we can then get the relevantvowel system by the following
relative rankings in Standard Dutch and Tilburg Dutch
low
a Standard Dutch Parse low or Parse labial
labial
low
b Tilburg Dutch Parse low Parse labial
labial
In Standard Dutch the constraint against low rounded vowelsissostrong
that it can cause at least one of the two incompatible features to disapp ear
In Tilburg Dutch on the other hand it is more imp ortanttokeep b oth
low and labial in the output than it is to av oid low rounded vowels
With resp ect to language acquisition wecansay the following I assume
that in the initial stage all faithfulness constraints are ranked very low and
all other constraints very high With resp ect to the three constraints under
discussion the Standard language therefore represents the unmarked case
A Tilburg child at a certain p oint will hear low rounded vowels and from
those she can conclude the reverse ranking that is needed for her dialect
Non dorsal long lax vowels The most dramatic di erence
between Standard Dutch and Tilburg Dutch is found in the realm of long
lax vowels Standard Dutch has only three of these vowels q in serre
s qr q in freule fr ql and q in zone z qn These vowels
haveavery limited distribution and are standardly considered to b e loan
A dialect with real length Tilburg Dutch
phonemes of Dutch b ecause they only o ccur in wordsofFrench origin I
will return to these Dutchvowels b elow
This typeofvowel o ccurs much more abundantly in Tilburg however
In this dialect we nd long lax vowels in a lot of lexical items not only
those of French origin Apart from that wehave a height distinction for all
three di erent places of articulation so that all in all we get the following
picture
front back
unrounded rounded
q yq q
bl qk knyqp sx qn
app eared button b eautiful
q q q
r qm j qn kl qr
rhyme onion ready
Phonetically what distinguishes these vowels form the ones in unam
biguously is length
In Standard Dutch all frontlongvowels are low lexically wehave q
but not q The di erence b etween Standard Dutch and Tilburg Dutch
thus might b e accounted for in the following way
a Standard Dutch Project N low low
b Tilburg Dutch low Project N low
The constraint Project N low requires long vowels to b e supp orted by
a feature low In Standard Dutch this constraint is undominated so that
all long vowels surface as low In Tilburg Dutch on the other hand lowering
of all long vowels is blo cked by the high ranking of the constraint against
insertion of low
Again the Standard dialect can b e assumed to b e the unmarked initial
stage of the language acquisition pro cess and a Tilburg child knows that it
has to reverse the ranking of the relativeFaithfulness constraint vis avis
Project N low once it hears a long non lowvowel
Dorsal vowels It seems that for all non dorsal vowel articu
lations wehave at least three p ossibilitie s in Tilburg Dutch tense short
lax and long lax The dorsal vowels fall somewhat outside of this system
however here wehave only twovariants a and Phonetically a could
be characterized as somewhat longer than but we can also see a as
tense and as lax All in all wehavetwo dorsal vowels instead of three
A dialect with real length Tilburg Dutch
tense long lax short
a
kans s
chance if
Wehave seen that the a pair b ehaves in a similarly exceptional way
b efore tautomorphemic r in Standard Dutch Unlike other tense vowels a
do es not seem to change colour in that environment at all My explanation
there was that q and a are phonetically to o similar This seems to b e
con rmed here in the sense that also in Tilburg where length and laxness
combine as phonetic features much more freely still no distinction can b e
made b etween q and a
The analysis of the a pair is further complicated by an idiosyncratic
rule of Tilburg which tenses the dorsal b efore nt mp and k braand
re brant br nt laand land lant l nt schaand shame sxant
sx nt raamp disaster ramp r mp baank bank ba k b k and
xt aacht eight axt xt naacht night naxt n xt The Standard
language has the lax varian t in all this cases The precise explanation for
these pro cesses remains to b e found
Stro op discusses similar more general pro cesses of vowel length
ening b efore homorganic nasal obstruent clusters in southern Dutchas
comp ensatory lengthening A nasal which is homorganic with an obstru
ent lo oses its indep endent place features b ecause it also slightly nasalizes
the preceding vowel it might seem to disapp ear altogether and the vowel
can take up its place We might also see a parallel esp ecially for the axt
clusters in Belfast Philadelphia and New York State tensing Fergu
son Dunlap Harris Lab ov which is triggered
by nasals fricatives and voiced stops It is still unclear however i why
in Tilburg only a are a ected ii what uniquely and non trivially
characterizes the nt mp k and xt contexts
Ileave these problems op en for further research since they seem to b e
su ciently indep endent of syllable structure It will b e clear that all these
considerations taken together make a a less suitable pair for testing
In most variants of German the alternation b etween tense and lax a has dis
app eared Vater Also the transcription and the phonetic realisation of tense lax
pairs is in some languages sometimes reversed Vater uses for the tense vowel
and a for the lax vowel I follow the Dutch tradition in this thesis
Ihave argued ab ove that Standard Dutch and Tilburg Dutch twaalf twelve mayb e
the result of a similar a tensing pro cess extending also to haalf in Tilburg Dutchbut
not in the Standard dialect The f in this case is the result of Final Devoicing of an
underlying v which surfaces in the plural twaalven In the next chapter wewillseethat
mayb e in Standard Dutch the context rz is also a trigger for tensing of a
A dialect with real length Tilburg Dutch
the morphologically triggered shortening and lengthening pro cesses under
discussion I will therefore put them apart until further discussion of their
status in the Tilburg vowel system p
Schwa The nal vowel of the Tilburg vowel system is schwa
It stands on its own b eing essentially featureless
schwa
r m
arm
I will not discuss the b ehaviour of Tilburg schwainthischapter I assume
it to b e basically the same as the b ehaviour of schwa in Standard Dutch
which will b e discussed in the next chapter
Why only lax vowels can b e long The rst question that
arises with resp ect to the Tilburg vowel set is why it is asymmetric in the
way it is Why is there a length contrast in the lax vowels but not in the
tense vowels
As a matter of fact this typ e of asymmetry is predicted by the constraint
Connect N lax as wehave seen in the discussion of r colouring Supp ose
that wehave a long vowel i e a vo calic no de linked to tworootnodes
o o
vocalic
Assuming that the two parts of the long vowel can only b e syllabi ed tauto
sylabically they cannot each head their own syllable the resulting struc
ture will always include a branching Nnode
N
o o
vo calic
However the Tilburg schwashows up one somewhat unexp ectedcharacteristic We
nd at least one word eweg away w x as a counterexample to the claim made for
Standard Dutch that no lexical word starts wit schwa
A dialect with real length Tilburg Dutch
Nbranches and therefore Connect N lax demands that the head should
dominate a feature lax Therefore if there is any asymmetry at all we
predict that this head is lax
There is however one p ossible exception In some dialects also the high
tense vowels can have long variants This may b e true for Tilburg Dutch
and Standard Dutchwhichhavesome loan words suchascentrifuge id
s ntrifyq z and team id tiqm It certainly is true for Limburg dialects
where there is a relatively large numb er of long high vowels Weijnen
Hermans a The interesting thing ab out this is that exactly
the high vowels do not have lax counterparts If we assume that in these
dialects the constraint against the combination of high of lax dominates
Connect N lax wegettheright result cf the table in Wewill
see indep endent evidence for this constraint ranking b elow
Vowel shortening In the immediately preceding section I have
shown that Tilburg Dutchhasvowel length In this section I will showhow
long vowels b ehave in morphological contexts I will show that quite a few
morphological pro cesses involve shortening Indeed this seems to b e true
for all pro ductive in ectional su xation pro cesses After su xation long
lax vowels shorten and tense vowels turn lax I will rst discuss some of
the relevant data An analysis will b e attempted in section
Diminutives The diminutive form of nouns is derived byadd
ingasu x k tj sk to the stem the feature content of the con
sonant dep ends on the stem nal segment Long vowels shorten in the
pro cess
a baos b oske b oss b qs b sk
b straot str otje street str qt str tj
c kn e en kn entj rabbit kn qn kn ntj
d b een bintje leg b qn b ntj
e b o ok b okske b elly byqk byksk
f steel stiltje shaft stel st ltj st ltj
g haand h endje hand hant h ntj
As can b e seen in g Tilburg Dutch also shows some traces of an umlaut
pro cess in diminutives The pro cess nowadays seems to b e highly morphol
ogized in this dialect and applied to a restricted set of stems only I will
not discuss the prop erties of the umlaut pro cess here cf Nijen Twilhaar
Wetzels and Bennink in prep and section of this
A dialect with real length Tilburg Dutch
dissertation for discussion of Dutch umlaut in dialects where it is more
pro ductive and also of the k tj ks alternations
If we concentrate on the non umlauting forms we see that long lax
vowels shorten to their short lax variants a e More interesting is
the b ehaviour of the tense vowel e in f The fact that this vowel
changes to in the diminutive probably has to b e seen as an instance of
the shortening pro cess that applies to the lax vowels Because tense vowels
in closed syllables do not exist the vowel changes to a lax version at the
same place of articulation Two p ossibilitie s are logically p ossible and
Wewould exp ect the to o ccur nowifwe assumed b oth e and
are mid vowels whereas is high Yet the ischosen here not the
This once again shows that is the mid lax vowel and islow
This raises the question what happ ens to the other tense vowels if they
o ccur in a shortening environment Consider the following examples
a lies liesje oat grass lis lisj
b tuut tuutje tire tyt tytj
c boek boekske b o ok buk buksk
d deuk d okske dent d k d ksk
e sloop slopke pillow case slop sl pk
The mid vowels all change to a lax variant changes to y and o to
which gives us an argument to consider also these vowels as mid and
and as low
The high tense vowels do not shorten at all It may seem therefore that
these vowels count as short in the system
Comparatives and sup erlatives Wenow turn our attention
to the comparative forms We do not nd shortening uniformly in the
comparative paradigm As a matter of fact we only nd shortening in
those cases where an ep enthetic d is inserted b etween the stem and the
comparativesu x er r This happ ens when the stem ends in l r or
n The same applies to Standard Dutch but only when the stem ends in
r
a sch oon sch onder b eautiful sx qn sx nd r
b f e en f ender ne f qn f nd r
c zwaor zworder heavy zw qr zw rd r
d geel gilder yellow el ld r
e gr oot gr otter big r qt r t r
A dialect with real length Tilburg Dutch
f h oog h ogger high h q h r
The forms e and f are exceptional They are the only forms with
shortening but without d ep enthesis It is very di cult to nd adjectives
ending in tense vowel fl r ng but from the few forms wehave wecan
conclude that the comparatives b ehavelike the diminutives in this resp ect
Shortening o ccurs also in sup erlatives The sup erlative su x is st and we
nd o ccasional forms like schonst and fenst The pro cess do es not seem to
b e pro ductive The o ccurrence of shortening dep ends on the frequency of
the adjective The more frequent the adjective the more likely shortening
b ecomes
Ordinal numbers The next forms to consider are the ordinal
numb ers Only the rst ten numb ers need to b e considered b ecause all
other numb ers either end in tiende which is already in the list or end in
a su x that is irrelevant b ecause its nal vowel is already short or schwa
tig ty honderd hundred duzend thousand etc
a een irste one qn rste
b twee twidde two twe tw d
c drie driede three dri drid
d vier vierde four viqr viqrd
e v e ef v efde ve v qf v fd
f zes zesde six z s z sd
g zeve zevende seven zev zev nd
h aacht aachtste eight axt axtst
i nege negende nine ne ne nd
j tien tiende ten tin tind
Only the rst ordinal is formed in a somewhat irregular way in Tilburg but
all other forms are derived byadding de to the cardinal numb er Although
eavery limited set of examples here vowel shortening of course wehav
b ehaves exactly as wehave come to exp ect the long lax vowel b ecomes
short the tense e b ecomes in twidde the high and low tense vowels
are not a ected by shortening and neither is e if it is protected by a nal
schwa
A dialect with real length Tilburg Dutch
Verba pr sentia sg Apart from the diminutives verbal
in ection o ers the most pro ductive paradigm of vowel shortening Below
Igive the in nitive and the p erson singular form of a selected number of
verbs and the complete paradigm of the present tense of eten eat
a o ote ot whistle yqt t
b k e eke k ekt lo ok k qk k kt
c slaope sl opt sleep sl qp sl pt
d k oope kopt buy k qp k pt
e ete it eat et t
f koome komt come kom k mt
g speule spult play sp l spylt
h begiene begient b egin b in b int
i zuuke zuukt lo ok for zyk zykt
j roetse roetst run ruts rutst
k verbraande verbraant burn v rbra nd v rbra nt
singular plural
k et w et
t g t
h t z et
According to Boutkan wealways nd shortening b efore fk p t l m
n g the Tilburg dialect has Final Devoicing We nd it incidentally b efore
the fricatives ff g rg r is called a fricativeby Boutkan b ecause of its
uvular pronunciation in the Tilburg dialect I have no real explanation for
this di erence b etween fricatives and other segments It might b e that again
wehave a re ex of tensing b efore fricatives also observed in Philadelphia
English and for axt clusters in Tilburg ab ove This tensing tendency would
sometimes b e stronger than shortening Below I will ignore the cases b efore
the fricatives and r and concentrate on the cases where shortening do es
o ccur
Other categories In the previous subsections wehave seen
that long lax vowels and mid tense vowels shorten in in ected forms if we
take in ection to b e a rather broadly de ned category including diminu
tives if the in ectional su x starts with a consonant There are a few
residual categories to consider where we also nd shortening pro cesses
First of all there is a class of lexicalised comp ounds as I will provi
sionally call them These are comp ounds whichhave acquired a sp ecialize d
A dialect with real length Tilburg Dutch
meaning and therefore might b e analysed as constituting a single entry in
the lexicon
a h o os h osbaos house landlord hyqs hysb qs
b ger eet geritschap ready instruments r qt r tsx p
c spaore sp orsp ot save money b ox sp qr sp rsp t
d t o on t onman garden gardener tyqn tynm n
With newly formed comp ounds we do not normally nd shortening Com
pare for instance d to the relatively new form t o onsentrum garden cen
tre which is probably not lexicalized and which is pronounced as tyqns ntrum
and not tyns ntrum On the other hand it is not crystal clear what ex
actly counts as a lexicalise d comp ound For instance while we do nd
shortening in erbezie strawb erry rb ezi lit earth b erry wedonot
nd it in e erpel p otato qrp l litt earth apple It is unclear what
constitutes the essential di erences b etween these two examples
Another category where we nd incidental shortening are prep ositions
and pre xes like the following
a neeve n e e next to nev n f
b aon on on qn n
c o ot ot out yqt yt
It is however very unclear what the p ossible conditions could b e on the
shortening pro ces applying to these forms b oth in derived words and in
prep ositional phrases It is p ossible that there is a relation to sentence
stress Furthermore shortening seems to b e optional in most cases For
this reason I assume these pre xes prep ositions to havetwo lexical forms
one with a long vowel and one with a short one
Finally Iwanttomention the form k e end k ender child children
nd r The er plural su x is the only su x that do es not start k qntk
with a consonant but nevertheless triggers shortening in k ender On the
other hand er is not the pro ductive morpheme for expressing the plural As
a matter of fact it only o ccurs on one or two forms aajer eggs Again I
assume that in this case the plural form is lexicalise d There is no shortening
b ecause the short lax vowel is underlying
The v f alternation in this form probably is connected to a lter excluding V CV
morphemes where V is a lax vowel and C v or z see p
A dialect with real length Tilburg Dutch
Analysis In this section I will showhow Tilburg ts into the
theory established earlier in this chapter Four problems will havetobe
dealt with In the rst place given the p ossibilityofoverlong syllables why
do es shortening o ccur at all In the second place whydoshorttensevowels
shorten along with the long lax ones Thirdly how should we classify the
di erentvowels in a su ciently constrained feature theory And nally
do wehave a case of real underlying length here These questions will b e
dealt with one by one b elow
Shortening in derived contexts Wehave seen that long lax
and tense vowels shorten if the syllable in which they o ccur gets overlong
b ecause of some morphological pro cess For most su xes this is not prob
lematic b ecause they consitute syllables If we assume a su x syllable
gets incorp orated in the proso dic word of the stem the stem nal degener
ate syllable loses its p eripheral p osition after su xation As wehave seen
ab ove degenerate syllables can only o ccur at the end of a word
The verbal su x t causes more problems Take for instance the verb
slaope sleep sl qp The stem of this form sl qp consists of a sylla
ble sl q plus a degenerate syllable p If we attach a t this cannot b e
incorp orated into the proso dic structure of the base I leaveouttheless
imp ortan t details in the following tree
N N
N N
s l p t
For reasons discussed ab ove there can only b e one degenerate syllable p er
word the t cannot even get its own degenerate syllable
If we shorten the vowel the p can b e incorp orated into the rhyme
and the t can get its own degenerate syllable
N N
N N
s l p t
The problem therefore is rather with Standard Dutch This reasoning seems to
indicate that in ectional su xes seem to have some clitic likestatusinthislanguage
from a phonological p oint of view
A dialect with real length Tilburg Dutch
This notion of an overlong syllable turns out to b e not very easy to for
malise however First of all we nd shortening also in forms like eten it
to eat It is not very clear why et would b e to o long Even if e do es not
allow consonants in the rhyme b ecause of Connect N lax we could still
make t extrasyllabic or extraproso dic It is di cult to see what would b e
wrong with this structure
Given that we nd lexical words in Tilburg like kaort map k qrt
with an extrasyllabic r and extraproso dic t as discussed for Dutchin
the previous section section it is even questionable that shortening
should have a p ositive e ect in sl opt you sleep sl pt from slaopen
sl qp
It seems as if the nal t of a stem can b e extraproso dic while this
option is not op en if the t formsasu xofitsown in Tilburg We
will also havetotakeinto account that Standard Dutchdoesnotmakea
di erence b etween derived and underived forms in this resp ect the second
p erson singular form of to sleep is slaap t pronounced as slapt with
tense long a
The standard pro cedure of accounting for this typ e of variation in
constraint based phonology is to assume there is a constraintwhichhas
su cient strength in one dialect but is less strong in the other I prop ose
the relevant constraint in this case is one whichsays that every morpheme
should b e parsed into proso dic structure at least partially
MorPa At least one element the most sonorous element of a
morpheme is incorp orated into a proso dic word
We will return to this constraint several times in the chapters that follow
I prop ose MorPa is ranked rather high in the Braban t dialects while
it has sunk to irrelevance in Standard Dutch In particular in Brabant
the constraintisranked higher than the constraintonvowel parsing In
the following brackets indicate Proso dic Word b oundaries since Standard
Dutchdoesnothavevowel length I assumed the relevantinteracting con
straint here is the one against insertion of lax in Tilburg Dutch it is the
constraint Parse V requiring cons ro ots to surface
Tilburg Dutch MorPa Parse V lax
a slaop t sl qp t
The pro cess is reminiscentofthevarious vowel shortening pro cesses in English which
also apply in derived environments only Myers has argued that all of these
shortening rules are instances of one pro cess repairing overlong syllables See Prince
Borowsky Yip for other fairly recent treatments of the English
phenomena
A dialect with real length Tilburg Dutch
Candidates MorPa Parse V
p
sl q p t
sl q p t
b kaort k qrt
Candidates MorPa Parse V
k q r t
p
k q r t
c kneupt kn p t
Candidates MorPa lax
kn p t
kn p t
d werk t
Candidates MorPa Parse V
p
w r k t
Standard Dutch lax MorPa
a slaap t
Candidates MorPa lax
sl p t
p
sla p t
b kaart
Candidates MorPa lax
k r t
p
ka r t
c werk t
Candidates MorPa Parse Vowel
p
w r k t
In this case wemight assume that the Tilburg Dutch ranking is unmarked
re ecting the initial state of Universal Grammar The Standard Dutch
child learns to rerank the constraint lax once it observes that the vowel
in a morphologically complex form like slaapt is not laxed contrary to exp ectation
A dialect with real length Tilburg Dutch
The question still remains what is the explanation for the di erence
between the rst p erson singular st qt and the third p erson singular st t
of the verb st oote to push st qt Why is the t allowed in one form to
remain outside of the proso dic word and why should it incorp orate in the
other form
Zonneveld has argued that Dutch has an empty theme vowel in
the case of the rst p erson singular This vowel could in Tilburg Dutchallow
the t to incorp orate into an onset in this form The same is not p ossible in
the third p erson singular b ecause that form do es not have a theme vowel
This analysis cannot b e adopted in the approach suggested here b ecause it
is unclear what the nature of the empty theme vowel would b e As wewill
see in the next chapter an emptyvowel ro ot no de is interpreted as schwa
and it is imp ossible to b e more empty than an emptyroot
Fortunately it is not necessary to assume this typ e of emptynodeeither
since the facts already follow from the assumptions made plus the fact that
in the case of the third and second p erson singular a su x t is added
a st oot st p erson
Candidates MorPa Parse Vowel
st q t
p
st q t
b st ot t nd dperson
Candidates Geminate MorPa Parse Vowel
st q t t
p
st q t t
st q t t
st q t t
st q t t
I added a constraint against geminate consonants in b Tilburg Dutch
do es not have geminate consonants under any circumstances and certainly
not at the end of the word so wemay safely assume this constraint has a
very high ranking
The b ehaviour of tense vowels Iwillnow discuss the b e
haviour of tense vowels in shortening pro cesses Given the nature of these
vowel the term shortening is of course rather infelicitous but I will nev
ertheless stick to it as a descriptive term Iwillnow rst explain the
b ehaviour of the mid vowels and then I will discuss the high vowels
A dialect with real length Tilburg Dutch
As wehave seen the mid vowels e and o change to y and
resp ectivel y The reason why they b ecome lax should bynowbeobvious
In the form neem t takes the t is forced into the proso dic word and
therefore the m is forced into the co da of the ne syllable This syllable
b ecomes heavy and b ecause of Connect N lax its head should b e lax
Boutkan o ers a diachronic explanation for some of the facts
discussed here The alternation b etween e which is the most common
alternation b etween mid tense and high lax vowels according to Boutkan is
explained by reference to Old Dutch According to Van Helten the
Wachtendonck Psalms a translation of the psalms which at presentisour
main source for Old Dutch grammar named after a th century owner of
the only copywehave had the following verbal paradigm for the present
tense
Verbal paradigm in the Wachtendonck Psalms
singular plural
on un
is it et
it unt ont int
Furthermore the Proto Germanic rule whichchanged an e to i b efore
a syllable containing i or j was still op erative in Old Dutch Wecan
therefore construct the following paradigm for gevan give in the language
of the Wachtendonck Psalms
e i i j
Paradigm of gevan in the Wachtendonck Psalms
singular plural
gevon gevun
givis givit
givit gevunt
Although the i of the su x disapp eared in later stages of Dutch some
dialects like Tilburg retained the i in the stem The alternation has
later b een generalized to a rule raising all tense vowels in the di erent
p ersons
Regardless of the value of this as a diachronic explanation we cannot
explain the synchronic facts of Tilburg Dutchthisway Howdoesthechild
which presumably do es not knowtheWachtendo ck Psalms acquire the pro
ductive pro cess of e alternation in the di erent in ectional paradigms
We nd the same alternation in some Standard German verbs as well geben gibt
give nehmen nimmt take etc
A dialect with real length Tilburg Dutch
Furthermore the alternation in present day Tilburg can also b e found in
lexical comp ounds even if no i is present in the second stem at all Also
synchronically the pro cess is no longer restricte d to the e pair which
is another unexplained fact under the diachronic analysis Finally itisnot
clear why the raising pro cess should b e restricted to tense vowels
It seems more reasonable to assume that simply is the lax variantof
e y of and of o The upshot of this is that and are low
vowels Before I will delveinto the consequences of this nding I prop ose
to rst take a closer lo ok at the high vowels which nishes our study of the
e ects of shortening on all classes of vowels
As wehave seen i y u do not alternate in shortening environments
a h ebeg i nt he starts
b h e st y rt he steers
c h e vl u kt he curses
The reason for this is that the constraint against lax high is inviolable in
Tilburg as it is in Standard Dutch On the other hand MorPa still forces
the stem nal segmentinto the co da The constraint that gets violated is
Connect N lax
b egient
lax
Candidates MorPa Connect N lax
high
b i n t
b n t
p
b in t
The system should of course work in the same way for words with high
vowels other than y and u
The Tilburg vowel system Wehavenow established the fol
lowing vowel system for Tilburg Dutch disregarding length for a moment
ATR ATR
high i y u
mid e o y
low
Ihave to admit that there is one verb mentioned by Boutkan fr eete to eat abun
dantly fr qt of which the second and third p erson forms is fr t although the form
fr t is also found This seems to b e an isolated case however which therefore in my
view should b e treated as an exception mayb e on a par with umlaut cases suchas
stad steden cities
A dialect with real length Tilburg Dutch
I assume features can combine freely but with resp ect to ap erture features
the following two output conditions apply
lax
a
high
b low lax
Wehave seen that b oth constraints are well motivated phonetically A
problem is that wenowhaveto t a in the system These vowels
are standardly assumed to b e low Phonetically they are certainly lower
than If weintro duce a fourth vowel height distinction we get the
strange situation that b holds for mid lowvowels but not for really low
ones This would weaken the phonetic motivation for such a constraintto
a large extent
I therefore prop ose to distinguish the a pair from the other vowels
not by an ap erture feature but by a place feature These are the only
vowels that are back but not rounded Adding place features to we
get the following picture
ATR
cor dor s
cor dors
lab lab
high i y u
mid e o a
low a
ATR
cor dor s
cor dors
lab lab
high
mid y
low
The fact that there are no high unrounded dorsal vowels p ossibly has the
phonetic motivation that dorsal vowels can only b e pro duced by op ening
the mouth to a considerable degree or that raising the back of the tongue
involves tongue ro ot retraction The question whether is mid or lowis
more di cult to answer Mayb e it is most reasonable to saythatitislow
b ecause we do not want to p ostulate a constraint against the dorsal low
combination b ecause that combination seems phonetically so desirable and
therefore a constraint banning it is unlikely to haveuniversal status If is
lowwe only have to p osit which is not an unnatural constraint Stan
dard Dutch has basically the same system b e it that it has an additional
A dialect with real length Tilburg Dutch
high ranking constraint against labial low onlyoverridden in the case
of diphthongs
dorsal low
If this constraintisranked b elow the constraint banning labial low com
binations we get the right result
Inowturntothevowel length parameter First I have already ex
plained whyonlylaxvowels can b e long I assume that the parts of a
long vowel can only b e tautosyllabic If this is true long vowels always
are in heavy syllables therefore by Connect N lax they should b e lax
The only tense vowels that could p ossibly b e lengthened without laxing
dep ending a little bit on the ranking of constraints are the high vowels
b ecause they have an indep endent constraintpreventing them from laxing
Indeed it is the case that a few English and French loan words can b e
found whichhave these long high vowels Smith et al team id
tiqm centrifuge dryer s ntrifyzj etc There are absolutely no such
words with non high tense vowels
In Standard Dutch three vowels are lax and long These are the vowels
in the th column of the table in on page whichIhave neglected
until now I prop ose to analyse these as q q and q resp ectivel y The
latter ones are normally analysed as yq and q b ecause it is assumed that
long vowels havetobevariants of existing short vowels In a phonologi
cal theory of interacting and comp eting constraints this assumption is no
longer necessary The representations q and q seem closer to the pho
netic reality I assume there is a constraintwhichsays that doubly linked
vo calic no des have to b e supp orted by low long vowels are lowered We
have already seen this constraintabove namely in the discussion of diph
thongs see page This constraint outranks the constraint against low
labial vowels which as I claimed ab ove is activeontheDutch short vowels
Standard Dutch therefore has three truly long vowels All three of these
are lax b ecause they are in a heavy syllable All three of them are low
b ecause of Project N low
Long vowels in other Brabant dialects Although most Bra
bant dialects have a morphologically conditioned pro cess of vowel short
Typ ologically I make the prediction that length either is found in b oth tense and
lax vowels when Connect N lax is ranked very low or it only o ccurs in the lax vowels
but not in the tense vowels with the p ossible exception of the high vowels the Brabant
situation It is predicted that no language can o ccur where only tense vowels havea
length distinction but no lax vowels Although this seems to b e in disagreementwith
the implicit assumptions of most phonologists I have not b een able to nd data that
shed light on the issue
A dialect with real length Tilburg Dutch
ening and also a much larger set of long vowels than Standard Dutch the
actual structure and content of the vowel set varies considerably Inthis
section I will brie y discuss two other Brabant systems Hofstade Dutch
Keymeulen and Taeldeman sp oken in the extreme south of the Bra
bant area and Antwerp Dutch Nuyts sp oken in central Brabant
In the latter dialect all long vowels are lax and all tense vowels are
short This is the exact opp osite situation of what is normally assumed to
b e the case in Standard Dutch Wethus nd words like the following all
examples have b een copied from Nuyts
i stipt prompt q qr scream
y dyrf dare q v qr re
u buk book Uq vUqt fo ot
q b qk bro ok e sp el toy
a kat cat q m q sleeve
The right generalisation can b e obtained if we assume that i Connect N
lax do es not visibly play a role in Antwerp Dutch and ii Antwerp Dutch
hasavariantof Connect N lax with lax in stead of low
Connect N lax Antwerp Dutch Vo calic features are at
tached to more than one ro ot they are supp orted by the feature
lax
In Hofstade Dutch the parameters tenseness and length are indep endent
In this dialect we nd a complete vowel system All tense and lax vowels
have b oth long and short variants In all there are therefore di erent
vowels in Hofstade some of which are listed b elow
i wit white iq wiql wheel
yq sxyqr barn y mys mouse
e b elt image eq veql much
m skirt q w qt wide
a kat cat q m q sleeve
As in Antwerp Dutch the constraint Connect N lax do es not playa
visible role in Hofstade The same is true for the constraint Project N
under any of its guises in this dialect however All these constraints are
ranked b elowFaithfulness Every combination of length and tenseness may
surface as they are sp eci ed regardless of syllabi cation with one exception
long vowels cannot o ccur in syllables that are also closed by a consonant
This however presumably follows from the general structure of rhymes in
whichatmosttwo p ositions are available not from any violable constraint
A dialect with real length Tilburg Dutch
The limited distribution of long vowels Trommelen and
Zonneveld haveshown that the loan phonemes q q and q
in Dutchhavea very limited distribution They only o ccur in either the
absolutely word nal sup erheavy syllable or in the p enultimate when
the nal syllable is headed byschwa q isvery rare
g ene embarassment z qn cr eme cream kr qm sc ene scene
s qn
freule noble lady fr ql
zone id z qn record id r k qr controle control k ntr ql
Trommelen and Zonneveld note that when a full vowel su xis
added to these forms the vowel shifts in qualitytowards the nearest by
vowel tense in an op en syllable lax in a closed syllable
a g ene embarassment z qn g en ant embarassing zen nt
or with reduction z n nt
b expert id ksp qr expert ise id ksp rt se
c controle control k ntr ql control eer to check k ntrol qr
d zone id z qn zon aal zonal zon al
In Trommelen and Zonneveld s analysis this distribution should b e ac
counted for by the following rule
foreign foreign
The loan phonemes of Dutchthus have a feature foreign under Trom
melen and Zonneveld s analysis this feature can incidentally b e replaced
by low under the analysis presented here The non lowversion of q will
have to shorten to e b ecause of the constraint low lax in combination
N lax The latter constraint forces the short vowel in an with Connect
op en syllable to b ecome tense and the former requires that then it also has
to lo ose its feature low The result is a frontvowel without lax or low
i e the tense front mid vowel
The counterpart of in a constraint based theory would b e a con
straint against the feature low on a vowel when followed by another full
vowel
Another way to lo ok at this fact however is the following In all the
examples given here the long vowels have primary word stress in the un
derived form After derivation stress shifts to the su x Therefore if we
Trommelen and Zonneveld adopt an analysis under whichschwadoesnot
head its own syllable I refer to the next chapter for details on this typ e of analysis
A dialect with real length Tilburg Dutch
assume that long vowels in Dutchmay only o ccur under primary stress the
alternations in also follow
Wenow exp ect long vowels also to show up b efore full vowels Indeed
a few cases are known from the literature although their number is admit
tedly small and limited to instances of long q
prairie id pr qri aero dynamica id qro b eta id b qta
In Tilburg Dutch the number of suchwords is somewhat larger as is ex
p ected b ecause long vowels in general are more frequent in this dialect In
all cases the long vowel gets primary stress
e uriejaant b ossy p erson qrijant
rinnewaosie damage r n w qsi
m ooniekaa concertina m qnika
p e eresol parasol p qr s l
sjaokies quiet s kis
Conclusion In this section I haveshown that the vowel system
of the Brabant dialect Tilburg Dutch can b e explained under the assump
tions made for Standard Dutch in earlier sections The assumptions re
garding syllable structure and extraproso dicity together with a stipulation
that all morphemes should at least partly b e incorp orated into the proso dic
word have b een used to explain the distribution of vowel length and mor
phological shortening pro cesses Even though the Tilburg system contains
six non dorsal lax and six non dorsal tense vowels I have argued these
two systems are not symmetric Tense vowels are high or mid lax vowels
are mid or low The dorsal vowels are set apart altogether They do not
show a distinction b etween long lax on the one hand and short on the other
and they similarly do not showanyheight distinction The argumenthas
b een carried over to Standard Dutch which is presumed to havetwoad
ditional comp eting constraints one which prohibits low labial lax vowels
and one which requires that all long vowels are low The fact that there
A problematic case for this analysis is the pair hygiene id hi ij qn hy
gien isch id hi ij e nis b ecause primary stress in this case is still on the former
long vowel Yet the su x isch shows a somewhat exceptional b ehaviour in the stress
system normally wewould exp ect it to b e stressed I have no sp eci c analysis for this
form but the exceptional stress b ehaviour might help to explain it Another explanation
Harry van der Hulst p c mightbethat isch more generally seems to require that
vowels immediately preceding it are tense witness pairs like alcohol lkoh l alcoholish
alcoholic lkoholis
According to Trommelen and Zonneveld the long q b efore r is the result
of r Colouring applied to i and words like b eta form a sp ecial stratum in the lexicon
Derivation of the Dutchvowel system
is an asymmetry b etween lax and high vowels on the one hand and non
high tense vowels on the other in that the rst class can o ccur in closed
syllables whereas the second cannot has b een argued to follow from the
general mechanisms of the system in particular of Connect N lax
Derivation of the Dutchvowel system
In this thesis I assume the theory of sp eci cation and undersp eci c ation
of segments in which in principle anycombination of features or feature
values can b e underlying Illicit feature combinations will b e ruled out by
grounded constraints like the ones summarized in the previous subsec
tion and repaired by deletion rules Features not present in underlying
representation will similarly b e lled in automatically This means no con
straints on underlying representations are needed as in other theories of
undersp eci cati on
For Dutch I assumed the following vo calic features coronal labial
dorsal low high and lax For the sake of simplicity I assume all these
features are monovalent but the following discussion would not haveto
change dramatically if they were not It would just have to b e complicated
Free combination of these features gives us di erent p ossible
inputs represented in the following table H high L low R RTR lax
c coronal l labial v dorsal a plus indicates that a feature is present a
blank that it is absent
H
L
R
c
l
v
H
L
R
c
l
v
H
L
R
c
l
v
H
L
R
c
l
v
The vowel in row has no features at all I will argue in the next
Derivation of the Dutchvowel system
chapter this is the schwa The vowels in the rows through only have
ap erture features no place features These would b e high lowandRTR
variants of schwa None of these o ccur in Dutch We need therefore to have
constraints to the e ect that ap erture features can only surface if they are
supp orted by place features The following is the structure of the vo calic
no de
vo calic
V place Ap erture
coronal labial dorsal high low lax
Aschwa do es not havea vo calic no de at all Some other vowels the tense
mid vowels only have place features but there are no vowels with only
ap erture features and no place features in the output i e high or lowmid
vowels these would b e the surface corresp ondents of row in the table
I assume this is the result of several constraints which require ap erture
features to b e accompanied by sp eci c place features for instance
a lax dorsal low lax
b high labial
c etc
These constraints should b e ranked higher than the constraints against
the insertion of dorsal coronal and labial In that case every ap erture
feature will always b e accompanied by a place feature and schwa will b e
the only p ossible vowel without a V place no de
Rows through represent the coronal vowels is e i and
will b e lled out with lax b ecause of the following constraint
ordering
a low lax
b lax
Row is Rows and have illegal feature combinations high
el y They will b oth surface as the diphthong lax and high low resp ectiv
i
Constraints like this express the idea that all vowels need a place feature when they
have an ap erture feature I will not go into the question to what extent such constraints are grounded
Derivation of the Dutchvowel system
Rows through represent labial vowels without dorsal sp eci cation
However all of these vowels except the ones that have a coronal sp eci
cation will acquire a sp eci cation dorsal I will assume this is the result
of the following two constraints
dorsal
a
coronal
b labial dorsal
The constraint against a combination of dorsal and coronal seems as
universally high ranked as the constraint against low high
These feature combinations surface then as the back rounded or front
rounded vowels Con icting ap erture features cause diphthongization
The rows in through all represent dorsal vowels Combinations
through and through do not surface b ecause they contain
con icting place sp eci cations Furthermore the dorsal vowels that have
a high marking will get labial sp eci cation and end up as u by the
interaction of the following constraints
a high labial
dorsal
b high coronal
coronal
The rows through also representback labial vowels Finally the
forms b etween and that do not have con icting ap erture features are
p ossible inputs yielding a and
The main di erence b etween Standard Dutch and Tilburg Dutchisthat
in the former dialect a constraint against the combination of labial and
low is active which has no e ect in the latter Another di erence is that
doubly linked vowels in Dutch always get laxed b ecause of Connect N
lax and lowered b ecause of Project N low whereas in Tilburg
Dutch they only get laxed
In all we can derive the Dutchvowel system by the following four blo cks
of constraints within each blo ck it is not p ossible to establish anyinternal
ranking
dorsal high low
a low lax
coronal low labial
This constraint has to b e ranked under labial coronal to make sure that underlying
i s do not surface as y
Because of their low ranking most constraints in the last blo ck are actually irrelevant
for the description of Dutch I added them b ecause their e ect can sometimes b e seen in other languages
Conclusion
b high labial labial dorsal lax dorsal high coronal
Parse dorsal Parse coronal Parse low Parse labial
Parse high Parse lax
c lax coronal low dorsal labial
d lax low dorsal labial dorsal lax
coronal high labial high etc
I refer to the top ology of the Dutch lexical phonology in app endix C
Conclusion
In this chapter I have discussed real vowel length as opp osed to the
phonetic length which is a result of tense vowels ending up in op en syllables
thatwe nd in Dutch and in BrabantDutch dialects It is quite inter
esting that long vowels and diphthongs always have a lax head regardless
whether they are underlying or the result of r colouring or diphthongisation
This fact is unexp ected from the p oint of view of a length based theory but
it is exactly what a lax based pro jection theory predicts Because in ad
dition the head of long vowels and diphthongs in Standar d Dutchislow I
prop osed to follow the tradition of letting the twohalves of a long vowel b e
dominated by the nucleus allowing me to formulate a pro jection constraint
on branching nuclei Project N low A similar constraint could also b e
used for Antwerp Dutch where all long vowels are lax and all short vowels
tense
Wehave also reached some results that are not directly related to the
theory of phonological pro jection For instance I haveshown howambi
syllabicity extrasyllabicity and catalexis can b e expressed in the theory of
proso dic structure used here I have also argued that lax is b est seen as
a tongue ro ot feature to b e group ed together with the ap erture features
high and low Finally the analysis of Tilburg vowel shortening gave evi
dence for a constraint MorPa to the e ect that all morphemes should at
least partly b e incorp orated into the proso dic structure All of these results
will turn out to b e quite useful for us in the chapters that follow
Derived schwainDutch
Intro duction
Schwa probably is the most well studied vowel of Mo dern Dutch Its ex
ceptional b ehaviour in syllable structure its stressles sne ss and its use as
a reduced and ep enthetic vowel have inspired phonologists to a varietyof
theories ab out this mysterious segment
In my view schwaprovides us with convincing evidence for the hyp oth
esis that vo calic feature comp osition and syllable structure are strongly
related Schwa has only one feature cons it therefore is a very defective
typ e of vowel and it heads a very defectivetyp e of syllable In terms of
the pro jection constraints there is no feature that allows schwa to o ccur in
the head p osition of anytyp e of proso dic constituent except for the ma jor
class features cons and p ossibly son itself I will show these two
facts are connected and that we can derive the latter from the former
In the next twochapters I develop a theory of Dutchschwathatiscom
patible with the general framework of this thesis At the outset it is useful
to make clear that I distinguish b etween three typ es of schwainDutch
This distinction is pretheoretical in the sense that we can empirically dis
tinguish b etween these typ es without prejudice as to whether one or more
of these typ es should have some sp ecial theoretical status indep endentfrom
the others
R schwa Stress less vowels alternate with schwainDutch as they do in
English However this alternation is much more restricted in the
former language than in the latter The restrictions are manifold
Some vowels never alternate at all some p ositions in the word do
not allow for reduction etc Also even if a vowel can reduce this
reduction is never obligatory in Dutch The more formal the style of
sp eech the less vowel reduction we nd I will therefore takestyle
dep endent alternation with a full vowel as a litmus test for r schwa
r for reduction
Intro duction
E schwa Words ending in a consonant cluster i e a closed syllable fol
lowed by a degenerate syllable can get an optional schwaintervening
between the two consonants Here I take the existence of a variant
without schwa as the deciding test for e schwa e for ep enthetic
R schwas are never e schwas and vice versa
Ihave to note however that the variants without schwaofwords like
melk milk Marc name hulp help etc sound overly formal to
manynativespeakers of Dutch including the author of this thesis
U schwa This is empirically a rest category If a schwa is neither an
u for underlying r schwa nor an e schwa I call it an u schwa
U schwa do es not alternate
The reason why I coined the somewhat abstract terms e schwa r schwa
and u schwa is that not every author agrees on the analysis that is im
plicit in terms like epenthetic schwa and underlying schwa For instance
some authors have argued that also u schwas are the result of ep enthesis
and or reduction Even those authors have to explain however why in some
contexts reduction and ep enthesis are obligatory while in others they are
optional I will discuss some previous theories on Dutchschwa in section
in the next chapter However I am not aware of any study trying to
approach all instances of schwa at the same time
Interestingly the three di erent instances of schwahave some phono
logical prop erties in common which set them apart from the rest of the
vowel system Most prominent among these prop erties no doubt is the im
p ossibility for either of these vowels to b ear primary stress but I will show
that there are other prop erties that put the schwas apart as well
On the other hand the three schwas also di er from one another in
signi cantwa ys For instance underlying schwa and ep enthetic schwa seem
to b e in complementary distribution
The fact that all schwas are clearly di erent from other vowels and the
fact that they slightly di er among themselves are the main issues of the
next twochapters This chapter will deal with r schwaande schwa the
next chapter will deal exclusively with u schwa
I will try to make the rst fact follow from the observation that schwais
a feature less vowel It is almost absolutely central to the vowel triangle
This is not true if wetakeinto consideration some very informal styles of sp eech
where vowels which can b e reduced in somewhat less informal styles are deleted For
instance the rst a in banaan banana can b e reduced to in some styles b nan
but in very informal styles of sp eech it can b e deleted altogether bnan From this
point of view the schwa is b oth an r schwa b ecause it alternates with a full vowel and
an e schwa b ecause it alternates with null
See however in the next chapter for a weakening of this rm statement
Intro duction
so it do es not have its own place or ap erture sp eci cations I will assume
the following prop ertyofschwa to b e the most central one from which
most of the others can b e derived
Prop erty Schwa is exceptional in the vocalic system because it bears no
phonetic features
The informal observation of prop erty gets its empirical underpinning in
Ko opmans van Beinum Her phonetic research has shown that
schwa sounds in natural continuous speech areconsiderably shorter than
the other short vowels no strong consonantal in uence exists on schwa du
ration schwa sounds display a spectral spread larger than any other vowel
and surrounding consonants seem to play a role with respect to the mid
point formant distribution of the schwa within the whole vowel system
Ko opmans van Beinum cf also Browman and Goldstein
on targetless schwa in English Esp ecially the latter two criteria seem to
me to b e esp ecially relevant indications that schwa do es not haveitsown
place sp eci cations
It is attractivetohave this prop erty as the most basic prop ertyofschw a
from the p oint of view of language acquisition Prop erty is a phonetic fact
that is op en to direct observation Most of the prop erties to b e discussed
b elow are more abstract It seems therefore more plausible to assume the
child observes the featureless nature of schwa and derives its other prop er
ties from this than to takeany of the other prop erties which are much more
dep endent on other more abstract prop erties of the Dutch phonology as
we will see as basic
I will therefore make this prop erty the starting p oint of the analysis for
all three o ccurrences of schwa From this single basic prop erty I will derive
other prop erties whichIhave found to hold for Dutchschwa The reader
can nd a complete list of these prop erties at the end of the next chapter
on page
Schwa do es not b ear any of the vo calic features high low lax
coronal labial or dorsal Jakobson Anderson I as
V V V
sume all these vo calic features are monovalent or otherwise only one value is
relevant for the constraints stated b elow In the feature geometry adopted
here prop erty would b e stated as follows
Schwa is a ro ot no de with no feature material dep endentonit
In many dialects schwa is p erceptually quite close to the mid front rounded vowel
y In other dialects it comes closer to e g in Rotterdam Dutch or e g in The
Hague Dutch I assume that these are matters of phonetic implementation
Intro duction
Following McCarthy I assume that the ma jor class features are re
alized on the ro ot no de directly An empty ro ot no de still has the ma jor
class features on it
This means we can still sp ecify the schwaasanelement b earing ma jor
class features Given these assumptions Dutchschwa gets the following
representation
cons
son
This is a maximally undersp eci e d segment if we assume ma jor class fea
tures are never undersp eci e d Clements If consonantal turns out
to b e non exsistent as has b een argued by Hume and Odden or de
p endent on the ro ot no de as defended byPadgett the ro ot would
consist of only son The interesting prediction wewould make then is
that there are exactly two totally empty segments son and son
b Ihave nevertheless decided not to work out this line of thought To
the contrary I will often use only cons as a representation of schwa
Now the schwabehaves in some ways as a normal vowel b ecause it
has exactly the same sp eci cation of ma jor class no des as other vowels
However in one very imp ortantway it is a defectivevowel it has no vo calic
class no de We will see that manyoftheschwa prop erties listed b elowcan
b e reduced to this defective nature Schwa often cannot o ccur where other
vowels can
Following this line of reasoning we should b e able to draw the same
conclusions ab out schwa in other languages that havethevowel in their
inventory I will therefore discuss the b ehaviour of schwaintwo languages
other than Dutch viz French and Norwegian in chapter
I try to make the di erences b etween e schwa r schwa and u schwa
follow from the di erent phonological levels in which they arise and from the
fact that di erent factors interact with the featurelessnes s of schwaineach
of these cases E schwa and r schwa are derived at later p ostcyclic levels
U schwa is underlying hence already present in the Lexical Phonology
This representation is very similar to the one prop osed bySzpyra and Bethin
for the Polish yer Yet Szpyra assumes that the ro ot no de is completely
empty i e also the ma jor class features are not sp eci ed In other words Polish yer for
Szpyra is Since I assume class no des do not exist without features it is imp ossible
to interpret such a structure under the assumptions defended here
This assumes that b is the consonantal counterpart of schwa In what follows it
will b e shown that this consonant as well as h indeed stands in a sp ecial relation
to the emptyvowel h could b e seen either as a continuantversion or as a sonorant
version of b i e mayb e even the glide of See Keating McCarthy
Pulleyblank Stemb erger Steriade for discussion of the glottal stop
and h as empty consonants
Prop erties of r schwa
The reason why it is imp ortant to distinguish b etween a Lexical and a
Postlexical phonology is not that derived r schwaande schwa sometimes
cannot o ccur where u schwa can This can b e seen as a particular instance of
the emergence of the unmarked McCarthy and Prince we cannot
derive marked con gurations but we can allow them to surface when they
are underlying b ecause deleting the vowel would b e even more costly The
real reason for distinguishing b etween Lexical and Postlexical phonology is
that in some cases u schwa cannot o ccur in environments where r schwa
and e schwa can One schwa is deleted in exactly the same con guration
where another schwaisderived Here I can only see one solution viz to
let u schwa b e present on a phonological level the lexical level where e
schwa and r schwadonotyet o ccur Atthislevel u schwa can then b e
deleted whereas at a later level the o ended constraints are no longer as
strong as they used to b e R schwa and e schwamay therefore arise at
these particular levels
In this chapter I concentrate on the two p ost lexical o ccurrences of
Dutchschwa r schwaande schwa I discuss these b elow one byone Each
time I rst formulate the prop erties of schwa in a more or less theory
indep endentway b efore I givemyown analysis
Prop erties of r schwa
It seems most convenient to start the discussion with r schwa the schwa
that alternates with a full vowel in unstressed p osition and therefore is
commonly analysed as a reduced variant of the full vowel I will supp ort
this analysis As I stated ab ove reduction of schwa is far more restricted in
Dutch than it is in English This is what makes Dutchr schwa particularily
interesting
Restrictions on r schwahave b een discussed in Martin No o
teb o om Bo oij Van Zonneveld
Ko opmans Van Beinum Neyt and Zonneveld Slo otweg
and Wester Kager Visch and Zonneveld Kager and
Trommelen and Zonneveld among others Most of the generative
phonologists in this list have concentrated on the relation b etween stress
p osition and reduction Other factors are sometimes mentioned but not
worked out
One of the reasons for this state of a airs p ossibly is that phonological
theory at present cannot incorp orate all relevant factors in a natural way
On the other hand from Giegerich Harris and Burzio wecan
infer that all conditions on reduction in Dutchmentioned here are also at least partly
active in some English dialects
Prop erties of r schwa
In the rst place reduction is heavily in uenced by extralinguistic factors
suchasword frequency and so ciolinguisti c parameters such as so cial class
Secondly someofthe internal factors are not very well understo o d For
instance reduction seems to b e easier b efore r than b efore other seg
ments I am not aware of any phonological theory that can explain this
centering e ect of r in a satisfactory way Inthischapter I will ignore
all extralinguistic factors except style register Of the linguistic factors
I only discuss those that are reasonably well understo o d and that can get
an explanation in my theory or anyb o dy else s I refer to Martin for
the completest description of facts and factors
Why do es an unstressed vowel reduce to a schwa I assume that ideally
vo calic features have to b e supp orted by stress and vice versa So if a vowel
dominates lax or labial or some other vo calic feature this vowel in a
certain way ideally heads a syllable whichinturnheadsafoot Inversely if
a syllable is stressed it tends to have a full sp eci cation for vo calic features
def
Connect Ft V Project V Ft Project Ft V N
dominates a vo calic no de i N heads a branching fo ot
I argue b elow that in a few cases Dutchdoeshave unary feet b ehaving in
a slightly di erentway
These two pro jection constraints are clearly related to Connect N lax
which I used and defended in the previous chapters Where Connect N
lax states that the feature lax preferably is dominated by the head of a
branching rhyme and that the head of a branching rhyme inversely domi
nates the feature lax these constraint state that i a vo calic no de prefer
ably is in the head of a fo ot and ii the head of a fo ot dominates a vo calic
no de see chapter for extensive discussion of this p oint
The question arises whether wehave to distinguish b etween di erent
vo calic features or to distinguish b etween di erentlevels of stress Wewill
see b elow that we do need to distinguish b etween twotyp es of metrical
p osition one which is in a fo ot but not its head and another whichis
outside all feet This distinction is to b e very imp ortant also for u schwa
Of course other constraints are interacting with Connect Ft V In
some style registers hardly anyvowel is reduced at all This shows that
full vowels can freely o ccur in unstressed p osition in that register I assume
astyle register is a separate grammar Generally it seems to b e the case
across languages that the more formal the style of sp eech the higher ranked
Cf Mo ortgat and Van der Hulst Charette Van der Hulst b
Kenstowicz b among others for other formalisations of similar intuitions
On rhythmic p ostlexical stress in forms like helder der e clear er INFL see section
Prop erties of r schwa
the faithfulness conditions Parse and Fill are I will not pursue this
conjecture here except that I will takeitasmy guideline for the study
of Dutch see section for some discussion In more formal styles of
sp eech the requirement to parse vo calic features gets more weight than
Connect Ft V If for instance an o ccurs in unstressed p osition
it is deemed more imp ortant to parse the feature bundle dorsal low lax
than to satisfy Connect Ft V
Because schwa has no vo calic features at all it is the most ideal vowel in
unstressed p osition according to the pro jection constraint The rst prop
erty of r schwa apart from its lack of features is now trivially explained
Prop erty R schwa is the reduction vowel
Vowel reduction is a consequence of Connect Ft V Since schwa has the
representation in it is the only vowel that do es not violate Connect Ft
V when it o ccurs in unstressed p osition From whichproperty follows
directly
Inowhave set up the basic mechanism for vowel reduction Iwill
pro ceed to discuss some restricti ons on it My main sources for the data
are Martin Bo oij and Kager
The rst two of these restrictions have to do with p eripheral p ositions
in the word R schwa is disallowed b oth absolutely word initially and ab
solutely word nally However for word initial p ositions this can b e shown
to b e related to some principle not directly referring to the p eriphery of
words
Word initial p osition In word initial onsetless syllables vow
els can never reduce not in anystyle of sp eech
elite elite el t l t egaal plain e al al
ovaal oval ov al v al agent p oliceman a nt nt
It can b e shown that the prohibition against schwa in onsetless syllables
is much more general not restricted to the rst syllable of the word It
seems that schwa cannot o ccur in onsetless syllables In addition wewill
see b elo w that r schwaavoids all word p eripheral p ositions This implies
For instance in French the more formal the style of sp eech the more liaison i e
parsing of consonants Encrev e In Turkish the more formal the style of sp eech
the less ep enthesis i e insertion of morphologically unsp onsored material
A pretonic tense vowel can sometimex reduce to a lax vowel esp ecially if this
vowel is a e g banaan banana banan can b e pronouncedas b nan Although this
pro cess is very interesting for instance b ecause it cannot b e straightforwardly describ ed
as reduction in my theory since the change from tense to lax involves addition of a feature
rather than subtraction I cannot discuss the phenomenon here
Prop erties of r schwa
there are two indep endent factors prohibiting r schwaword initially since
in this p osition schwa o ccurs b oth in an onsetless syllable and in the word
p eriphery I will rst discuss the former factor
Reduction in onsetless syllables is also prohibited inside the word This
is demonstrated in the words in Only words with an aV sequence
where V is the p otential reduction vowel are relevant b ecause tense vowels
other than a all can develop a glide which makes the following syllable
not onsetless
Maoist id mabo st mao st ma b st ma estoso id mab st oso
ma stoso ma b st oso
It is not clear whether a glottal stop is inserted b etween the twovowels it
seems at least to b e p erceptually much less salient than it is in German Yet
schwa reduction will b e blo cked regardless of whether we insert the glottal
stop or not A related restricti on seems to b e the one against reduction
after h
helaas alas hel as h las heraut herald her ut h r ut
hotel id hot l h t l habijt monk s fro ck hab it h b it
As we will see h and b pattern together in many other contexts It seems
not unreasonable to analyse them as empty consonants Neither of them
seems to have its own supralaryngeal place of articulation
At rst sight this might b e seen as a condition on syllable structure
It is sometimes supp osed that vowels immediately before h do not reduce
either Facts like the following are given in supp ort of this assumption
mahonie mahony mah oni m h oni alcohol id lkoh l lk h l
Yet the unacceptability of these facts is not as strong by far as that of the
examples in or As a matter of fact these examples sound quite
grammatical b e it very informal to me and I will treat them in this way
The same seems to b e true for reduction of the rst vowel in hiatus
Also this reduction do es not seem to b e as strongly blo cked as in the case
of the second vowel An additional problem in this case is that the only
sp ot where we could p ossibly exp ect any e ect at all is the place where the
stressless vowel is a cf If the vowel is any of the other tense vowels
e i y u o a glide can b e inserted lexically to resolve hiatus and the
vowel can freely reduce in the p ost lexicon cf
According to Kager the constraints against reduction b efore h or b hold
for word initial pretonic syllables only reduction in medial prevo calic vowels is only
slightly inhibited If this is true I have no explanation for the fact My own observations
do not really con rm Kager s p oint For me all the examples in and are
on the same level
Prop erties of r schwa
na ef naive na f n f coca n id koka n kok n
theater theatre te j a t r t j a t r oceaan o cean o se j an o s j an
The reduced forms do not seem to b e very bad I will therefore concentrate
on the cases of tautosyllabic b h and
Cohen et al observed that we can generalize over the hiatus
context and the p ositions adjacent to h Because in hiatus a glottal stop
is automatically inserted wemay assume that a syllable may not consist
of only empty unarticulated segments
a A ro ot no de is empty i it do es not dominate a place no de
b Contour If the head of a syllable is empty its onset maynot
be empty
In the chapters that follow in particular chapters and we will see
that the contour constraint is more general its actual formulation should
b e something like if the head of a syl lable is x its onset may not bex
This will also rule out illegitimate syllables suchas ji and wy The
constraint Contour is very closely related to the OCP with the syllable
as its domain but it di ers from that principle in twoways In the rst
place the OCP always refers to two adjacent features while Contour in
this case is violated b ecause of a lack of features In the second place a
syllable such as ji could also consist of one segment i linked at the same
time to b oth the onset and the nucleus cf also chapter This is ruled
out by Contour but not by the OCP
We can generalize this observation furthermore over all instances of
schwa E schwa and u schwa are never found word initially as the second
vowel in hiatus p osition or after h either We therefore have the following
prop erty of Dutchschwa
Prop erty Schwa cannot be tautosyl labic to h or b
Because h b or are never found in Dutch we
may assume the constraint Contour in b is ranked in a fairly high
p osition In particular it dominates the pro jection constraint demanding
reduction Connect Ft V As we will see b elow it will also dominate
the constraints on the o ccurrence of e schwa and u schwa
Word nal p osition Like absolutely word initial vowels abso
lutely word nal vowels cannot reduce to schwa
This is an adaptation of Noske s Onset Nucleus Contour Constraint
On the somewhat exceptional status of word nal schwa in English historical phonol
ogy see Harris
Prop erties of r schwa
to ee id t fe t f cola id k ola k ol
herrie noise h ri h r foto picture f oto f ot
In this case the e ect cannot b e attributed to the syllable or any other level
lower than the word As we will see b elow op en syllables generally favour
reduction and also the weak p osition of a tro chee is very favourable for this
pro cess We will also havetotakeinto account that the constraint against
word nal p osition is not shared byu schwa which o ccurs in word nal
p osition fairly often though it do es apply to e schwaaswell
It can also b e demonstrated that it is not nal syl lables that blo ck
reduction since word nal closed syllables do reduce whenever comparable
word internal syllables do cf e g radar id rad r rad r The e ect
is therefore to b e attributed to the segmental level
Prop erty A word nal segment cannot be r schwa or e schwa
What could b e the reason that p eripheral segments are so reluctanttobe
come schwa In the previous chapter wehave already seen another instance
of the particular b ehaviour of p eripheral segments word nal consonants
in degenerate syllables are not deleted This was then attributed to the
theory of Generalized Alignment The de nition of the generalized Align
mentschema of McCarthy and Prince b ex is copied in
b elow For some more discussion of this constraint family see app endix A
def
Align Cat Edge Cat Edge Cat Cat suchthat
Edge of Cat and Edge of Cat coincide
where Cat Cat PCat GCat Proso dic and Grammatical
categories
Edge Edge fRight Leftg
What do es it mean to say that the edges of two categories coincide In
all work done on Alignment Theory the relation seems to imply that the
categories involved share a segment Let us adopt this p osition The
segment I will take to b e a ro ot no de This seems reasonable b ecause ro ot
no des are the lowest phonological ob jects with proso dic a liation at least
in the approach defended here The edge lab el Right of Cat now refers to
x
the rightmost ro ot no de R of Cat i e the ro ot no de R Cat suchthat
x x
there is no ro ot no de S Cat S following R A similar de nition can b e
x
provided for the lab el Left
Wecannow formulate the following de nition of the coincidence rela
tion
On the p ossible role of nul l strings cf McCarthy and Prince b and It o and Mester
Prop erties of r schwa
An edge E of a category C coincides with edge E of category
C i E E and feature F dominated byE a feature F
dominated byE suchthatF F and F dominated byE
F dominated byE suchthatF F
In this de nition the relation is used as a primitive Two class no des
and features havetohave the same lab el feature value and index for the
relation to hold So what says informally isthattwo edges coincide
if they are exactly the same The crucial part of this de nition is the
requirement that all dominated features of the edge segments havetobe
the same This is a stipulation but in my opinion not a very unreasonable
one
We assume the underlying phonological representation of any morpheme
is a set of strings of segmental material What I assume then is that a
grammatical category immediately dominates a string of ro ots McCarthy
The morphology of to ee id therefore lo oks as follows feature trees
have b een con ated to IPA sym b ols
Stem
t f e
Crucially one of the morphological edge no des is e The Principle of
Consistency of Exp onence says that no changes in the exponenceofa
phonological ly speci ed morpheme arepermitted McCarthy and Prince
This means that the Stem no de will always have an e edge no matter
what we do
Reduction means that the vowel features of e are are no longer visible
for the phonology This implies that a Proso dic Word built on t fe will
not see coronal In other words the right edge of the Proso dic Word will
be By our de nition of coincide the edges of Proso dic Word and Stem
will therefore not b e aligned prop erly
If u schwa is really underlying schwa Alignment will not cause a prob
lem for that vowel b ecause in that case the edge of the morphological word
as well as the edge of the proso dic word will b e schwa Also word internal
reduction can of course not b e prohibited by Alignment whichonlylooks
at edges
Because word nal vowels are never reduced wemay assume that Align
dominates Project Ft Noweven if wewould want to distinguish b e
tween tw oversions of Align corresp onding to the two di erent edges like
Align itself might of course b e ranked b elow other constraints in particular the
constraint enforcing nal devoicing which also applies in word nal p osition in Dutch
Prop erties of r schwa
Align Left and Align Right all evidence in Dutchwould seem to p oint
in the direction of Align Left dominating Align Right This has as
a consequence that wehave an extra reason for banning reduction in word
initial p osition even if we do not insert a glottal stop at the b eginning of
the word
Vowel quality An interesting factor in reduction is the quality
of the vowel that has to b e reduced Some vowels reduce very easily other
vowels resist reduction under almost all circumstances In particular e is
virtually always reduced when it is in unstressed p osition with the excep
tions discussed ab ove while y hardly reduces at all If reduction means
getting rid of features which seems the most reasonable assumption to
make this leads us to the conclusion that some features seem to b e harder
to get rid of than others
The most prominent constrast can b e observed b etween A vowels and B
vowels The latter category reduces under very sp eci c circumstances only
However since the relevant feature lax interacts with syllable structure
which itself interacts with stress and since stress is an imp ortant indep en
dent factor in reduction the direct in uence of exactly this feature lax is
particularily unclear I will therefore put o its discussion until the next
few sections Here I will concentrate on the other features
Kager establishes the following reduction hierarchyforvowels
I havechanged his feature lab els in order for them to b e compatible with
the vowel theory develop ed in the previous chapter
a e high labial dorsal
b a high labial dorsal
c o high labial dorsal
d i high labial dorsal
e u y high labial dorsal
Ihave to add that some parts of the hierarchy are not really well established
For instance Kager remarks that hardly occurs in stressless
positions such as initial pretonic position and that its grouping with o
is therefore by necessity more based on theoretical extrap olation than on
empirical observation Also there is a lot of variation b etween individual
but not for instance in Yiddish cf Katz Lombardi and section in
the next chapter
For instance the pre x su x asymmetry in syllabi cation in German is taken to b e
an indication for this ranking in McCarthy and Prince In Dutchwe nd exactly
the same typ e of asymmetries as will b e brie y discussed in the next chapter
Prop erties of r schwa
sp eakers and within one sp eaker b etween di erentstyles of sp eechandeven
between individual words Sometimes reduction do es not really lead to a
pure schwa but to a schwa likevowel containing some phonetic traces of
the original colour I havechosen to nevertheless put the gradient facts into
a discrete scheme A discussion of style registers will b e taken up b elow
Two generalisations are particularly strong I will list them as sep erate
prop erties here
Prop erty R schwa alternates with e and less often with a and
o
Prop erty R schwa almost never alternates with a high vowel
By de nition these prop erties are prop erties of r schwaonly However
we will see in chapter that in other languages underlying schwas also
alternate with frontmidvowels
We see a clear correlation b etween markedness and reducibility e is
probably the most unmarked full vowel If all vo calic features are mono
valent this vowel only b ears the sp eci cation coronal the activevalue for
ap erture features high and low is F in the system prop osed here y
on the other hand is a very marked v owel with features coronal labial
and high
The feature high seems particularly activeinblocking reduction High
vowel reduction seems only p ossible in some very frequentwords like minuut
ziek music myzik m zik but normally it minute minyt m nyt and mu
is imp ossible It seems as if this feature has a particularly strong attachment
to the feature tree In a constraint based theory we can formalize this by
assuming the constraint forcing an underlying feature high to b e parsed
in the output is ranked very high higher than Project Ft in anystyle
register of Dutch
I will followDeSchutter and Kager in assuming that
round or labial as I call it is another feature blo cking reduction Re
duction of the labial mid vowels o and sounds quite informal Finally
a similar consideration leads us to p osit a constraint on parsing the feature
dorsal ranked very low but ab ove Parse coronal in order to capture
the di erence b etween e and a on the one hand and the di erence
between a and o on the other
Kager s hierarchy in can now b e re ected in the constraint
hierarchyin
Parse high Parse labial Parse dorsal Parse coronal
These constraints are never in direct con ict with one another so that
the ranking given here can only b e established on the basis of interaction
Prop erties of r schwa
with other constraints In chapter wehave seen some evidence for a high
ranking of Parse high as well as of Parse lax and Parse low on the
basis of diphthongisation Here we see additional evidence for this ranking
coming from the unrelated area of vowel reduction
It is very well p ossible that parts of this ordering have a universal char
acter For instance the low priority assigned to parsing coronal might
re ect the univeral unmarkedness of this feature It is well known that
in English the only non reducible vowels are high cf Fidelholtz
Giegerich for recent discussion The reduction hierarchy of Breton
Dressler also seems very similar to the one prop osed here for
Dutch high vowels do not reduce at all and the b est candidates for reduc
tion are e and a
I will suggest b elow that the relative ordering of Project Ft with re
sp ect to this hierarchy dep ends on style register In very informal styles
only Parse high dominates Project Ft sothatonlyhighvowels are
excluded from reduction In very formal styles al l Parse constraints dom
inate Project Ft except mayb e Parse coronal so that all vowel seg
ments other than e are exempt from reduction
Notice that it is imp ossible to distinguish b etween i on the one hand
and u y on the other in the approach presented here All three vowels
resist reduction b ecause this would violate Parse high If Kager is
right reduction of the latter twovowelsisworse b ecause they are round
in addition to b eing high Such a generalisation cannot b e captured in
the formalism of OT We cannot add the violations of several constraints
to weigh them against one other constraint If y u are ruled out in a
certain grammar they are ruled out b ecause they violate Parse high b e
cause this constraint dominates Parse labial Then i is out as well
We cannot rearrange the order of Parse high and Parse labial b ecause
then wewould incorrectly predict that o are very marked reducers to o
I am not sure whether Kager s distinction b etween y u and i
is so strong that it p oses a serious problem His main motivation seems to
b e that it is easier to nd words where i reduces fi guur gure bikini
vier river than to nd words where the other two reduce Yet id ri
also for i it is p ossible to nd numerous cases where this vowel seems to
never reduce at all and for y and u there are some cases of optional
reduction e g muziek music myzik m zik as Kager indicates
It mightwell b e that there simply are more frequentwords with stress les s
i than with stressles s y or u Given this lack of empirical clarity I
do not consider my nding that all high vowels should b ehavealikeasa
Cf Mohanan for a theory where violations against constraints can be
accumulated
Prop erties of r schwa
problem
Stress The next factor in reduction I want to study is the only
really strong condition on reduction in English only stressles s vowels are
reduced Stressless ness is one of the most central prop erties of schwa It
holds without exception for all instances of Dutchschwa and it is wide
spread in other languages as well
Prop erty R schwa e schwa and u schwa cannot occur in stressedpo
sition
Interestingly Dutchmakes a distinction b etween twotyp es of unstressed
p osition in vowel reduction This has b een the main topic of most of the
studies on reduction mentioned ab ove The facts can b e demonstrated on
the word fonologie phonology f onolo In very formal sp eech this word
is pronounced as just indicated Two alternative less formal pronuncia
tions are p ossible f on lo and f on l The latter one is even more
informal than the former What is crucially imp ossible is the pronunciation
f onol If the second vowel reduces the rst one should reduce as well
Bo oij
Various prop osals have b een made to deal with this p osition e ect Most
of these prop osals refer to suprasyllabic metrical structure Wehaveto rst
establish what the metrical structure of a form suchasf onolog e is b efore
we can account for the reduction facts Since this level of representation is
not the topic of this thesis I will not review the literature here
In a pro jection theory the fact that b oth no and lo are candidates for
reduction is an indication that they b oth are in a dep endent p osition at
some level of pro jection The fact that lo is more resistant to reduction is
an indication that it is still a head at some other level of pro jection where
no is still a dep endent For this reason I havechosen to use the following
representation I assume that the rst two syllables are joined together in a
fo ot like constituent and that the third syllable is joined to this constituent
into a larger constituent say a sup erfo ot
Prop erties of r schwa
Word
Ft Ft Ft
fo no lo i
I assume that something like the Strict Layer Hyp othesis holds Selkirk
all syllables are organized into feet all feet into sup erfeet all su
p erfeet into words The heads of sup erfeet are the b earers of primary and
secondary accent In the normal case a sup erfo ot consists of one binary
branching fo ot but in exceptional cases when there is a ternary pattern a
unary fo ot may o ccur in the dep endent p osition of a sup erfo ot This analy
sis mayormay not b e the most optimal one but at least it provides us with
the necessary terminology to distinguish b etween twotyp es of unstressed
p osition
The rst unstressed syllable is in the weak p osition of a fo ot and I will
call it a weak p osition The second unstressed syllable I will call semi
weak
Kager prop oses that there is a hierarchy based on metrical p osition
Kager his ex like there is a hierarchy based on vowel
quality In the formalism of pro jection constraints it is however imp os
sible to refer to the di erence b etween a weak and a stray p osition in the
appropriate way In order to account for the fact that no reduces more
easily than lo wewould need to saythat if a vowel lacks vocalic features
it occurs in the dependent position of a foot Yet this is neither a p ossible
pro jection constraint nor a p ossible weakness constraint
Prop erty Reduction to schwa is general ly easier in weak positions than
in a semi weak positions
Kager uses an analysis in which the second syllable of fonologie is in the weak
p osition of a fo ot and the third syllable is stray
i
fo no lo i
Prop erties of r schwa
Dep ending on one s metrical theory one can rephrase the statementof
this prop ertyinvarious ways One of the few frameworks in whichit
is di cult to make this typ e of distinction is Burzio who would
presumably assign a ternary fo ot without anyinternal structure to the rst
three syllables of fonologie Yet even in this case we could for instance
refer to a di erence b etween fo ot internal and fo ot nal p ositions
The idea is that if we reduce a vowel V we also obligatorily reduce all
vowels which are easier to reduce than V
First let us see what the reason is for reducing as manyvowels as p ossible
in the most formal register The imp ortant pro jection constraint here is of
course Project V Ft
Project V Ft N dominates a vo calic no de N heads a
branching fo ot
The b est output of fonolo i for Project V Ft would b e fon l i
b ecause all full vowels o ccur only in the heads of branching fo ot A candi
date like fon lo i violates Project V Ft b ecause it has a full vowel
the o of lo which is not the head of a branching fo ot For a similar rea
son also the candidate fonol i gives one violation of Project V Ft
and the candidate fonolo i gives even two violations
The di erence in reducabilitybetween the twotyp es of syllable has to b e
attributed to their di erence in p osition The syllable lo in is still the
head of a fo ot even though this fo ot do es not branch I prop ose to explicitly
distinguish pro jection constraintatfootlevel which refers to branching feet
and one which refers to feet in general The sp eci c constraintwillbeof
the form in but I will add to this a slightly more general constraint
Project V Ft N dominates a vo calic no de Nheadsa Foot
These two constraints Project V Ft and Project V Ft are in a so
called P aninian relation If Project V Ft is violated a syllable is
outside the head of a fo ot but not reduced Because it is outside the head
of a fo ot it is also outside the head of a branching fo ot Therefore there
will also b e a violation of Project V Ft The reverse is not true a non
reduced syllable in the head of a non branching fo ot will give a violation
Burzio himself notes a similar kind of distinction for English in the fo otnote
on page in English the rst unstressed syllable of t tama gouchi or panama
reduces much more easily than the second one Burzio also notes that similar
facts hold for syncop e cf memor zation mem rization Heeven refers to the historical
phonology of French Jacobs where Latin similitudine simlitudne sembletude
Two constraints A and B are in a P aninian relation if A applies to a prop er subset
of the candidates to which B applies Prince and Smolensky
Prop erties of r schwa
of Project V Ft but not of Project V Ft Concretely fon lo i
violates Project V Ft but not Project V Ft and fonol i violates
b oth constraints
In Dutch we can make an empirical di erence b etween weak syllables
and semi weak syllables Therefore wemay assume the ordering of these
constraints is
Project V Ft Project V Ft
Again this ranking can b e established only indirectly by studying the in
teraction with other constraints I assume this ordering is stable in Dutch
The only p ossible di erence b etween styles of sp eech is the ordering of
faithfulness constraints In the case of fonologie the relevant faithfulness
constraintisParse labial b ecause we are dealing with rounded mid vow
els There are three p ossible rankings of this constraintvis a vis the two
instances of Project V They corresp ond to three styles of sp eech
a formal Parse labial Project V Ft Project V Ft
b semi formal Project V Ft Parse labial Project V
Ft
c informal Project V Ft Project V Ft Parse labial
Wenow get the following tableaux for each of the relevantstyles
a formal
Candidates Parse labial Project Ft Project Ft
p
fo no lo i
fo n lo i
fo no l i
fo n l i
b semi formal
Candidates Project Ft Parse labial Project Ft
fo no lo i
p
fo n lo i
fo no l i
fo n l i
Given the P aninian relation just established if Project V Ft would b e ranked
higher than Project V Ft the e ect of the latter wouldbeinvisible What wewould
get is a language where unstressed syllables are reduced regardless of their precise met rical p osition
Prop erties of r schwa
c informal
Candidates Project Ft Project Ft Parse labial
fo no lo i
fo n lo i
fo no l i
p
fo n l i
The form fonol i could never b e generated byany ordering of these
constraints The reason is that this form violates all constraints while
there are always alternative candidates which fare considerably b etter I
will provide some more discussion on style registers b elow section
Closed syllables Reduction is much easier in op en syllables
than in closed syllables It is not altogether excluded in the latter category
esp ecially not if the closing segment is a sonorant or s Furthermore we
nd re exes of the same kinds of constraints that count for tense vowels
For instance the frontmidvowel reduces most easily Also word internal
reduction is much easier than p eripheral reduction and reduction adjacent
to h or a glottal stop is blo cked Nevertheless reduction of vowels in closed
syllables is more marked than reduction of vowels in op en syllables Given
the theory outlined ab ovewehaveseveral options to analyse this e ect
Prop erty R schwa prefers open syl lables
As we will see b elow the same prop erty holds for u schwa but not for
e schwa
In the rst place we could assume this is a vowel quality e ect lax
resists reduction just like high lo w or labial Wehave already
seen in chapter that diphthongization provided us with an argumentthat
Parse lax should b e ranked high
Alternatively we could analyse this e ect as the result of stress Closed
syllables are heavy and we could assume for instance that even though they
See also Burzio on English reduction and Hawkins Kenstowicz and
Kisseb erth on reduction in Carib Jacobs on reduction and vowel deletion
in the history of French all of which seem also blo cked by the closed nature of the
syllable
There is some variation b etween sp eakers with resp ect to the strength of prop erty
Some sp eakers such as Bo oij nd reduction of the underlined vowel in words
suchasbe nzine id portier p orter only p ossible if the following consonant is deleted
OK OK
b nz n b z ne p rt r p t r Others such as Kager do not
necessarily have to delete the consonant For them the forms b nz n and p rt r are
acceptable pronunciations
Prop erties of r schwa
are not stressed they still form their own abstract fo ot of some sort Under
suchahyp othetical approach they would always end up in a metrical
p osition that disfavours reduction as wehave just seen
Finally the constraint Connect N lax could b e broughttobearupon
this issue It could b e that this constraint is higher ranked than Project
Ft Therefore lax cannot get lost in a closed syllable even if this syllable
is unstressed
Given the very intimate relationship in Dutchbetween vowel quality
syllable structure and stress it seems imp ossible to nd empirical evidence
that would shed light on this issue There is a conceptual reason whyI
opt for the feature analysis lax is a feature that wants to b e parsed
like other features This reason is that the reduction or non reduction of
closed syllables is very muchstyle dep endent andIwant to argue that style
dep endency is always a matter of reranking of faithfulness constraints It
is di cult to establish a precise p osition for Parse lax in the hierarchyof
Parse constraints I will establish in the next section
Style registers Summarizing what wehave found so far wemay
say that some constraints on reduction are absolute in Dutch viz the con
straints on p eripherali ty and against adjacency to the placeless consonants
h and b Other constraints are relative viz the constraints related to
vowel quality and those referring to the precise typ e of unstressed p osition
wehave
As wehave already seen ab ove this variation is normally attributed
een three to style registers I followed Bo oij in distinguishing b etw
such registers Style I informal Style I I semi formal and Style I I I
formal
Normally it is assumed that eachsuchstyle register is a separate gram
mar This concept had to face some problems in earlier approaches based
on extrinsically ordered rules In the rst place style registers could di er
from one another in arbitrary ways in a more informal register rules could
b e i ranked di erently than in a formal register ii b e present while they
were absent in a formal register or iii b e absent while they were present
in a formal register Furthermore the discrete division of style di erences
Yet myintuition is that lax vowels vowels in closed syllables are at least easier to
reduce than high vowels If this is true at least the ranking Parse high Parse lax
holds
Bo oij used a somewhat di erent terminology
Tomy knowledge the rst to distinguish b etween style registers within generative
phonology was Harris Of course there has b een a strong tradition of studying
registers in pre generativework and for instance Selkirk refers to Fouch e
rather than to Harris when she intro duces the concept in her study of French liaison
See Dressler for discussion of the notion style register in generative grammar
Prop erties of r schwa
in for instance three registers is somewhat arbitrary It is not easy to see
whywe should not make a division into for instance ve or ten registers
Observationally an imp ortant di erence b etween more formal and less
formal styles of sp eech seems to b e that the former are closer to the under
lying forms while the latter allow for more reduction ep enthesis deletion
and other op erations In Optimality Theory this can b e easily expressed
The more formal the style of sp eech the higher ranked the faithful
ness constraints
Mayb e is a principle of universal style register comp etence I already
expressed the conjecture that it might b e the only principle This would
solve some of the problems with earlier approaches In the rst place the
relation b etween style registers would no longer b e arbitrary but regulated
by This would solve also an imp ortant learnability problem The
child only has to learn one phonological grammar the di erences in style
would b e regulated by the universal principle Furthermore weno
longer have to p osit an arbitrary number of style registers Wecanhavea
register where the faithfulness constraints are lowest in the hierarchy one
in which they are highest in the hierarchyandeverything in b etween
For Dutchwe established the following partial constrainthierarchies
a Align Contour Project Ft Project Ft
b Parse high Parse low Parse lax Parse labial
Parse dorsal Parse coronal
c Parse lax
Icontend that the rst two hierarchies can b e combined in many di erent
ways I decided to disregard lax vowels hence c here Every hierarchy
is p ossible as long as the partial orders of a and b are preserved
Kager s style registers are characterized as follows
Reduces in Weak p ositions Semi weak p ositions
e formal formal
a semi formal semi formal
o i semi formal informal
y y informal excluded
Laver identi es two other factors that might cross linguistically indicate in
creasing informality of sp eech rising pitch level and range and dropping loudness level
and range Sp eech rate on the other hand do es not seem to b e a contributing factor at all
Prop erties of e schwa
As discussed ab ove in our system i will always pattern with the other
high vowels rather than with o With this exception it is p ossible to give
the rankings that will generate these three registers
a Style I I I formal Align Contour Parse high Parse
labial Parse dorsal Project Ft Project Ft Parse
coronal
b Style I I semi formal Align Contour Parse high
Project Ft Parse labial
Project Ft Parse dorsal Parse coronal
c Style I informal Align Contour Parse high Project
Ft
Project Ft Parse labial Parse dorsal Parse coronal
More subtle di erences are p ossible in this system It mighteven b e that
vergenerate style registers For instance according to wenow somewhat o
Kager e always reduces and the high vowels in stray p osition
never do This means that Parse coronal has a xed highest p osition
it is always b elow Project Ft andParse high a xed lowest p osition
itisalways ab ove Project Ft Iamnotvery sure this observation is
strong and if it is whether this could not simply b e a so ciolinguistic
p erformance e ect Mayb e we could say some styles are so overly formal
or so overly informal that they are never used Yet it seems more plausible
to say that languages x an upp er and a lower limit for their faithfulness
constraints
Prop erties of e schwa
Of the three di erentschwas we distinguish here e schwa the real svarabakhti
vowel is the least well studied The reason probably is that on the one
hand its distribution is relatively narrow and on the other hand this o ccur
rence is somewhat problematic A typical instance of e schwa o ccurs in the
consonant cluster at the end of the word help id h l p A pronunciation
a sounds somewhat overarticulated in the ears of of this word without schw
some native sp eakers but it is not very clear howepenthesis improves the
word Of course on the one hand wedoawaywithamarked sup erheavy
syllable On the other hand however the result is a closed syllable with
schwa as its head which is also marked in Dutch Furthermore the sono
rant l has to b e re syllabi ed into the onset of the schwa syllable All
in all the slight improvement of syllable structure seems to demand a high price
Prop erties of e schwa
I will assume here that e schwa really is the result of ep enthesis an
uncontroversial p oint of view as far as I can see cf Berendsen and Zonn
eveld and De Haas for the most extensive discussion of e schwa
to date Below I will discuss the prop erties of this schwa one byone
The ep enthetic vowel is schwa The rst prop ertytobedis
cussed is that it is precisely schwa that is the ep enthetic vowel and not
some other vowel In other languages other vowels may indeed act as
ep enthetic vowels For instance the Icelandic ep enthetic vowel seems to
b e u the ep enthetic vowel in Spanish is e and in Yokuts it seems to b e
i see Archangeli for an overview As far as I know however all
languages that allowvowel ep enthesis and that haveaschwa on the sur
face the twovowels are the same So the fact that the ep enthetic vowel in
Yokuts is i is an indication that this language do es not haveanyschwaat
all
The reason whyschwaisafavourite default vowel should b e attributed
to its b eing essentially featureless For one reason or another we sometimes
need to insert a vowel which is not underlyingly present but wealways do
this in the most economic way This means in the rst place that wedonot
insert unnecessary features such as coronal or high If schwa is allowed
as a vowel in the language at all schwa therefore is the preferred ep enthetic
vowel
Prop erty E schwa is the epenthetic vowel
In Dutch the interaction of several constraints forces us to insert a vowel
between the l and the p of h lp In principle we can insert anyvowel
also an a or an oran y Yet each of these vowels will b ear a set
of vo calic features viz dorsal labial lax and labial coronal lax re
sp ectively Every feature constitutes a violation of a constraint against the
insertion of exactly that feature low labial lax etc Schwagives
a minimal violation of these constraints This I contend is the reason why
cross linguisti call y schwa is the preferred ep enthetic vowel just likeitisthe
preferred reduction vowel
It is however p ossible that di erentvowels play the role of default vowel at di erent
lexical levels In that case schwamight b e p ermitted only after the lexical level where
some other vowel has played the role of default vowel This would cause some complicated
pattern of surface violations of the generalisation made in the main text I am not aware
of any system that has b een convincingly shown to p ossess these prop erties
The same line of reasoning would lead us to p osit b asthefavourite ep enthetic
consonantinany language that allows the consonant at all I will not go into this matter
any further The glottal stop certainly is the ep enthetic consonant in Dutch
The ep enthetic vowel is not normally transcrib ed in Dutch writing If one wants to
Prop erties of e schwa
In languages whichdonothaveschwa a constraint should b e active
that bans this vowel The most straightfoward waytodothiswould b e the
following
EmptyRoot No empty ro ots are allowed
This constraint states that every vo calic ro ot no de needs to dominate a
vo calic no de which in turn needs to dominate a place no de and or an
ap erture no de whichhave to b e lled These constraints therefore demand
that even the default vowel in the language gets some minimal sp eci cation
On the other hand we could also use a pro jection constraint to do the same
work
Project N V N is the head of a branching N N dominates
avo calic no de
The advantage of this constraint clearly is that wehave used it already
Since vo calic ro ots usually o ccur in the head p osition of a syllable b oth
constraints have the same empirical e ect I cho ose to use Project N V
b ecause it is formally related to the other pro jection constraints and it ts
well within their family
The precise nature of the required feature sp eci cation under the vo calic
no de to satisfy Project N V dep ends on the weight the language assigns
to the di erent features In one language it might b e that insertion of a
place feature is su cient and that coronal is the most unmarked place
feature In such a language the default vowel will surface as e as in
Spanish except when the feature co o ccurrence restricti on coronal high
is ranked su ciently high in whichcasewe will get an i likeinYokuts
In other languages insertion of labial will lead to the smallest number
of violations and we will get an u like in Icelandic I predict that y can
never b e an ep enthetic vowel since it contains two place features coronal
and labial and there seems to b e no feature co o ccurrence restriction in
UG forcing the two features to o ccur together
mimic informal sp eech in writing usually one of two transcriptions is chosen viz either
hel le p oru hel lu p The rst sp elling is probably chosen b ecause e is the normal e
transcription also for u schwa mode mo d and the second sp elling u b ecause
the vowel normally transcrib ed by this letter is phonetically closest to schwa in most
Dutch dialects
Indep endent evidence for either of these constraints may come from the fact that
in languages likeTurkish or Yoruba in whichvowel harmony seems blo cked internal to
a morpheme schwa likevowels are often an exception to this blo ckage they can only
o ccur in harmonic contexts Clements and Sezer Pulleyblank Bennink in
N V or EmptyRoot preparation This might b e seen then as a result of Project
dominating the constraint that blo cks morpheme internal spreading For more discussion
on the constraint EmptyRoot see chapter and app endix A
Prop erties of e schwa
Thus for Spanish Yokuts Icelandic and Dutchwe get the following
rankings
a Spanish
N coronal labial Project
high coronal high
b Yokuts
labial Project N coronal
coronal high high
c Icelandic
N labial coronal Project
labial high high
d Dutch
coronal labial Project N
coronal high high labial high ordering irrelevant
BelowIgive the relevant rankings for an ep enthetic vowel in each of the
four languages just mentioned
a Spanish
Candidates Project N V labial coronal
u
o
p
e
p
i
high Candidates coronal high
p
e
i
b Yokuts
The ordering coronal labial seems very marked given the fact that coronal
place is in most languages the most unmarked place for b oth consonants and vowels It
seems indeed true that ep enthetic vowels in most languages are chosen from the set fi
e g Steriade i e either a frontvowel or a placeless vowel
Prop erties of e schwa
labial coronal Candidates Project N V
u
o
p
e
p
i
Candidates coronal high high
e
p
i
c Icelandic
Candidates Project N V coronal labial
p
u
p
o
e
i
high Candidates labial high
o
p
u
d Dutch
Candidates coronal labial Project N V
p
u
o
e
i
high Candidates labial high coronal high
p
This theory predicts that at least the vowels a e i o u can function
as an ep enthetic vowel in some language On the other hand it predicts
that a universally marked vowel like y can never b e an ep enthetic vowel
since it contains two place features coronal and labial and there seems
to b e no feature co o ccurrence restiction in UG forcing the two features to
o ccur together In other words there probably is no grounded constraint
Prop erties of e schwa
coronal labial parallel to labial high To the b est of myknowledge
this prediction is b orne out
Project N V thus constrains the ep enthesis of schwa as an empty
vowel It cannot in and of itself b e the reason for schwaepenthesis b ecause
a syllable with schwa violates Project N V justaswell as a syllable
without it I prop ose therefore that the real reason for schwaepenthesis
is another pro jection constraint viz Project N cons a constraint
requiring syllables to have at least a vo calic ro ot in their head
N cons N heads a branching N N dominates a Project
cons ro ot
Insertion of schwa is the most economical way of complying with Project N
cons for syllables which start out without anyvowel since schwa is ex
actly a cons ro ot
N Wethus could assume that there are twoslightly di erent Project
constraints at work in Dutch A di erent solution is also feasible how
ever This would b e to include cons in the set of vo calic features V
In that case and if we assume that the constraints against insertion of
place features are su ciently high we could replace Project N cons
N V i e the following constraint ranking has the same e ect by Project
N cons as Project
coronal labial dorsal Project N V
Wehave just seen that this ranking holds for Dutch For the sakeofsim
plicity and ease of reference I will keep however to the constraint name
Project N cons
E schwa do es not o ccur at the end of the word One of
the biggest mysteries of Dutche schwa is the following whyis h lp in
informal styles of sp eech improved to h l p and not to h lp The
latter certainly seems b etter from the p oint of view of metrical structure
it has less violations of Connect N lax It seems that e schwaisbanned
from absolutely word nal p osition
Wewould also have to assume that a vowel with a vo calic no de in addition to a
N V b etter than a vowel with only a vo calic ro ot in vo calic ro ot satis es Project
other words Project N V should b e a gradual constraint
As we will see in the next chapter there is cross linguistic variation with rep ect to
the allowance of ep enthetic schwainword nal p osition Many languages would prefer
word nal ep enthesis in this case Yet Moro ccan Arabic and Berb er are instances of
languages where ep enthetic schwa never o ccurs at the end of the word either
Prop erties of e schwa
Wehave already seen that r schwa has the same prop ertyofavoiding
the word edge prop erty page The explanation I havegiven for
that schwa can b e used for e schwaaswell I assumed the constraintthat
the edges of the morpheme and the proso dic word should coincide Align
ment is very strong in Dutch barring all kinds of changes of these edges
except nal devoicing The morpheme help ends in a p Therefore
the corresp onding proso dic word should end in a p as well It preferably
should not end in an ep enthetic schwa
Wehave already seen an additional blo ckade against ep enthesis of schwa
at the end of the word One reason for putting the nal p of help in a
degenerate syllable rather than leaving it unparsed is the constraint Align
as I argued in the previous chapter h l p is b etter than h lp b ecause
its morphological and phonological edges are b etter aligned At least one
of these two constraints outweighs the requirements on syllable structure
E schwa only o ccurs in the last syllable of the word This
constraintcontrasts with the previous one Together they have asane ect
that e schwa can only o ccur in closed syllables This makes it di erentfrom
the other schwas which preferably r schwa or even exclusively u schwa
o ccur in op en syllables
This generalisation has also b een contested byseveral authors such
as Kager and Zonneveld De Haas and Kager Ac
cording to these scholars there is one other environment for ep enthesis
VC C C i e b efore a word nal syllable headed by u schwa
They use this observation as an argument that u schwabehaves likeaword
edge They cite paradigms like the following to demonstrate the e ect I
underline e schwatocontrast it with u schwa in this example
m helmet b h l m r name c h l ma name a h l
w r m warm m r m r marble m r m t id
In my view the ep enthetic schwain h l m has a di erent status from
the one in h l m r As I already stated ab oveepenthesis is for many
sp eakers almost obligatory in the former case In the latter case however
pronunciations without schwa are p erfectly acceptable In a normal mo de
of sp eech mostofmy informants would say h l m but h lm r w r m
but marm r
So while technically b oth schwas should countase schwa and while
we should also explain whye schwa is almost absolutely forbidden b efore
This observation is con rmed by Kager
Prop erties of e schwa
a full vowel as in m rm t I will put apart the marmer cases Having
done this the following prop ertyofe schwa emerges
Prop erty E schwa only occurs in the last syl lable of the word
The reason for this seems to b e that the app earance of e schwa is closely
linked to the degenerate syllable E schwa is found only b etween a closed
syllable and a degenerate syllable
This result conforms the facts A degenerate syllable is very marked in
manyways For instance it lacks a head which is a violation of one of the
central principles of syllable constituent theory the pro jection constraint
Project N V
Degenerate syllables are formed at the lexical level Schwaepenthesis
seems in all resp ects to b e a p ost lexical or word level cf Trommelen
N cons gets quite strong in pro cess Assume therefore that Project
this level of phonology whilst it do es not visibly play a role in the lexical
phonology This will ban all degenerate syllables whichnow should get
vo calic heads An empty head is clearly not vo calic
The ideal syllable head is a vowel The most economic waytoget
a syllable head is byepenthesis of a vo calic ro ot For reasons discussed
ab ove this syllable head cannot follow the original content of the degenerate
syllable i e the p of help it therefore has to precede it Furthermore the
l is resyllabi ed into the onset of the newly created syllable I assume it
stays within the co da of the rst syllable and therefore gets ambisyllabic
N N
N
N
N
N
h l p
The p ost lexically created syllable violates some constraints that are oth
erwise unviolable in Dutch Most imp ortantly this is the only o ccurrence
These cases remain to b e a problem under any analysis I am not awareofany
prop osal that can describ e them in a satisfying way Itseemstobetrue however that
e schwa can only o ccur in a cluster which is p otentially a word nal cluster It can not
m since there is no draxm and byt l r since there is no b tl o ccur in dr
I assume sonority restrictions are still in force excluding the p ossibility of consonan tal heads
Prop erties of e schwa
of a closed syllable without a lax feature on the head If we assume p ost
lexically the prohibition against insertion of lexical features is very strong
much stronger than Connect N lax this apparent anomaly can b e ex
plained away More in general at several places b elow we will see that
pro jection constraints have lost their force at the p ostlexical level It seems
that p ost lexically the relation b etween feature content and proso dic struc
ture is lost More in general it is sometimes observed that for instance p ost
lexical rhythmic stress is less sensitive to subtleties of syllable structure
than lexical phonology it also do es not seem sensitive to the heavy light
distinction in syllables
Word internal contexts in whiche schwa do es not o ccur
There are three contexts in whichwe donot nde schwa contrary to our
exp ectations in the discussion so far This is the case if a degenerate syllable
is preceded by i an op en syllable ii a syllable that ends in a consonant
which is homorganic with the consonant in the degenerate syllable or iii
if the nal segment is a coronal obstruent These contexts are exempli ed
in a b and c resp ectively
a slaap sleep sla p sla b p groot big ro t ro t
b bank id b k b k ramp disaster r m p r m p
c hart heart h r t h r t
herfst autumn h r f st h r f st h r f s t
The third case is the simplest The fact that a coronal obstruentisinvolved
directly leads to the conclusion that there is no degenerate syllable in these
cases at all The coronal obstruent is simply outside the proso dic word al
together Sp eci cally it is not in a syllable therefore it do es not constitute
N cons Ep enthesis is therefore not necessary a violation of Project
The relevant economy principles for instance the constraint against in
serting a ro ot no de will therefore blo ck it as they blo ck the o ccurrence of
anyschwa adjacenttoavowel
I prop ose that in the rst two cases ep enthesis is blo cked by indep endent
factors viz the Onset constraint and a constraint against intro ducing
consonantal place features We will see b elow that also u schwa disapp ears
after and p ossibly b efore a full vowel and wehaveseenabovethatr schwa
did not arise immediately adjacent to a full vowel either
The explanation in the case of r schwawas the constraint Contour
If we assume this constraint outranks Project N cons we get the
following result Tomake things more explicit I draw the following tableau
for the word slaap sleep sla p
Prop erties of e schwa
N cons Contour Project
Candidates Contour Project N cons
p
sla p
sla p
In English a schwa is inserted b etween a dipthongized long vowel and
l or r example McCarthy Halle and one might argue
that a similar case can b e made for long vowels b efore r in some variants
of Dutchasisshown in see section in the previous chapter for
more discussion of the b ehaviour of tense vowels b efore r
feel j l feeling jl foal fow l goalie gowlij sure suw
assuring sur re fa j reit fa jr t
zwaar heavy zwa r duur exp ensive dy r boer farmer bu r
The reason might b e for instance that the high sonorancy of the liquids
and esp ecially of r makes them b etter co das than other segments I will
not however try to work this out
We are now left with the cases of homorganic clusters These include
at least the words ending in nasal obstruent clusters like bank id b k
b k and ramp disaster r mp r m p Wemay also wish to include
cases suchaskans chance k ns k n s and hand id h nt h n t
and even held hero h lt h l t All these latter cases however can also
b e analysed as including extraproso dic coronals A condition against schwa
ep enthesis in those clusters was indep endently needed to account for cases
suchasherfst autumn h rf st h rfs t
In previous analyses the resistance of these clusters against ep enthesis
y e ect See Selkirk for was usually ascrib ed to a geminate integrit
an overview The nasal and the obstruent shared a place feature
cons cons
nasal voice
Place of Articulation
No segment could intervene b etween the two halfs of this semi geminate
Yet under the assumptions made here there are two problems with this
approach
Prop erties of e schwa
In the rst place we could think of the No Crossing Constraint of Gold
smith But if schwaisanempty ro ot no de insertion of schwabe
tween the two ro ots in should not cause any line to b e crossed at
all
cons cons cons
nasal voice
Place of Articulation
An alternative might b e to argue that the relevant constraintisNo Skip
ping Van der Hulst and Smith Archangeli and Pulleyblank
No Skipping a feature that is asso ciated to and in the con
guration should also b e asso ciated to
No Skipping Project N cons
The result of schwaepenthesis in is a violation of No Skipping Yet
there is at least one way to repair this violation and wemightwonder why
this reparation is not made We could degeminate the cluster so that
weget
cons cons cons
nasal PofA PofA voice
If weepenthesize a schwainto this cluster this will cause a violation of nei
ther the No Crossing Constraint nor the No Skipping Constraint However
this form is worse than in at least one other asp ect It contains more
instances of the Place of Articulation no de and of course also of the place
features dep endent on that no de In other words it has more violations of
the constraints against coronal dorsal or labial As wehave already
seen the reason why schwa is ep enthesize d rather than some other vowel
was to avoid violations of exactly these constraints Here we see that they
are so strong that they can even block ep enthesis
This constraint is clearly related to the Contiguity constraint at the level of syllable
structure as discussed in the previous chapter
Some other crazy reparation p ossibilities not discussed in the text are linking the
consonantal place features to the schwa deleting the consonantal place features moving
the features from the nasal to the schwa etc Although all these steps would avoid the
violation of No Skipping they would obviously result in quite absurd constructions to b e
ruled out by indep endently needed and very strong feature co o ccurrence constraints
Summary and conclusion
A third logical p ossibility is also excluded by the constraints we already
intro duced ab ove This is to disasso ciate the Place features from one of the
two segments the nasal or the obstruent altogether In that case ep enthesis
will also not result in a violation of either No Skipping or No Crossing
However if we delink the place features from the obstruent the result will
be a b or h like element and wehave already seen that schwanever
surfaces next to such a consonant If we delink the place features from the
nasal it is not very clear what the result will b e Wemight either supp ose
that placeless nasals are blo cked altogether there is no nasal variantof
the glottal stop or the h or we might extend the interpretation of the
Contour to exclude schwa adjacenttoany place less segment
Style registers Even though ep enthesis in words like help and
kerk church is almost obligatory in a formal style of sp eechpronunciations
without schwa are still p ossible
Very formal styles of sp eecharecharacterized by a high ranking of
faithfulness constraints as I have argues at various places ab ove In this
case the relevant faithfulness constraint is the condition against insertion
of non underlying ro ot no des Normally this constraintisranked under
N cons the constraint forcing vowel ep enthesis In a very Project
formal style of sp eech the constraint raises to a very high level where
ep enthesis b ecomes imp ossible in any morphological or phonological con
text
Summary and conclusion
In this chapter I have discussed the two o ccurrences of p ost lexically derived
schwawe nd in Dutch a schwa that is derived by reduction from full
vowels in unstressed p osition and a schwa that is derived byepenthesis in
consonant clusters
In b oth cases I have argued that the relevant constraint is a pro jection
constraint Reduction follows from the fact that full vowels prefer to o ccur
in the head of a fo ot when they o ccur outside of that p osition in particular
the weak p osition of a fo ot every vo calic feature F means a violation of
a constraint whichsays if a segment dominates F the segment should b e
the head of a fo ot Since schwa is the minimal vowel it gives the smallest
p ossible numb er of violations of this constraint Ep enthesis on the other
hand is an e ect of a pro jection constraintwhichsays that every syllable
should haveavo calic head A degenerate syllable consisting of only a
consonant violates this constrain t We therefore prefer to insert a vowel
and if we do this in the most economic waywe again cho ose the minimal
Summary and conclusion
vowel i e the schwa
In b oth cases schwa app ears to b e a compromise b etween on the one
hand a requirementofhaving a vowel in a certain p osition and on the other
hand of havingafewvo calic features in that p osition as p ossible Making
this compromise has some cost in this case that wehavetoallow an empty
ro ot This empty ro ot cannot o ccur in all environments for instance not
after a hiatus or a h and this causes reduction and ep enthesis to b e
blo cked in those environments
It is quite striking that the blo cking environments for reduction and
ep enthesis are almost exactly the same In a rule based theory of phonol
ogy this similarity is di cult to capture since probably reduction and
ep enthesis should b oth b e the results of di erent sets of rules in such
a theory In the constraint based theory prop osed here there are simply
some constraints blo cking schwa to arise in some environments There are
other pro jection constraints which cause full vowels to reduce or schwas
to b e ep enthesized and these can b oth b e sub ordinate to the same surface
constraints
Similarly the constraint based analysis allowed us to express the e ects
of style of sp eech on b oth phenomena in a uniform way In b oth cases we
could say that the more formal the style of sp eech the closer the similarity
between input and output This means that in more formal styles of sp eech
w e exp ect less reduction and less ep enthesis Again this correlation is less
easy to capture in a rule based analysis
As for the representational side of the phenomena discussed here I
think these phenomena provide fairly strong evidence that schwas are re
ally emptyvowels Phenomena of reduction of schwa and ep enthesis of
schwa instead of some other vowel are b est to b e understo o d in terms of
feature loss and representational economy resp ectivel y The pro jection the
ory predicts that emptyvowels haveavery defective pro jection b ehaviour
In the next chapter I will show that this b ehaviour is evidentinDutch
underlying schwas as well
Dutch U schwa
Intro duction
U schwa is probably the most hotly debated of all o ccurrences of schwain
Dutch Its place in the vowel system has b een sub ject to quite some contro
versy Some authors haveeven denied a separate status in the Dutchvowel
triangle for this vowel arguing instead that it is a sp ecial case of either
e schwa or r schwa Although I will endorse an ep enthesis hyp othesis for
some of the facts b elow I will also argue that some other u schwas necessar
ily are underlying Indeed in a theory without constraints on underlying
representation it is di cult to ban underlying schwa s esp ecially if there
are schwa s on the surface If everything can b e input to the grammar also
emptyvo calic ro ots can
My idea is that underlying schwainmany cases is blo cked from surfac
N V the constraint against syllable ing for instance b ecause of Project
heads without a vo calic no de Only when there are other constraints that
force the schwa to surface an exception can b e made to the general rule
One instance of such a constraint is the constraint Project cons N
again a pro jection constraint this time forcing underlying cons ro ots to
o ccur in the heads of a syllable
Project cons N A segmentShasa cons ro ot S is the
N head of a branching
This constraint can also b e read as Parse V underlying vo calic ro ots have
to b e parsed It is hard to distinguish b etween Parse V and Project cons
N in Dutch b ecause empty vo calic ro ots when parsed always get parsed
into a syllable
Another p ossibility sometimes suggested e g in Van den Berg is that schwa
is a variantof y or to which it p erceptually is very similar The suggestion is not
worked out seriously byany recent author however and given the resistance of other
front rounded vowels to reduction it seems di cult to reconcile with any of the other
facts ab out e schwa
Prop erties of u schwa
The underlying nature of u schwa explains some of the di erences b e
tween this vowel and the two other schwas In the rst place u schwais
already presentatthelowest level of the phonology whichaccounts for the
fact that in some ways it is more restricted than the other schwas But on
the other hand u schwa is underlyingly present which means that it can
sometimes surface in contexts where the derivation of e schwa and r schwa
was blo cked
Prop erties of u schwa
Because of the extensive literature on this sub ject and b ecause there is
a rather large range of known facts I have decided to devote this whole
chapter to my discussion of this vowel In this section I will study the
prop erties of Dutch that have b een discussed in most recent literature and
that every theory of this vowel has to takeinto account I will defend the
hyp othesis that schwa s defective nature has as an e ect that it can only
head a minimal syllable viz a CV syllable without a co da or a complex
onset In the next section section I will contrast this theory to some
of its comp etitors
Syllable weigh t The rst prop erty of u schwa is already known
to us b ecause it is true also for r schwa but not for e schwa It is prop erty
rep eated here for convenience
Prop erty U schwa can occur in an open syl lable
The standard example for this prop erty is that whereas mica id mi ka
and Mieke girl s name mi k arevalid Dutchwords the same is not
true for mi k
Prop erty is esp ecially surprising under length theory if only long
vowels can o ccur in op en syllables schwa phonetically the shortest vowel
of all is not exp ected to o ccur in the same environment Yet tenseness
theory also do es not have a ready explanation for these facts b ecause schwa
is not exactly tense either
In the theory defended in the previous chapter however this problem
disapp ears The relation b etween vowel quality and syllable structure was
expressed by Connect N lax as follows
N lax N dominates lax i N branches Connect
See Trommelen and Bo oij a for the prop osal that schwa phonologically
is long
Prop erties of u schwa
If schwa is featureless it also do es not b ear the feature lax Connect N
lax therefore makes an even stronger prediction than what is listed here
as prop erty It predicts that u schwacanonly o ccur in an op en syllable
On rst sight this prediction may seem to o strong I will nevertheless
argue b elow that it is correct For now it should b e su cient to note that
prop erty is a corrollary of Connect N lax and the representation of
schwaasanempty ro ot
Wehave already seen that also r schwa had a preference for op en sylla
bles while the fact that e schwa o ccurred in closed syllables was the result
of indep endent and stronger factors An imp ortant di erence b e
tween u schwa and the two others is that the former can also o ccur in op en
syllables at the end of the word Wehave seen that r schwa and e schwa
never o ccur in that environment
mod fashion tant aunt lind linden tree oranj or
ange
If u schwa is really an underlying schwa this di erence is what we exp ect
The reason why e schwa and r schwawere banned from the p eriphery was
Alignment theory Metaphorically sp eaking the edges of morphemes have
to b e absolutely clean We are not allowed to reduce anything and we are
not allowed to insert anything if these op erations would change the edges
Aschwa which is underlying of course cannot b e banned from a p e
ripheral p osition in the same way As a matter of fact we predict that
p eripheral underlying schwas should always surface as such This gives us
the desired result for word nal schwa
U schwa does not occur word initially At this p oint the
question might arise whether the discussion in the previous subsection also
gives us a satisfactory explanation for word initial u schwa b ecause of the
following prop ertyofu schwa
Prop erty U schwa does not occuratthebeginning of the word
There are absolutely no Dutchwords like d m This prop erty is attested
in other languages likePolish French and German as well
See Hall for German and Wetzels for French RowickaandVan de
Weijer note that the observation can b e made for the Polish yer and Hermans
p c observes that it is true for Serb o Croatian yer as well Some discussion of non
Dutchschwa s will b e given in chapter
In Dutch the prop erty is restricted to lexical forms As a matter of fact most clitics
do start with schwa er r there em m him t t it es s once and een
n a Even p ost lexical schwa observes most of the other prop erties stresslessness
supp ort of sonorants coronals or k only no complex onsets See section for discussion
Prop erties of u schwa
In the case of r schwa and e schwa whichwere also banned from word
initial p osition I claimed there were twocontributing factors The rst one
was Align and the second was a prohibition against syllables of which
neither the head nor the onset haveany features
As explained ab ove Align cannot b e the explanation for a ban against
underlying schwainword nal p osition b ecause u schwa does o ccur in such
a p osition It seems therefore reasonable to assume it also cannot ban schwa
word initially Ihave already p ointed out that even if wewould assume
an asymmetry b etween Align left and Align right all evidence seems
to p oint in a direction where Align left is stronger than Align right
for instance the b oundary b etween pre xes and stems cannot b e crossed
by syllabi cation whereas the b oundary b etween stems and su xes can If
Align do es not work we are left with the second explanation The word
initial syllable seeks an onset This onset can only b e lled by a glottal
stop which results in an illicit completely empty syllable Such an illicit
syllable will also surface when we do not insert an onset segmentatall
Wehave already seen that the constraint against totally empty sylla
bles called Contour isinviolable in Dutch grammar Since it do es not
directly con ict with Onset its actual p osition in the hierarchyishowever
irrelevant for the present analysis as long as its dominates Align
Contour Align
d m
Candidates Contour Align
d m
p
d m
b d m
No matter where werank Contour with resp ect to Onset the form d m
will always come out as the optimal form as long as Contour Align a
ranking for whichwe will see some indep endent evidence b elow
As was the case with e schwa and r schwa we exp ect schwa to b e blo cked
also after h This exp ectation is b orne out indeed
a hV b h c h
a bra h m a bra h m a br h
There are no forms where we nd a lexical schwa after h
Some other segmental e ects Other prop erties of schwathat
seem of relevance for any theory of Dutchschwa are the following
Prop erties of u schwa
Prop erty Consonant clusters before schwa undergo epenthesis as if
they were syl lable nal
m helmet b h l m r name c h l ma name a h l
w r m warm m r m r marble m r m t id
Prop erty In some Dutch dialects sp is metathesized at syl lable end
and before schwa but not before ful l vowels Stroop
a v sp v ps wasp
b m s p l m p s l medlar
c s pi ri n p si ri n Aspirin
Both of these prop erties seem to show that the consonant cluster b efore a
schwabehaves as a syllable nal consonant cluster as has rst b een noted
by Kager and Zonneveld I will call this typ e of theory which claims
that schwa do es not license a full syllable the No Syl lable Theory
Unfortunately the scop e of these two prop erties is rather limited In
particular the variant of Dutch with metathesis b efore a schwa syllable is
anything but widespread Yet many dialects have metathesis at the end of
the word This seems to b e an argument against rather than in favour of
an analysis in whichwe generalize over schwaandword end
The same is true for the ep enthesis facts as wehave seen Ep enthesis
b efore schwa syllables seems to b e of a di erent nature than ep enthesis at
word end Another thing is that the No Syllable Theory has some problems
in explaining metathesis itself If schwa do es not form a syllable there is
no obvious reason whyitwould b e inserted at all
The next generalisation sometimes listed as a prop ertyofschwa Kager
and Zonneveld Kager is even more questionable
Prop erty x and diphthong plus r are excludedbefore schwa as
wel l as syl lable nal ly whereas they may occur before ful l vowels
The e ects of these constraints are illustrated in whichlikemany
other examples in this section are copied from Kager
a V V rV b V V r c V V r
i j i j i j
u ro ra u r ur
a xV b x c x
xo ra x l x
It is questionable whether these facts have to b e attributed to the presence
or schwa Indeed Gussenhoven has shown that these facts might
Prop erties of u schwa
b etter b e seen as fo ot level well formedness constraints The well formed
diphthong r and x sequences in and are separated bya
fo ot b oundary The stress patterns of aurora and angora are u r o ra
and x o ra resp ectivel y
Now the fact that these sequences cannot o ccur to the left of a schwa
can b e seen as a result of the fact that schwa cannot o ccur at the left edge of
a fo ot in other w ords since Dutch feet are left headed that schwa cannot
o ccur in the stressed p osition of a fo ot
The onset of schwa syllables Inow turn to a prop ertyof
Dutchschwa which is quite sp ectacular
Prop erty Schwa cannot have a complex onset
This prop erty is demonstrated in the following example
a ka tr l pulley b ka tr l c katr
dy plo duplicate dy pl dypl
Also this prop erty seems go o d evidence for the No Syllable Theory since
p ossible complex onsets also are not allowed word nally asissho wn in
c Clusters b efore schwabehave as co das not as complex onsets b e
cause they are co das if schwa do es not pro ject a syllable and if we further
more assume that all consonants preferably incorp orate into a syllable
In its original form Kager and Zonneveld prop erty was some
what stronger stating that consonant clusters b efore schwa cannot havea
rising sonority
More recent literature has brought up a signi cantnumber of counterexamples
words with rising sonority sequences b efore schwawhich are disallowed at
the end of the word to this version of prop erty however some of which
are quite systematic the following examples are from Kager and Zon
neveld
drachme drachma dr xm Ariadne name ari tn
butler id bytl r ritme rhythm r tm
Dafne name d fn franje fringe fr nj
kubisme cubism kyb sm partner id p rtn r
tarwe wheat t rw orgasme orgasm rx sm
ordner le rtn r rapalje rabble rap lj
dr xm bytl d fn etc
All of the words in have a rising sonority sequence b efore schwa In
terestingly these clusters not only cannot o ccur word nally but they also
are not p ossible onsets For instance the tl sequence of butler wealso
Prop erties of u schwa
nd in atlas id which is syllabi ed as at las tl is not a p ossible onset
in Dutch
Admittedly a signi cantnumb er of these words are fairly recentloans
Yet isme is a pro ductive stress attracting su x words like franje part
ner tarwe and rapalje seem to b e very much adapted to the Dutch sound
system and wealsohavetoexplainwhyaword like ritme is b orrowed in
exactly this form unlike for instance its English counterpart
The fact that words like are p ossible is a severe problem for the
No Syllable Theory Ifthel of butler is not licensed in the co da of the rst
syllable and schwa do es not pro ject its own syllable this segment should
b e licensed in some other way In that case it is di cult to imagine why
for instance the r of the imaginary but imp ossible katr l could not also
b e licensed in this other way
I b elieve that in contrast prop erty is fairly well establis hed for word
nal schwa Counterexamples are a few learned words like oeuvre and
timbre which at least in the ears of the present author sound very foreign
very French like the full vowel of the rst word still has its French quality
the full vowel of the second is even nasalized while nasal vowels do not exist
in Dutch phonology They seem to b e real loan words
Interestingly schwa in other p ositions in the word more in particular
word initially do es not seem to ob ey the same restrictions Here we nd
words like the following
plezier joy pl zir brevet certi cate br v t bretels braces
br t ls
It seems to me that the schwainvolved here really is r schwa All of these
forms have somewhat marked pronunciations with e instead of In
terestingly for some sp eakers the surface schwaof pr cies seems to b e
reinterpreted as u schwa These sp eakers pronounce the wordas p r cies
so with a simple onset and a sonorant segment in the co da
I p ostp one the question whyu schwa do es not license a complex onset
to section b elow Here I want to note that wealsohave to exclude the
logically p ossible syllabi cation k t r l but that this can b e accomplished
by indep endent means since lax vowels can never b e followed by p otential
complex onset clusters as wehave seen in section in chapter
The co da of schwa headed syllables The facts just mentioned
are very interesting but b efore we can start to analyse them we rst haveto
consider one other prop erty which is probably somewhat more controversial
than the previous ones
See chapter for discussion of words ending in Cj with C a coronal consonant
Prop erties of u schwa
Prop erty Schwa can license at most one sonorant in the coda
Prop erty a somewhat weaker version of prop erty implies that while
bek r b eaker and vog l bird are Dutchwords the same cannot b e said
for kag k or vog p This observation needs some motivation b ecause
it is usually ignored or even denied in the literature Cohen et al
Kager and Zonneveld
First of all observe that there are no words ending in C C ifC is
i j j
not a coronal obstruent
a va d rk
w n d lp
b lieverd darling li v rt
anders di erent n d rs
vervelend b oring v r ve l nt
Since coronal obstruents at word edge are arguably outside the Proso dic
Word the Proso dic Words in b end in schwa plus one consonant a
shows that after a schwa headed syllable a degenerate syllable is not al
lowed
Secondly the consonantinarhyme b efore a coronal can only b e a
sonorant Compare the words in b to the ones in a with schwa
plus a non coronal obstruent b elow and the ones with normal tense and
lax vowels in b
a a d pt
o b e l ks
k ta r kt
OK
b a d pt adept
OK
n ks nothing
OK
k ta r kt cataract
OK
haks squared
OK
feks shrew
If we assume prop erty is true this gap can b e explained
Inow turn to the words where schwa is followed by one consonant and
no coronal Prop erty claims that this consonant can only b e a sonorant
or a coronal obstruent which is not incorp orated into the proso dic word
but Kager and Zonneveld claim that virtually anything can follow
schwa They give the following list
Native words ending in t are hard to nd if any exist at all The ones listed are
the only ones I could nd and they almost all seem English loans I have no explanation for this gap
Prop erties of u schwa
n m s
wapen weap on bezem bro om gratis free
baken b eacon bodem b ottom vonnis judgement
zegen blessing bliksem lightning kermis fun fair
oefen exercise goochem clever dreumes to ddler
adem breath stennis trouble
k uw
havik hawk schwaduw shadow wereld world
monnik monk zwaluw swallow arend eagle
perzik p each zenuw nerve mosterd mustard
hinnik neigh peluw pillow mieters sup er
t f p l
lemmet blade tinnef garbage dubbel double
pocket pap erback sherri id koppel couple
racket id hennep hemp wikkel ller
ticket id tokkel pluck
There are on rst sight three classes of problems in this table for our pro
p osal Those are the words ending in k in p and in f resp ectivel y
Words ending in f and p are very rare The words ending in f apart
from the two mentioned by Kager and Zonneveld I have only found
gannef swindler are loan words from Yiddish or English Words ending
in p are p erhaps even more marked The only examples other than the
one Kager and Zonneveld mention are Wezep and Gennep b oth
names of towns I will put them apart for this moment only to brie y
return to them b elo w p
So the real problem is the relatively large class of wordsendingin
k and a class of words not mentioned by Kager and Zonneveld
probably b ecause it is hard to nd relevantwords that are uncontroversially
monomorphemic viz the class of words ending in x
x
gezel lig cosy z l x
I will not discuss the uw cluster here b ecause the problems related to this cluster are
not central to my present account It is included for the sake of completeness Trommelen
has argued that unstressed uw should b e derived from underlying w This
theory has some attractive prop erties which will b e brie y discussed in chapter
Apart from the words listed in this table there also is a large group of adjectives
ending in the denominal derivational su x lijk l k eerlijk erl k honest litt honour
ly heerlijk herl k delicious litt lord like gevaarlijk varl k dangerous litt danger ly
Prop erties of u schwa
keurig neat k qr x
heftig heavy h ft x
overhandig hand over ov rh nd x
korzelig cantankerous k rz l x
Most of these words ending in ig are adjectives although there are some
exceptional verbal forms suchasoverhandig probably historically derived
with the adjectivizing su x x which is still pro ductiveinmod
ern Dutch This etymology is now opaque b ecause the stems of the forms
in are no longer available as indep endent elements
I prop ose that in all these cases the schwa is not underlying but is
an instance of r schwa the result of a reduction pro cess applying to an
underlying b efore a dorsal obstruent in an unstressed syllable This
assumption is supp orted by the following facts of Dutch phonology
There are no Dutchwords ending with unstressed x or k on the
surface This can b e explained if there is a reduction pro cess applying
obligatorily Otherwise it is an unexplained gap in the system
In Dutch there are a few allomorphs of the agentive su x most im
p ortantly r and aar The morphological rule governing the choice
between the allomorphs considers the preceding syllable if this con
tains schwa the choice will b e aar otherwise it will b e r Compare
for instance wand l aar wand l r wanderer luist r aar
luist r r listener beoef n aar beoef n r practitioner
OK OK
with spr e k aar spr e k r sp eaker and bez t aar bez t r
o ccupant etc
r pre The words in k and g all select r instead of aar predik
dikaar preacher punnik r punnikaar someb o dy who knits with
There is one recalcitrantword havik hawk which seems to b e pronounced by some
sp eakers with in spite of stress on the rst syllable Zonneveld suggests that
the actual reduction rule should refer to the quality or length of the preceding vowel
reduction of p ost tonic short frontvowels would only apply when the immediately
preceding syllable is short
Yet the few Dutchverbs ending in m are problematic for this prop erty b ecause
they select r instead of aar ad m r someone who breaths bez m r someone
who uses a bro om sweep er
There also are some forms in n r etc but they have to b e analysed as lexical
exceptions For instance the word opener do es not just mean someone who op ens
but like its English alternative has the instrumental reading the instrumentby which
one can op en something This instrumental reading is not available for wandelaar
luisteraar beoefenaar etc
One nal remark is that agentive nouns formed on the basis of some stems ending
r schild r aar schild r r painter in r are somewhat irregular schild
timm r man timm r aar timm r r carp enter
Prop erties of u schwa
a sp o olknitter zaniker zanikaar someb o dy who nags frunnik r
frunnikaar ddler
If an adjective ends in a derivational su x r or n there are no
su xes m or l it cannot get overt in ection cf
het mo oi mo oi huis
the b eautiful agr house
the b eautiful house
het hout n hout n huis
the wood aff agr house
the wo o den house
de Amsterdam r Amsterdam r meiden
the Amsterdam aff agr girls
the girls from Amsterdam
The adjectivizing su x ig x do es not pattern with the schw asuf
xes nor do es lijk l k
a het baas x baas x meisje
the boss aff agr girl
the b ossy girl
b het eer l k eer l k meisje
the honour ly agr girl
the honest girl
Again this b ehaviour can b e explained if one assumes that ig has
the underlying structure not and similarily that lijk
l k not l k
The Dutch rst p erson singular pronoun ik uncontroversially has an
underlying if it is stressed In clitic p osition however this
can b e reduced to schwa
In very formal styles of sp eech the vowel in words as aardig nice
and monnik monk surfaces as I have to admit that the relevant
style of sp eechisvery formal and therefore I am not sure that this
pronunciation is not to b e attributed to in uence of sp elling
In some pragmatic contexts one can nd het bazig meisje and het eerlijk meisje
indeed In ectionless adjectives give some sp ecialized meaning to the DP Crucial is that
one can never say de opene deur
Prop erties of u schwa
Morphological comp ounding is a pro ductive pro cess in Dutch Nor
mally one of the segments s or is inserted b etween the two
memb ers of the comp ound Historically thismight b e related to the
fact that b oth segments also served as plural in ection and genitive
case but this seems no longer relevant in the contemp orary morpho
logical system Trommelen and Zonneveld
The choice b etween s and null is very complicated One of the
most imp ortant factors undoubtably is the phonological make up of
the rst element If this wordendsinavowel consonant group the
normal choice seems to b e a Dutch sp elling conventions re
quire that the schwa b e sometimes written as en If the rst element
of the comp ound ends in schwa consonantwe never nd as the
intermediary segment Most of the time they shownointermediary
segment at all b The only exception is the agentivesu x r
which combines with s c
Nound ending in k there are no nouns in x like monnik demand
a as the intermediary segment the signs denote morpheme
b oundaries just like most other p erson nouns ending in a syllable
headed by a full vowel
a mannenwerk m n w rk men s work
zonneschijn z n sx in sunshine
schapevlees sxap vles litt sheep s meat
mutton
b kleuterwerk kl t r w rk to ddler s work
heuvelrug h v l ryx chain of hills
bezemsteel b ez m stel bro omstick
havenloods hav n lots harb our pilot
c schilderswerk sx ld r s w rk painter s work
d monnikenwerk m n k w rk litt monk s work
tedious drudgery
This again shows that words ending in k pattern with other words
ending in full vowel plus consonant not with words ending in schwa
plus consonant
I conclude that there is evidence for a reduction from to inDutch
It probably is not a coincidence that the underlying lax vowel that reduces
so easily is In the previous chapter I have already argued that is
the lax counterpart of e and wehave seen in this chapter that e is the
vowel that is most susceptible to reduction
One could observe that most of these forms have a simple onset just like forms with
a real schwa This is unexplained by the reduction analysis since Dutchvowel reduction
Prop erties of u schwa
Ihave no explanation why exactly the dorsal obstruents trigger this
pro cess in so many cases Mayb e this is only a historical coincidence We
can analyse the words hennep gannef and even tennis and others in the
same way The tests listed in the main texts seem to p oint in this direction
wherever they apply For instance we nd tennis r tennis player in stead
of tennisaar Finally pronunciations like tenn s sherr f and henn p
seem to b e marked but not imp ossible in contrast to forms suchas ad m
breath ad r vein and op n id According to this criterion all
schwas b efore an obstruent are therefore r schwas
Degenerate and schwa headed syllables Inowhave explained
why all the cases of surface schwa obstruent sequences have to b e treated
in a way di erent from schwa sonorant sequences The former are cases of
reduction the latter can b e real cases of u schwa
Only the former are to b e seen as really unproblematically o ccurring at
the surface From this it follows that every theory of Dutchschwa should
trytogive an explanation for prop erty
The rst solution that comes to mind is that these word nal schwas
may b e followed by a degenerate syllable But then new problems arise
First why should this degenerate syllable only contain sonorants The
same is certainly not true for degenerate syllables following other vowels
Second why is the same prop erty also attested in non nal syllables cf
v r lies lose v k bies Third why are there no words like the following
vad rk vog lp kon k
These facts seem to indicate that the following prop erty holds for u schwa
Prop erty U schwa cannot b e fol lowedbya degenerate syl lable
This prop erty has to b e explained rst Why do syllables headed byu
schwa and degenerate syllables exclude each other I want to relate this
to the fact that in monomorphemic forms we do not nd sequences of
seems to b e hardly blo cked by complex onsets at all On the other hand since almost
all these forms are historically derivations with x I b elieve that the complex onset
test cannot b e used as a go o d test
Another p ossibility is that the schwainvolved actually is e schwa A relatively large
numb er of these words have a nasal b efore the schwawhich is not homorganic with the
obstruentfollowing it It can b e that in one way or another these words have underlying
nasal obstruent clusters which are marked for not b eing homorganic and that ep enthesis
applies almost obligatorily in these cases It remains a puzzle under that analysis why
h n p sounds more acceptable than r m I will not explore this p ossibility further here
Prop erties of u schwa
schwa headed syllables Forms suchas kab m l or kant r l are
absent from the Dutch lexicon
Ihave already argued that the favoured metrical p osition of schwais
in the weak p osition of a fo ot Although this has b een left implicit in the
discussion of the degenerate syllable in chapter the same should b e said
for this constituentaswell Word nal sup erheavy syllables almost always
attract stress as in specul a as biscuit emigr ant id just like syllables
followed byschwa Stress also falls normally on the syllable just b efore
schwa as in oranje orange
In the analysis presented here sup erheavy syllables as such do not ex
ist Sequences like laas in speculaas or grant in emigrant are analysed as
sequence bisyllabic units sp e ky la s e mi r n t parallel to the ranje
of oranje o r n j
Since the Dutch fo ot is left headed and since schwa headed syllables and
degenerate syllables prefer to b e in the weak p osition of a fo ot rather than
b eing unfo oted this probably means that a fo ot is built on the last full
syllable and the degenerate syllable in the case of so called sup erheavies
I assume therefore that b oth schwa syllables and degenerate syllables
have to b e in the weak p osition of a fo ot as a result of the following
instance of Project Ft
Project Ft V N is the head of a branching fo ot N domi
nates a vo calic ro ot
By logical transp osition Project Ft V is equivalent to the statement
N does not dominate vocalic features Nisintheweak position of a
branching foot In other words schwa has to o ccur in the weak p osition of
a fo ot
This constraint is the reverse of Project V Ft that I have discussed
in the previous chapter I assume it is inviolable at least in the lexical
phonology of Dutch Therefore every syllable that do es not have a normal
head i e either a schwa syllable a or a degenerate syllable b can
only o ccur in the weak p osition of a fo ot
Sequences of degenerate syllables are also absentbutwehavegiven an indep endent
explanation for this fact in chapter Furthermore descriptively sequences of a degen
erate syllable followed byaschwa syllable if the consonants in this schwa syllable are
coronal are also found e g aarde earth a r d I will return to these latter cases
b elow
Apparent exceptions to this generalization forms in which degenerate syllables o ccur
word internally or in which they are p eripheral but do not receive primary stress very
often include a high vowel which means they may b e analysed in a di erentway namely
as one exceptionally closed syllable not followed by a degenerate syllable as wehave seen
in chapter
Prop erties of u schwa
N b N a
N N
k k
This combined with the assumption that Dutch feet are tro chaic and max
imally binary Kager gives us the desired result Two incomplete
syllables cannot o ccur next to each other b ecause in that case they would
b oth have to app ear in the weak p osition of a binary fo ot Which is imp os
sible by de nition
If sequences of a schwa syllable plus a degenerate syllable are not al
lowed we still have to nd the reason why sonorants can follow u schwa
My hyp othesis is that in these cases the schwa is not the actual head of
the syllable It mayeven not b e present in the relevant part of phonology
at all Iwanttocontend that in the lexicon the level relevant for the
present discussion the nal syllable of theater id te ja t r is tr with
r the head
N
N
t r
The schwa is later ep enthesized in a way to b e discussed b elow An em
pirical argument for this analysis may come from level I non Germanic
su xes such as adjectivizing aal when combined with theatr this su x
gives tejatral not tejat ral Bo oij De Haas and Trommelen
Similarily we nd pairs suchas lter id N f lt r ltreer
to lter f ltr qr and monster id m nst r monstrueus monstrous
m nstry s On the other hand this criterion is not watertight since
there are many other pairs in whichtheschwa either is not deleted at
all or otherwise changes into a full vowel b efore exactly the same suf
xes puber adolescent pyb r puberaal like an adolescent pub ral
l apostolaat ap ostolate ap stolat ether pubral apostel id ap st
id et r etherisch ethereal eteris etc The morphology phonology
interface cannot b e sub ject of the present study
The reason why it is exactly the class of sonorants that is allowed to
followschwacannow b e understo o d Many languages that allow syllabic
All vowel initial derivational su xes are level I except for adjectivizing achtig
which is the only bisyllabic su x without schwa
Prop erties of u schwa
consonants restrict this set to the class of sonorant consonants The
same assumption could b e made for the lexical phonology of Dutch
Obligatory versus optional ep enthesis This leaves us with
the following two questions
i why is there obligatory ep enthesis at all
ii why are the sonorants after ep enthesis lo cated in the rhyme
b r rather than in the onset br We will see in chapter
that in French the latter option is chosen rather than the rst one
The rst question has already b een answered in the discussion of e schwa
There we assumed that at the p ost lexical level a constraint holds which
states that syllables havevo calic heads cf example on page
Clearly vo calic heads are missing from syllables with consonantal heads
as well as from degenerate syllables without anyheadatall Both are
therefore predicted to undergo ep enthesis in the p ost lexical phonology
Apotential problem is that there are very formal styles of sp eechwhere
no ep enthesis applies in degenerate syllables like help Even in those styles
of sp eech theater is still pronounced tejat r never as tejatr There is
a and ep enthetic u schwa a di erence b etween e schw
Wethus havetodraw a distinction b etween syllables without a head
which are allowed in some styles of sp eech on the one hand and sylla
bles with a consonantal head whicharenever allowed to surface from the
p ostlexical phonology
This is the p ointtointro duce the rst weakness constraintofthis
thesis The reader might recall that weakness constraints are formally the
opp osite of pro jection constraints The latter connect certain features to
head p ositions whereas the former connect certain features to dependent
p ositions
In chapter I will discuss a weakness constraint related to the feature
high In this case I prop ose that we can make the relevant distinction
between degenerate syllables and syllables headed by a sonorant consonant
with a weakness constraint which refers to the feature consonantal
Weak cons N A segmentSbears cons S app ears in the
dep endent p osition of N
Some Eastern Dutch dialects allow nasal consonants to stay syllabic also in the p ost
lexical phonology Most German dialects also allow syllabic liquids in p ositions which
corresp ond to r sequences in Standard Dutch
A similar distinction is drawn b etween Peak constraints and Margin constraints in
Prince and Smolensky Yet the theory presented there crucially hinges on the
notion of an intrinsinc sonority hierarchy whereas the one here do es not
Prop erties of u schwa
A syllable with a consonantal head violates but a syllable without a
head do es not If wenow assume the following ranking
N cons formal style Project N cons cons Weak cons
informal style
The weakness constraint dominates the constraint against insertion of schwa
i e a bare vo calic ro ot no de cons in all styles of sp eech It is therefore
always b etter to insert a schwa in syllables with consonantal heads For
degenerate syllables only the constraint Project N cons is relevant
and the p osition of this constraint with resp ect to the faithfulness constraint
cons is variable across style registers as wehave seen ab ove
Wethus have to conclude that the p ositioning of faithfulness constraints
in the hierarchy is not totally free Apparently wedohave to p osit some
b oundary on their p osition in this particular case at least an upp er b ound
ary for cons b elow Weak cons N
Again on complex onsets The answer of the second question
in lies in the interaction of several constraints we discussed b efore
Let us lo ok at the two p ossible outcomes of schwa ep enthesis into a head
p osition of tr
a N b N
N N
N N
t r t r
Given general conditions on syllable structure e g sonority restrictions
in the co da these are the only p ossible ways of providing a vo calic head
The reason why a should b e chosen at least at the end of the word
rather than b is clear bynow It is the same reason whyanytyp e of
non underlying schwa is forbidden to arise at absolute word edge Wehave
The need for another theoretical subtlety arises in the analysis of Dutch dialects
sp oken in northern and eastern parts of the Netherlands hinted in the previous subsec
tion In these dialects schwaepenthesis is obligatory in syllables headed by a liquid i e
r or l but it is blo cked if the head is a nasal i e m n or Nijen Twilhaar
These dialects thus have syllabic nasals on the surface but no syllabic liquids
The reason why nasals in some grammars are more acceptable than liquids despite there
lower level of sonority in other asp ects of syllable structure is an interesting issue not to b e explored here
Prop erties of u schwa
already seen that this is true for r schwa and e schwa It should b e true for
the particular ep enthetic schwa in tr typ e syllables as well The reason
for this was the constraint Align
This constraint do es not apply to non nal tr typ e syllables so that
in these cases other constraints b ecome relevantinthechoice b etween a
and b One imp ortant condition is of course Connect N lax Since
schwa do es not b ear any features it also do es not b ear the feature lax
Therefore its o ccurrence in a closed syllable is disallowed in the lexical
phonology and the representation in b should b e preferred
Yet the scarce data that wehave seem to p oint in a di erent direction
The reader might recall that some sp eakers of Dutchsay p rsis for precies
precise normally pronounced as pr sis I already argued ab ovethatit
might b e that in this case an original r schwa alternating with presis
is reanalysed as u schwa This would corresp ond to a change in syllable
p osition All schwas preceded by a complex onset seem to b e r schwa
The only other piece of evidence I have b een able to lo cate is that there
is a pre x ver of the relevant structure There are no pre xes starting
with a complex onset and schwa The string v r furthermore o ccurs in
anumber of words in which synchronically it no longer functions as a
pre x verstand exp ertise verlies lose verdriet grief
In so far as we are allowed to make generalisations over this very small
set of evidence we seem to get a rather than b There should
therefore b e another factor favouring the r in the co da over the r in the
onset Recently some researc hers have suggested that complex onsets
p erhaps have to b e licensed by a fo ot in a certain way Davis Van der
Hulst a Takahashi If this is true and if furthermore it is true
that schwa cannot pro ject its own fo ot this could explain whyword nal
and word medial word initial schwa syllables b ehave the same with resp ect
to syllable structure However the data remains scarce and therefore it
might b e to o early to draw de nite conclusions
Another question that now arises is whywe never nd syllables with
both a complex onset and a sonorantcoda In other words wehaveto
exclude the syllable typ e in b elow b oth word nally katr l and in
other p ositions tr lzir
Prop erties of u schwa
N
N
N
t r l
As far as I know there is not one single exception to this generalisation
neither word internally nor word nally Since degenerate syllables and
syllables headed by a real underlying schwaalsodonothave complex
onsets wemayformulate the generalisation that syllables without vo calic
features do not allow more than one branching N
Let us lo ok more closely at the syllable structure used in this dissertation
o designates the terminal no de of syllable structure the ro ot no de of
feature trees
N
N
N
N
o o o o
Nisdepen Wehave seen that the branchingness or non branchingness of
dent on a quality of the head viz the presence or absence of lax We
have also seen that vo calic features play a role in proso dic structure higher
than the syllable where total absence of vo calic place and height features
implies a weak p osition in a fo ot and vice versa Wehaveseenthateven
the internal structure of the N might b e dep endentonvo calic features It
seems only natural then to assume that vo calic features also playarole
in the branchingness of the highest level of syllabic X bar structure N
Because all full vowels allow complex onsets wemay assume the require
ment on the branching of Nisvery similar to the one on the branching of
higher order structure anyvo calic feature of the set f high low lax
coronal labial dorsal g is su cient but at least one of these features
Prop erties of u schwa
N We can therefore use the constraint is necessary to license a branching
Project N V encountered at various places ab ove
This is one of the twoversions of Project N that we encounter The
other one was Project N cons the constraint forcing at least some
vowel to b e the head of a syllable even if this vowel is schwa
To complete the set of p ossible Project N constraints I can mention
Project N F This constraintwould require syllables to have at least
some feature material in their head Since V Fand V cons ab
stracting away from secondary articulation on consonants for a moment
on which see chapter these three constraintsareinaP aninian relation
If a syllable lacks features from F initshead italsolacks the feature
cons or any feature from V Similarily if a syllable has a head of which
therootisnot cons this head also has no vo calic features from V
The three constraints thus corresp ond to the following markedness hi
erarchy
Candidates Project F Project cons Project V
no segmental head
p
consonantal head
p p
schwahead
p p p
full vo calic head
Ihave found no strong evidence for p ositing Project F when wehave
already p osited the other two constraints Wehaveseenabove that the real
distinction b etween degenerate syllables and syllables with a consonantal
N More head has to b e made byaweakness constraint Weak cons
discussion on this problem will b e provided in chapter
There are two N no des in We exp ect there to b e no di erence
between the higher and the lower syllable no de in If the lower
instance of Nisnotallowed to branch the higher instance similarily should
not b e allowed to do so Strictly sp eaking this means that schwa syllables
should not b e allowed to have an onset at all
This is not in accordance with the facts wehave found Schwa syllables
allow for onsets but not for complex onsets A lot of the argumentation
ab ove is based on this assumption
There is one constraint whichisrelevant for our present concerns
This is the constraint Onset rememb er that this constraint can also b e
The theory presented here is crucially based on the absence of an onset constituent
Onset based theories that do refer to notions similar to the ones here head branching
etc are presented byVan Lit and Charette
Prop erties of u schwa
formulated as a pro jection constraint as shown in chapter A syllable
wants to have a sp eci er an onset segment regardless of the feature
structure of its head Wehave already seen at various o ccasions ab ove
that this wish for an onset is very strong in Dutch In chapter we
will see that Onset can force a syllable to cross morpheme b oundaries in
many languages Supp ose wehave an underlying p otential syllable tr
Supp ose furthermore that Onset dominates Project N V inDutch
Four parses are relevant
Candidates Onset Project N V
tr
p
t r
p
t r
tr
If we leave b oth consonants unparsed we create an onset less syllable Al
though this is exactly the structure Project N V favours it is ruled out
b ecause it do es not meet Onset On the other hand if we parse b oth
consonants this will certainly satisfy Onset Butwe violate Project N
V twice unnecessarily so b ecause derivations where one consonantisin
the Onset and one is unparsed hence with one branching N only do also
exist Which of the latter two t or r willeventually b e chosen has
to b e decided by indep endent constraints for instance constraints having
to do with sonority slop es or Contiguity of parsing McCarthy and Prince
b Word initially Align will probably cho ose the rst consonant i e
in this case the t
N Taken together the three pro jection constraints Onset Connect
lax and Project N V have as an e ect that syllables headed byu schwa
or a sonorant consonant are core syllables i e syllables consisting of a head
preceded by one single consonant This typ e of syllable seems universally
unmarked It is interesting to see that in Dutch the segments with sub
minimal vo calic quality can only head this minimal typ e of syllable
Given the curious fact that no words exists starting with a sonorant schwa
m r etc wemay also b e led to conclude that in the cases at hand
t will b e chosen as the prop er output On the other hand as I have p ointed out
ab ove it is di cult to draw conclusions from the absence or presence of schwa syllables
with a certain onset given the overall scarcityofsuch syllables
In the previous chapter wehave seen that reduction in Dutch is not blo cked by
complex onsets b ecause reduction is p ostlexical There is at least one language how
ever where reduction schwa seems to b e sub ject to the same conditioning as Dutch
u schwa This language is Macushi Carib Hawkins Kenstowicz and Kisseb erth
Prop erties of u schwa
Schwa surrounded by identical consonants Schwacanthus
b e followed by sonorant consonants in Dutch It also can b e preceded by
every consonant except for h and the glottal stop Yet the following
sequences seem never to o ccur monomorphemically
m m r r l l
It is not true for instance that schwa cannot b e surrounded by consonants
b ecause other combinations do o ccur
r l r n r m r r
kerel guy toren tower harem id
korrel p ellet koren corn
m l m n m m m r
stamel stammer samen together marmer marble
rammel rattle amen id emmer bucket
n l n n n m n r
ponem face toner id
panel id binnen inside
l l l n l m l r
molen mill golem id trailer id
propel ler id
Only words ending in C m are hard to nd but this maysimply
son
re ect the fact that m sequences in general are rare Furthermore for
r m and l m sequences after lax vowels this probably has to b e related
to the complementary distribution of u schwa and e schwa section
More in general we can thus observe the following prop ertyofschwa
Prop erty C C theconsonants preceding and fol lowing schwa may
i i
not be the same exceptwhenC n
i
Both in derived and underived environments wedo nd n n sequences
n serves among other things as the most common pluralitymarker on
nouns binnen inside boon en b eans baan en jobs etc On the
other hand r r clusters even seem imp ossible to derive in those con
e morpheme which normally shows up texts For instance the comparativ
as r kwaad er angrier takes a der allomorph when combining with
Normally words in Macushi show an alternating pattern of schwas and full vowels
starting with schwa w nam ri mirror wan mar r mu mirror but this pattern
can b e disturb ed by closed syllables which do not reduce as well as syllables with com
plex onsets which also do not reduce Thus we nd sibm rik p e little now instead of
s bmir kip b ecause the rst syllable is closed and krat pe alligator now in stead of
kr tup b ecause the rst syllable has a complex onset
Prop erties of u schwa
a stem ending in r zwaar der zwaar er heavier There are no suf
xes l m so that we cannot test the validity of prop erty for
p olymorphemic domains with other consonants
In chapter we will see that prop erty is also attested for Norwegian
p olymorphemic words Now the explanation for prop erty isactually
quite straightforward under the theory defended here Since schwaisas
sumed to b e inserted only p ostlexicall y in these cases all these sequences
would lexically have the form C C with the two consonants in the onset
i i
and the nucleus of the syllable resp ectively Wehave seen go o d evidence
in chapter and also earlier in this chapter that there has to b e a contour
between onset and nucleus This explains whywe do not nd ji or h
sequences and now it can als help us explain the blo cking of ll rr mm
etc
As for the derived forms it is actually unclear to me why n n se
quences are freely allowed whereas r r are blo cked Wehave seen that
also in underived words we do nd n n sequences What should make the
coronal nasal so sp ecial is at presentamystery to me
Schwa after ng The next prop ertymay at rst sightseemto
b e a problem for the p oint of view just outlined
Prop erty mayoccur before schwa as wel l as syl lable nal ly whe
reas it is excludedbefore ful l vowels
This prop erty is illustrated in the following example
a V b c
o ra l
b o
As is shown by the example b o this pattern cannot b e reduced to a
constraintonwell formed feet This prop erty seems a rather strong argu
ment for the No Syllable Theory The distribution of is probably to b e
explained by a lter banning dorsal nasals from the onset p ositon This
prop ertysho ws that the schwainengel is the only vowel that do es not
attract this consonantinto its onset All other vowels force this consonant
in a p osition where it is banned by indep endent constraints
I assume that the su x ing in for instance verdring ing suppresion v rdr
litt surpress ing is underlyingly where the is optionally reduced to schwa as
it is b efore other dorsal consonants
Notice that the theory presented here is not compatible with an analysis of as
the placeless nasal as defended in Trigo If where a placeless ro ot wewould
exp ect it to avoidbeingintheonsetofschwa just like h and b
Prop erties of u schwa
This would explain the non existence of a ora the cannot b e in
the co da probably b ecause of the Syllable Contact Law but it also cannot
app ear in the onset If in a l the schwa has no syllable it do es not force
into its onset
Any theory contesting the No Syllable Theory will havetoprovide for
a explanation of this fact A rst observation we might makeisthat is
only found if the head of the syllable is a sonorant consonant more precisely
a liquid or n I have found no instances of plus real u schwa while
the numb er of examples with consonantal heads esp ecially l is su ciently
large
angel sting engel angel zwanger pregnant
honger hunger wingerd vineyard Engels English
tenger slender klungel bungler pingel higgle
The also has to b e preceded byalaxvowel aangel eengel etc
which probably is to b e explained by the fact that it can only o ccur in
co das I supp ose that the pattern therefore is the result of some constraint
blo cking to o ccur to the left of vowels Lacking any evidence on what
the form and nature of this constraint could b e however I will leavethis
matter op en for future research
The reasons whye schwa and r schwadonothaveproperty on the
other hand is quite clear Ep enthesis cannot apply after b ecause
can b e followed only by coronal obstruents or by consonants that are
homorganic to i e dorsals But wehaveseenthatepenthesis never
applies directly b efore coronal obstruents i e at the end of the proso dic
word or in homorganic clusters Reduction on the other hand cannot
happ en to full vowels in V clusters b ecause as wehave just seen
full vowels simply are never allowed to o ccur in these p ositions lexically
Postlexical reduction to schwaafter is thus bled by the lexical constraint
p osited here
U schwa and stress The next two prop erties of schwa are
related to the stress system
Prop erty Schwa is stressless
There are a few names ending in like Inge Slinge etc There are not
manywords ending in an n or ns jongen b oy tangens tangent and a few other
Ihave found no words in m or Many in ectional su xes start with schwa
and combine freely with words ending in showing that the lter s do not prohibit sequences crossing morpheme b oundaries
Prop erties of u schwa
Prop erty When schwa is the nal vowel of the word
a If schwa is preceded by one or moreconsonants stress is on the penul
timate syl lable
b If schwa is not precededbyconsonants stress is either on the penultimate
or on the antepenultimate syl lable
As we will see in the next chapter the stressles sne ss of schwa prop erty
is a recurring feature of languages that include schwa in their vowel system
Wehave also seen that it is a prop ertyofDutchr schwa and e schwa as
well
Prop erty is less widespread cross linguistic all y Together with prop
erty itsays that schwaisinvisible for stress In a way a syllable headed
byschwa b ehaves as an extrametrical syllable Normally stress in Dutch
is on the nal or pre nal syllable If the nal vowel of the word is schwa
however stress falls on the pre nal or the antep enult It is sometimes said
that schwa is stress attracting
The rst clause of this prop erty has already b een explained Project Ft
V required stressless schwatobeintheweak p osition of a binary fo ot This
implies the syllable immediately preceding schwa should b e in the strong
p osition of the fo ot It gets stress and since primary stress in Dutch is gen
erally on the nal fo ot this means it gets primary stress A schwa headed
syllable b ehaves like a degenerate syllable in this resp ect
The second clause is somewhat more problematic It describ es the
stress b ehaviour of words like Belgi e Belgium b l i j Itali e Italy
itali j Sicili e Kirgizi e Alexandri e al ks ndr j Veluwe v elyw
Betuwe b etyw the last two are names of regions in the Netherlands
er patrician patr sij r agrari er farmer a r arij r patrici
We do not exp ect there to b e a di erence b etween schwa preceded bya
closed or by an op en syllable In b oth cases the schwa should b e forced into
the weak p osition of a fo ot i e we exp ect all the words just mentioned to
b ehaveasAleksandri e But in fact the latter word b ehaves as exceptional
and is itself sub ject to regularization towards Aleks andri e in the sp eechof
manyspeakers Something seems problematic with these examples
A few things are o dd ab out the examples mentioned All relevant cases
haveahighvowel The reason for this is not very clear It cannot b e
b ecause these vowels are the source of glide insertion b ecause in p olymor
phemic forms glides are also pro duced after mid vowels zee en sea pl
ze j n reu en male dog pl r j n Furthermore all these examples
are either top onyms viz those ending in or names of p ersons viz those
ending in r The adjectival variant of these top onyms do es not include a
sc hwa Belgisch b l is Kirgizisch k r izis Veluws vely w s It might
Iamaware of one non top onymic example ending in weduwe female widow
Prop erties of u schwa
therefore b e that these nouns are morphologically complex formed bya
ro ot ending in i and a stress neutral su x consisting of schwaoreven
j The corresp onding adjectives then consist of the same ro ot plus s
We nd a few forms not ending in ij thatmight give some evidence for
this Spanje Spaans Spain Spanish spans sp nj As in many languages
the top onymic stratum is however full of exceptions for instance twoad
jectives corresp ond to Itali e Italisch Italic and Italiaans Italian which
makes it di cult to test this hyp othesis
Adjacency b etween schwa and full vowels Prop erty
again is found in Dutchaswell as in other languages that haveschwa
Prop erty U schwa cannot occur next to a vowel
In my discussion I want to distinguish b etween two cases viz prevocalic
and p ostvo calic schwa I will discuss these cases in this order
In monomorphemic forms we cannot see what happ ens if V
sequences are input to the grammar Such a sequence do es not surface
but there is no way to observe whether this is the result of schwa deletion
consonantepenthesis or some other means
We can directly see what happ ens if we add a su x starting with a full
vowel to a stem ending in u schwa One example stem is elite id and an
example su x is adjectivizing air qr The resulting adjectiveiselitair
e li t qr not e li t qr The schwa gets deleted or remains unparsed
I assume that after su xation wehavetwo relevant candidates e li t qr
where the a x forms an onsetless syllable and e li t qr where schwais
deleted and the t is parsed into the onset of the su x syllable The
w eduw In the rst place it is not clear whether this word is not morphologically
complex cf weduwnaar widower weduwnar In the second place it mightbethat
the second syllable of this word contains a schwa itself at some level of abstraction
Trommelen See Zonneveld for a line of reasoning similar to the one
presented here
Ihave also found a form ending in i r which do es not denote a p erson viz terri er
id
Dutch shows an asymmetry b etween pre xes and su xes in that the former but
generally not the latter form a syllabi cation domain on its own syllable b oundaries
do not generally cross the b oundary b etween pre x and stem whereas it do es cross the
b oundary b etween stem and su x although there is a class of su xes which also form
their own domain One e ect of this is that across the pre x stem b oundary wedo nd
V sequences e g ge eerd honoured qrt I refer to Bo oij
Langeweg and Van Oostendorp for discussion
Of course there are in nitely many other p ossibilities like leaving the full vowel of
the su x unparsed or inserting a glottal stop I assume these p ossibilities are ruled
out by constraints on the insertion and deletion of vo calic and consonantal material
resp ectively
Prop erties of u schwa
rst form violates Onset the second violates the requirement to parse
schwa i e the feature cons
Wehave already seen that Onset is virtually inviolable in Dutch We
may therefore assume the following hierarchy
N A vo calic cons ro ot should b e the head Project cons
of a syllable
Onset Project cons N Parse V
This gives us the following tableau for elite air
Candidates Onset Project cons N
e li t qr
p
e li t qr
It is b etter to ignore the emptyrootoftheschwa than to create an onsetless
syllable or a sequence of empty ro ots Since stem su x combinations seem
to b ehaveinidentical ways with resp ect to syllabi cation as monomor
phemic forms I assume that the b ehaviour of these two domains with re
sp ect to schwa deletion is also similar Also underlying V sequences
in monomorphemic forms will b e resolved by deletion of the schwa
I already discussed the most imp ortant reason for the ban on V
sequences in the chapter on e schwa and r schwa An empty syllable would
result in these cases i e a syllable with empty ro ots in b oth nucleus and
onset and insertion of a glottal stop do es not help b ecause the glottal stop
itself is an empty consonant Again it is di cult to test what happ ens to
underlying schwas that are underlyingly adjacenttovowels in monomor
phemic forms so that again wehave to take recourse to sequences arising
from su xation
The schwa seems to never surface in these cases but it is di cult to
nd really hard evidence for deletion of schwa There are three schwa initial
a xes an adjectival agreement marker r forming the comparative
on adjectives and n forming the plural of nouns and verbs As far as
bines freely with stems ending I can tell neither of these su xes ever com
in non stressed vowels they com bine with stems ending in stressed vowels
but these have b een analysed in chapter as ending in an empty consonant
underlyingly On the other hand there is no real evidence for schwa
deletion in these cases Often simply a di erent allomorph is chosen For
instance the plural su x after an unstressed vowel is neither n nor n
Ihave already argued that Project cons N has a similar e ect as Parse V
This consonantwould probably diasapp ear after su xation
Prop erties of u schwa
but s The plural of familie family fam li is fam lies not fam lie en
or fam lien cf genie genius z n genie en genies see Lieb er and
Bo oij for more discussion Similar patterns are to b e observed for
the other schwa initial su xes
Thus while the su xed forms do provide us with some evidence that
schwa cannot followfullvowels they do not inform us as to what exactly
happ ens phonologically when a V sequences arise they are a voided
by morphological means already
Complementary distribution of u schwa and e schwa The
next prop erty of u schwaIwant to discuss is prop erty
Prop erty U schwa and e schwa areincomplementary distribution
In all phonological contexts where wehaveu schwa we do not havee schwa
and vice versa As wehave seen ab ove ep enthesis applies b etween two
consonants if they are together at the end of a syllable and they do not
form a homorganic nasal obstruent sequence We can say that u schwa
o ccurs everywehere but there The question is whyu schwaisblocked in
the environments just mentioned
This prop erty is demonstrated in the following table
e schwa arm id rm r m
help id h lp h l p
kurk cork kyrk kyr k
u schwa al les all ls l s
ritme rhythm r tm r tm
ticket id t kt t k t
The correct generalisation seems to b e that we cannot haveu schwa i e
non alternating schwa b etween two consonants if these consonants could
b e the co da of a syllable and a following degenerate syllable
There are no words like arrem r m with a stable u schwa meaning
they could not have an alternative realization rm This fact may seem
There are a few monomorphemic forms of which the stem ends in a vowel which
use the form stem n to denote the plural and the in nitive gaa n go doe n do
zien see staa n stand The problem is however that these forms are monosyllabic
so that we do not exp ect these to need schwa deletion and furthermore that these are
strong verbs with a very exceptional paradigm also for other forms so that they cannot
b e used as evidence
Two names are problematic in this resp ect Wil lem William whichdoesnothave
apronunciation without a schwa w l m w lm and contrasts with lm whichhas
an optional e schwa f lm f l m and Col lem which similarily contrasts with olm elm
Prop erties of u schwa
quite curious but I b elieve the explanation b ehind this distribution is quite
straightforward Supp ose wehave an input r m
The rst two segments are parsed into one syllable What can wedo
with the other two They cannot form one closed syllable b ecause u schwa
is not allowed to o ccur in a closed syllable in the lexical phonology as
wehave seen in the previous section They also cannot each form their
own separate syllable b ecause this would involveaschwa syllable and a
degenerate syllable which are not allowed adjacent to one another b ecause
they b oth need to nd a place in a weak p osition of a fo ot
This means it is imp ossible to parse b oth segments One of the two
has to go There are two reasons why this should b e the schwa In the
rst place schwa is only a ro ot no de with one feature cons whichis
very easy to get rid of as wehave seen ab ove In the second place the
m is at the edge of the word whichwewanttokeep clear as muchas
p ossible b ecause of Align Theschwa do es not get parsed and m forms
a degenerate syllable with ambisyllabic r in its onset This means the
output of lexical phonology for r m will b e rm just as for rm
An underlying schwa has no e ect in these circumstances
Let us now turn to another pair in for a moment al les versus
als I assume the rst word has the underlying structure l s and the
second form has the underlying structure ls In b oth cases the s can
b e extraproso dic b eing a coronal obstruent This means that in the second
form there is no need for ep enthesis as wehave seen in the section on e
schwa In the rst form on the other hand the schwa stands at the end of
the proso dic word It can surface in a normal op en syllable with the again
ambisyllabic l in its onset There is no need to leave the underlying ro ot
no de unparsed Therefore the rst form will leave the lexical phonology as
l s and the second as ls The two forms are correctly predicted to b e
di erent
Schwa deletion To conclude this discussion of u schwa I have
to saya word ab out schwa deletion In some informal mo des of sp eech
schwa is deleted in some phonological environments This pro cess seems to
a ect all instances of schwa u schwa and r schwaalike if it applies at all
I assume e schwa is excluded from deletion b ecause it makes no sense to assume
aschwa is rst ep enthesized just to get deleted afterwards Since I assume that b oth
ep enthesis and deletion are p ostlexical phenomena and since the theory of derivations
used here is Optimality Theory in which they b oth are the result of output lters it is
not even p ossible to assume some vowel is rst ep enthesized and then deleted
I also have to remark that in the history of Dutch a rule of ap o cop e has applied
deleting some word nal schwas As a result of this we nd a few isolated cases of
alternation suchasHere heer lord That this is not related to the optional pro cess
discussed here is easily observable the rst form can only refer to the Lord in a religious
Prop erties of u schwa
In my description of the facts I partly follow Bo oij a but my
judgements of some of the facts are partly di erent Bo oij found the fol
lowing generalisation
Prop erty When a Dutch word has two consecutive syl lables with schwa
as their vowel the rst of these schwa s may be deleted
As wehave seen sequences of schwa syllables can only b e derived The
second schwa can b e in an a x or in an enclitic It do es not really seem
to matter however whether the rst or second schwa are u schwa e schwa
or r schwa The two considerations together lead us to the conclusion that
this pro cess is most probably p ost lexical
The context for the deleted schwamust b e obstruent liquid accord
ing to Bo oij a The second schwa can never b e deleted the nal n
in some of these forms is pronounced in some dialects but not in others
a tokkel en to pluck t k l n t kl n
b dobber en to oat d b r n d br n
c verbeter em correct him v rb et r m v rb etr m
d knu el en cuddle kn f l n kny n
e kietel en to tickle kit l n kitl n
Ihave to note here that there app ears to b e some disagreementonthese
data among nativespeakers of Dutch Although I have no problem in
accepting Bo oij s judgements mostofmy informants do not seem to agree
Henk van Riemsdijk p c notes that to his ear this typ e of elision sounds
more typical of p o etic language than of informal sp eech and Harry van der
Hulst p c feels that the liquid gets a syllabic head rather than part of a
complex onset I will nevertheless assume that the facts in are correct
Ben Hermans p c observes that in his dialect the elision is not p ossible
at all
In e we see that at this p ost lexical level of the derivation wecan
derive onsets which are not allowed at earlier stages There are no words
starting with tl
Bo oij observes that universally obstruent liquid clusters are the most
favoured typ e of onset In myspeech deletion is actually allowed b etween
anytwo consonants as long as the second is more sonorous than the rst
So I can delete a schwabet ween k n in teken en draw tek n n
tekn n or b etween m l in rammel en rattle r m l n r ml n
Tomy ear kietlen is in no way b etter than teeknen This probably means
sense whereas the second can also refer to a wordly lord or gentleman
Prop erties of u schwa
that in my grammar the notion p ossible onset is de ned even more lo osely
at the p ost lexical level than in Bo oij s grammar For me anything with a
rising sonority slop e is acceptable at this level
Bo oij a p oints out that it is not really necessary that the vowel
immediately following the deleted schwaisaschwa itself Full vowels can
also o ccur A more relevant generalisation seems to b e that schwadoes
not delete if the next vowel b ears stress geraamte skeleton ramt
beloven promise bloven beraad delib eration brat Some lexicalize d
cases are exceptions to this suchasgeloven b elieve lov n lov n
gereformeerd Calvinist ref rmert ref rmert terecht rightly t r xt
tr xt Wemay assume that these words have acquired a lexical entry in
which the rst schwa is no longer present cf also Standard German glauben
b elieve
Some sp eakers can also delete r schwas in syllables that are followed by
syllables with primary or secondary stress apparaat apparatus par at
p r at pr at kapelaan chaplain k pel an k p l an k pl an koppeling
clutch k p l k pl Again high frequency of words is a furthering
factor here and it might b e that these sp eakers have a separate entry for
these words
The generalisation I want to explain is thus the following
Prop erty Schwa can be deletedi itapp ears before a non stressed
vowel and between two consonants C C whereC has a lower sonority
i j i
than C
j
In the next chapter we will see that in Frenchvowel deletion is much more
common and much less restricted to certain styles of sp eech There I will
analyse the deletion in terms of a constraint Charette has termed
the Avoid Vowel Principle In the terminology used here the Avoid Vowel
Principle is a constraint against cons ro ots This constraintwillonly
a ect schwas if the Parse constraints on the vo calic features high low
lax coronal labial and dorsal are ranked ab oveit
I adopt Bo oij s a suggestion that the second restriction means
that schwa can only b e deleted when the resulting cluster is an onset to
the remaining vowel I cannot answer the question why the restrictions on
p ossible onsets are less strict at the p ost lexical level It seems generally to
b e the case that this typ e of restriction is lifted at later levels Borowsky
Mohanan
Consider the nal part of tokkel en The following two partial syllabi cations are p ossible
Prop erties of u schwa
a N N b N
N N
N N N
N
k l n k l n
In very formal styles of sp eech the rst form will b e chosen b ecause it con
tains more underlying material In less formal styles of sp eech faithfulness
b ecomes less relevant In particular the constraint Project cons N
forcing the surfacing of the feature cons is relatively lowranked as we
haveseeninvarious sections ab ove Other constraints b ecome more rele
vant such as the constraints against vo calic ro ots b is in this resp ect
more optimal than a b ecause it contains less vo calic ro ots i e less
schwas Wehave also already seen that the ranking of cons can vary
in the Dutch p ostlexical phonology corresp onding to style register
The reason whyschwa cannot delete b efore a stressed vowel is probably
to b e attributed to an integrity e ect Since Dutch has a tro chaic stress
system a pretonic schwa is separated from the following syllable by a left
fo ot b oundary It may b e that syllables cannot b e built over formerly
established fo ot b oundaries cf Rice
Conclusion Ihave tried to give a full accountofthebehaviour
of those instances of schwa that alternate neither with full vowels nor with
zero Although I found supp ort for the ep enthetic nature of u schwain
some cases namely when it is followed by a sonorant this is not neces
sarily true for all instances of u schwa Both sonorants and u schwacan
pro ject only a very limited typ e of syllable Full vowels can pro ject more
complex syllables obstruents cannot pro ject any syllable at all Neither
supp orts co das b ecause neither b ears a feature lax Neither supp orts
complex onsets b ecause Nnodeshave to b e supp orted byvo calic features
so any extra N no des apart from the one needed to satisfy Onset is
banned When ep enthetic schwa is inserted in a syllable headed by a sono
N is resp ected even if this means disresp ect for rant consonant this ban on
Connect N lax Thus p ostlexically wehave the ordering Project N
V Connect N lax
Like all other schwas u schwa can never o ccur in stressed p osition and
it even has a very strong preference for the weak p osition of a fo ot It
Prop erties of u schwa
seems to b e banned from other p ositions b oth the strong p osition of a fo ot
and the unfo oted p osition like a degenerate syllable I consider it one
of the more imp ortantinsights of recent theories of Dutchschwa like the
No Syllable Theory that they have drawn a parallel b etween degenerate
syllables and syllables headed byschwa This parallel is retained in the
prop osal made here
All these prop erties follow from the pro jection theory plus the assump
tion that schwa is totally empty As a consequence schwawillalways count
as a more marked syllable head than a full vowel b ecause it lacks a vo calic
no de but it will also count as less marked than a consonantal head or a
completely empty head
Since in the previous chapter I noted that it is attractive to take the
phonetic emptiness of the schwavowel as our starting p oint I will add a
note on the learnability of the system presented here We could assume
that initially all pro jection constraints are ranked very high Simplifying a
bit this means that the child exp ects to hear no schwa this b eing sucha
bad syllable head At a certain p oint she hears the word mode From this
word she can drawseveral conclusions i some faithfulness constraints the
constraints against insertion of features whichwould makethevowel a full
vowel and parsing constraints havetoberanked higher than Project N
V ii Onset Project N V She still do es not exp ect schwas with
complex onsets with co das or in stressed p osition Since she never encoun
ters suchwords and furthermore will never nd other evidence for a low
ranking of the relevant pro jection constraints she will continue to correctly
assume such forms do not exist
Assume furthermore a child encounters the word theater She then
knows that schwa can o ccur in closed syllables at least at some phonological
level If she has already encountered instances of e schwa ar m she
knows that it is most economic to assume that this level is the p ostlexical
N lax at that level anyway level b ecause she needs to rerank Connect
This leaves her with teatr for the lexical level Since complex onsets cannot
o ccur with schwa at that level they certainly also do not o ccur with empty
heads since b oth typ es of syllable violate Project N V in the same way
The child therefore adopts an analysis in which r is the nucleus of the
syllable The minimal change to get this would b e to assume that sonorants
can b e syllable heads Since the child never encounters any evidence that
also obstruents can b e such heads this will b e her nal conclusion
Previous analyses of u schwa
Previous analyses of u schwa
As I havementioned in the intro duction to this chapter Dutch u schwa
has b een the topic of several studies In this section I will discuss some of
these previous studies and compare them with my ndings I distinguish
roughly b etween three typ es of analysis Ep enthesis Theory according to
which u schwaisalways the result of ep enthesis Reduction Theory accord
ing to which u schwaisalways the result of reduction and the No Syllable
Theory according to which u schwa is underlying but for one reason or
another do es not pro ject a syllable
The Reduction Theory and the Ep enthesis Theory have the advantage
of global economy we indep endently need reduction and ep enthesis rules
to account for r schwa and e schwa resp ectively The problem is that we
nd u schwa in lo cations where neither r schwa or e schwa o ccur and vice
versa The No Syllable Theory has the advantage of making a lot of inter
esting and correct predictions But some of its predictons are wrong and
furthermore it has little or nothing to say ab out the similarities b etween
u schwaontheonehandande schwa and r schwa on the other
Some or mayb e most authors adopt mixed versions of these theories
For instance Zonneveld assumes that schwaisalways ep enthetic
except in absolute word nal p osition where it is the result of reduction
The p osition I defended in the previous sections can b e seen as a combi
nation of the insights of Ep enthesis Theory and the No Syllable Theory
Nevertheless it is useful to study the extreme p ositions to see their relative
merits and problems I will therefore discuss these three theories and some
of their representatives b elow each in their own subsection
Reduction Theory I will discuss two instances of a reduction
approachtou schwa Brink and Zonneveld
Brink Reduction Theory has probably b een intro duced
at least in generative grammar by Brink This scholar claimed that
e and inDutch are in complementary distribution e is found exclu
sively in stressed p osition while schwa is found exclusively in unstressed
p osition For reasons of economy of underlying representations then Brink
prop osed to take e as the underlying vowel in b oth cases and havearule
changing this vowel into schwa in stress environments
In earlier literature e g Cohen et al it is sometimes assumed that schwa
is not di erent from other vowels at all I hop e to have made clear that at least this
p osition is untenable
Previous analyses of u schwa
Wehave already seen that r schwa most often alternates with e m uch
more so than with any other vowel It has sometimes b een observed for in
stance by Kager that the reduction rule that allegedly applies in this
case has some prop erties that makes it di erent from the optional reduc
tion rule which results in r schwa For instance the latter do es not apply
word nally aswehave seen Yet a substantivenumb er of examples exist
whichshowword nal schwa like mode elite etc Another di erence is
that in the words just mentioned the nal vowel must b e pronounced as
schwa Even in the most formal styles of sp eech the nal vowel cannot b e
pronounced as mo de elite
In some Dutch dialects the schwahasaslightly fronted and lowered
pronunciation approximately as or I consider this to b e a matter of
phonetic implementation b ecause it applies to all schwas also for instance
to e schwa which nob o dy supp oses to b e derived from underlying e
Although these di errences b etween the two reduction rules make Brink s
analysis somewhat less attractive they are not necessarily crucial
ob jections against his theory which is stated in the SPE framework It
simply is not excluded that there are two di erent rules
Amuch stronger ob jection is that a lot of the o dd prop erties of schwa
remain unexplained under this account like the restriction to simple onsets
the prohibition of following degenerate syllables etc Of course eachof
these restrictions could in principle b e incorp orated into the reduction rule
but only at the cost of complicating the rule to an extent that has proven
to b e unacceptable to most authors
Zonneveld Interestingly Brink s theory has
b een partly revived in Zonneveld Zonneveld prop oses a threefold
analysis for u schwa
U schwa b efore sonorants is the result of ep enthesis just as in my
prop osal laid out ab ove my prop osal is inspired by Zonneveld
in this resp ect
U schwa in a xes is underlying schwa The distribution of this vowel
is restricted to these a xes it do es not o ccur in stems
Stem nal U schwa as well as r schwa is the result of reduction of
e
The di erences b etween r schwa and stem nal u schwa for instance the
absolute imp ossibilit y of pronouncing mo de is still unaccounted for in
It is also defended in Trommelen and Zonneveld
Previous analyses of u schwa
this framework Furthermore most of the sp ecial prop erties of u schwa are
still unexplained Zonneveld mentions the prop erties broughtup
in Kager and Zonneveld and as wehaveseenatvarious o ccasions
ab ove he explains away most of these This is not true however for most
of the new observations made here For instance while Zonneveld
seems right in disproving the claim that schwa syllables can have no onsets
at all he do es not account for the fact that the schwa syllable can havea
simple onset only
Also the admittedly small group of words ending in unstressed ee is
at rst sight a problem These are words like to ee mentioned earlier
Zonneveld observes however that these words are always bisyllabic
Polysyllabic forms like dominee vicar d omine do not count b ecause they
havestressontheantep enult For technical reasons in Zonneveld s
ideas ab out stress this means the nal e forms a fo ot of its own Poly
syllabic forms with p enultimate stress and nal stressles s e do not o ccur
However in my account it is also true that left adjacency to a stressed
syllable and b eing a coronal mid vowel form very favourable conditions
for reduction This do es not mean we cannot make a distinction b etween
r schwa and u schwa
The strongest argument against the reduction analysis of word nal
schwa in my view is that wehave seen that the grammar in all other cases
avoids deriving a schwa in the absolute word nal p osition The schwa
derived from e would b e an exception to this The only way to understand
this is that in a way stressles s e would b e much more o ensivethanany
other vowel Yet e is apart from schwa itself the most undersp eci ed
vowel It is unclear to me whyavowel with only a coronal sp eci cation
could b e so much more unnatural then an y or u that it can override
the otherwise inviolable prohibition against word nal derivation of schwa
One other somewhat surprising feature of Zonneveld s theory is
that real schwaisonlyallowed in a xes I am not aware of any other
convincing analysis in which the set of segments allowed in a xes is bigger
than the set allowed in stems Furthermore this prop osal fails to capture
the similarities b etween a xal schwa and the derived schwa in stems for in
stance its stressless ness the fact that it can only b e followed by a sonorant
etc
On the other hand the analysis of the Dutch stress system and the
parallelism b etween sup erheavies and schwa syllables seems very convincing
to me I have tried to incorp orate those features of Zonneveld s theory into
the one presented in this thesis
He mentions a few examples but these are either the clearly foreign forms uvre
and timbre with a foreign and a nasalized vowel resp ectively or names of French p eople
or lo cations Sartre Montmartre
Previous analyses of u schwa
Ep enthesis Theory Schwa can b e derived in a way other than
reduction Wehave seen that there is indep endent evidence that schwacan
also b e the result of ep enthesis Another p ossible way to go is therefore to
assume that all u schwas are the result of ep enthesis
Wehaveseenabove that this hyp othesis is succesful for the u schwa s
b efore word nal sonorants like theater and vogel bird It has b een pro
p osed for this context byDeSchutter Ho eksema Ter Mors
Van der Hulst and Van Lit Zonneveld and Hermans
a
In other contexts for instance word nally mo dulo coronal obstruents
the p osition is much harder to maintain The problem is that there are quite
a lot of near minimal pairs of which one form has a schwa while the other
one do es not
a sonate sonata sonat sonat
stramien canvas stramin stramin
als if ls l s
trein train tr in t r in
b paraat prepared parat parat
lawine avalanche lawin lawin
al les everything l s ls
terrein terrain t r in tr in
Wehave already seen that e schwa and u schwa are in complementary dis
tribution It is unclear how to distinguish b etween als and al les if all schwas
are the result of ep enthesis
For this reason most of the authors pleading for an Ep enthesis Theory
of u schwa either ignore the facts in or otherwise supplement their
theory with some additional theory on the schwas in For instance
Zonneveld assumes these schwas are the result of reduction and
Hermans assumes these schwas are extrametrical vo wels to which
some version of the No Syllable Theory applies
No Syllable Theory An ingenious and in uential theory ab out
the idiosyncratic b ehaviour of u schwa has b een prop osed by Kager and
Zonneveld According to this theory schwa is not able to head a
complete syllable If the bimoraicity requirement or Connect N lax is a
requirement on syllables it do es not a ect schwa This approachhasbeen
However the problem is that as we will see undeniably at a certain level of repre
sentation schwa has to form a syllable If we do not assume this syllable is bimoraic
Trommelen the problem is that even at that level short syllable nal vowels should b e excluded
Previous analyses of u schwa
further formally worked out in e g Kager and Hermans I
will henceforth refer to this family of approaches as the No Syllable Theory
The No Syllable Theory dep ends on level ordering It has to b e assumed
that the bimoraicity requirement applies only at an early level of lexical
derivation It can b e demonstrated that at a later level the consonants in
frontofschwa actually have to b e re syllabi ed into the onset in other
words that schwa has to form a syllable
Like many other languages Dutch has a Final Devoicing pro cess which
can b e provisionally formulated as the following rule see section for
a fuller discussion
Final Devoicing
son voiced
This rule explains whyaword like moed mud courage is pronounced
as mut Bo oij and Rubach argue that the Final Devoicing rule has
to b e assigned to the last stratum of Lexical Phonology cf Trommelen
The precise statement of Final Devoicing and its interaction with
other principles will b e discussed in section For now the formulation
given here is su cient
This implies that the Bimoraic Constraint has to b e checked at an earlier
r level Final Devoicing do es not apply to a word like moeder mud
mother This shows that at the level of Final Devoicing the d no longer
is uniquely in the co da of the rst syllable
I will now turn to several variants of the No Syllable Theory that have
b een prop osed These are Kager and Zonneveld Kager Her
mans and Nijen Twilhaar
Kager and Zonneveld According to the original pro
p osal of Kager and Zonneveld word nal schwa can form a sp ecial
typ e of app endix the terminology is from Halle and Vergnaud to
phonological words called a schwappendix This schwapp endix consists
roughly of schwa followed by an arbitrary numb er of consonants The
schwapp endix could b e preceded by a regular app endix consisting of either
one non coronal consonantoroftwo coronal consonants Both app endices
are optional C stands for a coronal obstruent
c
C C
o
word
C C C
c c c
It is true that there is one prop ertyofschwa syllables whichIdidnot
takeinto account in the previous sections For instance words like klooster
See Visser for an application of this theory to Frisian
Previous analyses of u schwa
monastery klost r are not uncommon in Dutch However the struc
ture of these words is not very clear under the accountgiven ab ove The
syllabi cation should b e something like klo s tr The problem is the s
There is no ro om for this segment in either the rst syllable whichendsin
a tense vowel or in the second one which cannot have a complex onset
It can also not form a degenerate syllable b ecause degenerate syllables and
syllables headed byschwa cannot b e adjacentaswehave seen I will return
to this problem in section b elow
Kager and Zonneveld s theory makes a lot of interesting predic
tions For instance it predicts the lack of complex onsets b efore schwa be
it by merely stipulating that a word nal schwa can only b e preceded by
an app endix consisting of one non coronal while the schwapp endix itself
can not start with a consonant the stress neutral and stress attracting
b ehaviour the non o ccurrence of some clusters b efore schwa Wehave seen
however that some of these alleged prop erties were later argued to b e less
essential to schwathanwas thoughtby Kager and Zonneveld On
the other hand some of the new prop erties listed ab ove cannot b e cap
tured in their framework like the obligatoriness of an onset
One imp ortantdrawback of the schwapp endix analysis is that it predicts
that u schwaisalways right p eripheral in the word Wehave seen that this
is simply not true and u schwa can o ccur virtually anywhere in the word
provided su cient other material is present to build a legitimate syllable
Furthermore the theory seems to b e di cult to incorp orate into a some
what more restrictive theory of extrametricality Usually it is assumed that
either a segment or some phonological constituentsuch as the syllable can
be marked extrametrical None of the two app endices conforms to this
criterion
More imp ortantly the organisation of the two app endices and esp ecially
of the schwapp endix in is largely arbitrary There is no sp ecial reason
whythevowel in this app endix can only b e schwa or why it cannot b e
preceded bytwo non coronal consonants instead of only one This means
that many of the most crucial prop erties are simply stipulated and some
imp ortant generalisations are missed
I think Kager and Zonneveld s pap er is empirically the most
imp ortantcontribution to the discussion ab out schwa that has b een made
hitherto It intro duced two ideas into the discussion In the rst place this
was the No Syllable hyp othesis the idea that schwas do not license syllables
or complete syllables In the second place this was the idea that schwas are
extraproso dic and fall outside of the proso dic word I think however that
Zonneveld has shown quite convincingly that the latter part of the
The Dutchword is not morphologically complex
Previous analyses of u schwa
theory is not tenable the most imp ortan t problem is that it cannot explain
why stress falls invariably on the antep enultimate syllable in words ending
in schwa Final stress in the proso dic word is otherwise not the regular
pattern in Dutch at all
Kager Most of the problems just mentioned were also
noted by Kager Instead of the schwapp endix he develop ed an
even stronger version of the No Syllable Theory According to Kager
the schwa is the only vowel that has no underlying mora All other vowels
have one or two moras dep ending on whether they are long or short Kager
assumes that at level I schwa do es not build a syllable at all b ecause only at
that level those rules apply that pro ject syllables from moras Furthermore
Kager assumes that the segments preceding the schwa get incorp orated into
the preceding syllable which thereby b ecomes sup erheavy The structure
of bende band thus b ecomes at level I
b n d
At later levels the d has to b e re syllabi ed into the onset of a newly
pro jected syllable headed byschwa in order to escap e nal devoicing
Because in Kager s theory word nal sup erheavy syllables attract stress
and Dutch feet are built on top of syllables the stress facts are readily
explained schwanever is stressed b ecause it is not in a syllable at the
relevant level of the derivation and the syllable preceding schwa is stressed
b ecause it is sup erheavy The fact that schwa do es not allow for a complex
onset is explained by the following lter
Complex Onset Condition Level
C C
cons
son
son
unless the two consonants o ccur in the same syllable
Kager s Complex Onset Condition is stated just in terms of obstruents
and liquids b ecause they form the only p ossible complex onset clusters
in Dutch At least one cluster tl is incorrectly included into this lter
the word butler bytl r butler is incorrectly predicted to b e ill formed in
Dutch
Previous analyses of u schwa
The reason for this exceptionalityof bytl r seems to b e that tl is the
only stop liquid sequence that is not a p ossible onset cluster for indep endent
reasons p erhaps having to do with the OCP
In order to account for this fact Kager would have to revise to
Complex Onset Condition Level
C C
i j
unless C and C app ear in the same syllable or C C is not a
i j i j
potentially well formed onset cluster
This version of the Complex Onset Condition says if some consonantclus
ter can b e an onset it must b e an onset Yet this formulation is not without
problems for Kager b ecause it crucially refers to the notion of onset
and there is no such thing as an onset in moraic phonology Kager
claims there is one piece of indep endent evidence for this constraint There
are no words of the form z b ra i e a lax vowel followed by a complex
onset followed by another vowel If ambisyllabicity is not a p ossibilityat
level I while the b has to b e in the co da of the rst syllable b ecause of
Connect N lax or b ecause every syllable has to b e bimoraic dep ending
on one s theory ab out the di erence b etween A vowels and B vowels
explains why this structure is ruled out The constraint therefore parallels
what wehave termed the Syllable Contact Law the unless clause would
probably now b e captured in terms of constraintinteraction
As wehave seen ab ove Kager noted as a problem for Kager and Zonn
eveld s theory that it predicts schwa syllables can only b e p eripheral
which is not true Yet Kager himself has to assume that at level I
remains unsyllabi ed in the middle of a word like Hengelo which actually
is a place name
Level I
h x l o
Schwa apparently is not incorp orated into the Proso dic Structure at this
level of representation yet This means that the Principle of Proso dic Li
censing in anyofitsformulations is violated as far as this structure is
concerned The fact that pre schwa consonants are incorp orated into the
previous syllable shows us that at least there is a tendency for all struc
ture to b e incorp orated in the existing material This in turn shows that
Previous analyses of u schwa
Proso dic Licensing at least should hold in some weaker form and that the
schwa is an exception to it
Kager and Zonneveld analyse this exceptional b ehaviour as a
consequence of a certain version of extraproso dicity but Kager do es
neither give an account for schwa s exceptional b ehaviour in this resp ect
nor of the segments that can followschwa In kater tom cat kat r the
r is not syllabi ed at level I
k a t r
Later on it gets syllabi ed together with the schwa One of the curious
facts now is that these unsyllabi ed segments are sonorants Under a moraic
view of syllable structure wewould probably wanttosay that these are in
away the moraic consonants But it is crucially imp ossible to saysucha
thing in Kager s framework If these consonants were to b ear moras
they would also come to b ear syllables b ecause that is what happ ens at level
I But in that case the whole argumentaboutschwa b eing amoraic would
break down I do not see a way to solve this problem
Another problem for Kager s approach is that it is still stipulative
as far as the nature of schwa is concerned Kager stipulates that schwa
do es not b ear a mora There are three typ es of vowels in this approach
monomoraic vowels like bimoraic vowels like o and amoraic vowels
like the O stands for the features of and o
o
O O
There is no formal relation b etween the phonetic and phonological empti
ness of schwa and its b eing amoraic One could just as well imagine a
language which has three schwas one amoraic one monomoraic and one
bimoraic or a language which has just a bimoraic schwa and amoraic a
i and u in its vowel inventory In other words there is no theory of the
relation b etween segmental content and number of moras
Furthermore as wehave seen e schwa and r schwa share some of the
prop erties of u schwa Kager do es not provide us with a theory of
e schwa which indeed seems hard to incorp orate into his analysis but he
Previous analyses of u schwa
do es give a partial analysis for r schwa from which it b ecomes clear that
for him this vowel is p ost lexical He do es not drawany relation b etween
the di erent o ccurrences of schwa U schwa is sp ecial b ecause it is amoraic
r schwa is sp ecial b ecause it is derived p ost lexically
Hermans a The relation b etween the lack of features
of schwa and its defective metrical b ehaviour seems to b e b etter estab
lished in an unpublished manuscript by Hermans a who furthermore
defends a less radical version of the No Syllable Theory
Hermans assumes that all pro jection no des of proso dic structure haveto
see certain nonphonetic features on another plane Since schwa is empty
it also lacks these features Therefore proso dic no des on top of such empty
ro ots are illegitimate Hermans assumes this holds for all no des but for
technical reasons there is an exception for the lowest one the nucleus
Schwa in Dutch can therefore only license a nucleus Hermans a
assumes the Dutchnucleus can only contain sonorant segments This ex
plains why only sonorants are allowed after a schwa It also derives Kager
and Zonneveld s claim that schwa syllables have no onset In or
der for an onset to exist a syllable no de is needed which cannot b e there
b ecause it is not licensed b yany feature Unfortunately aswehave seen
Kager and Zonneveld s generalisation is not totally adequate and
should b e replaced by schwa syllables have exactly one segment in their
onset Exactly this generalisation can not b e captured under Hermans
account
Also the non o ccurrence of schwainword initial p osition is di cult to
explain under this approach Hermans a assumption is that at a
later lexical level all schwas have to pro ject to a syllable whichinturn
needs an onset This onset needs to b e lled by a glottal stop but sequences
of unarticulated segments are assumed to b e blo cked Thus far this is more
or less the traditional approach Cohen et al which is also defended
here But it is unclear how the sequence of glottal stop plus schwacan
b e excluded under Hermans a analysis Because p ostlexically schwa
apparently builds a syllable it should have acquired the feature licensing
it But then it presumably is di erent from the glottal stop whichisnota
fully sp eci ed syllable head but a consonant
In general it seems that Hermans a theory can account for the
claims ab out schwa made in Kager and Zonneveld rather neatly Un
fortunately however most of these claims have b een shown to b e untenable or sub ject to revision
Previous analyses of u schwa
Nijen Twilhaar An interesting interpretation of the
analysis of schwa along the lines of Kager and Zonneveld has b een
provided by Nijen Twilhaar Nijen Twilhaar s pap er esp e
cially concentrates on the assumption that schwas are extraproso dic
Nijen Twilhaar p oints out that seemingly underived nouns ending in
schwa all select the common gender determiner de In dialects whichdis
tinguish b etween three genders like Hellendo orn Dutch they even are all
feminine He p oints out that non feminine nouns ending in schwa often
changed their grammatical gender while inversely feminine nouns some
times develop ed a word nal schwa
On the basis of this Nijen Twilhaar argues that these forms have
the following morphological structure
root
N
The schwa in this analysis is a ro ot selecting low level su x Ro ots cannot
o ccur with this su x Nijen Twilhaar assumes William s Right
Hand Head Rule holds for Dutch A consequence of this is that the su x
determines the gender of the noun If we determine that the ro ot initially
is the domain of syllabi cation wederive a lot of Kager and Zonneveld s
schwa prop erties
Interestingly it app ears that the nouns which are exceptions to Kager
and Zonneveld s generalisation like ritme and uvre select the de
terminer het in Standard Dutch and are neuter in Hellendo orn Nijen
Twilhaar assumes that these forms are monomorphemic so that schwabe
haves like a normal vowel and that there is a lexical redundancy rule like
If then neuter
W
Even though there are a few counterexamples and problematic examples
as noted by Nijen Twilhaar himself it seems that these observations
are basically correct They require further investigation b ecause they are
problematic for the approach defended here
On the other hand an approach along these lines has to face many
problems that are not addressed in Nijen Twilhaar In the rst
place this theory makes exactly the same predictions for nouns as Kager
and Zonneveld and therefore he has the same problems including the
prediction that word internal schwa should b ehaveentirely di erently In
the second place other than Kager and Zonneveld it also seems to predict
that word nal schwa followed by a sonorant consonantbehaves entirely
On the other hand there are exceptions to this generalisation viz words like kas
tanje chestnut and tarwe wheat
Remaining problems
di erently unless wewould b e willing to supp ort su x analyses for r
m n etc as well There seems to b e no evidence for this p osition The
criterion of determiner selection Nijen Twilhaar prop oses for do es not
seem to work
a de adder id het anker the anchor
b de deken the blanket het teken the sign
c de asem the breath het desem the leaven
d de stoppel the stubble het koppel the b elt
Yet phonologically these forms b ehave exactly like the forms ending in
In the third place it remains unclear whyschwa syllables in adjectives
and verbs b ehave similar to those in nouns except again if there would
b e evidence that in those cases the schwa app endices are su xes
Remaining problems
In the previous sections wehave put aside some technical problems either
b ecause their treatmentwould unnecessarily slowdown the discussion or
b ecause they needed some extra information we could not yet provide at
er useful the time In order to makemy prop osals more explicit it is howev
to also solve these problems I will brie y discuss eachoftheminthis
section
Final Devoicing The rst problem I want to discuss is Final
Devoicing The problems are mainly technical and involve the following
p oints In the rst place wehave to make sure that Final Devoicing applies
at the end of a word notwithstanding the fact that most other pro cesses
are prohibited to apply at word edge Secondly I cannot say that Final
Devoicing applies to obstruents in the co da b ecause under my assumptions
the nal underlying d of brood bread is in the onset of a degenerate
syllable and still Final Devoicing applies to this segmentaswell
The rst p oint is not really much of a problem in a theory where con
straints can b e highly ranked Wehave seen that Align the constraintthat
is resp onsible for the relatively static b ehaviour of word edges has a high
ranking in Dutch This do es not mean however that Align is inviolable
under all circumstances In the preceding sections wehave already seen
evidence that it is dominated by for instance the constraint Contour
causing d m to surface d m
We can assume that at least the constraint resp onsible for Final Devoic
ing call it FinalDevoice is ranked higher than Align
Remaining problems
FinalDevoice Align
The inverse ordering of these constraints is also attested viz in Yiddish
This language shows devoicing of syllable nal obstruents just likeDutch
However word nal obstruents exceptionally do not devoice
a red I sp eak red ret
b retst you sp eak retst redst
c shrayb I write sra jb sra jp
d shraypst you write sra jpst sra jbst
e ayz ice a jz
f ayskastn ice b ox a jskastn
g briev letter briv
h brieftreger p ostman briftregr
The di erence b etween Yiddish and Dutch is describ ed by Lombardi
by stipulating a nal exceptionality parameter which has the plus value
in the former language and the minus value in the latter We can now
understand somewhat b etter the nature of this parameter In Dutch the
necessity to devoice nal obstruents is absolute In Yiddish it is overridden
b y the desire to keep the edges of words clean
This brings us to the next p oint How can weformalise naldevoicing in
a theory of degenerate syllables It lo oks as if wehavetomake a disjunctive
statement for Dutch obstruents devoice in a co da of a regular syllable or
in the onset of a degenerate syllable Iwillhowever here again follow
Lombardi who argues that the relevant constraint has the following
form
FinalDevoice voice except in the environment
sonor sonor
Laryngeal
Other languages with the same parameter setting are Romanian and Serb o Croatian
according to Lombardi
Lombardi also claims that voiced is a privative feature a claim which I will follow
here
The except in the formulation makes us susp ect FinalDevoicing is actually the
con ation of several constraints at least one of which prohibits the feature voiced on
obstruents in anyenvironment and another which enforces an obstruenttobevoiced
when followed by a sonorant I will not work this out
Remaining problems
It is as yet formally unclear whyvoiced obstruents are restricted to this
environment preceding a sonorant but neither a segment in a co da nor a
segment in the onset of a degenerate syllable satis es this condition Both
are therefore correctly predicted to b e sub ject to Final Devoicing
Sup erheavy syllables b efore schwa The next problem con
cerns the class of words exempli ed by klooster monastery klost r The
problem with these words is that they seem to have a sup erheavy syllable
preceding a schwa syllable while I have argued that degenerate syllables and
schwa syllables can never b e adjacent to one another On closer scrutiny
the problematic forms app ear to fall into three classes
a klooster klost r monastery meester master mest r
heester shrub hest r langoeste langouste l ust buste
bust byst hamster id h mst r etc
b aarde earth waarde worth aarzel hesitate daalder dollar
gaarne readily vaandel banner
c pienter clever pint r boerde fabliau burd ottielje otil
la tilj
The segment s is the only segmentwhich can freely stay in extraproso dic
p ositions throughout the word b oth at the left p eriphery as in s troop
treacle and at the right p eriphery as in melaats leprous It can also
stay in an extrasyllabic p osition b etween two syllables b oth headed bya
full vowel ekstra In the latter case the syllable following s starts with a
t Exactly the same con guration can b e found in the forms in a so we
may assyme the s in this case is extraproso dic not the onset of a degenerate
syllable
The second class consists of words with an a It is interesting that
these forms contain an a followed by a sonorant and a voiced obstruent
or in one case a nasal In chapter wehave seen that in Tilburg Dutch
tenses in similar circumstances It maywell b e that here wehavea
re ex of a similar historical tensing pro cess whichwould just apply in
front of rz in Standard Dutch
The third class nally consists of words with high vowels Wehave seen
in chapter that these vowels do not havealaxcounterpart and therefore
can exceptionally o ccur in a closed syllable If this is true the examples in
I assume the t in pantser armour p nts r and similar forms is the result of a
phonetic rule of homorganic stop insertion The only counterexamples of whichIam
aware and which do not fall into either of these classes are deerne lass taveerne tavern
and heinde en verre far and near
English has the same typ e of exceptional words cf Harris
Remaining problems
c are not counterexamples to our claims at all since they consist of a
normal closed syllable followed bya schwa headed syllable
Post lexical u schwa Wehave seen that u schwa can app ear
in a right p eripheral p osition while this p osition is prohibited for r schwa
and e schwa Lexically u schwa cannot o ccur word initially however Even
this changes at the p ost lexical level In Standard Dutch a lot of clitics
have the form C where C is either a sonorant or a coronal consonantas
is shown in
m him n a r her there s once t it the
While these clitics b ehave as enclitic s if they can showing Onset is still
op erative if it can these clitics can phonologically still app ear in a phrase
initial p osition
a Een n man huilt niet
a man cries not
Real men do not cry
b Er r k omt iemand
There comes someb o dy
Someb o dy arrives
c Het t regent
it rains
d m m vertrouwen kan ik niet
him trust can I not
I cannot trust him
Some dialects also have an in ectional pre x usually denoting the p er
fect participle Examples are Katwijk Dutch next to the coast Overdiep
and Aalten Dutch next to the German b order Bennink p c The
following is an example sentence from Katwijk Dutch
Ik heb Krijn n zien
I have K case epenthetic n seen
I have seen Krijn
For one reason or another examples with phrase initial er like b are not ac
ceptable for some nativespeakers
Remaining problems
According to Overdiep p ersonal names and some other categories
in Katwijk Dutchhave a case in ectional element in the accusative In
this case the name Krijn b ears this in ection The immediately following
participle starts with a schwa In this case an n is ep enthesized according
to Overdiep As far as I can tell an alternativewould b e to assume that
the in ection is n and the n gets deleted in all contexts except b efore a
vowel In any case it can b e seen that Onset is still active although it no
longer prohibits word initial schwa
The pre x in these dialects corresp onds to a pre x in most other
dialects and the standard language Interestingly nonpre xal word initial
is also found for instance in the word geluk luck lyk and gera
nium id ranijym In cases such as these Katwijk Dutch and Aalten
Dutch also have and never
Van Haeringen observes that in Dedemsvaart Dutch another
eastern Dutch dialect even participles which are used as attributivead
jectives again showthe variant Bennink p c notes that the same
alternation holds in Aalten Dutch
a Wij hebt melk eko okt
we have milk co oked
Wehavecooked milk
b Wij hebt eerepp els eko okt
we have p otato es co oked
Wehavecooked p otato es
c Mij lust liever gebakken eerepp els as geko okten
me likes better baked p otato es than co oked
I like fried p otato es b etter than co oked ones
It seems then that in these cases the adjectival forms are formed already in
the lexicon where schwa initial words are not allowed whereas formation
of the real participles can b e p ostp oned until the p ost lexicon
Notice however that in the previous chapter wehave seen that the
derivation of schwa initial words is still blo cked p ost lexically The only
initial schwas that can surface are the underlying schwas that arise only
p ost lexically i e initial schwas in clitics and su xes In the p ostlexical
phonology it thus seems that Contour gets ranked b elow the constraint
Project cons N and or Align
Finally I have to note that p ostlexical derivation also allows schwato
b ear some weak form of secondary stress see for instance the stress pat
terns in words like gemakkelijk er e easy er AGR m k l k r
Remaining problems
Also the constraint Project Ft V seems to get a lower ranking b e
lowarhythmic constraint requiring every second syllable to have stress in
the p ostlexicon
Umlaut Finally I want to brie y discuss umlaut Many eastern
dialects of Dutch aswell as most German dialects have an active umlaut
ing pro cess which is triggered by one or more schwa containing su xes It
seems to b e a distinctivepropertyofschwa that only this vowel can trigger
umlaut in most of these dialects
Prop erty Schwa in su xes can trigger umlaut
Wehave seen that some remnants of the pro cess can b e found in Tilburg
Dutch but it seems to b e really pro ductive in for instance Schinnen Dutch
Province of Limburg Wetzels Hellendo orn Dutch Province of Over
ijssel Nijen Twilhaar and Aalten Dutch Province of Gelderland
Bennink in preparation All of these dialects have their own p eculiariti es
Ihavechosen to only discuss Aalten Dutch for practical reasons a native
sp eaker of this dialect was close at hand
Umlauting involves fronting of a vowel usually the nal vowel of the
stem Some examples are provided b elow all copied from Bennink in prepa
ration with some adaptations to t the phonological transcription system
used here
underived form derived form gloss
mus myz mouse pl
vust vyst st pl
b s bysk n bundle dim
b k b ksk n tray dim
Historically the umlaut triggering schwas are derived from frontvowels
mostly i Synchronically there is no sp ecial reason to see them as r
schwa s in present day Aalten Dutch since they never alternate with any
full vowel
The authors just mentioned all p oint out that this causes at rst sight
a somewhat curious situation b ecause schwa as a non coronal vowel trig
gers fronting i e coronalization of a preceding vowel I will basically
adopt the solution put forward by Nijen Twilhaar and supp ose the
representation of umlaut triggering schwaisasfollows
cons coronal
Remaining problems
The representation consists of two indep endent elements an emptyroot a
schwa and a feature coronal The diachronic development can b e seen
as the loss of all connections underlying asso ciation lines and class no des
between these two elements
The derived form muz willnowlookschematically as in
m u z
coronal
The rst constraint that is relevantisParse coronal the vo calic place
feature coronal has to b e linkedtoa vo calic place no de Logically sp eak
ing there are two options coronal can b e link ed to the u yielding myz
or it can b e linked to the schwa creating a lot of intermediary no des C
place vo calic V place and presumably yielding muze The former is
clearly the option that is selected
There are at least two p otential reasons why this is the case In the
rst place Aalten Dutchshows the e ect of a curious constraint against
nal coronal vowels in a word All words which in Standard Dutch end
inafrontvowel either have defronted this vowel or end in a schwain
Aalten The feature coronal can therefore not do ck on the nal vowel of
the phonological word but it has to go to the stem nal full vowel This
is approximately the line of reasoning defended by Bennink in preparation
Yet this reasoning cannot fully explain all facts of umlaut In the rst
place many other Dutch and German dialects that have umlauting do not
prohibit coronal vowels in word nal p osition So at least in those dialects
some other factor should b e at work In the second place umlauting in
words like vader ken father dim v d rk n a ects the full vowel not
the stem nal schwa If the reasoning just laid out would b e correct we
would exp ect the pronunciation vad rk n Yet this is not the case
It seems therefore more correct to supp ose that it is a characteristic
prop erty of the schwavowel as opp osed to other vowels that it can not
serve as an anchor p oint for the place feature regardless of whether schwa
is in a su x or a stem
I already indicated that in one sense it is more costly to attach a place
feature to schwathantoany other vowel The reason for this wehave
I assume schwa coronal e in op en syllables or in closed syllables in Dutch
Bennink in preparation rep orts that Aalten sp eakers refer to outsiders unsucces
fully trying to imitate the lo cal dialect as sp eakers with pien in de buuk pain in their
stomach pin nd byk To outsiders this seems an adequate Aalten phrase since it
contains non diphthongized versions of words which in Standard Dutch are pronounced
as p in and b yk resp ectively Lackofdiphthongisation is a prominentcharacteristic
of eastern Dutch dialects Yet the real Aalten pronunciation of this phrase would b e pine
in de boek pin nd buk
Remaining problems
already seen It is the constraint Project V Ft whichsays that un
stressed syllables should not dominate a vo calic no de Since the vowel in
the su x as well as that in a nal schwa headed syllable is unstressed
attachment of the feature coronal to that vowel will cause unnecessary vi
olation of that constraint Attaching of the feature to a full vowel can never
cause such a violation b ecause that vowel already has a vo calic no de hence
already violates the constraint against such no des in unstressed p osition
Also the umlauting b ehaviour of schwathus follows from this vowel s
minimal nature It has very little structure and therefore it is not able to
serve as a linking p oint for many features
Finally Ihave to discuss one additional case namely one in whichwe
add an umlauting schwa to a stem containing at least two backvowels An
example is foto photograph Given the preceding discussion it is not clear
whether the output should b e fot k n or f tok n Both rank eqally well
with the constraints mentioned b efore Howdowe explain that it is the
former output that is selected More in general why is the last full vowel
always selected rather than some other vowel
I prop ose the answer lies in the fact that the coronal feature is part of
a su x What this means is that it should b e aligned with the right edge of
the proso dic word on top of the stem as much as p ossible McCarthy and
Prince Akinlabi Wethus have a constraint whichsays that
the elements of the diminutive su x should o ccur as much to the rightas
p ossible Therefore the form foteuken is preferred over feutoken b ecause
in the latter case the coronal feature has shifted unnecessaril y far to the
left
For convencience I will call this constraint Suffix althoughitcanbe
formulatedasamemb er of the Alignmentfamily Align dim X PrWd
R
The ideal output for Suffix has coronal at the rightmost syllable of
the word Yet this constraint is outranked by the pro jection constraint
Project V Ft Wethus get the following tableaux assuming only can
didates which satisfy Parse coronal
N do es not head a branching fo ot N do es Project V Ft
not dominate a vo calic no de
Suffix The elements of a su x should o ccur maximally close to
the right edge of a proso dic word
Logically we should also exclude a realisation feuteuken in whichbothvowels are
umlauted I assume some kind of economyprevents this double linking Once Parse
coronal is satis ed multiple linking of this feature gives only unnecessary asso ciation lines
Conclusion
muz
Candidates Project V Ft Suffix
p
m yz
m u ze
vader k n
Candidates Project V Ft Suffix
p
v d rk n
v ad rk n
v ad rk n
foto k n
Candidates Project V Ft Suffix
p
fot k n
f tok n
fotok n
This concludes my discussion of umlaut U schwa can cause umlaut even
though it is not itself front b ecause it can b e accompanied by a feature
coronal
Conclusion
In this chapter I have discussed many prop erties of Dutch u schwa the
schwa that never alternates with either zero or full vowels It is argued
that only the schwa that o ccurs in op en syllables is really underlying The
other u schwa is the result of an obligatory rule of ep enthesis in syllables
which lexically havealevel in which a sonorant consonant is the head of
the syllable
In the lexical phonology schwa only licenses CV syllables I have argued
that the reason for this is that schwa lacking a vo calic no de cannot b e the
head of anybranching constituent at all it do es not license a co da or a
dep endent syllable at the fo ot level It also do es not really license onset
segments but since at least one such segment is forced to b e presentby
forces from outside i e the constraint Onset the result is a core syllable
Exactly the same reasoning causes also sonorant consonants to only o ccur
in CV syllables with V in this case the p osition lled by the sonorant
Pro jection constraints served to formally express all of these generalisations
Another consequence of the emptiness of schwaisthatitavoids the
neighb ourho o d of other emptysegments In particular it cannot o ccur in a
Table of Prop erties
syllable of which the onset is empty a glottal stop or h Again this was
expressed in terms of a pro jection constraint
The pro jection theory thus plays a crucial role in my analysis of schwa
Schwa is a defectivevowel this is the reason why it pro jects a very defective
typ e of syllable
In this resp ect there is a di erence b etween the lexical and the p ostlex
ical level Post lexical schwa can o ccur in manymoreenvironments than
lexical schwa as we can see in the comparison b etween e schwa and r schwa
on the one hand and u schwa on the other Post lexical schwa can o ccur in
a closed syllable it can have a complex onset etc All of these things are
not p ossible for lexical schwa
It seems then that certain top down pro jection constraints requiring
the heads of pro jection levels to b e supp orted byfullvowels havelostmuch
of their force in the p ostlexical phonology Other pro jection constraints
b oth top down the constraint against degenerate syllables requiring in
sertion of cons ro ots in ar m and b ottom up the constraint forcing
unstressed full vowels to reduce to schwa b ecause full vowels wanttooccur
in the head of a fo ot are very forceful in the p ostlexicon whereas they do
not playsuch a strong role in the lexical phonology I am not sure whether
anything more general can b e said ab out the di erence b etween the two
di erentlevels of analysis
Table of Prop erties
Below I give a table of all prop erties of DutchschwawhichIhaveidenti ed
in this chapter and the previous one For every prop ertyIhave indicated
p
whether the prop erty holds for each of the three typ es of schwa A
mark indicates that the schwa has that prop erty an asterisk that it do es
not have it a dash indicates that the prop erty is not relevant and a P of
preference indicates that the prop ertymay b e violated although schwas
which satisfy the prop erty surface in more style registers than schwas which do not
Table of Prop erties
Prop erty R schwa E schwa U schwa
p p p
Phonetic emptiness
p
Reduction vowel
p p p
Not tautosyllabic with h or b
p p
Not word nal
p
Alternates with e a o
p
No alternation with i
p p p
Unstressed
p
Not outside fo ot P
p
Op en syllables only P
p
Ep enthetic vowel
p
Only in word nal syllable
p
Not in word initial p osition
p
Allows ep enthesis b efore itself
p
Allows sp metath b efore itself
p
x V V r
i i
p
No complex onset
p
Only sonorantincoda P P
p p
degen syll
p
OK
p
C C
i i
p p
Stress on preceding syllable P
p p p
V V
p p
Compl distr u schwa and e schwa
p p
Schwa deletion
p p
Deletion only in p ossible onset sites
p
Umlaut
SchwainFrench and Norwegian
Intro duction
In the previous twochapters I havederived some prop erties of Dutchschwa
from the phonetic and phonological emptiness of this vowel From this it
was made to follow that schwa only pro jected to a core syllable in the
weak p osition of a fo ot Since this emptiness by de nition is the same for
schwa likevowels in languages other than Dutch we exp ect the phonological
b ehaviour of these vowels to b e similar also they should pro ject only a
minimal structure In this chapter I study two languages other than Dutch
with a schwa in their surface vowel inventory to nd out whether or not
this exp ectation is b orne out These languages are French and Norwegian
The reason whyI havechosen these languages is that their schwabehaves
di erently than Dutchschwaininteresting and signi cantways while still
the phonologies of these languages are not to o di erent from that of Dutch
so as to obscure the comparison
Some prop erties of schwa seem universal This is esp ecially true for its
stressless nes s Schw aavoids stressed p ositions not only in the languages
discussed here but also for instance in Malayalam K P Mohanan T
Mohanan German Hall Wiese Noske
Indonesian Cohn and many other languages It also seems true that
if a language has schwa on the surface and either vowel ep enthesis or vowel
reduction schwa tends to b e the ep enthetic or reduced vowel Finally word
p eripheral p ositions are disallowed for schwa likevowels in many languages
All of these facts ab out schwa can b e made to follow from pro jection theory
On the other hand some prop erties I mentioned in the preceding chapter
seem less universal and more typical of Dutch One of the problems studied
in this chapter is that schwa can have complex onsets in b oth French and
Norwegian Both languages havea word ordre pronounced more or less as
r dr I will try to explain this as an e ect of a minor reranking of some
constraints Essential is that neither language likeDutch allows a word
like ka tr l where a schwa headed syllable has b oth a complex onset and
acoda
French
In Pro jection theory we exp ect schwa to license only a very limited typ e
of syllable structure in every language This is a prediction which seems
to b e b orne out at least for the languages under discussion Yet it will b e
shown that languages may slightly di er as to what they consider to b e a
minimal onset
French
There is an overwhelming amount of literature on Frenchschwa I refer to
Anderson Basb ll Charette Dell
Durand Morin Noske
Schane Selkirk b Tranel
among others for discussion of the manyinteracting factors that
in uence the b ehaviour of this segment Interestingly the range of theories
for Frenchschwa put forward by these authors is approximately the same
as the one discussed in the previous chapter for Dutchschwa For instance
some authors e g Tranel have prop osed that Frenchschwa
is a oating melo dic element which is not attached to a skeletal p oint and
others have assumed it is an emptyskeletal p oint without melo dic content
e g Charette
It is imp ossible to discuss all known facts ab out Frenchschwainthis
chapter As compared to the fairly comprehensive discussion of Dutch
hwa in the previous twochapters the discussion presented here inevitably sc
is somewhat sup er cial I only discuss the observations that seem most rel
evant with resp ect to the discussion of Dutchschwa in the previous chapter
Within the terminology of the preceding chapter wecansaythatFrench
has instances of e schwa and u schwa but not of r schwa The language
do es not allowvowel reduction under any circumstance I will takethis
as an indication that French has a su ciently low ranking of the relevant
pro jection constraint relating vowel quality to heads of feet reduction in
Dutch takes place b ecause full vowels preferred to o ccur in heads of feet
and apparently this constraint is not visibly activeinFrench
Still this relation b etween feet and features is not totally irrelevant As
we will see there are a few contexts in whichschwa alternates with full
vowels esp ecially with and although it is usually asssumed that this is
not reduction to schwa but rather enhancement from schwa to a full vowel
the notion of a Fo ot can b e argued to b e relevant so that some version of
Pr oject Ft still has a role to playinFrench phonology
Underlying and ep enthetic schwa are di cult to distinguish in French
b ecause this language allows muchmoreschwa deletion than Dutch A
consequence of this is that almost all schwas alternate with zero Never
French
theless some convincing tests have b een develop ed to distinguish b etween
underlying and ep enthetic schwainFrenchaswe will see in section
and therefore I will pro ceed to assume this distinction
My discussion of French is organized in approximately the same wayas
my discussion of Dutch The most imp ortant prop erties of Dutchschwa
are listed and each time it is discussed in what resp ects the Frenchvowel
is di erent or similar I skip the prop erties of Dutchschwathatseemtoo
much connected to the rest of the Dutch phonological system For instance
the b ehaviour of diphthongs or b efore schwa cannot b e used as a test in
French b ecause this language do es not havediphthongs and uses only
marginally I do not discuss the prop erties of r schwa either since reduction
do es not o ccur in French On the other hand I will mention some prop erties
that are more sp eci c to Frenchschwa A complete graph of the constraint
interactions needed in the analysis of Frenchisprovided in app endix C
Of course schwainFrench like in all other languages by de nition has
Prop erty
Prop erty Schwa is exceptional in the vocalic system because it bears no
vocalic features
The variation in the phonetic realisation of schwaisvery large in French
In some dialects the realisation of this vowel is very close to the realisation
of the lax rounded coronal mid vowel For this reason some authors
have prop osed to give the twovowels the same representation e g Morin
Although it is in itself curious that the emptyvowel phonetically
is so close to one of the most sp eci ed vowels coronal labial lax I
will follow the authors mentioned ab ove and assume and havean
essentially di erent representation and that Frenchschwa has Prop erty
in most cases
Yet it seems that wehavetomake one exception It will app ear b elowthatschwa
deletion o ccurs muchmoreofteninFrenchthanitdoesinDutch Yet a few words havea
so called stable schwa a schwa that cannot b e deleted although schwas in comparable
nu id femel le female debout environments can instances of stable schwa o ccur in me
stand up etc Morin Durand Slater and Wise and Charette have
suggested that in these cases wedohave a full vowel probably rather than schwa
in the underlying representation Ifwewould want to implement this in the theory
presented here we could follow this suggestion stable schwa is not a schwa at all but
a full vowel Real schwas in French are empty ro ots In the remainder of this chapter
I will concentrate only on those instances of schwa that do alternate these are either
underlying schwas which can b e deleted or ep enthetic schwas which can b e inserted
The only worked out alternative prop osal byTranel and Noske holds
that normal schwa is a oating vowel not linked to a skeletal slot underlyingly while
stable schwaislinked to such a slot Such a prop osal cannot b e adopted in the framework
prop osed here For a general criticism of the idea that schwa is a oating vowel cf the
discussion of Kager s prop osal for Dutchschwa in the previous chapter
French
Again we exp ect all other prop erties of Frenchschwa to b e related to
this central prop erty This is trivially true for the rst prop ertyofE schwa
which is discussed in the next section
E schwa is the ep entheticvowel Prop ert y As noted
it is di cult to nd go o d evidence for schwas that are unambiguously
ep enthetic in French Because underlying schwas so often alternate with
zero it is hard to nd cases in which a surfacing schwaisunambiguously the
result of ep enthesis not an underlying schwa that happ ens to have escap ed
deletion
Still there are a few go o d reasons to distinguish b etween underlying
schwa and ep enthetic schwa In the rst place there are a few French
dialects e g Midi French sp oken in southern France in whichschwa
deletion is much less common than it is in Standard French Still in these
dialects some words showaschwa zero alternation which is not attested in
other words
For instance Durand gives fo c f k jib versus phoque f k
seal as a minimal pair In Standard Paris French b oth words are pro
nounced as f k It is imp ortanttokeep this distinction in mind since the
distinction b etween real consonant nalwords and words ending in schwa
in French is sometimes denied for instance in the Government Phonology
analysis defended by Charette
A similar test can b e designed for dialects that are closer to Standard
French It is usually assumed that underlying schwa in these dialects sur
faces obligatorily when adjacent to a so called h aspir e For instance Du
rand notes that phoque havane tobacco brown seal is pronounced
f k avan while fochavane tobacco brown jib is pronounced as f kavan
It is therefore reasonable to assume that f k has a schwathatislacking in
f k Still there are contexts in whichwords like foc surface with a schwa
aisepenthetic I assume at least this schw
In my view the most plausible case for schwaepenthesis is the context
between twowords the rst of which ends in a consonant cluster and the
second of which starts with an onset I have copied the following examples
from Noske and underlined the ep enthetic vowel
a CC C
p enible b un contact
painful a contact
a painful contact
Because Paris French is a so ciologically very prestigious dialect the twowords tend
to b ecome homophonous in Midi Frenchnowadays as well
French
c un index formidable
an index terri c
a terri c index
I consider the schwa in these examples ep enthetic b ecause there is no reason
to assume it is underlying in either the preceding noun contact or index
or the following adjective p enible of formidable the insertion of schwais
optional even in the context of an h aspir e word
It seems that coronal segments do not have the same exceptional sta
tusinword nal p osition in Frenchwhich they have in Dutch The most
imp ortantargument for this is that there are no Frenchwords ending in
three obstruents The nouns in exhibit the maximal numb er of obstru
ents allowed in word nal p osition words like contacts kts or indext
kst are not p ossible in French Coronals are not allowed after clusters
of non coronals either For instance there are no words like armt In
general words allowatmosttwo consonants in nal p osition regardless of
the place features on those consonants It therefore do es not seem unrea
sonable to assume Frenchwords allo w at most closed syllables followed by
a degenerate syllable at their end but no extraproso dic segments
It is nevertheless true that the nal segmentinthewords in is a
coronal Yet this seems to b e due to another e ect namely that in the case
of sequences of equal sonority consonants within one word at least one of
the two segments most often the second one is coronal For instance we
nd acteur actor akt r but not akp r or atk r I therefore assume
here that the coronals in are in the proso dic word unlike coronals in
similar p ositions in Dutch This assumption makes it easier to understand
why these segments trigger ep enthesis The Dutchword contact do es not
trigger ep enthesis in any phonological environment at all I assumed the
reason for this is that the coronal can b e outside the proso dic word where
it do es not need to b e in a syllable Apparently the t in French contact
prefers to b e in a syllable t I take this as an additional argumentthat
it is in the proso dic word
In environments like these the ep enthetic vowel is always schwainFrench
like it is in Dutch The reason for this is the same in the two languages
For one reason or another probably related to syllable structure given the
conditioning environment we need to insert a vowel Since Frenchallows
Apparent counterexamples like encre ink can also b e dealt with in the theory to
be worked out b elow b ecause they consist of a closed syllable followed by an onset
For the sake of completeness I mention that word initially we do nd extraproso dic
s s however station id stasj strict id strikt etc
This observation has b een made and given a formal account in K Rice who
prop oses a notion of binding for heterosyllabic consonant clusters where binds i
has a less marked place feature than and precedes cf also Harris
French
schwas on the surface it is most economical to insert a schwa All other
vowels would intro duce additional features that are not strictly necessary
and therefore forbidden for reasons of economy
Since the alternation here is b etween a degenerate syllable and a schwa
headed syllable wemay assume the relevant constraintisProject N
cons the constraintwhich forces syllables to havevo calic segments as
their heads just as in Dutch p ostlexical phonology
E schwa do es not o ccur at the end of the word Prop erty
This prop erty is di cult to test to the letter as is Prop erty E
schwa only o ccurs in the last syllable of the word The ep enthetic schwas
in o ccur typically b etween twowords so it is imp ossible to tell whether
phonologically they b elong to one word or to the other
Prop erty therefore do es not seem to b elong to the Frenchvowel gram
mar It is simply contradicted by the facts in In Dutchthereasonfor
not ep enthesizi ng in the word nal p osition was Alignment morphologi
cal and phonological edges had to b e aligned with one another as muchas
p ossible The second reason p ossibly was the constraint Final C forone
reason or another proso dic words prefer to end in a consonant
There is an indep endent reason to assume that Frenchhasalower rank
ing for the Align constraint than Dutch This constraint should b e ranked
lower in Frenchbecauseofliaison and encha nement These are two pro
cesses of re syllabi cation of the nal consonant of one word into the onset
of the rst syllable of another word The fact that resyllabi cation is al
lowed over a word b oundary implies that Align should b e ranked relatively
low For a general discussion of the di erences b etween Dutch English typ e
languages on the one hand and Frenchtyp e languages on the other see Pul
gram I will return to this b elow
Dutch for instance do es not haveany phonological lexical resyllabi
cation crossing a morphological word b oundary The reason for this presum
ably is that the constraint Align dominates Onset Consider for instance
the NP roodoog red eye
ro d ox
In the terminologyintro duced by Encrev e liaison is the pro cess that allows
word nal consonants to surface b efore vowel initial words while encha nementisthe
resyllabi cation itself
This statement needs to b e quali ed b ecause syllabi cationover the b oundary b e
tween a stem and one of a numb er of su xes is actually p ossible In McCarthy and
Prince s terminology we are referring to Align Left in this particular case The
same is true for the French example in
French
Candidates Align Onset
p
ro t ox
ro dox
Final devoicing has b een used here as a test for resyllabi cati on The rst
syllabi cation is preferred even if it includes a violation of the Onset con
straint b ecause repairing this restriction would make the proso dic structure
less well aligned with the morphological b oundaries
This apparently is not the case in French It seems wehavetoreverse
the ranking of Onset and Align in this language as can b e seen in forms
like petit ami small friend
Candidates Onset Align
p
p ti ta mi
p ti t a mi
This gives a piece of evidence that the Alignment constraintislower in
French than it is in Dutch It therefore is not very surprising that also
the ordering of the constraints Align and Connect N lax is reversed In
Dutch the ordering is Align Connect N lax
arm Dutch
Candidates Align Connect N lax
p
r m
rm
In French the ordering is reversed again Align is ranked lower than
Connect N lax
It seems unnecessary to distinguish b etween lexical versus p ostlexical phonology in
most asp ects of this particular analysis of French as opp osed to the case of Dutch This
however might b e simply a result of the less detailed nature of my discussion of French
as compared to the discussion of Dutch Also exactly with resp ect to the lo cation of
N lax in the hierarchy Frenchdoesseemtoshow some limited evidence for Connect
a distinction b etween lexical and p ostlexical phonology Lexically schwa can never o ccur
in an closed syllable but p ostlexicallywe will see that the second schwainaword like
devenait b ecame d v n can get deleted leaving the rst schwa in a closed syllable
d v n
Ihavechosen the example arm b ecause it happ ens to b e the same in French and
Dutch except for the p osition of schwa It is however p ossible that the Frenchword ends
in an underlying schwa rather than an ep enthetic schwa I b elieve this do es not a ect
the argumentation I also abstract away from the prop er way of treating the a pair
in French Dell and transcrib e all lowback unrounded vowels as a
French
arme French
Candidates Connect N lax Align
ar m
p
arm
The family of Alignment constraints is not totally irrelevantinFrench Li
aison do es not apply b etween anytwowords but basically only b etween
words that are in some higher order constituent for instance the phonolog
ical phrase Nesp or and Vogel Selkirk Kaisse It seems
therefore that instead of aligning morphological to proso dic words French
rather aligns syntactic to proso dic phrases Below I will argue that French
often seems to have a parameter setting phrase in most high ranking con
straints where Dutch has word
Aschwa can therefore only arise inside a phrase not at the p eriphery
of one The pronunciation k t kt or for that matter p enibl in
isolation as an indep endent phrase is therefore excluded since it has a
non underlying schwa causing misalignment of morphological and syntactic
b oundaries and proso dic b oundaries but the same is not true for the phrase
k nt kt p enibl in which the schwa only causes misalignment at the word
level which as I argued ab ove is irrelevantinFrench
This concludes my discussion of Frenchepenthetic sc hwa Let us now
turn to underlying schwa
U schwamust o ccur in an op en syllable Prop ert y
While this prop erty required some argumentation in the case of Dutchbe
cause it is not directly visible in that language where schwa can b e followed
by a sonorant on the surface in Frenchwe can immediately observe its
truth b ecause there are no forms with schwa tautosyllabic consonantin
that language So while there are words like those in a there are no
words like those in b
a tabl id mod fashion quatr four
b tab l mod m quat r
I argued that the reason why Dutch only allowed schwas or sonorant conso
nants in open syllables was the constraint Connect N lax Since schwas
and consonants havenovo calic features they do not have a feature lax
either and since a syllable can only b e closed if its head is lax the fact
that schwa and sonorants can only o ccur in op en syllabls is explained
The role which Connect N lax plays in the vowel grammar of French
is somewhat complicated It is stronger with resp ect to the constraint
French
Align than it is in Dutch as wehave just seen In chapter I p ointed out
that some southern French dialects like Languedo c French show a clear
e ect of this constraint all mid vowels e and o are laxed to
and in these dialects Yet the syllable structure constraints on schwa
are not limited to these dialects As far as I know schwa cannot o ccur in
closed syllables anywhere in the French language area
Traditionally two pro cesses are used in the analysis of Standard French
phonology which seem to b e related to the constraint Connect N lax
Both pro cesses are known as Closed Syllable Adjustment CSA the
rst pro cess is called e ADJ byTranel and is resp onsible for the
alternations in the second pro cess is called ADJ by the same author
and resp onsible for the alternations in I have added the forms in to
show that there are also instances of underlying segments that do not
alternate with e or atall
e
esp erer hop e inf sp ere esp ere I hop e sp r
c eder cede inf sede c ede I cede s d
lever stand up inf l ve l eve I stand up l v
mener lead inf m ne m ene I lead m n
r ever dream inf r ve r ev I dream r v
f eter celebrate inf f t f ete I celebrate f t
Traditionally it has b een assumed that there are three underlying vowels
e and The vowels in and are the result of laxing
The trigger for laxing in these cases is the fact that the underlying schwa
or e o ccurs in a closed syllable
This analysis ts with the theory prop osed in the previous chapter The
alternation b etween tense e in op en syllables and lax in closed syllables
is exactly what wehave come to exp ect In Dutchthelaxvowel alternating
with e was rather than but given the fact that Standard French
do es not haveany height contrast in the lax vowels at all it is not altogether
unexp ected that in this language should surface as the lax counterpart
to e Igive the representations of e and b elow
Qu eb ec French seems to have Y U but these really b ehaveashighlaxvowels
phonologically They o ccur only in closed syllables and whenever they alternate they
do so with high vowels See D echaine for more discussion of the very interesting
Qu eb ec alternations
French
e
cons cons
V place
Ap erture V place
coronal
lax coronal
The only real problem is that laxing a ects only the unrouded mid vowel
even though French has a lax back rounded vowel and a lax front
rounded vowel on the surface Why do not o and show the
e ect of laxing The examples in furthermore showthatConnect N
lax only works in one direction in French although e cannot o ccur in
closed syllables can o ccur in op en syllables
There are ways of technically solving the rst problem The second
problem will b e addressed b elow For instance we could assume there is a
feature co o ccurrence constraint against the combination of lax and labial
Such a constraint seems to b e su ciently grounded in the phonetics given
the close connection b etween laxness and lowness and the more general ban
on low labial vowels It could b e that this feature co o ccurrence constraint
is ranked higher than Connect N lax so that laxing will not takean
e ect in closed syllables that are headed by rounded vowels On the other
hand the feature co o ccurrence restriction should b e ranked lower than the
relevant Parse Feature constraints so that underlying and can
surface
labial
Parse labial Parse lax Connect N lax
lax
Inow turn to schwa alternations The fact that the emptyvowel is
also sub ject to Closed Syllable Adjustment is predicted by Connect N
lax As wehave seen this constraint has the e ect that schwa can only
o ccur in op en syllables This restriction to op en syllables can havevarious
e ects Supp ose that in French the constraint Connect N lax is ranked
su ciently high It should at least b e ranked higher than the constraint
against insertion of lax as wehave just seen
Connect N lax then forces insertion of a feature lax under a schwa
just as it did under an e in a closed syllable Mere insertion of lax
in the representation of would mean however that we get the following structure
French
cons
vo calic
Ap erture
lax
As wehave also seen for Dutch Frenchdoesnotallowvowels with only
a sp eci cation for Ap erture features In b oth languages ap erture features
need to b e accompanied by a place feature Universally coronal seems
to b e the most unmarked place feature For these reasons the following
representation is preferred over
cons
vocalic
Ap erture V place
lax coronal
This is the representation of Wemight therefore say that Pro jection
theory predicts Closed Syllable Adjustmentinexactlytheway it happ ens
In Dutch the coronal mid vowels are the ones that are most likely to al
ternate with schwa e g e and can b e most easily reduced Under
di erent conditions the same typ e of alternation is found in French and
in Norwegian as we will see shortly b elow It also seems true that in
languages whichlack surface schwa frontmidvowels play the role of that
vowel The theory presented here claims that schwa is a truly undersp eci
ed vowel and that coronal mid vowels are the least marked vowels This
seems to yield the correct result in all cases at hand
A di erence b etween French and Dutch is that underlying can sur
face in op en syllables in the former language but not in the latter This
could b e attributed to a di erence in ranking of Parse lax with resp ect
N lax in Dutch Connect N lax Parse lax so that to Connect
lax vowels that end up in an op en syllable lose their underlying feature
In French on the other hand Parse lax Connect N lax so that un
derlyingly lax vowels always surface as lax even if they end up in op en
syllables
Laxing in the head of a fo ot It is relatively easy to show
however that Closed Syllable Adjustment alone is not enough to explain
the alternation b etween and e on the one hand and on the other
French
Since the work of Selkirk it is known that metrical factors might
also b e involved Schwa do es not only surface as in a closed syllable but
also when followed by another syllable headed byschwa
In Southern French dialects for instance m ene is still pronounced as
m n Similarily aword like s evre I wean s q vr cf s evrer wean
inf s vre in which the lengthening of the rst vowel seems to indicate
that the syllable is op en still has the variantof e also laxes in
this environment and so do the other mid vowels in Midi French It would
be interesting to see whether a No Syllable Theory as discussed for Dutch
in the previous chapter could make the required predictions Given the
ob jections raised against such a theory I will not try to work this out
In other Frenchdialectsvowels in op en syllables are lengthened so that
wehave a contrast b etween serve id s rv on the one hand and S evres
id s qvr on the other The fact that the latter is lengthened
shows that it is in an op en syllable This syllable nal lengthening is even
more apparentinQu eb ec French where the long vowels get diphthongized
Charette
Qu eb ec French
i u
a r ever dream ra ve r qve p alir to b ecome pale pa lir
i
paqlir pr etrise priestho o d pra triz pr qtriz
u
encadrer to frame aka dre akaqdre
u
b partir to leave partir pa rtir merci thank you m rsi
i
ma rsi
Still there are no words suchas s q vr or s e q vr This shows
in my view that neither schwa nor e can o ccur in a stressed syllable even
when this syllable is op en
This argumentthus shows that it is not p ossible to reduce laxing in
theheadofafoottoCSA Wecouldnowwonder whether it is p ossible
to reduce the CSA to laxing in the head of a fo ot The Queb ec data of
Charette seem to p oint in that direction
i u u
a b ete stupid ba t b qt rage madness ra z ra z
i i
r eve dream ra v r qv pr etre priest pra tr pr tr
u u
b palpe feel palp pa lp p arc park park pa rk
i
verte green fem v rt va rt
Vowels followed by one consonantorby a consonant group which can b e an
onset lengthen as if they are in an op en syllable whereas vowels followed
See Charette for a more fully develop ed theory along the lines of Selkirk and
Basb ll Tranel and Verluyten for criticism
French
by a consonant cluster of decreasing sonority do not lengthen at all I think
this is convincing evidence that word nal consonants can form degenerate
syllables as Charette also argues But in that case wemightbeled
to think that indep endent reference to the closed syl lable is not necessary
We do not nd b et or b t b ecause in these cases e or would o ccur
in a branching fo ot not b ecause they are in a closed syllable
The reason why I think that the notion closed syl lable is still needed is
that we ndFrenchwords suchas p rdri but not p rdri or p erdri
There is no obvious wayinwhich the rst syllable in these forms constitutes
a footofitsown
Selkirk argues that the French fo ot is usually monosyllabic In
addition she claims that thereare instances however where the French
foot is morecomplex I submit that yet another principle is at work
in the de nition of the foot in French a principle according to which a
syl lable containing may join together with a preceding syl lable and with it
constitute a foot Notice that this is a prop osal of con icting constraints
since a fo ot built according to the last principle will by necessityviolate
the monosyllabicity requirement on feet This prop osal has several e ects
some of them having to do with schwa deletion which I will discuss b elow
section One other e ect is that twoschwas form a fo ot next to one
another at the end of a word
F
s w
s vr
Selkirk prop oses that in the strong p osition of a fo ot is sub ject to
arulechanging it to Although this analysis is very interesting and I will
follow it in spirit I think it fails to capture one imp ortant generalisation
In this theory the fact that it is exactly schwa that is adjoined to other
syllables to form a bisyllabic fo ot and the fact that schwachanges to
when it is in the strong p osition of a fo ot are formally unrelated Intuitively
b oth pro cesses seem to b e triggered by the fact that schwa cannot b e in
a stressed p osition In a constraint based analysis this typ e of statement
is easier to express we simply say that an unstressed p osition in a fo ot
implies schwa and vice versa than in the rule based approach of Selkirk
where the generalisation has to b e made bywayoftwo formally
indep endent rules
In Optimality Theoretic terms we could say that there is a constraint
French
forcing stress to b e as much to the right as p ossible I will use the shorthand
Align Foot for the constraintwhich has the following formal sp eci cation
h
Align Foot Align F R PrWd R
h
I used the diacritic sign F to indicate the head fo ot of the word I assume
every Proso dic Word has a designated fo ot which functions as its head
says that the head fo ot of the word in French should b e at the rightedge
of that word Since primary stress in Frenchisnever further awayfrom
the word edge than one schwa headed syllable I assume this constraint
is undominated
Because French do es not have reduction I have assumed that the pro
jection constraint Project V Ft whichsays that vowels in unstressed
p osition should not contain vo calic features hence b e schwa is ranked rela
tively lowinFrench To b e more precise all constraints on feature parsing
are ranked higher than this constraint Underlying vo calic features have
to b e parsed in French even in unstressed p osition We therefore have the
following partial ordering
Parse Feature Project V Ft
In this example Parse Feature stands for the set of all parsing constraints
on vo calic place features fParse coronal Parse dorsal Parse labialg
and mayb e also the ap erture features Parse lax Parse high Parse
low This set of constraints seem to cluster together in French it is not
p ossible to establish an internal ranking within this set
Yet I will show that one version of Project Ft the version which
forces the strong p osition of bisyllabic feet to contain vo calic features Project Ft
V is still visibly op erativeinFrench
def
Connect Ft V Project Ft V Project V Ft Nis
the head of a branching fo ot i N dominates lax
I also assume French feet are left headed The following constraint is there
fore undominated
I will assume that French has only one fo ot p er word since there is no evidence
for secondary stress and furthermore sequences of schwa syllables do o ccur in non nal
p osition e g genevois genevian z n vwa This can only b e explained under my
assumptions if neither schwa forms the head of a fo ot which in turn implies under the
assumption that French feet are binary and left headed that b oth are unfo oted
If this assumption is right the Alignment constraint in can b e replaced bya
constraint on plain feet without reference to the head of the proso dic word If Align F
R PrWd R is undominated it can only b e satis ed by candidate outputs whichhave
one fo ot at the outer right edge
This is a technical p oint of little relevance to the argumentation in the text
French
Trochee Feet are left headed
Trochee and the Connect Ft V constraints together haveasane ect
that the ideal binary fo ot contains a full vowel and a schwa in that order
also in French
We can observe that the nal syllable of a word is always stressed unless
this syllable contains schwa in which case the stress is on the p enult
Wecannow extend the hierarchy just established with two additional
constraints FootBin requiring feet to b e binary and lax Both con
straints should b e ranked relatively low FootBin b ecause stress is so often
on the nal syllable in this supp osedly tro chaic system which means the
nal fo ot contains only one syllable and lax b ecause the feature lax is
inserted in l eve m ene appel le etc
Parse Feature Align Foot Trochee
V FootBin lax Connect Ft
This constraint ranking gives us the right results for forms like paru ap
p eared containing only full vowels ordre order with one full vowel and
schwa and appel le call with one full vowel and twoschwas In the fol
lowing tableaux I have put the undominated constraints together in one
column for convenience In this column I have indicated the actually vio
lated constraintforevery candidate
pary
lax Candidates Undominated Connect Ft V FootBin
p a ry
p
pa ry
p a r Parse Feature
p a ry Align Foot
rdr
Candidates Undominated Connect Ft V FootBin lax
p
r dr
r dr
r dr
I will turn to the sp ecial case of enclitics b elow
The ranking b etween FootBin and lax is established in the tableau in b elow
Compare the optimal candidate with p l
It is p ossible that fo ot binarity is not a violable constraint see app endix A If that
is true and if furthermore French feet are tro chaic words ending in a stressed syllable like
par y should b e assumed to have an extra empty syllable and FootBin in these tableaux
should b e replaced by another constraint probably one against empty syllables
French
ap l
Candidates Undominated Connect Ft V FootBin lax
a p l Align Foot
p
a p l
a p l
a p l
a p l
a p l
The question nowiswhywewould turn schwainto the lax vowel inthe
head p osition of a fo ot and not for instance into the tense vowel e even
though the latter vowel has fewer features hence is less marked It seems
that it is not su cienttohave the constraint Connect Ft V ask for any
vo calic feature to supp ort the head of a fo ot Connect Ft V inFrench
demands a supp orting lax feature just like Connect N lax
To conclude this section consider what happ ens with a form ending in
e suchasappelez call ap l e
ap le
lax Candidates Undominated Connect Ft V FootBin
a p le Align Foot
a p le
a p le
p
a p l e
a p l
a p le
One minor problem arises in the analysis of this form In the rst place
the reader mightwonder why the actual output a p l e do es not vio
late the constraint Connect Ft lax For this it is imp ortanttosee
that Connect Ft lax like all pro jection constraints refers to heads of
branching constituents Since the e in this case is not in a branching
constituent it can by de nition never violate this constraint
Although French cannot apply reduction to create an optimal fo ot of
a full vowel and schwa it can in some cases ll an underlying schwa
with vo calic material to comply with this requirement The constraint
Connect Ft lax can therefore b e seen to play a role even in a language
without reduction
U schwa do es not o ccur at the b eginning of the word
Prop erty As I noted in the discussion of Dutch this prop erty seems to
French
b e close to universal LikeDutch Frenchdoesnothaveanyword of the
form I assume the same solution for French as I did for Dutch It
is b etter to leave the schwa unparsed than to satisfy the Onset constraint
by inserting a glottal stop In other words I assume the ranking for b oth
languages
N Contour Parse Vowel Project cons
It is b etter to leaveavowel unparsed than to create an empty syllable
This reasoning app eared to extend to h as well in Dutch schwawas also
absent from p ositions adjacent to this segment It is not p ossible to test this
hyp othesis in French b ecause this language do es not have h consonants
on the surface An exception has to b e made for some Northern French
dialects and for Wallo on in which this h b ehaves like Standard French h
aspir e
The phenomenon of h aspir e aspirated h in general seems to b e of
particular interest As wehaveseenabove haspir e words blo ck deletion
of underlying schwa while at the same time they do not trigger schwa
ep enthesis See Tranel for a p ossible solution of this problem
Consonant clusters b efore schwa cannot b e p ossible com
plex onsets Prop ert y This seems to provide us with the clear
est di erence b etween DutchschwaandFrenchschwa French has lots of
words with complex p ossible onsets b efore schw aasisshown in example
When these words are b orrowed and adapted to the Dutch system
the liquid shifts its p osition
French papavre p oppy papavr cadre id kadr c able id
kabl deb acle id debakl etc
Dutch papav r kad r kab l debak l etc
In Dutch the constraint against empty heads in closed syllables Connect N
lax is ranked lower than the constraint against empty heads in syllables
with an onset Project N cons Apparently inFrench the constraint
ranking is reversed
Input papavr
a Dutch p ostlexical
Candidates Project N V Connect N lax
p
pa pa v r
pa pa vr
French
b French
Candidates Connect N lax Project N V
pa pa v r
p
pa pa vr
Neither French nor Dutch allows for a sequence tr l In Dutch this se
quence is disallowed b ecause of Project N V schwa headed syllables do
not have complex onsets In French it is disallowed b ecause of Connect N
lax schwa headed syllables do not havecodas
Schwa is stressless Prop erty Other than the previous prop
erty this generalisation holds for FrenchasmuchasitdidforDutch I have
already explained why and howschwa is stressles s in French since part of my
explanation for the schwa alternation dep ended on this prop erty Here
as in Dutch the relevant constraintwas Connect Ft lax connecting
feet to vo calic feature material
As in Dutch clitics form an exceptional domain for some schwa prop
erties in French In the latter language enclitics can get word stress
donne le
d n l
give him
When wehave a cluster of schwa syllable clitics stress is on the last one
donne me le
d n m l
give me it
giveittome
These forms are exceptional in twoways Assuming for a moment that the
stem clitic combination is the prop er domain for word stress wehaveto
explain why the stress falls on the nal rather than the pre nal schwa given
the ranking laid out ab ove wewould exp ect don m l or d n m l
instead of and in the second place given the fact that the last clitic
is stressed why do esn t the vowel surface as
The null hyp othesis in these cases would b e that French stem clitic
combinations do indeed form the domain for the normal stress rule I think
the most imp ortant observation to make is that in all these cases is that all
French enclitics are monosyllabic Nowwemight assume that Frenchfeet
are sub ject to some requirementofIntegrity Kager b
French
Integrity AFoot may not cross the b oundary b etween stem and
clitic
Supp osing the Integrity requirement is undominated we get the following
candidate evaluations for donne le and donne me le
d n l
Candidates Undominated Connect Ft V FootBin lax
d n l Align Foot
d n l Integrity
d n l Integrity
p
d n l
d n l
d n l Integrity
d n m l
Candidates Undominated Connect Ft V FootBin lax
d n m l Align Foot
d n m l Integrity
d n m l Integrity
p
d n m l
d n m l
d n m l Integrity
The only serious candidates are those in which the fo ot is built on top of
the schwa syllable only These candidates by necessityhave a monosyllabic
fo ot Since Connect Ft V is a constraint on branching feet it is not
applicable to these forms Insertion of a feature lax therefore do es not
help in anyway but gives a violation constraint against the insertion of
this feature
Schwa cannot o ccur next to a vowel Prop erty This
is again a prop ertywhichFrench shares with Dutch As a matter of fact
the prop erty shows up even more strongly in the former language than in
the latter In Dutch schwa and full vowel can b e adjacent to one another
if they are separated by a morpheme b oundary to b e more precise the
b oundary b etween a stem and a pre x
b e argumenteer
b b r ym nt er
give arguments for
French
In Frenchschwas do not surface in environments comparable to those in
either
a re faire
r f r
do again
b re amener
r mne r mne
to bring back
c p etit e
ptit
small fem
d joli e
z li z lij
pretty
The facts are somewhat more complicated than this example might suggest
One imp ortant problem is that nal schwa do es not surface at all in many
dialects of French
Before I discuss these problems however I rst have to make more
explicit my analysis of monomorphemic forms In these forms Frenchdoes
not showany V or V sequences
I assume French and Dutch are exactly the same in this resp ect For
Dutch I argued that the main reason for the ban on vowel schwa sequences
was the Onset constraint combined with a constraint against sequences of
unarticulated segments like glottal stops and schwas Wehaveseenabove
that the same typ e of argumentation should hold for Frenchaswell given
that French do es not havewords starting with a schwa
Wecannow turn to monomorphemic forms I have already p ointed
out that French and Dutch di er with resp ect to the force of Alignment
constraints In Dutch morphological word b oundaries havetobewell
aligned with their phonological counterparts This caused ep enthetic schwa
to b e inserted inside the word rather than at the edge In French on the
other hand morphological Alignment constraints are not that strong This
has as one e ect that ep enthetic schwa can o ccur at the edge of the word
in the latter language
Wehave already seen that there is additional evidence for this relative
strength in Frenchversus Dutch Frenchallows for resyllabi cation over
A further problem is that re is the only schwa nal pre x in French and that the
analysis of this pre x is not unproblematic De Jong and Hietbrink Yet pro clitics
suchasne not je I le him etc also b ehaveintheway indicated here
French
the word edge liaison whereas Dutch do es not I prop ose that here we
have a similar e ect b oundaries b etween stem and a x are resp ected in
Dutchasmuch as p ossible
What I want to argue then is that in languages likeFrench or Italian
Onset is ranked ab ove Align I also assume that Onset is ranked ab ove
Parse Vowel the constraint forcing vo calic ro ots to b e parsed In a form
like n a pas do es not have we then get the following tableau I used square
brackets to indicate the b oundaries of the lexical words
Candidates Onset Align Parse Vowel
n a pa
p
n a pa
Dutch shows an asymmetry b etween pre xes and su xes with resp ect to
schwa deletion as discussed in the previous chapter French do es not have
this asymmetry at all For su xes we nd exactly the same pattern
elite air
Candidates Onset Align Parse Vowel
e li t r
p
e li t r
An interesting contrast b etween DutchandFrench can b e observed if we
consider the b ehaviour of the in ectional su x it happ ens that this suf
x marks adjectival agreement in b oth languages when it follows a stem
ending in a vowel As wehave seen in the previous chapter Dutch zee en
sea s surfaces as zej The schwa do es surface in the case of tran
sjunctural vowel schwa sequences I argued that the reason for this is a
constraint called MorPa which forces every morpheme to get some reali
sation in proso dic structure at least partly If is the only phonological
element in a su x it is this su x that should b e realized
Yet in French joli e pretty surfaces as z li not as z lij What
this should mean then is that the MorPa constraint is not as strong in
French as it is in Dutch
There is indep endent evidence for this di erence b etween the twolan
guages In Dutch all su xes always surface even if they consist of just a
coronal consonantlike t the third p erson singular marker on v erbs or s
a plural marker on nouns slaap t sleeps slapt adder s id d rs
Wehave seen that the Tilburg dialect of Dutcheven forces these consonants
to b e inside the proso dic word if p ossible so that we get sl pt instead of
sl qpt
French
In French on the other hand many in ectional a xes do not surface ex
cept under very sp ecial circumstances One instance is the plural marker s
It is imp ossible to nd a phonological re ex of this grammatical marker on
the form bon s good b if pronounced in isolation even though phono
tactically there is no reason why this form should not o ccur in Mo dern
French The s only surfaces in liaison contexts bons amis go o d friends
b zami There is therefore indep endent evidence to assume that French
do es not attachmuch imp ortance to the phonological parsing of all in ec
tional a xes Unlike Dutchitdoesnothave a reason to treat the nal
schwainjoli e di erent from any other schwa
Schwa deletion As I have indicated several times ab ove schwa
deletion in Frenchismuch more common than it is in Dutch Noske
distinguishes b etween six typ es of schwa zero alternation One of these
is ep enthesis discussed in sections and ab ove Two others are
prevo calic and p ostvo calic sch wa deletion discussed in section The
remaining three contexts are
Schwa deletion in a two sided op en syllable This alternation
is perhaps the most general one according to Noske A schwain
a two sided op en syllable is a schwa in the context VC CV or
VC OLV O obstruent L liquid In other words schwahasto
b e in an op en syllable has to have a simple onset and b e preceded by
an op en syllable Ihave copied the following examples from Noske
a tu devenais you b ecame
i tyd vn
ii tyd v n
iii tydv n
iv tydvn
b Henri devrait partir H would havetoleave
c ridvr partir
i ari d vr partir
ii aridvr partir
d Jacques devrait partir Jacques would havetoleave
On variations of this p ointbetween Standard FrenchandQu eb ec French see Charette
French
i zakd vr partir
ii zakdvr partir
Schwa deletion in phrase nal syllables Apart from this most
general typ e of alternation Noskeshows that wehave to distinguish
twocontexts in which underlying schwas do not surface both at
the p eriphery of the phrase Phrase nal schwa is deleted without
exception in Standard French
a la terre est plate the earth is at lat r plat
b la route est longue the road is long larut l g
Schwa deletion in phrase initial syllables Although somewhat
more marked and under somewhat more restricte d conditions phrase
initial schwas can also b e deleted
a Revenez demain Come back tomorrow r v nedm
padbil b Tefaispas de bil Don t worry t f
I will discuss Noske s rst two cases b elowbutIhavevery little to say
ab out his last instance of schwa deletion Noske remarks that this is the
most marked form of deletion I basically accept Noske s analysis that more
complex onsets are allowed at the b eginning of the phrase than in other
p ositions just like more complex onsets are allowed word initially than in
other p ositions in other languages for instance in Polish cf Gussmann
Below I will argue that the phrase in Frenchplays the role of the
word in Dutch or English in many other resp ects
According to Charette the reason whyschwas in Frenchhave
a tendency to not surface at all is the Avoid Vowel Principle
Avoid Vowel Principle
An emptynucleus remains uninterpreted whenever p ossible
The di erence b etween an interpreted and an uninterpreted nucleus in
Charette s terminology for all practical purp oses is the same as that b etween
aschwa headed and a degenerate syllable in our terms Within the formal
ism develop ed here this constraint can therefore b e stated as cons the
ban on vo calic ro ots The whenever p ossible clause in Charette s formu
lation is an indication that there are constraints which are in con ict with
the AVP
In the chapter on DutchIhave already discussed the most imp ortantof
these constraints For instance we need to make sure that only schwacan
French
b e deleted The reason for this is that all other vowels havevo calic features
like coronal labial high etc which need to b e parsed Apparently the
parsing of these features is more imp ortantthantheAvoid Vowel Principle
but the parsing of ro ots i e cons features is not otherwise schwa
would never b e deleted either
Parse Feature cons Parse Vowel Project cons N
I use Parse Feature as a shorthand for the set of related constraints
Parse coronal Parse labial Parse low etc
Another constraint that is of relevance is the constraint against degener
ate syllables Project N cons Degenerate syllables are disfavoured in
French except in phrase p eripheral p ositions Vowels can therefore only
b e deleted if their surrounding consonants can b e parsed in a di erentway
N cons cons Project
I assume that schwa deletion is a p ost lexical pro cess in French Given the
fact that two sided op en syllable deletion is p ossible only when the schwa
syllable is preceded by an op en syllable I think there is no ob jection against
accepting Noske s analysis that the consonant preceding underlying schwa
gets syllabi ed into the co da of the preceding syllable It is signi cantthat
the de nition of this two sided syllable seems to b e exactly cast in such
awaysoastomake this resyllabi cati on p ossible we can have rid vr
but not zak d vr b ecause the latter form necessarily contains a non
p eripheral degenerate syllable
A problem might seem to arise in the case of devenait when pronounced
as d v n b ecause now the rst syllable is closed while its head still sur
facesasschwa not as Yet wehave already seen in chapter that
in Dutch p ostlexical phonology where we insert schwaincontexts like
theat r the constraint Connect N lax can get lower ranked so that it
is no longer op erative in cases like this For French we could for instance
assume that p ostlexically dorsal lax dorsal Connect N lax We
will see b elow that Norwegian actually shows dialectal di erence with re
specttothispoint some dialects are exactly likeFrench and Dutchin
reranking the order b etween the constraints but in other dialects the rank
ing holds at every level of phonological analysis disallowing schwatooccur
in any closed syllable on the surface
We can now turn to word nal schwa deletion The interesting thing
ab out this deletion site is that it allows schwa not only to b e deleted in a
two sided op en syllable but also when preceded by a closed syllable or
when the schwa headed syllable has a complex onset Deletion is even p os
sible when schwa has a complex onset and is preceded by a closed syllable
French
as in Chartres name of a town sartr Since we are dealing with word
nal syllables the schwa syllable is actually alternating with a degenerate
syllable The nal syllable of quatre in J en ai quatre I have four of them
therefore is either a or b
N a
N
N
t r
b N
N
N
t r
The form in a is chosen inside a phonological phrase and b efore a con
sonant in other words in exactly the same environmentasschwaepenthe
sis as in qua tr copains four friends whereas the form in b is chosen
phrase nally
Wehave already seen that the Alignment constraints do not playarole
at the word level in French b ecause liaison causes syllabi cation over the
word b oundary and ep enthesis applies at the edge of words but that an
Alignment constraint is necessary at the level of the syntactic and phonolog
ical phrase again in order to restrict application of liaison whichdoesnot
apply to segments in anytwo adjacentwords Yet even phrasal alignment
cannot explain the deletion of schwaifwe assume that at least in some cases
the underlying schwa is nal For instance the phonological representation
lat r plat and the morphological and syntactic representation la terreest
plat e the earth is at do not seem particularily well aligned
We therefore have to nd a high ranking constraint dominating at least
Align the phrasal alignment constraint I use as a shorthand for the
phonological phrase McCarthy has prop osed a constraintwhich
ant to us in his Optimality Theoretic analysis of Eastern might b e relev
Massachussetts r insertion This constraintisFinal C
Final C A Proso dic Word has to end in a consonant Align PrWd
R C x
French
McCarthy argues basically that in the Eastern Massachussetts dialect of
English an r is inserted in vowel nal words in order to satisfy this con
straint spar idear etc
It seems that a similar constraint could b e at work here although French
would have a di erent parameter setting for Final C the phonological
phrase instead of the proso dic word and use a di erent strategy to reach
its aim and goal Instead of inserting a default consonant Frenchwould
prefer to delete vowels The following constraint ranking could achieve the
appropriate result
Parse Feature Final C Align Parse V
The ranking Final C Align Parse V has as a result that it is b etter
to delete phrase nal underlying schwa in order for a word to end in a conso
nant than to keep syntactic and phonological representations well aligned
The constraint ranking Parse Feature Final C on the other hand
ensures that only schwa can b e deleted not some full vowel which has more
phonological features
A parameter The phonological phrase in French seems to as
sume a lot of the tasks that are played by the phonological word in lan
guages like Dutch or English Pulgram This is true at least for
syllabi cation primary stress and ep enthesis The Alignment constraints
that are parametrized for the word level in Dutch seem parametrized for
the phrase lev el in French
Dutch English French
ep enthesis Align PrWd X Y X Align X YP X
syllabi cation
h h
stress Align F R Word R Align F R R
Final C Final C Word Final C
I assume now that like the two alignment constraints also Final C is
parametrized In Eastern Massachussetts English it is a constrainton
the level of the proso dic word Yet in French also this task is taken over
by the phrase it is phonological phrases notwords that have to end in a
consonant
In this particular place it seems preferable to depart slightly from the
ortho dox optimality theoretic p oint of view If the traditional way of lo ok
ing at things of Pulgram is right and there are languages in which
several pro cesses play a role at the phrase level which all play a role at
the word level in other languages there is reason to intro duce a limited
principle of parameter setting into the theory
French
Such a revision of the theory would also b e more attractive from the
p oint of view of language learnability UG would have some subset of
constraints all of which could b e sp eci ed for either the phrase or the
word These constraints would refer to a parameter ranging over e g fword
phraseg and all these constraints would b e set in the same setting by the
language learner The child would observe for instance that Dutchdoesnot
haveany plausible case of phonological resyllabi cation and therefore she
would set the parameter in every constraint in the appropriate way
On the other hand if it would turn out that there are mixed systems
sometimes referring to words and at other times to phrases wemight still
have reason to b elieve that all four constraints in are presentinUni
versal Grammar and that it is just an accidentthatFrenchcho oses the
phrasal variants of the alignment constraints and Dutchcho oses the word
variants Extensivetyp ological study would b e needed to settle this p oint
Assuming Final C takes precedence over Align hence by transitiv
ity also over Project N cons in French as it did in Dutch and assuming
that la terre est plate forms one phonological phrase we get the tableau in
Parse Features
Final C
Align
N cons Project
cons
Parse vowel Project cons N
la terre est plate the earth is at
Candidates Final C Align Project N cons
p
la t r pla t
la t r pla t
The question now is what happ ens to forms in which the nal schwa headed
syllable contains a complex onset as in feminine autre other Given
the fact that the pronunciation ot with deletion of the r as well as
of the schwa satis es all constraints just mentioned just as well as the
pronunciation otr it ends in a consonant it leaves one schwa unparsed it
has a degenerate syllable etc and since also the constraint Project N
cons banning unnecessary onset segments from empty headed syllables
is active at some level why do es not this rst form surface
For some dialects the answer to this question is a rmative For instance
Breton French Lo dge seems to allow exactly two pronunciations of
quatre four
French
kat
katr
Phrase internally schwa can surface and r can surface with it so that we
get the form in At the end of the phrase the schwa do es not surface
p erhaps b ecause of Final C But then there is no reason in Breton French
for the r to surface assuming the constraint Contiguity blo cks the
parse ka t r
In Paris Frenchwe can distinguish b etween several style registers In
informal sp eechwe nd kat at the end of the phrase but in more for
mal registers we nd katr The di erences b etween style registers I have
argued in the previous chapter can b e describ ed as a reranking of Faith
fulness conditions the more formal the style of sp eech the higher ranked
the Faithfulness constraints In the case at hand it could b e that in Paris
French there is a very formal style of sp eech in which for instance Parse
consonant is ranked higher than Project N cons In this style of
sp eech underlying consonants would always surface even if that means
that they surface in the complex onset of a degenerate syllable whichis
otherwise a very marked con guration
Candidates Final C Parse C Project cons
kat
p
katr
katr
Dell has p ointed out that only an analysis which assumes that nal
obstruent liquid sequences are onsets of degenerate syllables can explain
why the liquid is deleted in exactly these words and not in others For
instance an analysis assuming an optional rule deleting word nal sono
rants like the following can not explain why the sonorantisnever deleted
in word like parle sp eak or rhythme rhythm
In the theory presented here parle and rhythme are syllabi ed as pa r l
and rit m resp ectivel y there is indep endent reason to assume rl and tm
are not p ossible onset clusters so that the same problem do es not arise
There is no need to simplify the onsets of these degenerate syllables b ecause
they are already su ciently simple
For the sake of completeness I should mention that Dell lists a few clusters
which can o ccur b efore full vowels but not b efore schwaorword nally These are tn
French
Conclusion plus a note on learnability Despite the sup er
cial di erences b etween French and Dutch I haveshown that the grammars
of the two languages are only minimally di erent The basic di erence is
one in parameter setting What the word b oundary is for Dutch is the
phrase b oundary for French This accounts among other things for the fact
that Dutch has ep enthesis only within a word and French only within a
phrase
As for the ranking of constraints only a few rerankings were needed
In the rst place in French Connect N lax do es not a ect the rounded
vowels whereas in Dutchitdoes
labial
a Dutch Connect N lax
lax
labial
b French Connect N lax
lax
In the second place since in French lax vowels can o ccur in op en syllables
wehave the following reranking
a Dutch Connect N lax Parse lax
b French Parse lax Connect N lax
Because French do es not have reduction al l Parse feature constraints
are inviolable This also means that there is no way to establish an internal
ranking b etween di erentversions of Connect Ft lax
Finally the stress system of French is somewhat di erent from that of
Dutch In Dutch the feet are always binary b e it that sometimes they
include a catalectic syllable In other words Fo ot Binarity is undominated
in that language In French this constraint is dominated at least by In
tegrity by Connect Ft lax and by Align Foot
In other words b oth Connect N lax and Fo ot Binarityhavesunkto
a somewhat lower place in the hierarchyinFrench But their e ects are
still visible so that we cannot say that these constraints are parameters
switched o in French Wethus have found evidence in favour of parameters
ranging over a set of values suchasfword phrase g while at the same
time wehave evidence against straightforward binary parameters of the
on o typ e In comp ensation the Parse feature constraints havegot
ranked higher in French than was the case in Dutch
ethnologue ethnologist tl athl ete id kn acn e id b z objet ob ject dm
admirer admire and gz exil id All of these clusters are word internally analysed
as co da onset sequences It is not clear to me what would disallow them to o ccur b efore
aschwaorword nally
Norwegian
It seems to me that a child should b e able to acquire these rankings
fairly easily As already discussed at various places ab ove I prop ose that
at an initial stage the child assumes that all constraints are inviolable except
for the faithfulness constraints which are ranked very low cf section
in the rst chapter A Dutch sp eaking child can learn that Project
Ft constraints are ranked fairly low when confronted with di erentstyles
of sp eech in which full vowels do not alternate with schwas Given the
unpredictable nature of the full vowel in these contexts it is easy to see that
these alternations are due to reduction But in reduction Parse feature
constraints are always violated and Pro jection constraints are probably the
only p ossible causes for reduction On the other hand a Frenchchild can
learn that Connect N lax is ranked relatively low when she observes that
there are tense rounded vowels in closed syllables and lax vowelsinopen
syllables e g at the end of the word A French sp eaking child can acquire
the low ranking of Fo ot Binarityby observing that there are twotyp es of
feet sw and s and that the latter pattern o ccurs regularly i e cannot
b e treated as an exception as is the case in Dutch It is interesting that
the Frenchchild do es not have to learn anything particular ab out schwa
while the Dutchchild only has to see that it alternates with other vowels
Everything else follows from the empty nature of this vowel
Norwegian
Inow turn to Norwegian This language also has a schwa with some in
teresting prop erties but unlikeforFrench and Dutch there is no tradition
of studying these prop erties For the data in this section I have mainly
relied on the pioneering study on Norwegian syllable structure by Kristof
fersen and myown research on a Norwegian dictionary Haugen
Norwegian is interesting for several reasons Wehave seen that French
typ e schwas can have a complex onset but no co da table and Dutch
t yp e schwas can have a co da but no complex onset tafel Norwegian
has b oth typ es of schwa tig r ordr It do es not however haveschwa
headed syllables with b oth a complex onset and a co da katr l Norwegian
dialects furthermore di er from each other in interesting ways in closed
syllables some dialects have instead of schwa and in word nal op en
syllables some dialects haveanemptynucleus rather than schwa
I will concentrate here on Norwegian ep enthetic and underlying schwa
Ihave not b een able to nd su cient evidence for vowel reduction in any
Norwegian dialect It is true that unstressed syllables in the Germanic
stratum of the lexicon are almost always headed byschwainmanydi
Norwegian
alects of Norwegian while they are headed by e in others but as far as I
know there is no evidence that this schwa in the former group of dialects
has to b e derived from some underlying full vowel As far as ep enthetic
schwa is concerned I will concentrate only on those cases whichinDutch
I called sub cases of u schwa the schwawhichoccursword nally b etween
an obstruent and a sonorant consonantasin tiger Neither ep enthesis nor
reduction therefore seems at issue in Norwegian
Again the starting p ointformy analysis of Norwegian is schwa s Prop
erty
Prop erty Schwa is exceptional in the vocalic system because it bears no
vocalic features
Kristo ersen notes that in some dialects of Norwegian schwais
somewhat more fronted and sounds almost like a coronal mid vowel e
or Yet in most dialects the distinction b etween those vowels and schwa
seems to b e su ciently clear
Schwa is the ep entheticvowel Prop ert y Although
Norwegian do es not haveschwaepenthesis of the Dutch hel p typ e some
dialects show evidence for ep enthesis in words which underlyingly end in an
obstruent followed by a sonorant consonant as in Dutch theat r Kristof
fersen p oints out that Oslo Norwegian has a lot of words ending in
syllabic sonorants the sup erscript ed numb ers indicate the stressed syllable
of the word the typ e of tonal accent asso ciated to the word in question
known in the Scandinavian literature as accent and accent
n k kl key doqvn lazy kl simple b kn cymbals
laqkn sheet vimpl streamer m nstr pattern
In standard Norwegian orthography the words in question are transcrib ed
with an e between the obstruent and the liquid This schwa is also pro
nounced in many southern Norwegian dialects which therefore have the
following forms
k n laqk n vimp l m nst r n kq l doqv n k l b
Kristo ersen argues that these schwas should b e analysed as ep enthetic
for the following reasons In the rst place if the schwas were underlying
we could not explain why they are always followed by a liquid and never
by an obstruent given that wehaveaword like hasel hazel has l why
are there no words like has t for Oslo hast As wehave seen in
Norwegian
the discussion of Dutch ep enthesis theory can explain this fact given the
minimal assumption that only sonorants can b e nuclear heads
A further observation supp orting ep enthesis is that there are no monomor
phemic words with identical sonorant consonants preceding and following
the vowel Thus while there are derived forms like mur er mason mTrer
in Oslo there are no phonologically similar monomorphemic forms This
can b e understo o d in a theory with ep enthesis in terms of the Contour
constraint dominating Parse C as wehave seen for a similar observation
on Dutchinchapter
The fact that the ep enthetic vowel is schwa in these cases should not
come as a surprise As in Dutch and French there are reasons for inserting
avowel slot Since schwa is the most minimal vowel intro ducing the
smallest numb er of features this is the vowel that is chosen
Ep enthetic schwa do es not o ccur at the end of the word
Prop ert y Norwegian is likeDutch and di erent from French in its
selection of ep enthesis site Apparently in word nal CC clusters Norwegian
c ho oses the p osition between the two consonants rather than the p osition
following them Consider for instance the resp ectivewords for tigre in each
of the three languages
a French ti gr
b Dutch tij r
c Norwegian ti g r
In section I connected the di erence b etween French and Dutchto
other di erences b etween these two languages French has pro ductive ex
ternal sandhi while Dutch do es not French seems to select the phrase
rather than the word as a domain for stress etc I suggested that there is
a parameter like mechanism at play where various constraints which are
formulated at the word level in Dutchhave b een formulated at the phrase
level in French
Given the Norwegian ep enthesis site wenow predict that this language
should b e likeDutch with resp ect to sandhi as well The word internal
ord The schwaisbrought ab out by a relatively high ranking for Align w
parameter setting of this language therefore is word
The fact that there even are no t clusters means that the status of word nal
coronals in Norwegian is di erent from their status in Dutch Wehyp othesizedthatin
Dutchschwa is allowed to o ccur b efore coronals b ecause these were extrasyllabic and
therefore did necessarily not close the syllable In Norwegian apparently these coronal
consonants are parsed into the proso dic word whenever they can thereby closing the
syllable and forcing its vowel to o ccur as wherever this is p ossible
Norwegian
In this light it b ecomes relevant that Norwegian is like the Romance
language Italian a language which arguably has the parameter setting
phrase just likeFrench in that on the surface it has b oth long vowels
and long consonants and that it has a requirement that stressed syllables
should b e bimoraic this is due to a constraint I will call Prokosch Law
in chapter Furthermore vo calic length and consonantal length are in
complementary distribution in b oth languages in the sense that we can de
rive the one if we p osit the other in the underlying representations In the
literature on b oth languages there is a debate whether we should p osit
vowel length or consonantal length as underlying
Connected to this is Italian Raddoppiamento sintattico ifaword ends
in a stressed vowel the consonant of the following word gets lengthened
if the twowords are in the same phonological phrase Italian thus is
likeFrench in that it has external sandhi The Align word constraintis
ranked accordingly low
Interestingly Norwegian has a pro cess that is phonologically very similar
to Raddoppiamento Johnsen KayeandVergnaud Yet this typ e of
Raddoppiamento never crosses the b oundaries of the syntactic X It
therefore seems reasonable to exp ect Align word is ranked su ciently
high in Norwegian so that we predict word internal rather than word nal
ep enthesis indeed
Schwamust o ccur in an op en syllable Prop ert y Nor
wegian b ehaves exactly the same as Dutch with resp ect to this prop erty
ep enthetic schwas are forced inside a closed syllable for reasons of Align
ment Real underlying schwa on the other hand always o ccurs in an op en
syllable otherwise wewould still also exp ect has t
There are however some dialects of Norwegian which disallowschwain
closed syllables even in vowel ep enthesis One instance of such a dialect is
Bo d Norwegian sp oken in a town in Northern Norway In this dialect the
Oslo forms in are pronounced as follows
m nst r n k l doqv n k l b k n laqk n vimp l
In light of the discussion of French closed syllable adjustmentaboveitisof
course quite interesting that the vowel in the closed syllable surfaces as
Raddoppiamento Sintattico Norwegian Style as Johnsen KayeandVergnaud
call it applies b etween the two parts of a comp ound The initial consonant of the second
comp onent is lengthened if the rst comp onent ends in a stressed vowel This raises of
course the question what exactly de nes the domain of this pro cess since it is usually
assumed the two parts of a comp ound are X category elements Apparently the two
parts of a comp ound do not countas words in the sense of Align word The most
imp ortant observation however is that there is no external sandhi
Norwegian
For Standard Dutchwe assumed that the reason whyepenthetic schwas
other than underlying schwas can surface in closed syllables is that at the
p ost lexical level where these schwas surface the Connect N lax con
straint gets ranked very low The same should b e said ab out the southern
Norwegian dialects whichhaveepenthetic schwa
Bo d Norwegian is di erent from these languages in that Connect N
lax still plays a role in this dialect In particular Connect N lax dom
inates the constraint lax On the other hand Bo d Norwegian is similar
to Dutch and southern Norwegian in that it do es need to haveavowel
in every syllable unlike Oslo Norwegian whichallows syllables headed by
sonorants at every level
Bo d Norwegian
a Project N cons Connect N lax lax
b laqkn
lax Candidates Project N cons Connect N lax
laqk n
laqkn
p
laqk n
Southern Norwegian
a Project N V lax Connect N lax
b laqkn
Candidates Project N cons Connect N lax lax
p
laqk n
laqkn
laqk n
Oslo Norwegian
a cons Project N cons
b laqkn
Candidates Project N cons cons
laqk n
p
laqkn
laqk n
I do not transcrib e phonetic di erences other than the ones that concern us here In
particular there are variations across dialects in the pronunciations of aq and k which I do not note
Norwegian
The Oslo dialect is di erent from the others in that it has the Avoid
Vowel Principle cons ranked very high also at the p ostlexical level
Ep enthetic vowels are therefore disallowed regardless of their quality
Schwa do es not o ccur at the b eginningoftheword
Prop erty It seems that this prop erty is among the most universal
prop erties of schwa Norwegian do es not allowword initial schwas either
Apart from the Contour constraint against syllables with unarticulated
segments in b oth onset and nucleus one other factor seems at work in
Norwegian to prohibit word initial schwas
The fact is that schwa in Norwegian is found almost exclusivel y in the
Germanic stratum of the lexicon and that stress is on the rst syllable of the
word in this stratum Since schwa is unstressed in Norwegianjustasmuch
as in DutchandFrench this has an e ect that we hardly nd anyschwain
word initial syllables at all even if these syllables do have a normal onset
As a consequence of the fact that most words in the Germanic stratum are
bisyllabic and are almost in complementary distribution
This solv es the rst problem just mentioned The second problem is
why the vowel in VCCV sequences like eple is always short I prop ose that
N V isstill the reason for this is that also the strong version of Project
activeinNorwegian Complex onsets b efore a schwa syllable are avoided
wherever this is p ossible and syllables are maximally bimoraic
According to nativespeakers intuitions eple is syllabi ed as p l not
as e pl Wemight therefore assume that although the strong version of
Project N V isranked fairly lowinNorwegian it is still ranked higher
than the Syllable Contact Law
N V SCL Project
epl
Candidates Project N V SCL
eq pl
p
p l
If p ossible Norwegian still prefers schwa syllables to have simple onsets
even if complex onsets are not completely disallowed as we will discuss
now
Norwegian has the Germanic pre x be which contains a schwa Yet pre xes are not
necessarliy stressed as we will see in the next chapter
N V should also outrank the relevant faith If vowel length is underlying Project
fulness constraints
Norwegian
Consonant clusters b efore schwa cannot b e p ossible com
plex onsets Prop ert y As far as this prop erty is concerned Nor
wegian seems to b e for one part likeDutch and for another likeFrench
While ep enthetic schwas do not allow for complex onsets tiger tigler
underlying schwas in op en syllables do monstre pattern venstre left
etc Still there are no schwa syllables with b oth a complex onset and a
co da katr l Furthermore in words ending in V CC like eple
i
the vowel seems to b e always short These are the facts that I wantto
explain in this section
First let us turn to the di erence b etween schwa headed syllables and
syllables headed by a sonorant which get an ep enthetic schwa p ostlexi
cally What I just observed is that the latter cannot have a complex onset
whereas the former can In other words a and b are allowed but
c is not
a N b N c N
N N N
N
N
p l p r p
l r
In chapter I argued that we should distinguish b etween two pro jection
constraints one stating that syllables should have heads with vo calic fea
N V and another stating that syllables should haveat tures Project
N cons least a cons ro ot as their head Project
a Project N V N branches N should dominate a vo calic
feature
b Project N cons Nbranches N should dominate a vo
calic ro ot cons
The two onset segments in a givetwo violations of Project N V
b ecause they are not supp orted byavo calic head but they do not cause any
violations of Project N cons b ecause at least their head is a vo calic
ro ot On the other hand the two onset segments in c violate b oth
versions of Project N twice b ecause the head do es not haveanyvo calic
Ionceagainwanttopoint out that I am discussing underived forms only Polymor
phemic forms like epler apples plural do o ccur
Norwegian
N feature and it do es not haveavo calic ro ot The twoversions of Project
are in a P aninian relation b ecause a violation of Project N V will also
give a violation of Project N cons but not vice versa
We can establish the following constraint ranking
Onset Project N cons Parse Consonant Project N
V
N V has suchalow ranking in Norwegian that it Actually Project
seems to me that there is no reason to assume it is there unless one wishes
to stick to the idea that all constraints are present in the grammars of all
languages
Ihave noted earlier on in this chapter however that it mightbede
sirable to intro duce some idea of a parameter into the theory In the case
of the pro jection and weakness constraints one could imagine that the
feature parameter could b e set for every individual pro jection parameter
Thus French has the parameter setting lax for Project N and Project
N Norwegian has the Ft but the parameter setting cons for Project
same parameter settings for Project N and Project N but presumably
a di erent setting for Project Ft I think at least in the data presented
in this thesis there is no real counterevidence to this claim although we
have to assume that Dutch lexically has the setting V for Project N and
p ostlexically the setting cons Project N V will b e attributed a mi
nor role in the analysis of p ostlexical Rotterdam Dutch presented in the
next chapter but it is not clear whether this has rep ercussions for the anal
ysis of the Standard Dutch p ostlexicon cf the arguments for constraint
ranking in app endix B
Onset has to b e undominated b ecause all schwa headed syllables in
Norwegian have onsets Wenow get the following tableaux for a c
resp ectivel y
a pl
Candidates Onset Project cons Parse C Project V
p
pl
p l
p l
p l b pr
Norwegian
Candidates Onset Project cons Parse C Project V
p
pr
p r
c plr
Candidates Onset Project cons Parse C Project V
plr
p
p l r
p
p lr
p l r
In the case of underlying plr either the p or the l will survive The
exact decision will b e made by constraints on Contiguity of parsing Align
ment and on the sonority slop e The actual decision is not very interesting
I think and therefore I will leave the topic here
Alternation with degenerate syllables According to Kristof
fersen most masculine and feminine words whichareschwa nal in
most Norwegian dialects are pronounced without schwa in northern East
Norwegian In the terminology used in this dissertation these forms end
in a degenerate syllable rather than a schwa syllable
Two questions now arise In the rst place why do es this dialect prefer
degenerates over schwa syllables And in the second place whyisthis
preference restricted to the masculine and feminine gender
The answer to the rst question is quite obviously theAvoid Vowel
Principle cons Apparently this principle is ranked much higher in
northern East Norwegian than it is in any other dialect In other dialects
the Avoid Vowel principle is ranked b elow the parsing constraint Parse
N cons In the north east the ranking is exactly Vowel and Project
the opp osite
a moqn NE Norwegian
Candidates Parse Vowel cons
moq n
p
moq n
b moqn other dialects
cons Candidates Parse Vowel
p
moq n
moq n
Norwegian
To see why neuter nouns are di erent it is imp ortant to note that Norwe
gian seems to b e likeDutch in one imp ortant resp ect when a neuter words
ends in a schwa syllable the latter almost always has a complex onset
Wehave seen in section that Nijen Twilhaar observed this
for loan words in Dutch In Norwegian the correlation seems to hold also
for the Germanic stratum
The question therefore reduces to the question whywords with a com
plex onset resist deletion of the schwa Here again the distinction b etween a
cons and a V parameter setting of Project N b ecomes relevant Schwa
headed syllables are the only weak syllables with a complex onset in Nor
wegian Degenerate syllables are not allowed to have complex onsets as we
have seen and neither are syllables headed by sonorant consonants
Conclusion and another note on learnability The di er
ences b etween Norwegian and Dutch orFrench do not seem to b e very
great Norwegian like Dutch seems to treat the word edge as the sp ecial
edge as opp osed to the phrase The only real di erence b etween Dutch and
Norwegian seems to b e that in the former language sonorantandschwa
heads of syllables b ehave in exactly the same way Most imp ortantly they
b oth disallow complex onsets In Norwegian on the other hand evidence
emerges that we should draw a di erence b etween schwa headed syllables
and syllables headed by sonorants the former do allow for complex onsets
whereas the latter do not
Again this distinction is fairly easy to learn Hyp othesizing that in the
initial state neither schwas nor sonorants license an onset the Dutchchild
do es not have to learn anything in particular The Norwegian child will
however one day hit up on say theword ordr e At that p oint she will know
that she has to relax the condition on schwa headed syllables rank it b elow
Faithfulness say I assume in cases like this the child reranks only the
N V The most sp eci c version of a given family inthiscase Project
weaker version stays high up in the hierarchy and will not shift from there
b ecause there is no evidence to that e ect
Notice that there is no way to get to a system in which sonorants have
a complex onsets whereas schwadoesnot Assoonasachild would hit
up on a form kabrl she would have evidence to rerank Project N V to
alower p osition But this would also make p ossible schwa headed syllable
with complex onsets The reader can check that there is no way to rank the
constraints given here in suchaway that we get a system in which katrl
The hyp othesis here is that children rst get acquainted with somewhat more formal
style registers There seems to b e some evidence for this hyp othesis indeed Paula
Fikkert p c
Conclusion
is allowed but katr isnot
Also the distinctions b etween the di erent dialects of Norwegian seem
to b e learnable As in the discussion of the distinctions b etween French
and Dutchabove I assume that all constraints are ranked very high in
the initial stage of the acquisition pro cess except for faithfulness In this
case it means that the Bo d dialect represents the initial stage of the
acquisition pro cess The child from Oslo observes that in lakn there is
a syllable without a vowel thus hyp othesizing that a constraint against
inserting vowels should b e ranked higher than the constraint Project N
cons Achild from Southern Norway observes that lak n contains a
violation of Connect N lax She therefore assumes that a constraint
against the intro duction of lax is ranked higher than this constraint Since
there is no reason for her to assume that Project N cons also has
to leave its undominated p osition this automatically implies the ranking
N cons lax Connect N lax Project
Conclusion
Kager and Zonneveld remarked that it would probably b e di cult to
apply the No Syllable Theory they prop osed for Dutch to other Germanic
languages as well It is in myeyes a curious thing that although default
vowels are a well studied topic in many languages comparative studies are
rare one exception is Noske In this chapter I haveshown that de
spite the surface di erences the grammars needed for Dutch French and
Norwegian schwa are very similar given an idea of reranking in addition to
some minimal idea of a parameter setting In all three languages this vowel
only pro jects a very simple typ e of syllable in accordance with the predic
tions of the Pro jection Theory The exact nature of the syllable dep ends
on other parts of the grammar but these di erences are indep endently
observable
A Vowel Glide Alternation in Rotterdam
Dutch
Intro duction
In the previous chapters I have discussed the relation b etween typical vo
calic features such as lax and low and proso dic structure In this chapter
I will examine under which conditions vowels can o ccur in an extracephalic
p osition in the syllable In particular I will review a vowel glide alternation
in the Rotterdam dialect of Dutch In this dialect i alternates with j
under certain conditions which will turn out to b e interesting also from
the p oint of view of my theory of schwa I will show that all alternations
are the result of an underlying i sequence in whichtwovowels are in
comp etition
The diminutive su x in the Rotterdam dialect of Dutchhasinteresting
distributional prop erties some of which it shares with its counterparts in
other Dutch dialects For instance Standard Dutch has the alternating
forms pje kje tje and etje
raam pje koning kje la tje ding etje
window dim king dim drawer dim thing dim
little window little king little drawer little thing
This alternation is sub ject to phonological conditions in Standard Dutch
the form of the su x is dep endent on the phonological shap e of the stem
With a few di erences to b e discussed b elow the same paradigm is found
in the Rotterdam dialect In particular the consonantal alternations in this
dialect are almost the same as in Standard Dutch
I will not go into the details of the alternation b etween etje tj as
attested in dingetje and the other forms since it already received a consid
erable amount of attention in the literature Kruisinga Cohen
Ewen Gussenhoven Trommelen Van Zonneveld Van
der Hulst The form etje is found after stressed syllables ending
Intro duction
in a lax vowel and a sonorant consonant It is often assumed this is a
morphologically conditioned instance of schwa insertion
It is also commonly assumed that tje is the underlying representation of
the diminutive and that the other allomorphs are derived from it by phono
logical rule in particular by place assimilation of the initial consonantto
the previous consonant I will argue b elow that the underlying representa
tion of the rst segment of the diminutiveisslightly di erent from that of a
coronal stop but for the moment let us assume that the su x underlyingly
is tje indeed
Of central interest for this chapter is that the forms raampje and kon
ingkje are normally realized as rampi and kon ki in the Rotterdam
dialect so with a high frontvowel instead of j Rogier De Wilde
van Buul Oudenaarden whereas voetje and dingetje are always
realized as vutj and d tj In other words the form j shows up
after a coronal stop and the form i shows up in other contexts
In some cases the t is optionally deleted In t deletion environments
the word has two p ossible phonetic forms one with a t and the diminutive
realized as j and one without t and the diminutive realized as i the
former pronunciation sounds more a ected
kast tje k stj k si
The same alternation can b e found in other contexts We nd it for instance
at least diachronically in monomorphemic words like garage xarazi Fr
gara z horloge h rlozi Fr rlo z
A note of warning is in order here For the sake of simplicity I abstract
away from many details of the so ciolinguistic reality In particular I allow
myself the abstraction of assuming the existence of an indep endent Rotter
dam grammar In the real world probably all sp eakers of the Rotterdam
dialect know b oth variants kon kj and kon ki and they know the
former is more in accordance with the standard variantofDutch
In formal sp eech of Rotterdam sp eakers the variantwith j is found
and even preferred I consider this to b e a matter of di erentstyle regis
ters and I will supp ose a di erentstyle register means a partially di erent
grammar as I have also argued in c hapter The more formal style register
will b e discussed in section
Furthermore the i j alternation although it is seen by many native
sp eakers as the most prominentcharacteristic of their dialect Oudenaar
I do not discuss the deletion pro cess in this chapter See De Vries et al
Trommelen and Van Hout for grammatical and so ciolinguistic discussion
of this pro cess which is quite wide spread across Dutch dialects
The representations as given here do not corresp ond to the actual phonetic reality
They will b e improved b elow
Intro duction
den is certainly not unique to Rotterdam Dutch For instance I
have found some re exes of it in Tilburg Dutch monomorphemic words
as well menezie man ege m nezi Standard Dutch French m ne z
lozzie watch l zi Standard Dutch French h r l z medol lie medal
j
m d li Standard Dutch French med l e etc Historical alternations
similar to these actually seem to b e quite wide spread over the Dutchdi
alect area moreover there are instances of j changing to i as well as i
changing to j Weijnen The alternation is furthermore quite wide
spread in the diminutive forms in the Dutch dialects sp oken in Holland
It is also found in some contexts in Frisian as has b een noted by Cohen
et al my translation
One can also nd the i of middei afterno on moandei mon
day as j so m dj m ndj one could consider this
j to b e a form of refraction of i We nd other examples
of less accented i s that surface as j in kofje co ee plysje
p olice man naesje nation Because wehave made to o lit
tle investigations into this sub ject matter we cannot saymore
ab out it
According to Cohen et al in Frisian the alternation works in the opp osite
direction I just suggested for Rotterdam Dutch i changes to j in some
contexts The conditioning environment at rst sight is also di erentfor
Frisian and Rotterdam Dutch b ecause stress do es not seem relevantinthe
grammar of the latter See Bo oij for a discussion of Frisian breaking
within the framework of Non Linear Phonology Nevertheless wehaveto
takeinto account that underlying schwa is inherently unstressed so j is
always unstressed also in Rotterdam Dutch
Yet in one context the alternation seems to b e really typical of Rotter
dam sp eech The second p erson singular clitic je is normally realized as
i when it cliticiz es to another word except when this other word ends in
a coronal stop or a vowel
a h b j h bi haveyou
b m j m i mayyou
c hat j hatj hate you
d l st j l stj l si likeyou
With resp ect to the clitics one nds remnants of the pro cess in other Holland Dutch
dialects Amsterdam Leiden The Hague among others but often only non pro ductively
in frequent combinations suchashebbie haveyou assie if you etc
I will argue b elow that the stem nal obstruenteventually surfaces as not as t
but I abstract away from that here for simplicity
Intro duction
e z i j z ij z ii z iji said you
f j at j at i at you go
It do es not matter whether je is used as a p ossessive pronoun a re exive
pronoun or any other of its pronominal roles De Wilde Van Buul
The syntactic conditions under which cliticizati on can take place are quite
intricate but they are not at issue here It seems that the preceding word
has to c command the clitic in order for the pro cess to take place Henk
van Riemsdijk p c For example the form ie can b e used in phrases like
those in a and b but not the one in c I underline the relevant
clitic
a Do et een sjaal ommie nek
do a scarf around your neck
put a scarf around your neck
b Ommie oma komp mo ete we stil zijn
b ecause your grandmother comes must we silent be
Wemust b e silent b ecause your grandmother will come
c Azzie alles wat je heb je ie oma mo e
if you everything that you have grandmother must
your
geve b e je snel arm
give are you so on poor
If you havetogiveeverything you havetoyour grandmother
you will so on b e p o or
In a the clitic is either in the sp eci er of the DP NP complement je nek
of the prep osition om or it is the D dep ending on the syntactic analysis of
this construction which need not concern us here In either case the clitic
is separated from the prep osition by at least one syntactic category The
syntactic b oundary is even stronger in b In this case om probably is
in C and ie is emb edded in the complex sp eci er of the IP complement
of this C In b oth a and b the preceding word is a head and je
is in the sp eci er of the sp eci er of a complement of that head In c
the word preceding je is emb edded in the direct ob ject of a verb whereas
The term c command is used here as a descriptive term It remains to b e worked out
whether this alternation has to b e describ ed in terms of a so called direct syntax approach
Kaisse or rather in terms of a separate level of phonological representation Selkirk
Nesp or and Vogel Inkelas and Zec
Some of my informants rep orted a di erence in judgementbetween a and b
the latter is considered to b e somewhat worse than the former In general the e ect
seems however to b e so exceptionally light that I do not knowhow it can b e taken into
account
Intro duction
je is the sp eci er of the indirect ob ject of the same verb No phonological
cliticiz ation is p ossible
In any case I assume that the fact that the phonological alternation
involved o ccurs across syntactic b oundaries is an indication for its b eing
p ostlexical
The conditioning then is mainly phonological In e it is shown that
the clitic also do es not showupinits i variantafteravowel In this
environmentitisalways realized as j The same is true for phrase initial
p osition as is demonstrated in f
Wecannow ask the following questions with regard to the diminutive
su x
Why do es i alternate with j and vice versa Given the theories of
A vowels versus B vowels and of schwa presented earlier we should b e
able to describ e the b ehaviour of these segments in a satisfying way
It will also b e shown that alternation b etween i and j is a problem
if we assume that i is long
Why do es the rst segment of the diminutive su x assimilate to the
last segment of the stem That this typ e of assimilation is typical for
the diminutive su x only can b e seen if welookatthebehaviour of
asu xwithavery similar underlying form te t or the circum x
ge te t The t in these a xes do es not assimilate at all
ge b o om te leng te zwak te ge b o ef te
ge tree te long te weak te ge scoundrel te
timber length weakness rabble
Why do the coronal stops b ehave di erently with resp ect to the j i
alternation from all other segments b oth in the diminutive and in the
second p erson singular clitic
The study of the i j alternation brings together many of the topics
addressed in the previous chapters Furthermore it also intro duces a few
new topics some of which are directly relevant to the central concerns of
this dissertation The most imp ortantoftheseis gliding Whathappens
when vowels with all their vo calic features which make them so suitable
for syllable headedness app ear in a dep endent p osition such as the onset
or the co da
Finally the interesting interaction b etween consonantal and vo calic fea
tures that surface in this dialect will also b e the ob ject of inquiry
The second p erson clitic
In this chapter I argue that the answer to the last two questions in the
case of the diminutive is basically the same and is related to the under
sp eci cation of the rst segment of the diminutive su x This segment
obligatorily assimilates in its place features either to a segment on its left
or to a segment on its right
Finally Ialsowant to discuss the di erence b etween Standard Dutch
and Rotterdam Dutch From my analysis it will follow that b oth variants
have the same underlying forms for the second p erson singular clitic and
the diminutive su x Yet the two dialects b ehave di erently on the surface
so that wehave to account for this di erence in the phonological grammar
In section I rst lo ok at the facts of the second p erson clitic b efore
I analyze the diminutive in section The rst will b e analysed as i
and the second as bi underlyingly withtwovowels comp eting for a
place in the nucleus and the coronal vowel i interacting with coronal
consonants In section I brie y compare the Rotterdam facts to a
di erent glide vowel alternation Sievers s Law and I will show that this
comparison provides us with an extra argument against underlying vowel
length in Dutch Some more detailed problems will b e discussed in section
which concludes this chapter
The second p erson clitic
As far as I know the i j alternation in the Rotterdam Dutch second
p erson clitic henceforth S has previously b een analysed in a generative
frame only byVan den Berg and by Berendsen Both authors
take j to b e the underlying form of this clitic and derive i from it
by rule I will return to these prop osals b elow In this chapter I follow
these scholars in their basic prop osal and assume that b oth forms have the
underlying form i They di er only in their resp ective syllabi cation
a j N b i N
N N
i i
In a the glide j is represented as the high vowel i o ccurring in a
consonantal p osition This do es not seem a very controversial assumption
I will turn to the exact feature geometry of i b elow The representation
of Dutchschwaasanemptyvo calic ro ot given here has b een argued for in
chapters and The question is whether this schwa has to b e identi ed
as an underlying schwa in whichcaseitwould b e unparsed in the form i
The second p erson clitic
or as an ep enthetic schwa In the latter case b would consist of a high
tense vowel only I will argue b elow section that the former option
should b e chosen b is a high vowel followed by an empty ro ot Until
that p oint I will simply assume this representation to b e correct
I claim that i actually is the underlying representation of S in the
Rotterdam dialect Its precise phonetic representation is dep endent on its
syllable structure
Wherever this is p ossible the rst segmentof i is analysed as a nu
cleus i is a full vowel and all full vowels are preferably syllabi ed into
anucleus Wethus get the following structure for hoef je must you and
waar je where you
huvi N N wari N N
N N N N
h u v i w a r i
Schwa is also a vowel of course and therefore also wants to b e in a nu
cleus for instance b ecause of Project cons N On the other hand in
chapters through I have argued extensively that schwa is not as unprob
lematically a syllable nucleus as full vowels In particular do es not have
avo calic no de Therefore it do es not seem improbable that when i and
comp ete for a p osition in the nucleus the former wins in the unmarked
case This will b e technically worked out b elow
Whenever an onsetless syllable would o ccur by the realisation as i
however at the b eginning of a phrase or after a vowel the form j is
chosen instead Compare the forms in to the ones in b elow
huvi N N wari N N
N N N N
h u v i w a r i
The grammar cho oses whatever syllabi cation is optimal in a certain con
text Normally this optimal segment is i b ecause this syllabi cation gives
us a syllable with a maximally sonorous nucleus we will see b elowthat
this can b e formalized in terms of Project N V but after a vowel or
word initially it is j in order to avoid onsetless syllables that would yield
violations of the Onset constraintIhave used already at several p oints in
this thesis
We seem to have a fairly straightforward case of constraint interacti on at
hand First and foremost Onset decides but if wehave several candidates
The second p erson clitic
which tie on Onset wecho ose the candidate with the optimal nucleus the
one with a full vowel rather than schwa In the remainder of this section I
will discuss the di erent phonological contexts one by one and explain the
b ehaviour of i in those contexts
Hiatus We exp ect underlying i to surface as j whenever it
o ccurs phrase or word initially or after a vowel I will refer to these three
contexts as a hiatus context Let us rst examine lexical forms to see
whether this prediction is b orne out in that stratum
The word initial p osition is usually considered to b e an exception to
the Onset constraint Languages whichdonotallow for word internal
hiatus still allow for onsetless syllables in word initial p osition If Dutch
were to allow the same exception wewould actually predict i rather than
j Onset would b e irrelevant and therefore the constraint preferring full
vowels would decide
As wehave seen in the discussion of schwainchapter the Onset
requirement plays an imp ortant role in lexical words also in the absolutely
initial p osition I argued in chapter that words like d m are excluded
from the Dutch lexicon b ecause the rst syllable needs an onset this onset
can only b e ep enthetic b and b sequences are not allowed The work
ing of Onset in word initial p osition plays a crucial role in this line of
reasoning
But even if it were p ossible to formalize an exception mechanism to
Onset in the appropriate way the problem remains that we ndboth
words starting with j and words starting with i The former class is
much smaller but it contains at least one everyday Dutchword p erfectly
acceptable to all nativespeakers Apart from some forms morphologically
related to and probably derived from the second p erson singular clitic
jezelf yourself jewelste intensi er litt your b est we only nd the list
in a as opp osed to the much longer incomplete list of words starting
in i in b
I only considered words without stress on the rst syllable Otherwise the list in
b could havebeenmademuch longer but stressed schwa syllables do not exist for
indep endent reasons amply discussed in chapters through ab ove Furthermore for
the name Jeruzalem an alternative pronunciation jeryzal m is also p ossible for some
sp eakers so that we might conclude that in this case wehave r schwa rather than u schwa
The second p erson clitic
a jenever gin b idee idea
Jeroen b oy s name Irene girl s name
Jeruzalem id Iran id
icoon icon
idyl le idyl
il lusie illusion
ivoor ivory
etc
We do not nd any kind of alternation among these forms Neither the
pronunciation inever of jenever nor the pronunciation j de of idee are
ever found in Rotterdam Dutchorany other dialect that I am aware of
This is an imp ortant argument for an underlying distinction b etween i
on the one hand and i on the other I assume the forms in the second
column in start simply with i without a following schwa The forms
in the rst column start in contrast with i
Wehave already seen in chapter that the following two constraints
hold for Dutch
a Onset Project F N A syllable should have an onset if
the head of an N no de dominates any feature that no de should
branch
b Project N V if N branches N dominates a vo calic no de
Onset wants the sequence to b e parsed as j b ecause that sequence has
an onset while the output i do es not On the other hand the parse
j violates the constraint Project N V b ecause it contains a branching
N syllable no de with a schwa head The parse i do es not violate Project
V b ecause it has no branching syllabic no de at all and furthermore its head
dominates a vo calic no de with features high and coronal
The only case I know which could p otentially b e attributed to this phenomenon
is the variation b etween Izabel Old Testament and Jezabel New Testament in the
so called Statenvertaling the th century State translation of the Bible whichforms
the foundation for Mo dern Standard Dutch Yet the latter word is normally pronounced
as j ezab l or j ezab l so with stress on the rst syllable and the vowel realized as e
rather than Yet this may b e a re ex of the fact that Tib erian Hebrew had more or
less the same alternation as wehaveinDutch Bat El
The reason why the latter forms are so uncommon probably has to do with an
observation made in chapter fo otnote p if schwa is in the rst syllable it
is always preceded by one of the consonants in f t b v d g These segments do not
form a natural class under any feature analysis of Dutch The reason for their patterning
together here seems to b e a historic one b ecause there are or have b een pre xes of the
form t v r and There has never b een a Dutchpre x j
The second p erson clitic
Wehave also seen in section p that the Onset constraint
N V This assumption explained whyschwa licensed outranks Project
onsets but not complex onsets Here we can use it to derive the di erence
between i and j
Candidates Onset Project N V
p
j
i
This constraint ranking causes the j parse to b e chosen b oth phrase
initially and in intervo calic context After a consonant Onset b ecomes
irrelevant b ecause the consonant can ll the required onset p osition of i
I conclude that the sequence i willword initially surface as j Log
ically sp eaking this do es not tell us anything ab out S at all It is crucial for
the analysis that clitics qua phrasal a xes do not turn up in the phonology
until the p ostlexical level This explains for instance why a lot of clitics can
start with schwa even if they are phrase initial The moment they enter
the grammar the Onset and Contour constraints have lost much of their
force The following constraint ranking holds true
Parse V Onset Contour
But if this is the case it mightwell b e that p ostlexicall y the ordering of
the two constraints just mentioned Onset and Project N V is re
versed Aquick lo ok at the data reveals however that this is not the
case Whenever S o ccurs phrase initially or after a vowel it surfaces
as j This means that Onset in Dutch p ost lexical phonology still has
to dominate Project N V even if the former constraint is no longer in
violable as is the case in the lexicon I assume therefore that the ranking
Onset Project N V holds b oth lexically and p ost lexically in Rotter
dam Dutch
Hiatus after high vowels One typ e of hiatus p osition is of
sp ecial interest This is the case where i follows a high vowel in the
stem In that case the question is whether it is the stem nal high vowel
Tobevery precise the exact lo cation I mean is the rst syllable slot in an Intonation
Phrase or Utterance in the sense of Nesp or and Vogel and Selkirk in all
other p ositions S b ehaves as a phonological enclitic
Wehave seen indep endent evidence that this ranking holds in chapter where it
help ed us explain why sla p sla p with an underlying form slap it is b etter to
have an empty headed syllable than to have a syllable without an onset
The second p erson clitic
or the su xal i that undergo es gliding Wehave to distinguish b etween
three typ es of con guration a i j b V j where V is y or
u and c V V j so after a diphthong of which the second half
i j
is always high in Dutch All these environments are of interest b ecause
high vowels b efore a hiatus normally develop a glide to satisfy the Onset
requirement in all of these cases
The facts of i i are simplest In these cases none of my informants
OK
could pro duce the i alternantof S zie je see S zi i zi ji zi j
The reason for this probably is a lter excluding ji sequences combined
with the relative high ranking of Onset Indep endent evidence for such
a lter comes from the fact that there are also no Dutchwords with ji
sequences derived from underlying ji and that glide insertion is also
j OK j
blo cked b etween two underlying i s Shiiet Shiite s i jit s i it
In fact wehave to rule out two logically p ossible con gurations to ban
ji the i abbreviates a feature tree in the following representations
a N b N
N N
i i i
These are the two p ossible structures for ji sequences There can either
be two indep endent segments or there can b e one and the same segment
linked b oth to onset and nucleus Since ji sequences do not o ccur on the
surface in Dutch at all wehave to rule out b oth structures
I assume the relevant constraintisContour
Contour If the head of a syl lable is its onset may not be
We used this constraintinchapter to rule out C C and h sequences
i i
and onsetless schwa headed syllables Here wetakeaninterpretation where
linked to i if the head of a syllable is linked to i its onset might
not b e linkedto i This rules out b ot a and b as required In
order to get the result that zi ji zi j wehave to assume the following
ranking
Contour Project N V
Although there are words with j as wehave seen in chapter
Iwant to thank MiekeTrommelen for this observation
The second p erson clitic
We could sp eculate on whether should not b e formulatedsoastoalso
exclude some other non o ccurring sequences such as syllable initi al dl
tl bw or pw
ant manner klant client tlant a tr
b praat talk kwaad angry pwaat
In these cases as well mere reference to the OCP is not su cient In
addition we need a constraint ruling out doubly linked features in the onset
What these examples show is that something like the constraint Contour
is also crucially needed to rule out sequences of homorganic onsets The
prop er formulation could b e something like if an onset segment is the
fol lowing segment may not be and that at this level the following ranking
holds
Contour Parse Consonant
It is not clear why onset segments have this prop erty that they cannot b e
followed by homorganic segments but as a working hyp othesis Contour
will turn out to do its work reasonably well
Inow turn to hiatus after the other high vowels y and u The facts
are not very clear and judgments seem to di er among sp eakers Most of my
informants allow insertion of a glide w Wim Zonneveld p c has drawn
myattention to the fact however that other sp eakers prefer the j form
in these cases Ihave not b een able to nd evidence that this di erence
corresp onds to any geographic or so ciological b order line b etween the two
classes of sp eakers
I crucially assume here that the value coronal is present on coronal consonants and
that it is not a default value undersp eci ed at any relevantlevel of representation I
also assume r is not sp eci ed as coronal although I will not take up the issue of the
exact representation of this segment
There is one technical problem with this sp eci c formulation however I represented
catalexis in chapter with an empty consonant h or b in a degenerate syllable If
this is correct we can have degenerate syllables without a head bur not schwa headed
syllables without an onset This would argue in favour of a constraint if the nucleus of
a syl lable is the onset may not be The facts in would havetobeaccounted
for in a di erentway
Wehave already seen indep endent evidence for this ranking b e it indirectly
N V in example ab ove For wehave seen that Contour dominates Project
N V dominates Parse Consonant in the lexicon b e and weknowthatProject
cause this help ed us explain why sup erheavy syllables are not allowed word internally
ma t ka ma t ka underlying representation matka a form with a word
internal sup erheavy syllable is worse than a candidate in which the consonant is left
unparsed
My written sources do not discuss this context They do not give relevant examples
of high vowel i sequences either
The second p erson clitic
doeje do S nu je now S
dialect duj nyj
dialect duwi nywi
The facts of dialect are what wehave come to exp ect wi is a reasonably
normal syllable in Dutch wie who wiel wheel etc and there is no
reason why it should not o ccur The facts of Zonneveld s dialect are
therefore somewhat problematic for myaccount It mightbethatinthis
dialect homorganic glide insertion is blo cked until a later phonetic level
so that we only havea choice b etween nyi and nyj in whichcasewe
pick of course the latter A somewhat di erent suggestion will however b e
made b elow section
A similar distinction arises in the case of diphthongs After i we
always nd j After u the judgments are variable According to
De Wilde Van Buul the choicebetween je and ie hereseems to
bedependent on the level of civilisation of the speaker Lower so cial class
p eople say houwie p eople with a higher class background say houje De
Wilde Van Buul suggests that the crucial factor is that in lower class sp eech
the u is pronounced more like a consonant
This suggestion might indeed lead us to the right solution In section
I will show that there may b e evidence for some di erences b etween
the feature structure of the glide w and of the high vowels u and y
The relation b etween w and these vowelsisdi erent from the relation
between j and i which seem to b e exactly the same in all resp ects except
for syllable p osition It might therefore b e that glide formation of w is
in some way more costly than the glide formation of j I will makethis
idea of costliness more precise in section b elow This costliness would
blo ck glide formation in the higher so cial class variant of Rotterdam Dutch
For now I will assume that there is a constraint forbidding glide formation
of w in the relevantcontexts which is ranked higher than the constraint
demanding an optimal syllable nucleus since h uwi is less well formed
than h uj
In the lower so cial class the representation of diphthongs may b e dif
ferent from that of Standard Dutch altogether ei and au sound more like
It is very di cult to construct reasonably natural contexts where i follows y
b ecause the latter diphthong is not very frequent I found only twoverbal stems ending in
it krui wheel and verbrui incur someb o dy s displeasure These forms are disregarded
in my written sources My informants assert in these cases they would prefer j even
if they di er on the realisation of S after au I will ignore the dipthong y in the
following discussion b ecause it is hard to nd go o d testing evidence
Furthermore there seems to b e a di erence b etween lexical items and function words
or otherwise a frequency e ect zouwie should you nouwie nowyou or houwie hold
you seem much more acceptable than brouwie brew you Again I have nothing to
say ab out these e ects
The second p erson clitic
a j and aw resp ectivel y Therefore the nal segmentofau already has
to be a glide and there is no extra cost to putting it into an onset In other
words the representation of hou in this dialect is as follows
N N
N N
h a w
The formulation of the constraint against glide formation in section
will b e such that it do es not a ect this representation
After coronal stops The choice for j in hiatus is intuitively
quite plausible The quest for onsets has many diverse e ects acrosss lan
guages as wehave seen This variation might b e one of them
More problematic is the question whywe do not nd h di for had je
had S or d ti for dat je that S In lexical words the syllables di
and ti have no exceptional status at all
dynamisch dynamic dinamis direct id dir kt
divers various div rs
tiran tyrant tir n titaan Titan titan tyfoon typho on tifon
Even if wewould p osit a separate stratum of clitics di and ti cannot b e
disallowed b ecause they are the regular shap es of the third p erson singular
masculine clitic cf section
I ignore Dutch y for reasons mentioned in fo otnote
I do not fully discuss the facts of coronal non stops in this chapter but the data
seem to b e very interesting The fricatives allowtwo forms kas tje little case can b e
pronounced k se with a palatalized sound or k si For l and n the facts are even
more complex We nd a wide range of idiolectal and lexical variation According to
my informants the lexical verb kennie can S sounds b etter than the auxiliary bennie
are you for instance but b oth forms have a preferred form without the nasal cosonant
with a nasalized vowel and j for the clitic k j b j The same is true for l
verveelie v rveli sounds somewhat b etter than wil lie w li but the informants would
prefer to say v rvej and w j
De Wilde Van Buul notes there also is some dialectal variation with resp ect
to these facts For instance in Vo orschoten a town approximately kilometers to the
north of Rotterdam of which the dialect has b een studied byOverdiep wedo nd
j after n while the preceding v owel still nasalizes and frequently also after the
other coronal consonants and after r There is no reason to assume r is coronal in
Rotterdam Its pronunciation is uvular
I cannot o er an explanation for the facts in these other dialects if only b ecause more
precise details of their system are unfortunatelylacking
The second p erson clitic
On the other hand the sequence tj is not found in any lexical form
outside the class of diminutives As a matter of fact the sequence Cj
is nearly non existent with some regulated classes of exceptions to b e dis
cussed in section
I will rule out these con gurations by the following ad hoc constraintat
this p oint but I will give a more principled explanation based on Project N
V below p
C j
N N
Given all these considerations it seems hard to explain then why tj is
preferred over ti in the case of clitics
Now as a matter of fact the representation tj in the case of stem S
or diminutive forms do es not provide an adequate re ection of the phonetic
facts It is more correct to saythatthetwo coronal consonants merge
j
into one palatalized consonant t Van den Berg Tobeeven more
precise my impression is that with manyspeakers the primary articulation
j
sounds almost like a palatal This segment has to b e banned from
lexical words in which it do es not o ccur
A similar e ect has b een addressed in Hermans b discussing
palatalization in Limburg Dutch which is triggered by a totally di er
entenvironment but this need not concern us here Hermans adopts the
mo del of Clements in whichvo calic and consonantal place features
app ear on di erent tiers Hermans assumes the representation of a palatal
ized consonantisschematically as follows
cons
C place
coronal V place
i
coronal
i
The two place features are crucially coindexed Hermans b observes
that the place of constriction for coronal vowels is ususally more back than
that of coronal alveolar consonants If two features are coindexed they
have to get the same interpretation phonetically Therefore the coronal
feature under C place is interpreted as palatal rather than alveolar in this
case
Later work Hume Clements and Hume has worked out a
slightly di erent view on the interaction b etween consonantal and vo calic
features These authors argue that consonantal and vo calic place features
The second p erson clitic
app ear on the same tier Such a theory would give us the representation of
palatalized coronals in b In my opinion this view allows us to express
Hermans observation even more elegantly without the use of coindexa
tion on features In this case it is literally the same feature that acts as
a consonantal and a vo calic place feature It can only get one phonetic
interpretation Since alveolar vowels do not exist or at least since Dutch
do es not allow them this interpretation can only b e a palatal one
Van den Berg working in an SPE like framework claimed that
j written by him as ju b ecause he assumed schwawas the unstressed
variantof u was the underlying form and i was the result of
a reduction rule con ating the feature matrices of two segments into one
new matrix
This reduction was blo cked after a coronal stop b ecause there was a
palatalisation rule extrinsically ordered b efore j reduction whichcon
j
ated the t and j into a palatalised t Because j did not survive
the palatalisation pro cess as an indep endent element palatalisation bled
j reduction Berendsen o ers basically the same analysis but
stated in somewhat more mo dern terms I prop ose to also adopt this anal
ysis once again mo dernizing the technical apparatus and using insights of
mo dern syllable structure and autosegmental theory
I therefore assume that the two segments t and i are merged In the
following representations I mark features that are directly dep endenton
the C place no de with a subscript C and those that are directly dep endent
on the V place no de with a subscript V
a Input
cons son cons
C place C place
coronal vo calic
C
V place
coronal
V
This is necessary b ecause the coronal feature in the output has to countina
sense as a valid parse of b oth the consonantalandthevo calic coronal feature Wecan
always reconstruct the indices from the representation however a place feature that is
directly dominated by a C place no de has an index C and a place feature that is directly
dominated by a V place no de an index V
Presumably wealsohavetomention in the sp eci cation of the input the fact that the
coronal vowel should b e high b ecause similar palatalisation is not found with e
this condition is discussed in section On the conditions on the consonantsee
the discussion on p b elow
The second p erson clitic
b Output
cons son cons
C place
vocalic
V place
C
coronal
V
A few prop erties of this pro cess have to b e discussed here In the rst place
why do es it only a ect coronal stops and not for instance p and k The
most straightforward thing to do seems to b e to relate this to the fact that
the j is also coronal We therefore havetwo coronal segments adjacent
to one another This is an OCP violation whichmight b e seen to trigger
the pro cess at hand In other words wehave evidence for the following
ranking of constraints
OCP Empty root Parse Vocalic root
In the case of p and k there is no external reason for creating the
complex segment in b In the framework adopted here if there is no
reason for doing something we are not allowed to do it b ecause doing
something will always cause violation of the constraints that demand input
and output to b e maximally similar
In the second place in what waycanwe analyse the output of this
pro cess as b etter as more optimal than its input A rst reason seems to
b e the OCP Yet there are other ways to avoid OCP violations that do not
leave the vo calic ro ot unparsed One instance of an alternative structure is
given b elow
cons son cons
C place
C place
vocalic
V place
coronal
See also section on why only coronals can b e palatalized in underived forms
The second p erson clitic
Variations on this structure are also logically p ossible For instance an
other geometric p ossibility is that the V place no de links to the consonant
j
creating a doubly linked structure t i
This form leaves the original vowel slot parsed while it solves the prob
lem with the OCP In this resp ect then it seems to do even b etter than
b There is however one resp ect in which it do es considerably worse
syllable structure This form has the structure of a consonantandahigh
vowel glide which share material
Given our ndings presented ab ove there is no legitimate way to parse
these ob jects into syllable structure We cannot parse the two segments
as a complex onset b ecause as wehave seen j cannot o ccur in a complex
j
onset We also cannot parse them as a onset rhyme sequence t i either
b ecause as wehave seen onset segments may not share place features with
afollowing no de this was the constraint I called Contour
So the only way to satisfy b oth the OCP and conditions on syllable
structure is by creating b This means that at the level we are discussing
now which I assume to b e p ostlexical since we are dealing with clitics
these constraints outrank whatever constraints are violated by the phonetic
j
output t I will consider one such constraint SecArt from Rosenthall
In I give an informal de nition of this constraint
SecArt No segmentmay b e sp eci ed b oth for C place and V
place features
Postlexical grammar
Contour C j
N N
SecArt
As a matter of fact SecArt in my view should b e derived from two
constraints which can b e written semi formally as follows see app endix A
for some more discussion
C F A cons ro ot do es not dominate a vo calic feature
V
V F A cons ro ot do es not dominate a consonantal feature
C
Since every ro ot is sp eci ed as either cons or as cons these twocon
straints taken together have the same e ect as SecArt Furthermore
the separation into two constraints can help us derive the e ect that sec
ondary articulation is much more constrained on vo calic ro ots than it is on
consonantal ro ots We might assume that V F is ranked very high
C
j
I found one verb starting with t i tjielpen cheep I consider this form to b e an
j
onomatop o eia Furthermore this verb is often rendered as tjilpen t l
For secondary articulation on vowels in Swedish cf Clements
The second p erson clitic
The constraintC F plays the role of SecArt in the case of palatali
V
sation We will also see that there is evidence for some constraints which
are remotely related to these ones
We can now draw the following tableau for haat je
Candidates Contour C F C j
V
N N
j
ha t j
p
j
ha t
ha tj
ha ti
j
ha t i
It might b e an imp ortant assumption that this ordering de nes the p ost
j
lexical grammar As noted ab ove lexical forms with t are not found
at all There seems to b e hardly any evidence for underlying complex seg
ments We could therefore assume that the structure preserving constraint
C F is ranked topmost in the lexical grammar An imaginary input
V
ti r n will then surface as tir n
Candidates C F Contour NC Nj
V
j
t j r n
j
t r n
tj r n
p
ti r n
j
t i
This would explain the absolute non existence of forms like tjeran I
return to the lexical grammar in section b elow where we will see that
the situation is slightly more complex there
For now I prop ose to concentrate on the p ost lexical grammar I have
to discuss one other prop erty of the pro cess sketched in the fact
that it applies only to coronal stops There is no palatalisation of any
other coronal consonantinDowntown Rotterdam We nd ken ie ken je
wil lie wil je etc
Wehave to lo ok somewhat more closely at the representation of the
coronal segmentinventory to nd the reason why t and d are sp ecial
among coronals Ihave listed the relevant parts of the structure for all
Rotterdam coronals b elow
I once again p oint out that I do not consider r a coronal in Rotterdam Dutch
although this assumption do es not really a ect the argument
The second p erson clitic
s t
cons son cons son
C place
C place continuant
coronal
coronal
z n
cons son cons son
C place Laryngeal continuant C place nasal
coronal voice coronal
d l
cons son cons son
C place
C place Laryngeal
coronal
coronal voice
lateral
t is the simplest coronal consonant since it do es not b ear any feature apart
from place assuming cont do es not exist or is the unmarked feature
value cf Lombardi The intuition therefore is that t is the only
consonant that can supp ort some extra complexity in the form of secondary
articulation
One way of formalizing this idea is as follows All coronal consonants
except t and d have some consonantal feature a feature that normally
only go es with consonants
It is not unnatural to assume there are constraints banning combinations
of vowels with lateral or less strongly ofvowels with nasal Although
there clearly are languages with nasal vowels it is well known that nasality
is less marked for consonants than for vowels For instance languages with
nasal vowels always have at least one nasal consonant while the opp osite
is not necessarily true More in particular Rotterdam Dutch only has
nasal consonants no nasal vowels Furthermore many languages have nasal
vowels only in the context adjacent to a nasal consonant which seems to
indicate in those cases the nasality feature is in some way licensed by the
consonant cf Hu man and Krakow for an overview of the literature
on nasality
The second p erson clitic
We therefore need to sp ecify the constraints in a through c
V place
a
lateral
V place
b
nasal
V place
c
cont
V place
d
voice
d is the only consonant apart from t that allows for a secondary articu
lation in Rotterdam The constraint d whichwould b e needed to rule
j
out d is not very deeply grounded in the phonetics Although vowels
are not usually contrastively sp eci ed for voice d do es not seem a
constraintwewould allow under the Grounded Phonology Hyp othesis cf
section
j
On the other hand de consonantin had je had S ha seems to
be voiceless It cannot b e that Final Devoicing has applied rst b ecause
that is not the case in for instance hebbie have S or any other form
Voiced palatalized consonants therefore really seem to b e excluded
If intro duction of d were the only way of establishing this wewould
have to abandon the Grounded Phonology Hyp othesis b ecause a more
reasonable grounded feature co o ccurrence constraint seems to b e one of
the following form
V place voice
The problem raised here has a much wider scop e than palatalisation It
is w ell known that sonorants in general have to b e undersp eci e d for the
feature voice For instance the phonetic voicing of sonorants is invisible
for the working of Lyman s Law in Japanese It o and Mester
It o Mester and Padgett have suggested that these and similar
facts should not b e analysed in terms of non grounded constraints suchas
In the formulation of these constraints I havechosen to use the V place no de as the
no de that may not co o ccur with the consonantal features Mayb e I could just as well have
chosen the vo calic no de or each of the vo calic place features coronal labial velar
v v v
etc The reason that I have not chosen the latter option is that it would have tripled
the numb er of constraints The reason that I have not chosen the vo calic no de is that
height features seem to b e even stricter in their preference for vowels than vo calic place
features Di erences b etween high and low secondary articulations are rarely found if
at all cf Van de Weijer
The second p erson clitic
d but rather in terms of complemented by a theory of licensing
the core of which is given b elow
Licensing Cancellation If F G then F G
If the sp eci cation F implies the sp eci cation G then it is not the
case that F licenses G
License voice voice must b e licensed
If the Licensing condition dominates the condition that V place voice we
get the required result that segments with a secondary articulation cannot
b ear the feature voice
dj
Candidates License V place voice
p
j
t
j
d
The same is true for c a constraint against the combination of con
tinuant and vo calic quality do es not seem to b e grounded in the phonetic
reality The continuant contrast as far as I know do es not play a role in
the vowel phonology of any language Again this might b e attributed to
the interaction of an implicational constraint and a requirement on licensing
of the feature continuant
V place continuant
License continuant continuant must b e licensed
If we assume that the constraints against the combination of vo calic place
features with any consonantal non place feature plus the licensing con
straints on voice and continuant are undominated we predict only t is
susceptible to palatalisation t do es not b ear any feature except for coro
nal In order to make the right prediction ab out d wehave to assume that
Cf Steriade for a similar prop osal in whichhowever the term licensing in
a di erent almost the opp osite way in her theory obstruents cannot license voice
whereas sonorants can
As I have noted ab ove there are dialects which allow palatalization of other coronal
consonants than t My impression is that a dialect has most often palatalization
of s and z then of n and nally of l If this is correct it suggests that the
constraints in have some inherentweight ranking the higher on the list the heavier
the constraint and License continuant is relatively low The interesting thing is that
vowels are continuants and that they can b e nasal in some languages but not lateral
This re ects exactly the same hierarchy
The second p erson clitic
d has a slightly di erent status than the other feature co o ccurrence re
strictions in While the other features simply blo ck palatalisation in
the case of d it is voice itself that loses the battle While in the other
cases the constraints forcing parsing lateral nasal and continuant are
ranked higher than the feature co o ccurrence restrictions Parse voice is
ranked lower This low ranking of Parse voice seems intuitively justi ed
b ecause Dutch has several devoicing pro cesses such as Final Devoicing and
ProgressiveVoice Assimilation Lombardi Bo oij a while it do es
not have delateralisation denasalization or decontinuantisation
Third p erson singular clitic Inow turn to the third p erson
singular masculine nominative clitic S whichisvery similar to S pho
netically This clitic is realized as i ti or di dep ending on the phono
logical context but never as j tj or dj
realisations verb stem gloss
si sti zdi sj s z is he
l si l sti l st likes he l stj
m xi m xti m di m mayhe m dj
k ni k nti k ndi k nt can he k ndj
I follow Berendsen among others in assuming that di is the un
derlying form of this clitic Rules of voice assimilation degemination and
t deletion apply to it partly optionally yielding the variation in I
will not discuss these pro cesses here
Nowif die has the underlying form di it will always b e syllabi ed as
in
The phonetic similaritybetween S and the i variantof Sprovided the explanation
for the sp ecial b ehaviour of S after coronal stops according to Overdiep and De
Wilde Van Buur If S would b e i after coronal stops one could no longer hear
the di erence b etween S and S uit ie whistle S uit tie whistle S yti
This functional account did not explain however why the i form of S is not found in
environments where no confusion with S is p ossible either likein Roept Piet j i
moeder niet do esn t Pete call your mother where the underlined element can only b e
a p ossessive pronoun b ecause S can only b e the nominative and the sub ject p osition
is already lled by Piet Neither do es it explain why the same phenomenon o ccurs
in the diminutivewherewe equally nd a distinction b etween coronal and non coronal
consonants but where most of the time no confusion is p ossible
Bo oij a argues that what is optional here is phonological encliticisation Incor
p oration or Chomsky adjunction of the clitic into the proso dic word would b e optional
in the case of die and dat that All phonological rules mentioned would apply obliga
torily within the domain of the proso dic word however
The diminutivesu x
N
N
d i
If the underlying representation of S is di palatalisation will not take
place b ecause it would pro duce a complex segment in the co da where such
segments are not allowed It is therefore absolutely essential for this analysis
that the structure of S is di not di
The diminutivesu x
Ifollow the tradition established by the authors mentioned in the rst
section of this chapter and supp ose the underlying form of this su x is tje
i e ti
If the structure of S were di this su x would di er from the diminu
tiveinavery minimal way
diminutive
cons son cons cons
C place C place
coronal vocalic
V place Ap erture
coronal high
S
cons son cons cons
C place
C place Laryngeal
vo calic
coronal voice
V place Ap erture
coronal high
The two elements would only di er in the presence versus absence of the
feature voice Yet S always app ears as di whereas the diminutiveshows
the same variation as S The Rotterdam forms are raampie rampi but
The diminutivesu x
j
katje k k ti The di erence in voicing can not explain the di erent
b ehaviour b ecause wehave just seen voice is the only consonantal feature
apart from coronal that do es not blo ck palatalisation
Level ordering cannot provide us with an explanation either The i
j alternations discussed until now are p ost lexical Wewould then exp ect
S to pattern like S to the exclusion of the diminutive but the facts are
exactly the opp osite of this
Actually it is somewhat surprising that the diminutivebehaves likea
clitic While it is not unreasonable to assume that syntactic clitics and
in ectional su xes app ear only at this level the same is not immediately
j
obvious for the diminutive su x The fact that we nd not present
in derived lexical forms in diminutives is an indication that these su xes
have some semi in ectional status
However this may b e it seems that the only way to represent the phono
logical di erence b etween the diminutive and S is by incorp orating it into
the lexical representation one ends in a schwa the other do es not Assum
ing the diminutiveenters the grammar at the same time as the clitics viz
p ostlexically its b ehaviour is as exp ected
Ineedtosaymorehowever ab out the t in this diminutive b ecause
its b ehaviour is somewhat di erent from that of other t s in other envi
ronments This consonant assimilates in all its features to the preceding
consonant This is not the usual pattern for su xes starting with t I
givesomemorerelevant facts b elow
diminutive te S Past tense
raam pi ruim t kom ti verzuim d
little window space comes he neglected
koning ki leng t zing ti verleng d
little king length sings he lengthened
j
ko e auw t zei di raad d
little cow fainting t says he guessed
j
kat breed t haat ti haat t
kitten breadth hates he hated
The t of the diminutive su x is the only segment that shows this p eculiar
b ehaviour Normal derivational su xes like t in the second column of
do not assimilate at all Real p ost lexical elements like S and the
past tense su x show di erenttyp es of voicing assimilation but no place
assimilation The diminutive has place assimilation but no overt voicing
assimilation
See for instance Sp encer for some discussion of the hybrid status of diminu
tives in the distinction b etween derivation and in ection in many languages
The diminutivesu x
Wehave to formally express the fact that the obstruent in the diminutive
su x is extremely sensitive to place assimilation One waytoachievethis
is byhaving the segment totally unsp eci ed for place features Van der
Hulst The rst segment of the diminutivethus is not a coronal stop
but simply a voiceless stop without a place no de Since it is voiceless it
presumably also do es not have a Laryngeal no de It therefore is an empty
consonantal ro ot I have argued that the empty consonantal ro ot also is
the representation of a glottal stop Yet in the case of the diminutive
the glottal stop app ears in a somewhat p eculiar phonological context viz
b efore a p otential onset and in some cases also after a consonant The
glottal stop in Dutch only surfaces as a last resort to satisfy the Onset
requirement
I assume therefore that the diminutive su x is sp ecial in the sense that
it is the only morpheme in Dutch where a glottal stop is underlying This
explains the exceptional b ehaviour of the rst segment of this su x in the
Dutch lexicon
Supp ose furthermore that there is a lter which states that all con
sonants have to b e sp eci ed for a place feature Since glottal stops are
very limited in their distribution this do es not seem a very controversial
assumption I will return to glottal stops in hiatus contexts b elow
In order to satisfy the lter against empty consonantal ro ots a weakness
constraint stating that segments in the weak p osition of a syllable need to
have some place sp eci cation the obstruent in the diminutive assimilates
to the preceding consonant if there is one cf
Weak N C A segment in the weak p osition of a syllable needs
to have a consonantal place feature C C fcoronal dorsal
c c
labial g
c
cons cons
C place
This accounts for the raampie and koninkie cases which end with the fol
lowing representation the o s indicate ro ot no des
o o o o o o o o o o o o o o
r a m p i k o n k i
labial dorsal
To b e maximally accurate this constraint should b e replaced bya setofweakness
N C Weak N C g constraints fWeak
The diminutivesu x
N V will Since the obstruent can form an onset the constraint Project
select the optimal candidate
Candidates Onset Project N V
ram pi
ram p j
C j Notice that Project N V can replace the ad hoc constraint
N N
against Cj clusters in all the relevant cases It correctly do es not blo ck
Cj clusters in words with a full vowel suchasfjord
If no consonant is adjacent to the unsp eci ed segmentlikeinkoe tj
little cow or ring tj the segment has to get its place feature in a
di erentway
I prop ose that in this case the place feature is assimilated from the
i instead This typ e of vowel to consonant spreading is blo cked bya
constraintwe already encountered
C F A cons ro ot dominates a consonantal place feature
C
If Weak N C C F the b est thing we can do in koe tje is assimilate
C
the place feature of the vowel
Weak N C C F
C
Candidates Weak N C Monolog C F
C
ku bi
p
j
ku t
j
ku p
Ihave assumed the emptyrootreceived its place feature from the following
i not from the preceding u The reason for this is probably related
to the Crispness of Edges constraint prop osed by It o and Mester
The exact ordering of this constraint with resp ect to Weak N C cannot
b e established I call this constraint Monolog
Monolog The edges of a morphological domain should b e crisp
no feature should b e linked b oth to an edge segment of that domain
and to an element outside of the domain
Weak N C Monolog C F
V
Conceptually this constraintisvery closely linked to the Alignment constraints but
cf It o and Mester for arguments that it is not
The diminutivesu x
The edges of are not crisp b ecause the feature labial crosses one of
them the edges of on the next page are
o o o o o
k u p i
labial
The obstruentthus gets the place feature of the following vowel This
feature happ ens to b e coronal This is the reason why the rst segment
of the diminutive shows up as a coronal palatalized obstruent
cons son cons
C place
vo calic
V place
coronal
Now the result of this assimilation pro cess is of course the same as the result
the palatalisation pro cess discussed earlier koe tje will get the following
structure assigned to it
o o o o
j
k u
Inow turn to a somewhat more complicated case kat tje inwhich the
diminutive is preceded by a coronal consonant Here we exp ect the placeless
Again the variantwherewe create a palatalized consonant and leavethevocalic
ro ot unparsed is preferred over the variant where the vowel ro ot p ersists
i cons cons
C place
vocalic
V place
The reason for this is syllabi cation as discussed ab ove The sequence in i cannot b e
syllabi ed in any reasonable way b ecause of Contour
Sievers s Law
obstruent to assimilate to the place of the preceding consonant just likein
bak tje
The empty ro ot assimilates to the preceding obstruent If nothing
further would happ en this would make the form undistinguishable from
raampie and koninkie with resp ect to the i j alternation wewould
exp ect the form k ti
o o o o o o
k t t i
coronal coronal
Dutch has an indep endently motivated degemination pro cess simplifying
to so that will b e preferred over
o o o o o
k t i
coronal coronal
But here t and are still adjacent to one another other This is there
fore a violation of the OCP It can b e lifted by assimilating t and i in
the now familiar way As a result we end up with the following structure
o o o o
j
k
This structure violates neither Weak N C nor the constraint against
geminates nor the OCP norany of the syllable structure constraints We
j
get the pronunciation k as a result
Sievers s Law
The alternation b etween vowel and glide which is the topic of this chapter
is similar in some ways to Sievers s Law a phonological pro cess that has
b een op erativeinmany Indo Europ ean languages likeVedic and Gothic
Sievers Dresher and Lahiri Lahiri and Van der Hulst
Riad Rice Calabrese Although a full discussion of the
A complicationmight arise in the case of words like baan tje little job and paal tje
little eel In these cases n l t and i are all predicted to assimilate to one
another but this would mean that also the n and l should undergo palatalisation
Ihave the impression that this is true for n but not for l
Sievers s Law
complicated facts p ertaining to Sievers s Lawwould take us to o far a eld
it is interesting to compare the Rotterdam phenomena to real instances of
Sievers s Law I will concentrate on Gothic which is particularily interest
ing b ecause it p ossessed real phonological long vowels My main source
is Riad who discusses the following examples among others
a so okjis so okiis you seek
mikiljis mikiliis you glorify
sip o onjis sip o oniis you are a disciple
b nasjis nasjis you save
sto o jis sto o jis you judge
Riad argues following Vennemann among others that
syllable structure is the determining factor in all of these cases He as
sumes that there is an archiphoneme I which surfaces as either i or j
dep ending on the surrounding syllable structure
Thus the analysis of Sievers s law reduces to the characteriza
tion of syl labi cation in Gothic and Germanic Given archiphonemes
Sievers s law synchronical ly becomes a question of derivation
where I whenever it ends up as onset is realizedas j and
whenever it is not part of the onset is realized as a vowel
Riad
This is almost the same characterisation as I gave for Rotterdam Dutch
If we assume that the archiphoneme I in our present approach stands
for a nonsyllabi ed i we can adopt Riad s prop osals without
much further di culty
In the examples in a the optimal syllable structure is the following
I translate Riad s structures to the formalism adopted here I will return
to the representation of long vowels b elow
so okiis
Riad also has an underlying j di erent from the archiphoneme for glides
which do not have a predictable p osition in the syllable Here I p osit no such
archiphonemes for the Rotterdam vowel glide alternation Whether or not this could
also apply to Gothic is an op en question cf Calabrese
I represent Gothic long vowels as two ro ot no des that are b oth linked to the nucleus
no de so that sup erheavy syllables are represented as single syllables I havedonethis
to change as little as p ossible of Riad s representations Riad uses a mora theory even
though strictly sp eaking these representations are somewhat inconsistent with the theory
Ihavesketched in the previous chapters I am convinced however that with a few minor
adaptations these structures could b e adapted to the representations used here
Sievers s Law
N N
N N
N
N
s o o k i i s
mikiliis
N N N
N
N N
N
N
N
m i k i l i i s
The consonant b efore ii can b est b e analysed as part of the onset The
alternativewould b e to put this consonantinto the co da of the preceding
syllable but in general consonants prefer a p osition in an onset to a p osition
in a co da In this case the nucleus is the b est p ossible p osition for the ii
cluster if we assume that the complex consonant clusters of the form Cj
are marked in Gothic
The situation is somewhat more complicated in the case of nasjis On
rst sight this form p erhaps do es not di er to o much from sookiis We could
assume that the rst s in nasjis is in an onset We then get the following
structure
nasiis
N N
N N
N
N
n a s i i s
Sievers s Law
Yet Riad argues that in these cases wehavetotake Prokosch s Law
into account Prokosch According to this Law stressed syllables
should always contain at least two mora s i e two p ositions in the rhyme
in the Germanic languages cf the discussion of Norwegian in chapter
Main stress in Gothic is on the rst syllable so this rst syllable should
conform to Prokosch s Law This is no problem in the word sookiis b ecause
the long vowel already do es o ccupytwo p ositions The same story do es not
hold for the rst syllable of nasjis however For this reason the s should
be drawn into the rst syllable and we end up with the following structure
nasjis
N N
N N
N N
n a s i i s
In the form mikiliis whichwe discussed ab ove the syllable ki can stay light
b ecause it is not stressed
Disregarding the e ects of Prokosch s Law which is no longer op erative
in Dutch dialects the pro cesses in Gothic and the Rotterdam dialect show
up a remarkable resemblance There is however one subtle di erence that
wehave to explain in Gothic the long i alternates with ji in Rotterdam
Dutch the same vowel alternates with j
I think the key to the solution lies in the observation wemadeabove
Wehave just seen that a long o o in Gothic is able to let a syllable satisfy
Prokosch s Law This means that in Gothic a CV V syllable counts as
i i
heavy A Dutch i is not phonologically long but has a structure as a
Yet in Gothic long i has the really long two ro ot structure of b
a o
i
b o o
i
If weinterpret the two no des in b an onset and a rhyme the rhymal
segment will b e linked to i and also b e phonetically interpreted as such
The sequence as a whole will b e interpreted as ji on the surface The
Other issues
lter Contour against shared structure in tautosyllabic onset and rhyme
is apparently was less strictly observed in Gothic than it is in Dutch
Indirectly this could give us another argument against length in Dutch
If high tense vowels were really long it would not b e explained whyDutch
do es not b ehave like Gothic On the other hand the di erent ranking
of Contour can b e another explanation for the di erence it could b e
that long iq cannot break to ji b ecause that syllable is not p ossible for
indep endent reasons
Other issues
The main p oint of this chapter a statement of the conditions under which
i glides to j b efore a schwa in Rotterdam Dutch hasnow b een made
In this section I want to discuss some problems which arise if we treat the
Rotterdam Dutch i j alternation in the waysketched here Most of
these problems have already b een noted in the preceding sections but they
still await further treatment
In the rst place wehave the facts of Standard Dutch where wedonot
nd these alternations In the second place wehave to study the relation
between i and j in lexical forms where we nd semi minimal pairs
like Hanny name h ni h nj versus Sp anje Spain sp nj sp ni
Subsequently we also have to discuss the reason why the ultimate form of
S is i not i In the fourth place wehavewords like Itali e where
schwa follows i on the surface Finally in section I discuss the clitic
we which has almost the same structure as je viz it consists of a glide
followed byschwa but which do es not alternate with u or y and inversely
the p olite second p erson pronoun u which never alternates with w
Clitics and the diminutive in Standard Dutch The fact that
pakkie little b ox has a more optimal syllable structure than pakje in Rot
terdam Dutch raises the question why the sp eaker of Standard Dutchdoes
not cho ose this same optimal form Another way of putting the same ques
tion is why do es a Rotterdam sp eaker who starts sp eaking in a formal
style register all of a sudden prefer the less optimal syllable structure of
pakje ab ove pakkie
In previous chapters esp ecially in chapter I have argued that the
di erence b etw een style registers seems mainly to b e that Faithfulness con
straints are more observed in the more formal styles of sp eech
It is therefore crucial that S and the diminutive but also lexical words
like garage contain a schwa In informal Rotterdam Dutchthisschwa
the emptyvo calic ro ot no de remains unparsed This means that the con
Other issues
straint forcing the parsing of this ro ot is su ciently low at least lower than
Project N V In more formal styles of sp eech and in the Standard lan
N guage Parse Vowel and the equivalent constraint Project cons
is ranked much higher at least higher than Project N V
For a form like stok tje little stick this has the following e ect
stok tje formal style of sp eech
Candidates Parse Vowel Project N V
p
st kj
st ki
The di erence b etween Standard Dutch and Rotterdam Dutchisthus that
in the former dialect the constraint Parse Vowel do es not get lower than
N V inanystyle register The Rotterdam child has evidence Project
for a relatively high ranking of the Project N V constraint when she
observes that a form which surfaces in other contexts as j surfaces as i
here A child growing up in an environment where p eople sp eak Standard
Dutchnever encounters such evidence
A complication in the analysis of S is that the ie versus je forms
b ehave di erently with resp ect to Final Devoicing
heb je hebbie
h bj h bi
h pj h pi
These facts are discussed in Berendsen according to whose
intuitions Final Devoicing is optional in the case of j According to the
intuitions of the present author however and according to Bo oij the
variant with a voiced obstruent is more marked and certainly not b etter
than for instance lob j for loop je with an underlying voiceless plosive
I susp ect that in these cases the app earance of the feature voice is the
consequence of something like a minor phonetic assimilation pro cess
C j C j
j
voice
If we disregard this optional rule there are two p ossibilitie s h bi and
h pj In other words w e nd Final Devoicing in j forms but not in i
forms
In the theory prop osed here this do es not havetobetaken as an in
dication that the i form is more clitic li ke than the j form It can b e
Other issues
reduced to syllabi cation hebbie is naturally syllabi ed as h bi where
the b o ccurs in the onset of a full syllable On the other hand heb je
could b e syllabi ed as h p j It is true that this syllabi cation violates
some syllable structure constraints like the Syllable Contact Law On the
other hand it satis es the constraint against Cj clusters in onsets which
wehave to assume outranks the SCL
Project N V SCL
heb je formal style register
Candidates Project N V SCL
p
h p j
h b j
I should add here that Rotterdam S is the only clitic that do es not normally
show the e ect of Final Devoicing Other clitics b ehaveasifFinalDevoicing
applied b efore them heb em have em h p m heb een boek havea
b o ok h p nbuk This might b e related to the fact that in these other
clitics the vowel is a schwa cf Gussenhoven Bo oij b for recent
discussion of phonological cliticizati on in Dutch
Lexical forms Wehave seen that it is not necessary to sp ecify a
lexical di erence b etween i and j in S S or in the diminutive su x In
all those environments the di erence follows directly from resyllabi cati on
We can nowwonder whether there is any need to sp ecify this distinction
in any other part of the phonological system of Dutch Since wehave
lo oked at all relevant a xes and clitics the only place left to lo ok are the
monomorphemic stems
If we study other p ossible p ositions in the monomorphemic words a few
p ossible cases are found
garage ara z vs quasi kwazi
garage id
col lege k le z vs cohesie kohezi
class cohesion
schoelje sxulj vs koelie kuli
scamp co olie
Spanje sp nj vs Hanny h ni
Spain name
bonje b nj vs ponny p ni
ructions id
Other issues
Although none of these pairs are p erfectly minimal the distinctions seem
to b e fairly trivial and irrelevant They are quasi minimal pairs In each
of these cases the phonological environment directly b efore i and j is
exactly the same It seems as if wehave found cases where weneedto
sp ecify a distinction In one way or another wehave to indicate that the
words in the left hand column cannot b e pronounced as sxuli sp ni etc
even though the theory seems to predict otherwise
It has to b e noticed that in each of these cases it is exactly a coronal
consonant that app ears to the left of the j cluster I have found one class
of words with a non coronal consonant preceding j in Standard Dutch
hachje li afje snufje ikje puikje The corresp onding Rotterdam forms
all have ie These words have to b e regarded as a kind of idiomatic
diminutives This hyp othesis is supp orted by the fact that these forms are
all nouns and all have the neuter gender they cho ose the determiner het
het hachje het snufje etc The diminutive su x also makes a noun into a
neuter de viool the violin het viooltje the small violin This is not true
for the forms where je is preceded by a coronal They are not always nouns
and if they are nouns they mayaswell have common gender de kastanje
de ratatoelje de plunje de garage etc
Wemay therefore assume that in these cases the feature has b een linked
lexical ly to the consonant and to i The lexical sp eci cations of bonje and
ponny then are as follows in the case of p ni it do es not matter whether
the underlying form ends in i or in i this is the reason whyIhave
put a nal emptyrootinbrackets
a b nj
o o o o o
b n i
coronal
b p ni
o o o o o
p n i
cor cor
Van Haeringen p oints out that the word familie family has an interesting
history in Dutch in this resp ect In the oldest stages of Dutch the word was most
probably pronounced as fa mi li j In the nineteenth century the pronunciationwas
fa mi lj Since then fa mi li has b ecome the standard form although fami q lj is still used in some dialects
Other issues
We correctly predict that there are no lexical forms like herje I assume
r do es not have a coronal feature This form would get the following
representation
o o o o o
h r
coronal
The feature coronal is not doubly linked here Therefore this form can b e
syllabi ed as h ri without problems This is exactly what happ ens we
do havewords like herrie but no words like herje
For the rest we only nd j forms after a long vowel as in ooievaar
stork o j var i after a long vowel is also found but only in the words
na ef naive naif coca n id kokain and ru n ruins ry w in In
the case of na ef even if the underlying form would b e nai f the schwa
would b e followed by f which it do es not normally tolerate in its rhyme
as wehaveseeninchapter
In the case of ru ne and coca ne if the underlying forms would b e
ryi n and kokai n wewould get twoschwa syllables next to each
other in a monomorphemic word Again apparently the prohibition against
this typ e of con guration Project Ft V is lexically higher than On
set in Standard Dutch
In the immediately preceding discussion I have assumed that the obstruent
j sequences all are clusters in the literal sense Logically sp eaking it is also
p ossible that the j element is underlyingly sp eci ed as a secondary articu
lation on the consonant As a matter of fact the segments in and also
j j
for instance in words like jury id z yri and jam id z m seem single
segments on the surface with multiple articulation rather than sequences of
consonants
j
Why then are there no monomorphemic forms that surface as p k
j
or h f
Since only coronals allow a coronal secondary articulation the deciding
factor should b e the place feature The constraint which is resp onsible for
this is closely related to some constraints wehave already seen
ConsPlace C F A place feature dominated by a cons ro ot
C
no de should b e directly dominated by a C place no de
Notice the last word is a loan word of which the rst segment has b een adapted to
j
the Dutch phonology I assume this is an argument for considering z as an underlying
Dutch consonant
Other issues
This constraint if dominating the relevant Parse feature constraints
has as an e ect that secondary articulation is only allowed if primary and
secondary articulation are homorganic Consider for example the pruned
j
representations of x and z
j
a x
cons
C place
dorsal V place
C
coronal
V
b z
cons
C place
vo calic
V place
C
coronal
V
Representation a is ruled out by b ecause the coronal feature is
dominated by cons but not directly dominated by a C place no de In
b the only place feature coronal is dominated by cons and a place
no de The fact that it is also dominated by a V place is irrelevantfor
this constraint This is the reason why only coronals can have a coronal
secondary articulation
Of course this do es not exclude labial consonants with a labial secondary
articulation I do not discuss dorsal secondary articulation since its theo
w w w
retical status seems so marginal Yet segments like p b or f donot
seem to o ccur in Dutch For one reason or another non coronal secondary
articulations seem more marked than coronal secondary articulations
At the end of this section I have to brie y touch on three remaining
j
observations In the rst place we only nd tense vowels b efore z
j j
araz ar z this might b e related to the observation made on page
that vowels b efore fricatives are almost always tense In the second
place we nd only lax or high vowels b efore lj and nj sxulj sxolj
sp nj spanj indicating that in these cases wehave real clusters nj
Other issues
or lj the sonorant closes the preceding syllable while the yo d is syllabi ed
in the onset It seems that palatalized fricatives and nasal j are allowed
but palatalized nasals and fricative j sequences are not Finally exactly in
j
the case of lexical forms we nd no t forms thismaygive an indication
that we still need to refer to the notion Derived Environment cf Polg ardi
for discussion of issues relating to this in Optimality Theory The
segments t i fuse to a consonant with secondary articulation only when
they are separated by a morpheme b oundary Ihave not b een able to nd
a satisfying analysis for any of these three observations
The underlying form of S is not i Secondary articulation
formation in my analysis leaves an unparsed vowel ro ot viz the ro ot of
the underlying i The output of secondary articulation formation on
j
for instance the sequence si s without a following schwa will b e as
follows
cons cons
C place
vo calic
V place
coronal
j
In isolation this sequence would b e pronounced as s This raises the
question whywe need to assume si is the underlying form at all Why
couldn t the underlying form of S b e i rather than i
Ihopetohave made it su ciently clear in the preceding discussion that
it is essential to distinguish b etween underlying i and i b ecause other
wise we cannot explain the di erence b etween jenever and idee orbetween
the diminutive and S Single i should not b e p ermitted to surface as j
otherwise wewould incorrectly predict idee to surface as j de
I therefore assume that if a place no de is parsed and its corresp onding
ro ot no de is also parsed the place no de should b e linked to the ro ot no de
On the one hand this assumption seems somewhat ad hoc since it rules
out exactly the con guration that is discussed here On the other hand
it do es not seem to b e very unreasonable to assume a constraintlikethis
If two elements are linked underlyingly and if b oth elements are parsed
they should b e parsed together Naturalness is a notion that is di cult
to formalize or to measure but I do not think the assumption just made is particularily unnatural
Other issues
A problem connected to the underlying representation of S is whyonly
underlying high vowels are sub ject to gliding It seems to me that there is
evidence for a weakness constraint of the following form
Weak N high A vowel V is in the weak p osition of a syllable
V dominates high
In chapter wehave seen evidence for another weakness constraint refer
ring to the feature consonantal There is su cient evidence for weakness
constraints referring to high at di erentlevels of representation For in
stance in many languages high vowels are the only ones that can o ccur in
the weak p osition of a diphthong and Kenstowicz b has shown that
there are languages in which high vowels preferrably o ccur in the weak
p osition of the fo ot
Of course in a fully worked out theory we will exp ect constraints for
mally similar to Weak N high but referring to purely consonantal fea
tures which also do not o ccur in the head p osition of a syllable or fo ot
Weak N high expresses the intuition that high vowels are the most
consonantal the least sonorant among vowels see chapter for a fuller
discussion of this issue
High vowel followed byschwa One other fact that seems prob
lematic for the analysis prop osed here is formed bywords containing ij
like Belgi e and patrici er patrician
The problem is that for these forms the most logical underlying rep
resentation is i with the j inserted as a homorganic glide However
we cannot select this representation b ecause it would surface in the system
given here as i so that wewould incorrectly predict pronunciations such
as b lxi patrisi q r etc at least in the Rotterdam dialect
Wehave seen in chapter however that exactly these same words b ehave
as derived rather than monomorphemic forms with resp ect to stress If
they were really monomorphemic wewould predict stress on a p enultimate
syllable rather than on the antep enultimate where it actually resides My
The solution for this problem was that these words really are derived
schwa app endix in these cases constitutes a su x
This prop osal also helps us to understand the b ehaviour of the i
cluster in this case In chapter I prop osed a constraint MorPa rep eated
here
MorPa At least one element the most sonorous element of a
morpheme is incorp orated into a proso dic word
The form b l xi exists as the diminutiveof Belg Belgian
Other issues
If the schwain Belgi e is a su x of its own it should b e parsed into a
Proso dic Word according to MorPa This means that the parse b lxi
is unacceptable b ecause it leaves the whole su x unparsed The schwa
should b e parsed and in order to satisfy the onset requirement the gram
mar intro duces a glide that is homorganic with the preceding high vowel
b lxij
st p erson plural clitic Je is not the only clitic consisting of
a glide schwa in Dutch The other one is the rst p erson plural clitic
P we w The rst segment of this clitic is the labio dental glide w
but this clitic still do es not showany alternation with a labial high vowel
u or y at all dat we that we is never pronounced as d tu or
d ty but always as d tw or d t m with the allomorph m As
farasIknow no dialect of Dutch shows alternation of w with a high
vowel although j i is found in some restricted form in manyDutch and
Frisian dialects
On the other hand the p olite second p erson singular or plural pronoun
is u y There is no reduced or clitic form of this pronoun In particular it
do es not alternate with w Tomyknowledge there is no dialect in which
na u naby is ever pronounced as naw Although I call u a p olite
pronoun I feel this gap cannot b e attributed to the fact that u is only
used in formal sp eech In some idiolects u can b e used to address p ersons
with whom the sp eaker has reached a high level of intimacy like parents or
grandparents Nevertheless a pronunciation w isnever found Of course
a linguistic explanation for this gap is more easily found If the underlying
form of the p olite pronoun is y rather than y we predict the surface
to b e y This leaves the problem of P untouched however
It seems that wehaveanasymmetrybetween the two glides j and
w There are more reasons to consider the relation b etween i and j
amuch closer one than that b etween u or y and w j seems to b e a
more real glide than w in the sense that there are many arguments for
considering j really a variant of i in consonantal p osition while w seems
to b e slightly more consonantal
While j can never b e in a bip ositional onset at all jraak kjaad
w can act as an obstruent like element in onsets such as the one in wraak
revenge wrak or as a sonorantsuchasinkwaad angry kwat
I use the term glide in a lo osely descriptiveway Most of the asymmetries noted here
have b een observed byTrommelen who prop osed to assign w to the class of
liquids She also observed that intervo calic C w clusters are always syllabi ed as onsets
etui id e twi aquarium id a kwa ri jym
The distribution of wC clusters in onsets in Dutchisvery limited in twoways In
the rst place the number of words containing such clusters is very limited and in the
Other issues
On the other hand w and j do pattern together in certain ways In
particular this is true if we assume w is the glide version of y For
instance wehaveseeninchapter that a constraint against homorganic
glide vowel sequences a ects b oth ji and wy while the clusters b ecome
p erfectly regular if we replace the glides by normal consonants lier lyra
liqr dier animal diqr vuur re vyqr etc
The same typ e of restriction holds on glides after vowels While word
nal sequences like ej ij uw are excluded in contrast to ew iw
uj sequences like et it up are not
Although going into the matter in much detail would lead us to issues of
segment representation that fall to o far outside of the scop e of this thesis
I think one of the reasons for this o dd b ehaviour may b e the labio dental
character of w My preliminary prop osal is that this segment has a
coronal dental secondary and a labial primary place of articulation while
it still has a vo calic ro ot no de
cons
C place
labial V place
coronal
This representation would distinguish w from vowels and from j imme
diately b ecause the latter segments only have a secondary vo calic
place of articulation Clements Given an appropriate theory of p os
sible onset sequences which I will not try to develop here this gives us the
p ossibility to distinguish b etween w and j in onset clusters
On the other hand w is still a glide in the sense that it is a vo calic
no de dominating a vo calic place no de This makes it p ossible to formulate
the constraints on glide vowel vowel glide sequences just outlined
I nd it di cult to bring empirical evidence to b ear up on the represen
tation in I havechosen to make coronal the vo calic place feature
b ecause it allows me to formulate a pro jection constraint against gliding in
a fairly straightforward manner
second place in many dialects the w in these clusters hardens to the labio dental v
It might b e that the coronal feature is even absent phonologically and only the
result of phonetic enhancement The representation given here however might b e seen
as the result of a constraint to the e ect that having more than one place feature under
any place no de is dispreferred Ben Hermans p c
Furthermore some dialects for instance in Flanders and in Dutch Limburg havea
bilabial glide For these dialects I would b e forced to assume a representation where the
coronal feature has disapp eared and the labial feature is linked to b oth tiers
Other issues
N all vo calic features should b e in the head of an N Project V
except the feature coronal
V
Constraint excludes all vo calic ro ots in the onset except for the front
unrounded vowels and w Interestingly it can b e seen as a weaker version
of Onset Project F N
It seems furthermore justi ed to put the labial feature in C place p o
sition b ecause in its interaction with other consonants it b ehaves as labial
rather than coronal pw clusters are excluded b ecause of the OCP and
Contour but tw clusters do o ccur in Dutch as the examples twee two
twe twist dispute tw st and twijg twig tw ix may attest
Another prediction wemake is that w can lexically o ccur in the nu
cleus of a syllable just like nasals and liquids cf chapter Post lexically
w headed syllables would b e sub ject to schwa ep enthesis just like sylla
bles headed by sonorant consonants Now there are no syllables ending
in w in Dutch Yet Trommelen has shown that word nal yw
clusters in forms like schaduw shadow sxadyw b ehavelikeschwa headed
syllables in many resp ects for instance syllable structure and stress It
could therefore b e that in this case vowel ep enthesis is accompanied by
a total assimilation of the place features from w to the sup er cial y
head
Finally Iwant to p oint out that this analysis do es not imply that there
are no homorganic glides after u or y After u or u we usually
nd a bilabial u After y and y we nd either j or bilabial u
dep ending on the dialect and some sp eaker dep endent factors which are
hard to formalize
a douane customs duuan
b duo id dyjo dyuo
c bui en shower s b yj b yu
The except clause makes us susp ect that this constraint con ates some more basic
constraints for instance a general constraint against vo calic features in consonantal
p osition interacting with a constraint on licensing of coronal Aninteresting asp ect
V
of this is that here wehave another instance of the sp ecial status of coronals in the
syllable in the sense that coronal can o ccur in a p osition in which other place features
are not licensed See chapter for some additional discussion
There is a constrast b etween forms with a tense vowel and forms with a lax vowel
In the former case ep enthesis is obligatory schaduw sxadyw sxadw in the latter
case it is prohibited murw mellow m rw m ryw erwt p ea rwt rywt
The former case can b e explained but I have no explanation why there is no ep enthesis in the latter case
Conclusion
Conclusion
In this chapter wehave studied typical vo calic elements o ccuring in extra
cephalic p ositions I think the analysis of the vowel glide alternation has
given us the clearest example of an advantage of theories which use the
idea of relative strength of constraints over other views of grammar If we
put twovowels next to each other these get into comp etition for b eing the
head of a nucleus If one of the twovowels is a schwa and the other is a full
vowel the latter normally wins This is something wehave already seen in
chapter In this chapter I have shown that certain external circumstances
mayfavour the surfacing of schwaover that of the full vowel Twosuch
circumstances have b een shown to exist in Dutch These were the Onset
constraint and the OCP These two constraints seem to overrule the normal
desire of full vowels to b ecome syllable nuclei
With resp ect to the Pro jection theory the main result of this chapter
h are com was that wehave to p osit a class of Weakness constraints whic
plementary in their formulation to the Pro jection constraints The latter
say that a strong p osition in the proso dy is connected to a certain feature
the former that certain other features are connected to weak p ositions
It app eared that vowels which o ccur in such a p osition are typically
high vowels and most typically i Dutch w has b een argued to b e more
consonantal than j Furthermore wehave seen some evidence for the
hyp othesis put forward by Clements that consonants and vowels
have the same place of articulation features This hyp othesis can help us
explain palatalisation The reason furthermore why the voiceless stop t
palatalizes more readily than any other coronal consonantisthatitisina
sense the simplest consonantwhichprovides us with evidence either for a
theory of feature markedness or for monovalent features
The Pro jection ConstraintFamily
Intro duction
In this thesis I have shown in what way features on the head segmentof
some proso dic constituent and the structure which is built on top of that
segment can in uence each other It app eared that in particular aperture
features like high lax and low are relevant for proso dic structure
This nding in itself is not new On the one hand theories that work
with sonority scales often assign separate p ositions on the scale to high
mid and lowvowels where high vowels get a lower sonority and lowvow
els a higher sonority Most often these sonority scales are then used to
explain facts that are related to syllable structure such as gliding and
diphthongization On the other hand rules of metaphony changing
vowel height in stressed or unstressed p osition have b een prop osed for in
stance in the literature on various Italian dialects
Yet this insight never seems to have received a systematic treatment I
think that one of the advantages of the approach defended here is that it
o ers a more uniform formal means to express the relation b etween vo calic
features and proso dic structure This allowed us to develop for instance a
more interesting view on the relation b etween laxness and syllable closed
ness or on the degenerate syllable which is built on top of the empty vowel
The theory presented here is couched within a framework of constraint
based phonology Ifwe do not constrain the notion of p ossible constraint
such a theory is not particularily restrictive
This statementobviously needs some quali cation Since in standard
Optimality Theory the set of constraints is supp osed to b e universal and
nite it is not strictly necessary for the individual constraints to b e for
mally related to one another in any sense In principle it is p ossible that
Universal Grammar consists of a numb er of totally unrelated constraints
See Kenstowicz for the application of the sonority hierarchy to stress
phenomena
Pro jection and weakness
However given the ndings of phonological theory it seems more plausible
that the set of constraints is mo dular sub divided into subsets of tightly
connected constraints It also seems metho dologically b etter to assume
such a clustering of constraints As long as linguists do not agree on the
exact content of UG and all other things b eing equal the use of a constraint
A in a particular analysis seems to b e preferable over the use of a constraint
B when A is formally related to other constraints C and D whereas B is
not
The most imp ortant new constraint families whichIhave prop osed in
this thesis are the pro jection family and the weakness family These fam
ilies relating feature content to proso dic structure form the core of the
prop osal made here
Pro jection and weakness
I will start out by listing all the pro jection and weakness constraints that
Ihave prop osed in this thesis
Connect N lax p N dominates lax N is the head of a
branching N
Pro ject V Ft p N dominates a vo calic no de N is the head
of a branching Fo ot
Pro ject V Ft p N dominates a vo calic no de N is the head
of a Foot
N is the head of a branching Foot N dom Pro ject Ft V p
inates a vo calic no de
Pro ject N V p N is the head of a branching N N dominates
avo calic no de
Pro ject N cons p N is the head of a branching N N
dominates cons
N F p N is the head of a branching N N dominates Pro ject
some feature F F
Pro ject F N Onset p N dominates some feature N is the
head of a branching N A syllable with some feature in its head
should have an onset
Pro ject N low p N branches N dominates low
N p N dominates a vo calic feature N is in the Pro ject V
head of an N
Pro ject cons N p N dominates cons N is the head
of a branching N All vowels are heads of syllables
Weak N high p A vowelVisinthedependent p osition of a
N V dominates high syllable
Pro jection and weakness
Weak cons N p A segment S has the feature cons Sis
in the dep endent p osition of a syllable N
Weak N C p A segmentintheweak p osition of a syllable
needs to have a consonantal place feature C C fcoronal dorsal
c c
labial g
c
This family of constraints forms the core of the prop osal made in this
thesis It is a consequence of the syllable headedness hyp othesis that the
feature structure on the syllable head determines to a large extent the
structure of the syllable The general form of most of the constraints the
pro jection constraints listed here is as follows
x
a b ottom up N dominates a feature F N branches
x
where N f N N N Ft g F some feature set S
x
b top down N branches N dominates a feature F
x
where N f N N N Ft g F some feature set S
In some cases there is no reason to distinguish b etween a top down and
a b ottom up version of the pro jection constraints In those cases I called
the relevant constraints Connect X Y Yet in other cases ranking con
siderations led us to treat b ottom up and top down as constraints with
indep endent lo cations in the hierarchy For instance in French tense mid
vowels do not o ccur in closed syllables which I to ok as an indication for a
su cient high ranking of Project N lax but in the same language lax
mid vowels can freely o ccur in op en syllables I assumed that this shows
that Project lax N is ranked lower
Strictly sp eaking this is not necessarily true We could also assume that
while in Dutch Connect N lax dominates all Faithfulness constraints in
Frenchwe draw a distinction b etween Parse lax on the one hand and
lax on the other
a Dutch Connect N lax Parse lax lax
N lax lax b French Parse lax Connect
Connect N lax dominates the constraint against insertion of lax so
that tense vowels in closed syllables get laxed without problem Yet Connect N
lax is dominated by Parse lax so that underlyingly lax vowels staylax
even if they o ccur in op en syllables
Given this p ossibility of di erently ranking the twotyp es of Faithfulness
it might turn out to b e unnecessary to di erentiate b etween b ottom up and
top down pro jection It would therefore b e interesting to try to recast the
analyses given in this thesis using only Connect X Y constraints Yet
Pro jection and weakness
since this would complicate the ranking arguments considerably Ihave
decided to only use Project X Y constraints wherever there is no real
evidence for an active Project Y X
Normally pro jection constraints refer to heads of branching constituents
Only at the level of the fo ot wehave seen constraints referring also to heads
of non branching constituents
Furthermore in a few cases wehave seen evidence for constraints of the
following form the weakness constraints
x
a b ottom up N dominates a feature F N is in the weak
x
p osition of an N constituent
x
where N f N N N g F some feature set S
x x
b top down N is in the weak p osition of an N constituent
N dominates a feature F
x
where N f N N N g F some feature set S
We need to draw a distinction b etween pro jection constraints and weakness
constraints only in a theory that uses monovalent features
For binary features the weakness constraint N dominates a feature
x x
F N is in the weak p osition of an N constituent equals the
x
pro jection constraint N branches N dominates a feature F and
x x
N is in the weak p osition of an N constituent N dominates a
x
feature F equals N dominates a feature F N branches It might
b e relevant that the weakness constraints used here refer either to cons
or to high and that high and consonantal are two features for
which it is notoriously hard to design a unary feature analysis
Again it is p ossible that there is no real need to distinguish b etween
b ottom up and top down weakness but for ease of exp osition I have
decided to distinguish b etween the two
Several remarks hav e to b e made ab out these constraintschemes First
it seems that they are b est interpreted in a theory of proso dic structure in
whichevery pro jection no de is a projection of its head in the most literal
i i i
sense of the word for every N a pro jection of N N has exactly the
i
same features as N
There are several theories of syllable structure whichhave the required
prop erties In this thesis I havechosen to adopt an X bar like structure
which is similar in some resp ects to the theory of Bare Phrase Structure
prop osed by Chomsky In this theory the X bar lab els are seen as
mere mnemonic terms A structure like a is a mere way of representing
the structure in b in an insightful way
Pro jection and weakness
a N b
N
N
N
b l k b l k
We could also assume that feet are pro jections of syllables to higher levels
By transitivity feet and proso dic words are therefore headed by syllable
heads in the typical case byvowels
There are other ways of formalising the same idea that heads pro ject to
higher levels at which they enter into lo cal relations with other elements
Examples of such formalisations are Dep endency Phonology Anderson and
Jones Anderson and Ewen and bracketed grid theory Halle and
Vergnaud The equivalence b etween these formalisms and the typ e of
X bar trees used here has b een established in Van Oostendorp In
this thesis I havechosen to use X bar theory It will b e more di cult
to incorp orate these constraints into a theory using moras b ecause the
notion head seems foreign to this theory and b ecause the structures used
in ortho dox moraic phonology are not binary branching
There are two parameters in the de nitions and the level of
n
pro jection X and the feature F In this chapter I brie y discuss the range of
p ossibiliti es that arise if wecombine the di erent settings for the pro jection
level parameter and for the feature parameter Not every logically p ossible
combination is discussed and I have to admit that I havenoteven found
all logically p ossible combinations of parameter settings attested In the
discussion of the previous chapters it has turned out to b e convenientto
refer to the following sets of features
V fhigh low lax coronal labial dorsal cons son g the
v v v
set of vo calic features
F the set of all segmental features
Avery interesting alternativewould b e to assume that feet are built on top of N
or N no des The reason for this could b e that by de nition no further pro jection is
N This would help us explain why onset segments often do not count p ossible on a
for stress and it would also provide us with an interesting view of fo ot structure when
combined with the ideas sketched in app endix A
The fo ot level
For more discussion see section b elow I will now discuss the particular
instantiations of a and b for eachlevel of the proso dic hierarchy
The fo ot level
The rst level I want to consider is the fo ot level On this level we exp ect
the following pro jection constraints
Nheads a Project F Ft N dominates a feature F F V
a branching fo ot
b Project Ft F N heads a branching fo ot N dominates a
feature F F V
One version of a is the constraint Project V Ft which I used in
the analysis of Dutch and an instance of b is Project Ft V again
applying transp osition As wehave seen in the case of the Connect Ft
V constraints any vo calic feature under the vo calic no de is able to attract
in the case of a or license in the case of b a fo ot
Normally pro jection constraints refer to heads of branching constituents
Only at the level of the fo ot wehave seen constraints referring also to heads
of non branching constituents In the analysis of Dutchwehave seen evi
dence for a non branching counterpart of a Project V Ft
In chapter wehave seen evidence for a version of the constraint
Project Ft in Frenchwhich refers to the feature lax
Project Ft lax French Nheadsafoot N dominates a
feature lax
The reverse constraintmight b e resp onsible for mid vowel neutralisation
e ects in other languages and dialects For instance in Brazilian Portuguese
It is di cult to nd similar evidence at the word level I have not b een able to
nd any evidence that pro jection constraints mightbeatwork at levels higher than the
proso dic word such as the phonological phrase the Intonation Phrase or the Utterance
Mayb e one might think of the assignmentofFo cus tones as the result of such a constraint
x
N is the head of an Intonation Phrase N dominates a H tone but there do not
seem to b e languages in which the o ccurrence of for instance low tones is restricted to
the head of any phrasal category and furthermore instances of N dominates a H tone
x
N is the head of an Intonation Phrase are unknown to me
It is usually assumed that Dutchisquantity sensitiveasI have discussed in chapter
The analysis presented here raises another p ossibility which is not explored here viz
that Dutch stress is sensitive to the feature lax Again here wehave an instance of
the close connection b etween laxness quantity and closedness I nd it hard to nd
go o d empirical evidence for or against stress theories based on the closed versus op en
distinction of syllables on the one hand and stress theories based on the tensedness versus
laxness of vowels on the other
The fo ot level
lax vowels can only o ccur in stressed p osition In weak p ositions suchvowels
might b e neutralized towards their tense counterparts Wetzels
a m ro mor amos I we reside
b m ve mov ei move sing plur
In the case of Portuguese rhyme structure seems unimp ortant for the fea
ture lax what matters is rather fo ot structure
We might assume this pattern is the result of the interaction b etween
constraints on fo ot structure and all dominating Parse lax in Brazil
ian Portuguese
Project lax Ft Portuguese N dominates a feature lax
N heads a branching fo ot
Metrical Wellformedness MWF Project Ft Parse lax
m r o
Candidates MWF Project Ft Parse lax
p
m ro
m r o
m oro
m r amos
Candidates MWF Project Ft Parse lax
m ra mos
m r amos
p
mo r amos
In this thesis wehave not come across any evidence for a weakness con
straint at the fo ot level Yet there are generalisations which could b e stated
in these terms In all relevant cases I have found the feature to which
Weakness refers is high or cons
Weak Ft high A vowelVisintheweak p osition of a fo ot V
dominates high
In the discussion of Chamorro on page wehave seen that mid vowels in
that language surface as high in unstressed p osition This might b e seen as
an e ect of
Another e ect of the same constraint can b e seen in Northern Salentino
Italian Calabrese Exactly as in Chamorro mid vowels can only
surface as such in stressed p osition otherwise they are high
The fo ot level
sg kr e u s ntu kan osku
pl kri ti amu si nt mu kanu s s mu
b elieve feel recognize
When we follow Burzio wemighteven have evidence for weakness
constraints p ertaining to high vowels and consonantal syllable heads in
English Burzio distinguishes b etween three classes of syllables which are
relevant for English word stress Heavy LightandWeak Weak syllables
are in a sense the lightest syllables for stress in manywords they are not
metri ed at all and when they are incorp orated into a fo ot this fo ot itself
b ecomes weak i e it cannot get primary word stress Whether or not a
syllable is weak is dep endent on its head Burzio claims that Weak
syllables either haveanempty head in the case of degenerate syllables a
consonantal head ble cle etc or a high vowel as its head ly ry
A particularly well known example of a feature which stands in a sp eci c
relation to stress is of course tone In many languages tones are aligned
with the most prominent syllable in a fo ot a word or a phonological phrase
This is expressed for Serb o Croatian in the following ways by Hermans
a and Zec resp ectively
a Hermans a Tone Blo cking A tone maynotbe
governed i e it may not o ccur in the dep endent p osition of
either a syllable or a fo ot
b Zec Align High tone with the head of a fo ot
Tomethesetwo conditions seem to b e equivalent We could formalize b oth
constraints in the pro jection framework as follows
Project high tone Ft Serb o Croatian N dominates a high
tone N heads a fo ot
Actually the situation is a little more complex since high tones maynot
occurintheweak mora of a stressed syllable either so that wemay
assume that Serb o Croatian also has a tone related constraint at the level
N of
Finally Serb o Croatian also requires all feet to b e accompanied bya
tone at some level of representation This means that we also nd a tonal
counterpart of Project Ft in this langugae
Project Ft high tone Serb o Croatian N heads a fo ot N
dominates a high tone
See also for instance Goldsmith Peterson and Progo and Peterson for similar principles in Nguni
The fo ot level
The theory of pro jection and weakness constraints predicts that only fea
tures of the head of a syllable may a ect the p osition of that syllable in a
fo ot One very well known counterexample to this claim is Pirah a Everett
and Everett In this language syllables with a voiceless consonantin
their onset count as heavier than similar syllables with voiced onsets which
in turn are heavier than onsetless syllables Stress falls on the heaviest of
the nal three syllables I underlined stressed syllables
kaa gai word biisai red kaibai monkey etc
The fact that it is the feature voice which counts as relevantisinteresting
for at least two reasons In the rst place I assume that vowels which
apparently are the only syllable heads in Pirah a do not p ossess a sp ec
i cation for voice Now some work in morphology has concentrated on
the concept of a RelativizedHead Selkirk Di Sciullo and Williams
According to this work features of non heads may p ercolate up to
the pro jection level but only provided that these features are not present
in the head The Pirah a stress facts may provide us with a phonological
example of this
The second reason whyitisinteresting that it is the feature voice that
seems to count as relevant is that there is a strong correlation b etween
laryngeal features and tone Esp ecially the voicing of onset consonants
seems to b e the most common source of tonogenesis Kingston and Solnit
Authors like Halle and Stevens Yip Bao
Duanmu Van der Hulst c have suggested to treat these two
classes of features as one gesture For instance Yip prop oses the
following partial feature geometry
Laryngeal
Register H L voice Glottal Ap erture
c g s g
In this mo del High and Low tone on the one hand and voice are basically
the same thing More sp eci cally one could saythat voice High
upp er and voice Low upp er Other mo dels are p ossibly also
See Everett for a discussion of the interaction b etween stress and vo calic tone
in Pirah a
Yip assumes Pitch no des dep endent on Register in order to account for four
way heightcontrasts These I have omitted b ecause they do not seem to b e particularily
relevant to the present discussion Furthermore c g constricted glottis and s g spread glottis
The N level
feasible but in any case the relation b etween voicing and tone seems to b e
rather well established
Wethus can reduce the relevant stress constraint to the version of
Project N suggested for Serb o Croatian in ab ove
The N level
The level b elow the fo ot is the syllable level Here we nd the following
constraints
N N dominates a feature F N heads a a Project F
branching N
b Project N F N heads a branching N N dominates a
feature F
All pro jection constraints at this level seem to refer to branching Nnodes
I therefore leave out the subscript
An instance of constraint b is Project N V inchapters and
the constraint a has an instantiation that has come to b e known in
the literature as the Onset constraint Project N V can b e satis ed by
anyvo calic feature just like the constraints on the fo ot level The Onset
constraint on the other hand seems much stronger in the sense that any
syllable needs an onset even when it is headed by a sonorant consonant
or an emptyvowel
At the level of the syllable wealsohave evidence for a weakness con
straint
N high A vowelVisintheweak p osition of a syllable Weak
no de N V dominates high
This is part of the constraint HighWeak the constraint that states that
only high vowels can o ccur in extracephalic p osition A similar constraint
Weak N cons mightsay that only consonants may o ccur in onsets and
After this reduction only one problematic case remains to b e solved This is Madi
Madi Davis Go edemans p c The stress rule in this language seems to b e of
approximately the following form stress is on the rst syllable except when the second
syllable contains a coronal obstruent in its onset in which case stress is on that syllable
If this is true it is problematic for the accountpresented here in twoways In the rst
place if vo calic and consonantal place features are the same we cannot invoke the notion
of a Relativized Head In the second place place features do not seem to play a role in
prominence relations in any other system I know
N level
N that consonants o ccur only in the dep endent p ositions of Weak cons
syllables
In order to capture the observation that consonants typically o ccur in
onsets we could also add the following weakness constraint
Weak cons N A segment S dominates F F a consonantal fea
N ture S is in the weak p osition of an
The Onset constraint Project F N is among the strongest pro jection
constraints in one imp ortant resp ect any phonological feature is su cient
to force a syllable to have an onset It is unclear whythe Onset constraint
should b e so forceful Wehave encountered a weaker version of this con
straintinchapter viz Project V N al l vo calic features should bein
the head of an N Wehave also seen a constraint Project cons N a
constraint forcing all vowels even schwa to b e heads of syllables
In general the pro jection theory is b etter in explaining the symmetries
between the Nlevel on the one hand and the lower and higher pro jection
levels on the other than in explaining the asymmetries between those cat
egories Some of these asymmetries are i Project N never seems to
refer to lax or low whereas pro jection at lower or higher levels do es
ii onsets do not or hardly ever contribute to weight iii the relation
between voice on onset obstruents and tone on nuclei do es not seem to b e
mirrorred in the relation b etween nucleus and co da iv in many languages
co da p ositions allow only a subset of the consonants that are allowed in
onset p osition v the Contour constraint also seems quite typical for
onsets constraints against the homorganicity of co da segments and the
preceding nucleus seems far less widespread The theory would probably
have to b e supplemented by a theory on consonantal p ositions The pro
jection theory is intended to b e part of a mo dular phonological system of
which a theory of consonantal p ositions should b e another part
N level
The most imp ortant instance of a pro jection constraintonelevel b elow the
syllable level is what I have called Connect N lax
Connect N lax
a Project lax N N dominates lax N heads a branching
N
b Project N lax N heads a branching N N dominates lax
Nuclear level
Again all instances of Project N seem to refer to branching rhymes In
Dutch the two parts of Connect N lax seem inseparable visa vis their
overall p osition in the hierarchy butwehave seen that in Frenchonly
b is visibly active ininteraction with constraints on place In b oth
languages the constraint refers to the feature lax and the same is true for
Andalusian Spanish and for Polynesian languages like Eastern Javanese and
Tagalog In many languages the only vowels that are allowed in the weak
constituent partofadiphthong are the high vowels If falling diphthongs
form a branching N this might b e describ ed byaWeakness constraintof
the following form
Weak N high A vowel V is in the weak p osition of an N no de
V dominates high
As we will see in the following section it can also b e that diphthongs are
represented as branching nuclei In that case the N in this de nition should
b e replaced byN In chapter wehave seen evidence for a weakness
constraint Weak cons N This constraint said that consonants should
o ccur in a dependent p osition of the rhyme not in its head and it banned
syllabic nuclei in the p ost lexical phonology
In the literature we sometimes nd a constraint against co das the
NoCoda constraint There are in myviewtwo p ossible ways to view
the relation b etween this constraint and the pro jection constraints at the
N level Either we can see NoCoda as a formulation of Connect N lax
in combination with a constraint against the insertion of lax or wecan
see b as the result of the constraint ranking a NoCoda
Nuclear level
The lowest level of pro jection is the nucleus If we assume that this level
can branch just as well as any other level and that we need to distinguish
between a nucleus and a rhyme we might supp ose that also here wehave
pro jection constraints of the following typ e
a Project low N a segment S dominates low Sheadsa
branching N
b Project N low A segment S heads a branching N S
dominates low
There would however b e one p otential problem for this reformulation we could
not explain why there are langugaes whichdohave the feature lax and for instance
lax harmony while at the same time only allowing for op en syllables
Features
It is very likely that the nucleus is the domain for long vowels and diph
thongs cf for instance Harris for a discussion of English long vowels
If this holds true the constraint NoLongVowel can b e reduced to b
In Antwerp Dutch all vowels are lax if and only if they are long Un
der a branching nucleus analysis of long vowels this could b e analysed by
assuming a and b with reference to lax instead of low are very
strong in this dialect
What is interesting ab out the branching nucleus analysis is that falling
diphthongs show up the prominence asymmetries that wehavecometo
exp ect the rst comp onentvowel is generally low and or lax and the sec
ond comp onent is high and tense For instance Philips lists the
feature changing rules needed to account for all falling diphthong forma
tions mentioned in Hayes
Middle English iq uq ei ou Add high left
New High German iq uq ai au Add low left
Icelandic eq oq q e o Add ATR right
Central Swedish iq yqTq uq ij yw T u Add cons right
Czech eq oq ei ou Add high right
In all cases the dep endent part on the right gets either high cons or
ATR and the head to the left gets high or low
Atechnical problem that has to b e solved under an analysis assuming
branching nuclei for long vowelsisthatabranching nucleus has to count
N in all resp ects but not vice versa the branching nucleus as a branching
apparently is enough to satisfy Connect N lax it contributes to syllable
weight in the same way as a branching N and furthermore N and N cannot
branch at the same time otherwise wewould exp ect sup erheavy syllables
suchas k ilk in Standard Dutch As always there are ways of solving
this problem for instance by making the non co o ccurrence of branching N
Ninto a separate constraint dominating all others Another and branching
option would b e to intro duce a government relation into the theory and
derive the non multiple branchingness e ect from a stipulated adjacency
and directionality requirement of this relation Kaye Lowenstamm and
Vergnaud
Features
The following features or sets of features are referred to in the pro jection
and weakness constraints used in this thesis
Rising diphthongs need to have a separate treatment cf Sluyters for a
p ossible approach in which a subnuclear constituent is prop osed the peak
Features
def
a F the set of all phonological features
def
b V F the set of vo calic features f high low lax coronal
dorsal labial g
c lax low cons son upp er voice V
d high V
It is imp ortant to constrain the p ossible value for the F parameter to these
values since probably we do not exp ect for instance the place feature of
avowel to a ect its pro jection capacities in whichcasewewould predict
languages in which rounded vowels can only o ccur in stressed p osition or
in a closed syllable
The fact that ap erture features and more sp eci call y high and low
are relevant for syllabi cation has b een implicitly accepted in most of the
literature Many theorists who employ a sonority scale to describ e the
facts of a language and who distinguished among vowels on this scale as
signed a lower sonorityvalue to high vowels than to mid vowels or low
vowels
The sonority di erence is captured in the approach defended here by
assuming the metatheoretic generalisations in
a The features lax and low cons son play a role in
pro jection constraints not in weakness constraints
b The feature high and cons playsaroleonlyinweakness
constraints
This expresses the intuition that high vowels are in a sense more consonantal
than the non high vowels while low andlax vowels are in a sense the most
ideal heads of syllables and feet
In purely tree geometric terms it is not very clear why ap erture features
playsuch an imp ortant role in the pro jection of syllables It is relatively
easy to show that the ap erture no de and the place no de should b e sisters
since some languages show a pro cess of total vowel assimilation spreading
all vo calic features to the exclusion of the non vo calic features and there
is no evidence that the ap erture features are for instance linked higher in
the feature tree than the vo calic features Harry van der Hulst p c has
suggested that mayb e ap erture features play the role of a head in the
structure of a vowel Since the pro jection of an elementby de nition is
In order to go again one step further and explain we could either refer to pho
netic grounding or try to explore the hyp othesis put forward in Van der Hulst c
that phonological features haveaninternal structure which in the case of high would in
clude more consonantal elements and in the case of low and lax more vo calic elements
Features
the pro jection of the head of that element this would explain the asymme
try b etween ap erture features and place features If such a theory can b e
worked out in the appropriate way so that for instance we can still distin
guish the pro jection prop erties of e which has at least one vo calic feature
but no ap erture feature and which has neither an ap erture feature
nor a place feature it would b e an interesting extension of the pro jection
theory defended here Possibly such a theory would refer to the notion of
a Relativized Head
In section I have argued that also tone and voice play a role in
pro jection constraints In all wehave the following feature geometry with
the no des which are relevant for pro jection of metrical structure in b old
face
cons son
Laryngeal C place
vo calic
Ap erture V place
All features except for the Place features seem to b e able to playarole
in pro jection constraints From a phonetic p oint of view we could mayb e
say that the relevant features except the ma jor class features all havea
primary in uence on the F level buthightoneandvowel heighthave
the opp osite e ects high tone attaches to head p ositions whereas high
vowels tend to o ccur in dep endent p ositions even though the twotyp es of
heighthave comparable e ects on F The question therefore has to remain
op en
A further sp eculation in chapter has b een that the feature parameter
might b e set for the pro jection constraints at every di erentlevel For
instance in Frenchwehave found evidence for Project Ft lax but not
for Project Ft V whereas in Dutchwehave found evidence for the
This is true under the assumption that mid vowels are undersp eci ed for ap erture
The same typ e of problem holds for a particle approachtovo calic features with resp ect
to the comparison of high vowels whichbearnoAelement or op en feature and therefore
are presumably undersp eci ed for op enness and schwa
One p ossibility suggested in Van der Hulst c is that approximately the
b oldfaced feature sets here act in a sense as heads of the segment Head features would
then b e allowed to p ercolate up to higher proso dic levels of pro jection whereas this
would b e much more di culty for non head features In this resp ect it mightalso
b e relevant that segments with a place sp eci cation but without ap erture i e full
mid vowels such as e and o are much more common than vowels without a place
sp eci cation but with ap erture instances of the latter might b e
Constraints con icting with pro jection and weakness
latter constraint but not for the former No cases were found of con icting
or otherwise interacting Project Ft constraint It might therefore b e
more plausible that languages have at most one typ e of Project Ft
constraint and that a parameter is set
Constraints con icting with pro jection and weak
ness
The relation b etween vowel quality and syllable structure cross linguisti cal ly
seems to b e su ciently weaksoastohave gone undetected in most of the
literature In many languages the qualityofvowels seems to have hardly
any relation to proso dic structure at all Under the assumptions presented
here all the constraints presented are supp osed to have a universal force
They are present in the grammar of every language b e it that in some
grammars they can b e forced to b e violated
It is therefore imp ortant to detect the constraints that crucially con
ict with pro jection and weakness The most reasonable candidates are
the Faithfulness constraints In order to satisfy a pro jection or weakness
constraintwe often have to insert or unparse features Take for example the
constraint Connect N lax and let us supp ose this constraintisranked
very high and wehave an input domp with a tense vowel and a sequence
of consonants which can only b e syllabi ed as a co da plus an onset All
candidate outputs for this input which satisfy Connect N lax violate
some faithfulness constraint
domp
Candidates Connect N lax Faithfulness
domp
d mp lax
do m p Parse p
do m p Parse m
Parse p and Parse m are shorthands for parse constraints on all of the
features of these two segments In a language in which all of these faith
fulness conditions dominate Connect N lax the output domp will b e
selected
In some cases pro jection constraints can also work against each other
I will givetwo examples from the analysis of Dutch to illustrate this p oint
In the Dutch lexicon wehave Connect V Ft Project N V The
latter constraint among others disallows schwa headed syllables but the
former requires exactly such syllables in unstressed p osition Thus the head
Conclusion
N V in order to satisfy of an unstressed syllable can violate Project
Connect V Ft
The second example is from the p ostlexical phonology In that stratum
wehave to insert schwa in a syllable with a consonantal head suchas
vergiet strainer vr it in order to satisfy Project cons N Wecan
N V do this in twoways vr or v r The former violates Project
the latter Connect N lax We select the latter b ecause Project N
V Connect N lax
Conclusion
In this chapter I have discussed the notion p ossible pro jection or weakness
constraint Some of the e ects that fall under this notion have already
b een known for a long time For instance it is well known that high vowels
are the most likely among all full vowels to o ccur as the weak part of a
diphthong as a glide in an onset or in the weak syllable of a fo ot Similarly
consonants are very marked syllable nuclei and if they do o ccur as such
the syllable in most languages has to b e unstressed Generalisations such
as these can b e captured in a theory of weakness constraints
Lax and lowvowels on the other hand in most languages do not partic
ipate in glide formation do not o ccur as the weak part of a diphthong and
get stressed wherever they can Suchvowels are the most typical syllable
heads They also require a lot of extra structure for instance in Dutch
where they only o ccur as the heads of branching rhymes Generalisations
such as these can b e captured in a theory of pro jection constraints
Mid tense vowels are in b etween high vowels and lowvowels They
can o ccur in head p ositions but mostly they do not require anybranching
in many languages mid vowels can also b e in a glide p osition or in the
unstressed syllable of a fo ot There is go o d evidence for assuming they lack
anyvowel height sp eci cation since they alternate quite freely with schwa
This is what the theory presented here predicts
A Constraints and Families of Constraints
In this app endix I list all the constraints used in this thesis with the ex
ception of the pro jection constraints that have already b een dicussed in
the previous chapter I have adapted most of the constraints listed here
from the literature they serve asa background to the pro jection theory
Their interest for this thesis lies in their interaction with the pro jection
constraints not so much in their own sp eci c formulation
Here I show that some of these constraints can b e organized into fami
lies sets of constraints that are derived from one common ancestor by the
instantiation of certain parameters I will discuss each of these families in a
separate section The constraints which do not participate in any family are
group ed according to the phonological mo dule to which they b elong such
as autosegmental theory It is not my goal to give an extensive meta theory
of constraints in this app endix see McCarthy and Prince but I want
to giveanoverview of the typ es of constraints that seem to b e needed in
the study of a larger fragment of a grammar
For each constraint listed in this app endix I rst give the name I used in
the previous chapters plus the de nition of the constraint and a reference
to the rst time the constraintwas mentioned
A Proso dic well formedness
In this section I discuss those constraints on proso dic structure that I needed
in my analysis and that are not pro jection or weakness constraints I dis
cuss such constraints on syllable structure on fo ot structure and on word
structure each in their own section
A Syllable well formedness Apart from pro jection and weak
ness constraints w e turned out to need the following constraints on well
formed syllables
Syllable Contact Law SCL p In any sequence C C C ex
a b a
ceeds C in sonority
b
A Proso dic well formedness
Contour p If a segment in the onset is a following segmentmay
not b e
Contiguity p Given a string X Y if there is a syllable such
that are dominated by should b e dominated by
Co da h p h must b e in the onset
FinalDevoice p voiced except in the environment
sonor sonor
Laryngeal
These constraints are memb ers of di erent families The SCL is probably
one of a family of constraints on consonant clusters Similar constraints not
used in this thesis might b e one that demands a rising sonority in syllable
onsets and the government and binding constraints of Kaye Lowenstamm
and Vergnaud Rice and Harris whichhave
as an e ect for instance that clusters of consonants share a place feature
or otherwise that one of the two consonants is coronal It mightbethat
a theory of consonantal phonology in the syllable if worked out in the
appropriate way can also accomo date Contour within this family
The constraint Contiguity seems related to the Autosegmental No
Skipping constraint cf section A It might also b e p ossible to reduce
this constraint to an alignment constraint but this will not b e attempted
here Ab out FinalDevoice see Lombardi ab out Coda h Ihave
already suggested that it might b e the result of some interaction of weakness
constraints
Although this thesis deals mainly with asp ects of syllable structure
there are quite a lot of syllable structure phenomena ab out which I had
nothing to say For instance apart from the SCL I have not intro duced
any constraint dealing with the interaction of sonority place and voicing
features of consonants in clusters Clearly such constraints will b e needed
in a full edged theory of syllable structure that also wants to explain why
the consonants in an onset have a rising sonority why in clusters of two
consonants normally the consonants agree in place or else one of the twois
coronal etc Another asp ect of syllable structure ab out whichIhave had
very little to say are the asymmetries b etween onset and co da It is unclear
in this theory why some segments can o ccur in onset but not in co das As
I already p ointed out in chapter the pro jection theory should b e seen as
a mo dule in a broader theory of proso dic structure dealing also with these and other asp ects of syllable structure
A Proso dic well formedness
A Foot well formedness In addition to constraints on syllable
well formedness wehave also used three general constraints on the well
formedness of feet
Fo otBin p Fo ot no des are binary branching
Tro chee p Feet are left headed
Prokosch Law p The head syllable of a fo ot should b e heavy
Prokosch Law describ es an asp ect of metrical structure that is not other
wise studied in this thesis although it is recurrent in the literature the
fact that heads of proso dic structure tend to b e more complex and longer
than non heads I refer to the work of Dresher and Van der Hulst for a
more complete discussion
With resp ect to FootBin one could wonder whether this constraint
plays the role of a violable constraint or not All proso dic trees except
mayb e the word tree used in this thesis are binary branching at most also
at the syllable level The same seems to b e true under most recent analyses
for morphological and for syntactic structures suchasKayne and
Chomsky One might therefore supp ose that the binary branching
nature of X bar structure is a general prop erty of the linguistic calculus
the Gen function of Optimality theory rather than a violable constraint
The same might b e true for the constraint Trochee According to
Kayne and Chomsky structures of one pro jection are univer
sally left headed a while structures of two pro jections are left headed
at the lowest level and right headed one level up b
a X b X
X Y Y X
X Z
The syllable structures used in this thesis have exactly the same general
form According to the ndings of Hayes tro chaic systems
always consist of one simple pro jection only they either dominate two
mora s or two syllables Iambs on the other hand have a complex structure
b ecause their head needs to b e heavy in other words to dominate two
moras
In the analysis on Frenchwehave used the violabilityof FootBin as one of our
analytical to ols weallowed for the creation of degenerate feet
Alternatively one might try to derivethetrochee from a metrical principle requiring
all stressed syllables to b e followed by an unstressed syllable following Hung
A Proso dic well formedness
a Moraic trochee
F
b Syl labic trochee
F
A Proso dic well formedness
c Iamb
Ft
Mo dulo the lab els these structures corresp ond exactly to the general X bar
structures used for syllable structure here and prop osed for morphological
and syntactic structure in Kayne and Chomsky If this is
true there also might b e no violable tro chee constraint
A Word well formedness Wehave prop osed the following con
straints to deal with the syntax phonology and morphology phonology
interfaces and with proso dic word wellformedness
Align Word p Align PrWd X Y X
Align p Align X YP X
h
Align Foot p Align F R PrWd R
Final C p A Proso dic Word or phonological phrase has to end in
a consonant Align PrWd R C x
Su x p The elements of a su x should o ccur maximally close to
the right edge of a proso dic word Align Suffix R PrWd R
Monolog p The edges of a morphological domain should b e crisp
no feature should b e linked b oth to an edge segment of that domain
and to an element outside of the domain
MorPa p At least one element the most sonorous element of a
morpheme is incorp orated into a proso dic word
The rst ve constraints can b e captured within the formal framework
of Generalized Alignment as prop osed by McCarthy and Prince
According to these authors the basic scheme for Alignment constraints is
as follows
def
C E C C suchthatE of C and E Align C E
of C coincide
where C C MCat PCat
E E fLeft Right g
MCat the set of morphosyntactic categories fP A N AgrS DP
g
PCat the set of proso dic categories f Ft PrWd g
In the X bar theory used here moraic tro chees and iambs are built on top of the
segments dominated by N and syllabic tro chees on top of N no des See Hermans a
for a full development of this kind of theory
A Proso dic well formedness
Elsewhere Van Oostendorp I have argued that it only makes sense
to refer to left versus right edges in the case of proso dic categories such
as feet and syllables In the case of morphosyntactic categories it do es not
make sense to drawsuch a distinction Although this argument is in fact
immaterial to the discussion in this thesis I nevertheless have omitted the
reference to word edges in the case of Align Word
Although the constraint Monolog was also supp osed to b e a sp ecial
instance of Alignmentby McCarthy and Prince It o and Mester
haveshown that this is not in fact the case and that this constraint
which they call Crispness should instead b e allowed to play a separate
role in the grammar
A problem for the de nition of Alignmentgiven here is that it refers to
the following structure as well aligned
A B
a b c
This is true b ecause the right Edge of A can b e b and the left Edge of B
can also b e b In most cases however what wereallywanttohaveisthat
only the following structure counts as well aligned
A B
a b c d
In order to get this result McCarthy and Prince intro duce the notion
of an optional null string which can b e p osited b etween b and c and which
forms the formal Edge of A and B Yet this still do es not solve the problem
of since null strings are never obligatory It o and Mester s
notion of Crispness solves the problem in the sense that is assumed to
satisfy Alignment indeed but to violate Crispness I havegiven the name
Monolog to this constraint but actually it is known in graph theory as
the Single Mother Condition
The MorPa constraint nally expresses the intuition that all mor
phemes should participate at least partly in the core proso dic make up of
aword Wehave seen that this constraintmight b e parametrized for dif
ferent su xes in the sense that the incorp oration of some su xes seems to
b e more urgent than that of others
A Autosegmental Representations
A Autosegmental Representations
Autosegmental phonology is not a topic which is discussed in any detail
in this thesis Still we needed to use several constraints on Autosegmental
representations These I have listed b elow
No Skipping p a feature that is asso ciated to the segments and
in the segment string should also b e asso ciated to
C F p A cons ro ot do es not dominate a vo calic feature
V
V F p A cons ro ot do es not dominate a consonantal fea
C
ture
C F p A place feature dominated by a cons ro ot no de should
C
b e directly dominated by a C place no de
NoLongVowel p Vo calic no des maynotbelinked to more than
one root node
Harmony p Avowel V should have the same sp eci cation for
RTR as the vowel W that o ccurs on the left of V if V and W are
in the same proso dic word
We nd one small family of constraints following the general scheme in
a An cons ro ot dominates an consonantal place feature
b An consonan tal place feature is dominated byan cons ro ot
C F and V F are instantiations of b after transp osition and
V C
under the assumption that cons are mutually exclusive feature values
ConsPlace is an instantiation of a the only other p ossible constraint
would have the following form
V F Avo calic ro ot dominates a vo calic place no de
V
Because all ro ots are either vo calic or consonantal I assume the ma jor
class feature cons cannot b e undersp eci ed following Clements
the constraints V F and C F together imply EmptyRoot A
V C
vo calic ro ot needs to dominate a vo calic place feature and a consonantal
ro ot a consonantal place no de Empty ro ots are therefore dispreferred we
do not need to rule out emptyrootsby an indep endent constraint
Apart from these constraints relating ro ots with place features wehave
used two other autosegmental constraints the OCP and the No Skipping
constraint Both constraints haveplayed an imp ortant role in autosegmen
tal theory at least since Goldsmith The No Skipping constraint
furthermore seems to b e formally related to the syllable well formedness
constraint Contiguity
A Feature Co o ccurrence and Licensing
The constraint NoLongVowel althoughusedinchapter do es not
seem to b e strictly necessary in the analysis of Dutch Standard Dutch
do es have long vowels but these are all low We already have a constraint
which has this same e ect viz Project N low a memb er of the family
of pro jection constraints If this constraintisranked higher than the con
straint against insertion of the feature low it app ears that we do not need
to p osit a separate constraint against long vowels Consider for instance
the tableau for an input q
q
Candidates Project N low low
q
p
q
Although this is suggested in the main text of chapter it is not true for
all Dutch idiolects that the long rounded vowels are also lowered This
might b e due to a slightly di erent ranking of the constraint against the
combination low labial in the twovariants of Dutch
q DutchA
labial
Candidates Project N low low
low
q
p
q
q DutchB
labial
low Candidates Project N low
low
p
q
q
This argument do es not imply that cross linguisti cal ly wewillnever need
a constraint against long vowels cf Rosenthall but the evidence for
such a constraint in the phonology of Dutch is scarce if presentatall
A Feature Co o ccurrence and Licensing
In order to derivethevowel set of Dutchwe had to p osit the following set
of constraints all place features refer to features under the vocalic no de
A Parsing
dorsal high high low
dorsal low high coronal
coronal lax low labial
high labial labial dorsal lax dorsal lax low low lax
V place voice V place continuant
License
All of these constraints are grounded in the sense of Archangeli and Pul
leyblank they are p osited as primitives in phonological theory and
should b e derived from phonetics although of course it might b e p ossible
that in a more fully worked out phonological theory for instance one
that treats the internal structure of features along the lines of Van der
Hulst c they can b e seen as phonological theorems instead
The status of the constraint against the co o ccurrence of dorsal and
coronal is dep endent on one s assumptions regarding the status of dorsal
in the phonology of vowels It is sometimes claimed that there are reasons
to assume that dorsal is always presentonvowels If this is true this
constraintwould play no role at all
In chapter I referred to the theory of Licensing as put forward by It o
Mester and Padgett I refer to that pap er for a fuller explication of
this theory
A Parsing
I assume there are twotyp es of faithfulness constraints constraints which
require input prop erties to b e visibly present in the output Parse F
and constraints which require output prop erties to b e arising from input
prop erties The latter will b e discussed in the next section the former will
b e discussed here
The following parse constraints have b een used in this thesis
Parse Consonant A consonantmust b e parsed into syllable structure
Parse Vowel Avowel must b e parsed into syllable structure
Parse Ro ot A ro ot must b e parsed into syllable structure
Parse High Parse Tense Parse Labial
Parse Low Parse Coronal Parse Dorsal A feature has to b e parsed
into a segment
Faithful Underlying syllable structure may not b e a ected
Most of these constraints are plain straightforward parsing constraints I
have assigned the related Fill constraints of Prince and Smolensky
to the family of feature co o ccurence restricti ons high coronal etc to
the set of constraints on autosegmental structure Emptyroot and to
the family of pro jection constraints
A Constraints against unnecessary structure
Finally there is a constraint Faithful which demands that all input
syllabi cation is resp ected Assuming that underlying forms do not have
any syllable structure as wehave done throughout this thesis this means
that this particular constraint Faithful can only b e seen to workinthe
p ostlexical phonology
Ihave shown that it is di cult to distinguish b etween Parse V and
the pro jection constraint Project cons N if vo calic ro ots are parsed
they are usually parsed in the head of a syllable and if they are parsed
outside of such a p osition this could b e attributed to some Parse feature
constraint The test here would b e gliding of schwa i e wewould need to
nd a language where schwa needs to b e parsed and therefore it surfaces as a
glide for instance h This would provide us with evidence for Parse V
indep endentofProject cons N
A Constraints against unnecessary structure
One of the rst constraint families prop osed by Prince and Smolensky
was the family of constraints that militate against structure struc
The only memb ers of this family that have b een used here are the con
straints against features and against proso dic no des
low lax velar coronal labial cons cons
These constraints serve to blo ck the unnecessary insertion of features in
this thesis The formulation given here blo cking all o ccurrences of a given
feature in the output is very much dep endent on the assumption that
linguistic derivations satisfy the principle of Containment monotonicity
all elements of the input are present in all candidate outputs This means
that for instance an underlying low feature will always have a violation of
the constraint against low in all candidates This violation will therefore
b e irrelevant The same is not true however for a feature low that is
inserted in the course of the derivation b ecause there will always b e an
alternative derivation in which this feature is not inserted
McCarthy and Prince have prop osed to abandon the principle of
Containment and to allow constraints that explicitly compare inputs and
outputs instead If such a theory turns out to b e more fruitful than the one
based on Containment the formulation of these constraints will havetobe
adapted in suchaway that they ban inserted instances of a given feature
features that are presentinagiven output candidate but not in the input
A Ad Ho c Constraint
A Ad Ho c Constraint
In the course of the analyses we had to have recourse to one constraint
that could not b e motivated in any principled way In the ideal theory of
phonology this constraint will b e replaced by more principled explanations
R Colour p CV r inside the fo ot
r Colour says that r cannot o ccur after a short vowel inside a fo ot pre
sumably this should b e related to the high sonorityofr in a certain way
but it is not clear to me how this can b e accomplished
B Arguments for ranking
In this app endix I list all the arguments for the constraint ranking in Stan
dard Dutch that have b een collected in this thesis I have limited myself
to this language b ecause it is the most extensively discussed grammar I
assume that it is relatively easy for the interested reader to reconstruct
the arguments for French Norwegian Tilburg Dutchorany of the other
languages discussed here
Ihave limited myself to strictly lo cal ranking arguments That is to say
that if I have found evidence that i A B ii B C and iii A C I
have only listed the arguments for the rankings i and ii Evidence that
A C is often discussed in the main text of this thesis
In App endix C I give the top ologies of the fragments of Standard Dutch
whichhave b een discussed here as well as those for Rotterdam Dutch and
French The items in this app endix have b een listed in alphab etical order
For every ranking I mention a the ranking itself and b an example
comparison of two candidates for some input which is crucially established
by the ranking In this comparison I used the symbol A B meaning is
less harmonic than
In most cases I have only listed the ranking arguments that are abso
lutely clear from the main text This means the ranking argued for here is
only partial The top ologies argued for here will havetobeemb edded in the
complete constraint ranking in the eventual complete Dutch phonological
grammar Where there was no reason to distinguish b etween Project X
Y and Project Y X I hav etaken them together under the name Con
nect X Y
B Top ology of the Dutch lexicon
Align Onset
ro dox rot ox ro d o
Co da h Parse lax
m h ma h m h
Pro ject Ft V Align
ka t r k kat r katrk
B Top ology of the Dutch lexicon
Connect Ft V Su x
vad r v d r vad r umlaut
Connect N lax Align
m ma m
Connect N lax Harmony
mb mba mba
Connect N lax Parse lax
m ma m
Contour Align
d m d m d m
Contiguity Pro ject N V
r m p r mp r mp
dorsal
Parse dorsal Parse coronal
coronal
lax dorsal labial dorsal high coronal low dorsal
Dorsal coronal vowels do not exist
FinalDevoice Align
h b h p h b
Fo otBin Align
ka t r k kat r katrk
Geminate Parse C
tt t t
high
low
lax
high high
lax lax
i
coronal coronal
high
Parse high or Parse low
lo w
high high
or
low low
high high high
low low low
high coronal coronal
high i high
high labial labial
High vowels tend to surface as round
labial
high labial
coronal
y i i
labial dorsal dorsal
Rounded vowels are normally dorsal
lax MorPa
sl p t sla p t slap t
lax Pro ject N V
m n ma n man
lax dorsal dorsal
Lax vowels are normally dorsal
low lax low
B Top ology of the Dutch p ostlexical phonology
low r Colour
No lexical colouring of vowels b efore r
low
Parse loworParse labial
labial
Low rounded vowels do not surface
low dorsal dorsal
Lowvowels are normally dorsal
low lax lax
lowvowels are always lax
N Parse V Onset Pro ject cons
e li t qr e li t qr
Onset Pro ject N V t t
Parse C MorPa
w r k t w r k t w rk t
labial
Parse coronal Parse labial
coronal
i u y y
Parse F F where F flax low high coronal dorsal labial g
v v v
All these features normally surface
Parse high high labial labial
high high
high
labial labial
labial
Parse labial Parse lax
lax
o
low Parse lax Parse high
high high
lax i lax
coronal coronal
high high
lax lax
i
coronal coronal
Pro ject N low low
high
i i lax
coronal
low
DutchA Pro ject N low
labial
low
Pro ject N low DutchB
l abial
q q DutchA
q q DutchB
Pro ject N V Parse C
ma t ka ma t ka matka
tr l t l tl
B Top ology of the Dutch p ostlexical phonology
Align Pro ject V Ft
B Top ology of the Dutch p ostlexical phonology
t f t fe t fe
N lax lax Connect
eq r q r e r
Contour Pro ject V Ft
l t elit elit
h las helas
high
Connect N lax
lax
i q r iq r i r
lax dorsal Connect N lax
k r lax k k r k k rk
low Pro ject N low
q r q r e r
N V Onset Pro ject
ai aj a i
Parse F F where F flax low high coronal dorsal labial g
v v v
All these features normally surface
Parse high Parse low Parse lax
Parse labial Parse dorsal Parse coronal
Reduction hierarchy
Parse high Pro ject V Ft
High vowels do not reduce
Pro ject cons N Contour Onset
m m m
N cons cons formal style register Pro ject
h l p h l p h lp
Pro ject N cons SCL
he b j h p j h b j
N V Connect N lax Pro ject
tr t r tr
Pro ject V Ft Pro ject V Ft
Weak syllables reduce more easily than unfo oted syllables
Pro ject V Ft Parse coronal
Unstressed e is almost always reduced
r Colour Connect N lax
i r iq r i r
Weak cons N cons Pro ject N cons informal style
OK
tejatr teatr vs h lp h lp
C Ranking schemes
In this app endix I give four graphs representing the four most extensively
discussed grammar fragments of this thesis those of the Standard Dutch
lexical and p ostlexical phonology ofFrench and of Rotterdam Dutch
Ihave not included all undominated constraints or very trivial rankings
The intention of this app endix is merely to lay out the more complicated
transitive ranking arguments
C Top ology of the Standard Dutch lexicon
C Top ology of the Standard Dutch lexicon
C Top ology of the Standard Dutch p ostlexical phonology
C Top ology of the Standard Dutch p ostlexical phonol
ogy
C Top ology of the French phonology
C Top ology of the French phonology
C Top ology of the Rotterdam Dutch phonology
C Top ology of the Rotterdam Dutch phonology
References
Akinlabi A Featural Alignment Manuscript Rutgers University
Anderson J and C Ewen Principles of Dependency Phonology Cam
bridge University Press Cambridge UK
Anderson J and C Jones Three Theses Concerning Phonological Rep
resentations Journal of Linguistics
Anderson S The Analysis of FrenchSchwa or How to Get Something
from Nothing Language
Anderson S A Morphous Morphology Cambridge University Press
Cambridge UK
Angoujard J P and M Denais Le Pluriel Bris e en Tigrigna Langues
orientales anciennes Philologie et Linguistique Not consulted
cited in Buckley
Aoun J Is the Syllable or the Sup ersyllabl e a Constituent MIT Work
ing Papers in Linguistics
Archangeli D Underspeci cation in Yawelmany Phonology and Mor
phology PhD Dissertation MIT Published by Garland Press Outstanding
Dissertations in Linguistics Series
Archangeli D and D Pulleyblan k GroundedPhonology MIT Press
Cambridge Mass
Arono M WordFormation in Generative Grammar MIT Press Cam
bridge Mass
Arono M Morphology by itself Stems and In ectional Classes MIT
Press Cam bridge Mass Linguistic Inquiry Monograph
Bao Z On the NatureofTone PhD Dissertation MIT
Bagemihl B Syllable Structure in Bella Co ola Linguistic Inquiry
Baker M Incorporation A Theory of Grammatical Function Changing
University of Chicago Press Chicago
Bakker D and G Dibb ets eds GeschiedenisvandeNederlandse
Taalkunde Malmb erg Den Bosch
References
Basb ll H Metrical Theory and the FrenchFo ot in W Dressler
and O Pfei er eds Phonologica Akten der Vierten Internationalen
Phonologietagung pp
Basb ll H Sur l Identit e Phonologiq ue du SchwaFran cais et son R ole
dans l Accentuation et dans la Syllabation in Verluyten
Basb ll H DistinctiveFeatures Syllable Structure and Vowel Space
in P M Bertinetto M Kenstowicz and M Loparco eds Certamen Phono
logicum II Papers from the Cortona Phonology Meeting Rosenberg
and Sellier Torino pp
Bat El O Onset Violation in Tib erian Hebrew in J Merchant J Pad
gett R Walker eds Phonology in Santa Cruz Distributed by Linguistic
ResearchCenter UCSC pp
Beard R Morpheme Order in a Lexeme Morpheme Based Morphology
Lingua
Bennink C in prep Harmony and Disharmony in Hungarian and Turkish
PhD Dissertation Tilburg University
Berendsen E Final Devoicing Assimilation and Sub ject Clitics in
Dutch in H Bennis and W U S van Lessen Klo eke eds Linguistics in
the Netherlands Foris Dordrecht pp
Berendsen E The Phonology of Cliticization PhD Dissertation Univer
sityofUtrecht Also app eared with Foris Dordrech t
Berendsen E and W Zonneveld Nederlandse Schwa invo eging op z n
Deens Spektator
Berg B van den Foniek van het Nederlands Van Go or Zonen Den Haag
Berg B van den Een Fonologische Reductieregel in de Randstad De
Nieuwe Taalgids Also in H Hulshof red Transformationeel
generatieve grammatica in artikelen Tjeenk Willink Groningen
Berhane G Issues in the Phonology and Morphology of Tigrinya PhD
Dissertation Universit eduQu eb ec a Montr eal Not consulted cited in
Buckley
Berwick R The Acquisition of Syntactic Know ledge MIT Press Cam
bridge Mass
Bethin C Polish Syl lables Slavica Columbus Ohio
Bo oij G Klinkerrreductie in het Nederlands Leuvense Bijdragen
Bo oij G Dutch Morphology A Study of WordFormation in Generative
Grammar Foris Dordrecht
Bo oij G Generatieve Fonologie van het Nederlands Sp ectrum Utrecht
Antwerp en
References
Bo oij G Fonologische en Fonetische Asp ecten van Klinkerreducti e
Spektator
Bo oij G Lexical Phonology Final Devoicing and Sub ject Pronouns in
Dutch in H Bennis and F Beukema eds Linguistics in the Netherlands
Foris Dordrecht pp
Bo oij G On the Representation of Diphthongs in Frisian Journal of
Linguistics
Bo oij G a The Phonology of Dutch Oxford University Press Oxford
Bo oij G b Cliticization as Proso dic Integration the Case of Dutch
Manuscript Vrije Universiteit Amsterdam Toappearin The Linguistic Re
view
Bo oij G and R Lieb er On the Simultaneity of Morphological and
Proso dic Structure in S Hargus and E Kaisse eds Studies in Lexical
Phonology Academic Press San Diego pp Phonetics and Phonol
ogy
Bo oij G and J Rubach Morphological and Proso dic Domains in Lexical
Phonology Phonology
Bo oij G and J Rubach Postcyclic versus Postlexical Rules in Lexical
Phonology Linguistic Inquiry
Borowsky T Topics in the Lexical Phonology of English PhD Disserta
tion University of Massachussetts
Borowski T On the Word Level in S Hargus and E Kaisse eds
pp
Borowsky T J It o A Mester The Formal Representation of Ambisyl
labicity Evidence from Danish NELS
Boutkan D Klinkerwisseli ng in het Tilburgs Taal en Tongval
Boutkan D and M Kossmann Het Stadsdialekt van Tilburg Klank en
Vormleer Manuscript Leiden University
Brink D Problems in Phonology A Generative Phonology of Dutch PhD
Dissertation University of Wisconsin
Browman C and L Goldstein Targetless schwa an Articulatory
Analysis in G Do cherty and R Ladd eds Papers in Laboratory Phonol
ogy II Gesture Segment Prosody Cambridge University Press Cam
bridge UK pp
Buckley E On the Status of Vowel Length in Tigrinya Manuscript
UniversityofPennsylvani a
Burzio L Principles of English Stress Cambridge University Press Cam
bridge UK
Calabrese A Metaphony in Salentino Rivista di GrammaticaGenera
tiva
References
Calabrese A Towards a Theory of Phonological Alphabets PhD Disser
tation MIT
Calabrese A Sievers Law in Gothic A Synchronic Analysis with some
Notes on its Diachronic Development The Linguistic Review
Carrier J The Interaction of Morphological and PhonologicalRules in
Tagalog A Study in the Relationship between Rule Components in Gram
mar PhD Dissertation MIT
Charette M Conditions on phonological government Cambridge Uni
versity Press Cambridge UK
Chierchia G Length Syllabi cati on and the Phonological Cycle in Ital
ian Journal of Italian Linguistics
Chomsky N Remarks on Nominalizati on in R Jacobs and P Rosen
baum eds Readings on Transformational Grammar Ginn Waltham
Mass
Chomsky N A Minimalist Program for Linguistic Theory in K Hale
and S J Keyser eds The View from Building MIT Press Cambridge
Mass
Chomsky N Bare Phrase Structure Manuscript MIT Cambridge
Mass
Chomsky N The Minimalist Program MIT Press Cambridge Mass
Chomsky N and M Halle The Sound Pattern of English MIT Press
Cambridge Mass
Chung S Transderivational Relationshi ps in Chamorro Language
Cinque G A Null Theory of Phrase and Comp ound Stress Linguistic
Inquiry
Clements N Principles of Tone Asso ciation in Kikuyu in N Clements
and J Goldsmith eds Autosegmental Studies in Bantu Tone Foris Dor
drecht pp
Clements N The Geometry of Phonological Features Phonology Year
book
Clements N Towards a Substantive Theory of Feature Sp eci cation
NELS GLSA Amherst Mass
Clements N On the Representation of Vowel Height Manuscript Cor
nell University Ithaca New York
Clements N The Role of the Sonority Cycle in Core Syllabi ca tion
in J Kingston and M Beckman eds Papers in Laboratory Phonology
Between the Grammar and the Physics of Speech Cambridge University
Press Cambridge UK pp
References
Clements N a Vowel Height Assimilatio n in Bantu Languages Working
papers of the Cornel l Phonetics Laboratory Cornell University
Ithaca New York
Clements N b Place of Articulation in Consonants and Vowels A Uni
ed Approach Working Papers of the Cornel l Phonetics Laboratory
Cornell University Ithaca New York
Clements N The Sonority Cycle and Syllable Organization in W
Dressler H Lusch utzky O Pfei er and J Rennison eds Phonologica
Proceedings of the th International Phonology Meeting Cambridge
University Press Cambridge UK
Clements N and E Hume The Internal Organization of Sp eech Sounds
in J Goldsmith ed The Handbook of PhonologicalTheory Basil Blackwell
London
Clements N and J Keyser CV Phonology A Generative Theoryofthe
Syl lable The MIT Press Cambridge Mass Linguistic Inquiry Monographs
Clements N and E Sezer Vowel and Consonan t DisharmonyinTurkish
in H van der Hulst and N Smith eds The Structure of Phonological
Representations Volume I I Foris Dordrecht pp
Cohen A Het Nederlandse Diminutiefsu x een Morfonologische
Pro eve Nieuwe Taalgids
Cohen A C Eb eling K Fokkema A van Holk Fonologie van het
Nederlands en het Fries Inleiding in de Moderne Klankleer Nijho Den
Haag
Cohn A Stress in Indonesian and Bracketing Paradoxes Natural Lan
guage and Linguistic Theory
Cohn A and J McCarthy Fo ot Alignment and Apparent Cyclicityin
Indonesian Manuscript Cornell Universty and University of Massachus
setts
Davis S Topics in Syl lable Geometry PhD Dissertation Universityof
Arizona Tucson
Davis S Syllable Onsets as a Factor in Stress Rules Phonology
D echaine R M Stress in Qu eb ecois Evidence from High Vowels in L
Dobrin L Nichols and R Ro drigues eds Papers from the th Regional
Me eting of the Chicago Linguistic Society CLS Chicago pp
Dell F Les R egles et les Sons Herrmann Paris First edition
second edition
Dell F Consonant Clusters and Phonologica l Syllables in French Lin
gua
Dell F and M Elmedlaoui Syllabic Consonants and Syllabi cati on in
Imdlawn Tashliyt Berb er JALL
References
Di Sciullo A M and E Williams On the De nition of Word MIT Press
Cambridge Mass
y
Dixon R Some Phonological Rules of Yidin Linguistic Inquiry
Dogil G La Phonologie p eut elle Renoncer aux Traits Distinctifs de
Classe Sup erieur in B Laks and A Rialland eds Architecturedes
R epresentations Phonologiques Editions du CNRS Paris
Donegan J J Miller On the Natural Phonology of Vowels PhD Dis
sertation Ohio State University Published by Garland Press
Dresher E en A Lahiri The Germanic Fo ot Metrical Coherence in Old
English Linguistic Inquiry
Dresher E and H van der Hulst Head Dep endent Asymmetries in
Phonology in H van der Hulst and J van de Weijer eds Leiden in
Last HIL Phonology Papers I Holland Academic Graphics The Hague
pp
Dressler W Allegroregeln Rechtfertigen Lentoregeln Sekund are Phoneme
des Bretonischen Innsbrucker Beitr age zur Sprachwissenschaft Inns
bruck
Dressler W Metho disches zu Allegro Regeln in W Dressler and F
Mares eds Phonologica pp Fink Verlag Munich
Duanmu S AFormal Study of Syl lable Tone Stress and Domain in
Chinese Languages PhD Dissertation MIT
Dumas D Qu eb ec French High Vowel Harmony Proccedings of the
Chicago Linguistics Society
Dunlap E Tensing in Lexical Phonology in E Dunlap and J Pad
gett eds UMOP Papers in Phonology GLSA Amherst Mass pp
Dunlap E Issues in the Moraic StructureofSpanish PhD Dissertation
University of Massachussetts
Durand J Generative Phonology Dep endency Phonology and Southern
French Langue et Stile
Durand J French Liaison Floating Segments and Other Matters in a
Dep endency Framework in J Durand ed Dependency and Non linear
Phonology London Cro om Helm pp
Durand J Generative and Nonlinear Phonology Longman London
Durand J C Slater and H Wise Observations on Schwa in Southern
French in C Slater J Durand and X Bate eds French Sound Patterns
Changing Perspectives Asso ciation for French Language Studies no
University of Essex Not consulted cited in Durand and Charette
Edgerton F Sievers s La w and IE Weak grade Vo calism Language
References
Encrev e P La Liaison Avec et Sans Encha nement Phonologie Tridi
mensionel le et Usages du Fran cais Editions du Seuil Paris
Everett D and K Everett On the Relevance of Syllable Onsets to Stress
Placement Linguistic Inquiry
Evers A and R Huybregts The Cycle in Dutch Its Relevance to the
Theory of Grammar in A Kraak ed Linguistics in the Netherlands
Van Gorcum Assen pp
Ewen C The Phonology of the DiminutiveinDutch A Dep endency
Account Lingua
Ferguson C Short a in Philadelphi a English in M Smith ed Studies
in Linguistics in Honour of G L Trager Mouton The Hague pp
Not consulted cited in Harris
Fidelholtz J Word Frequency and Vowel Reduction in English Pro
ceedings of the Chicago Linguistics Society
Fouch e P Trait edePrononciation Fran caise e Edition Klinksieck
Paris
Fudge E and D Gilb ers On Underlying Vowels Manuscript Groningen
Univ ersity
Giegerich H Metrical Phonology and Phonological Structure Cambridge
University Press Cambridge UK
Giegerich H English Phonology An Introduction Cambridge University
Press Cambridge UK
Gilb ers D Phonological Networks A TheoryofSegment Representation
PhD Dissertation University of Groningen
Ginneken J van De Phonologie van het Algemeen Nederlands Onze
Taaltuin
Goldsmith J Autosegmental and Metrical Phonology Basil Blackwell
Oxford UK and Cambridge Mass
Goldsmith J ed The Last Phonological Rule Re ections in Constraints
and Derivations University of Chicago Press Chicago and London Studies
in Contemporary Linguistics
Goldsmith J ed The Handbook of PhonologicalTheory Blackwell
Oxford and Cambridge
Goldsmith J K Peterson and J Progo Tone and Accent in the Xhosa
oaches to Verbal System in P Newman and R Botner eds Current Appr
African Lingusitics Foris Dordrecht pp
Grimshaw J Extended Functional Pro jections Pap er presented at
Utrecht Lexicon Workshop
Gro ot A W de De Wetten der Phonologie en hun Betekenis voor de
Studie van het Nederlands Nieuwe Taalgids
References
Gussenhoven C Het Nederlandse Diminutief su x Schwa insertie An
ders Bekeken De Nieuwe Taalgids
Gussenhoven C Cliticization in Dutch as Phonological Word Forma
tion Journal of Linguistics
Gussmann E Resyllabi cati on and Delinking Linguistic Inquiry
Haas W de Partial Syllabi ca tion and SchwaEpenthesis in Dutch
Gramma
Haas W de and M Trommelen Morfologisch Handboek van het Ned
erlands Een Studie over Verschil lende Vormen van Woordformatie SDU
Den Haag Aan het Woord
Haeringen C van Iets over het Vo orvo egsel in het Verleden Deelwoord
Driemaandelijkse Bladen
Haeringen C van Herverfransing Mededelingen der Koninklijke
Akademie van Wetenschappen Nieuwe reeks KNAW Amster
dam Also app eared in Gramarie
Haeringen C van De Plaats van ie o e en uu in het Nederlandse Klink
erstelsel in Gramarie Van Gorkum Assen pp
Hall T German Syllabi catio n the Velar Nasal and the Representation
of Schwa Linguistics
Hall T Syl lable Structure and Syl lable relatedProcesses in German
Niemeyer Tubingen
Halle M On DistinctiveFeatures and their Articulatory Implementa
tion Natural Language and Linguistic Theory
Halle M Resp ecting Metrical Structure Natural Language and Lin
guistic Theory
Halle M Rules or Constraints The Case of Eastern Massachussetts
r Manuscript MIT
Halle M and N Clements AProblem Book in Phonology MIT Press Cam
bridge Mass
Halle M and M Kenstowicz The Free Element Condition and Cyclic
versus Noncyclic Stress Linguistic Inquiry
Halle M and K P Mohanan Segmental Phonology of Mo dern English
Linguistic Inquiry
Halle M and K Stevens On the Feature Advanced Tongue Ro ot
Quarterly Progress Report Research Lab oratory of Electronics
Cambridge Mass
Halle M and K Stevens A Note on Laryngeal Features Quarterly
Progress Reports Research Lab of Electronics MIT pp
References
Halle M and J R Vergnaud An Essay on Stress MIT Press Cambridge
Mass
Hargus S and E Kaisse eds Studies in Lexical Phonology Academic
Press New York etc Phonetics and Phonology series editors S An
derson and P Keating
Harris J James Spanish Phonology MIT Press Cambridge Mass
Harris J James Integrity of Proso dic Constituents and the Domain of
Syllabi cati on in Spanish and Catalan in K Hale and S J Keyser eds
The View from Building MIT Press Cambridge Mass pp
Harris J John Towards a Lexical Analysis of a Sound Change in
Progress Journal of Linguistics
Harris J John Segmental Complexity and Phonological Government
Phonology
Harris J John English Sound Structure Blackwell Oxford
Hart J t Fonetische Steunpunten De Nieuwe Taalgids
Hayes B a The Proso dic HierarchyinMeter inP Kiparsky and G
Youmans eds Rhythm and Meter Academic Press Orlando
Hayes B b Comp ensatory Lengthening in Moraic Phonology Linguistic
Inquiry
Hayes B Diphthongizatio n and Coindexing Phonology
Hayes B Metrical Stress Theory Principles and Case Studies Manu
script UCLA
Haugen E Norwegian English Dictionary a Pronouncing and
Translation Dictionary of Modern Norwegian Bokmaal and Nynorsk
Bergen etc Universitetsforlaget First Edition Consulted d Edi
tion
Hawkins W Patterns of Vowel loss in Macushi Carib International
Journal of American Linguistics
Hawkins J and A Cutler Psycholinguis tic Factors in Morphological
Asymmetry in J Hawkins ed Explaining Language Universals Basil
Blackwell London
Heeroma K De Plaats van Ie Uu Oe in het Nederlandse Klinkersys
teem Nieuwe Taalgids
Helten W van Middelnederlandsche Spraakkunst Groningen
Helten W van Die Altniederfr ankischen Psalmenfragmente die Lip
sius schen Glossen und die Alts udmittelfr ankischen Psalmenfragmente
Groningen New York
Hermans B On the Representation of Quasi long Vowels in Dutch and
Limburgian in R Bok Bennema en R van Hout eds Linguistics in the
Netherlands John Benjamins Amsterdam
References
Hermans B a DutchSchwa Manuscript Tilburg University
Hermans B b Het Onderscheid tussen Zware en Lichte Lettergrep en in
het Nederlands Spektator
Hermans B a The Composite NatureofAccent With Case Studies of
the Limburgian and Serbo Croatian Pitch Accent PhD Dissertation Free
University Amsterdam
Hermans B b Palatalisatie in een Limburgs Dialect Dissimilati e als
Gevolg van de OCP in G Bo oij and J van Marle eds Dialectfonologie
P J Meertensinstituut Amsterdam
Hewitt M Vertical Maximization and Metrical Theory PhD Dissertation
Brandeis University
Ho eksema J Wortels Stammen en de Schwa TABU
Ho op er J J Byb ee An Introduction to Natural Generative Phonology
Academic Press New York
Hout R van De Structuur van Taalvariatie Een Sociolingu stisch On
derzoek naar het Stadssialect van Nijmegen PhD dissertation Tilburg Uni
versity
Hu man M and R Krakow eds Nasals Nasalization and the Velum
Academic Press San Diego etc pp Phonetics and Phonology
Series editors S Anderson and P Keating
Hulshof H ed Transformationeel generatieveGrammaticainArtikelen
Tjeenk Willink Groningen
Hulst H van der Syl lable Structure and Stress in Dutch Foris Dordrecht
Hulst H van der Ambisyllabici ty in Dutch in H Bennis and F
Beukema eds Linguistics in the Netherlands Foris Dordrecht pp
Hulst H van der a Radical CV Phonology Manuscript Leiden Univer
sity
Hulst H van der b Core Syllables in Dutch handout of talk presented
in Tilburg
Hulst H van der c An Intro duction to Radical CV Phonology in
S Shore and M Vilkuna eds Suomen Kielitieteel lisen YhdistykenVu
oskirja Yearb o ok of the Linguistic Asso ciation of Finland SKY
Helsinki
Hulst H van der and J van Lit Lettergreepstructuur GLOT
Hulst H van der and N Smith On Neutral Vowels in K Bogers
H van der Hulst and M Mous eds The Phonological Representationof
ht pp Suprasegmentals Foris Dordrec
References
Hulst H van der and N Smith The Varieties of Pitch Accent Systems
in H van der Hulst and N Smith eds Autosegmental Studies on Pitch
Accent Foris Dordrecht pp
Hulst H van der and J van de Weijer Vowel Harmony in J Goldsmith
ed pp
Hume E Front Vowels Coronal Consonants and their Interaction in
Nonlinear Phonology PhD Dissertation Cornell University
Hume E and D Odden Against consonantal Talk presented at IWPS
Durham
Hung H The Rhythmic and Prosodic Organisation of Edge Constituents
PhD Dissertation Brandeis University
Hyman L A Theory of PhonologicalWeight Foris Dordrecht
Inkelas S and D Zec eds The Phonology Syntax Connection University
of Chicago Press Chicago
It o J Syl lable Theory in Prosodic Phonology PhD dissertation Uni
versity of Massachussetts at Amherst App eared in Garland Outstanding
Dissertations in Linguistics Series
It o J and A Mester The Phonology of Voicing in Japanese Linguistic
Inquiry
It o J and A Mester Alignment and Crisp Edges Talk presented in
OTS Workshop on Proso dic Morphology
It o J A Mester and J Padgett NC Licensing and Undersp eci catio n
in Optimality Theory Manuscript University of California Santa Cruz
Jacobs H Non linear Studies in the Historical Phonology of French PhD
Dissertation Katholieke Universiteit Nijmegen
Jakobson R Observations sur le Classement Phonologiqu e des Con
sonnes in N Ruwet ed Essais de LinguistiqueG en erale vol Editions
de Minuit Paris
Jakobson R Kindersprache Aphasie und Allgemeine Lautgesetze in
Selected writings volume I Mouton The Hague pp
Johnsen L J KayeandJ R Vergnaud Raddoppiamento Sintattico
Norwegian Style Handout AVH Seminar Trondheim
Jong D de The Sociolinguistic Aspects of French Liaison PhD Disser
tation Free University of Amsterdam
Jong D de and M Hietbrink The Morphonology of the French Pre x
RE in R Bok Bennema and C Cremers eds Linguistics in the Nether
lands Amsterdam John Benjamins pp
Kaisse E ConnectedSpeech Academic Press Orlando
Kager R A Metrical Theory of Stress and Destressing in English and
Dutch Foris Dordrecht
References
Kager R DutchSchwa in Moraic Phonology in M Ziolkowski M
Noske and K Deaton eds Papers from the th Regional Meeting of the
Chicago Linguistic Society Chicago CLS Volume I I The parasesssion on
the syl lable in phonetics and phonology pp
Kager R a German Quantity Class notes Universit at Salzburg Oc
tob er
Kager R b Generalized Alignment and Morphological Parsing Talk
given at the OTS Robustness Conference Utrecht Octob er
Kager R Consequences of Catalexis in H van der Hulst and J van de
Weijer eds Leiden in Last HIL Phonology Papers I Holland Academic
Graphics The Hague pp
Kager R E Visch and W Zonneveld Nederlandse Wo ordklemto on
Ho ofdklemto on Bijklemto on Reductie Voeten GLOT
Kager R and W Zonneveld Schwa Syllables and Extrametricalityin
Dutch The Linguistic Review
Kaisse E ConnectedSpeech Orlando Press New York
Katz D AGrammar of the Yiddish Language Duckworth London
Kaye J Lexical Phonology and Phonological Phenomena In W
Dressler H Lusch utzky O Pfei er and J Rennison eds Phonologica
Cambridge University Press Cambridge pp
Kaye J and J Lowenstamm De la Syllabici t e in F Dell D Hirst J R
Vergnaud eds Forme SonoreduLangage Hermann Paris pp
Kaye J J Lowenstamm and J R Vergnaud The Internal Structure of
Phonological Elements A Theory of Charm and Government Phonology
Yearbook
Kaye J J Lowenstamm and J R Vergnaud Constituent Structure and
Government in Phonology Phonology
Kayne R The Antisymmetry of Syntax Manuscript CUNY
Keating P Undersp eci cation in Phonetics Phonology
Kenstowicz M a Phonology in Generative Grammar Basil Blackwell
London
Kenstowicz M b Sonority Driven Stress Manuscript MIT
Kenstowicz M and Ch Kisseb erth Generative Phonology Description
and Theory Academic Press New York
Keymeulen L and J Taeldeman Tussen Fonologie en Morfologie de
TaalenTongval Vokaalverkorting in een Brabants Dialekt
Kiparsky P Remarks on the Metrical Structure of the Syllable in W
Dressler O Pfei er and J Rennison eds Phonologica Akten der
Vierten Internationalen Phonologietagung pp
References
Kiparsky P From Cyclic Phonology to Lexical Phonology in H van der
Hulst and N Smith eds The Structure of PhonologicalRepresentations
Part I Foris Dordrecht pp
Kiparsky P Some Consequences of Lexical Phonology Phonology Year
book
Kiparsky P Catalexis Manuscript Stanford University
Ko opmans Van Beinum F Vowel Contrast Reduction AnAcoustic and
Perceptual Study of Dutch Vowels in Various Speech Conditions Academis
che Pers Amsterdam
Ko opmans Van Beinum F Ako estische en Perceptieve Asp ecten van
Klinkercontrastreducti e en de Rol van de Fonologie Spektator
Ko opmans van Beinum F J What s in a Schwa IFAProceedings
Also app eared in Phonetics
Krejnovich E F onetika Nivskogo Jazyka Leningrad
Kristo erson G The Syl lable Structure of Norwegian Do ctoral Disser
tation Troms University
Kruisinga E De Vorm van de Verkleinwo orden Nieuwe Taalgids
Lab ov W Principles of Linguistic Change Volume I Internal Factors
Blackwell Oxford UK and Cambridge Mass Language in Society
Lahiri A and V Evers Palatalisati on and Coronality in C Paradis
and J F Prunet eds The Special Status of Coronals Academic Press
San Diego pp
Lahiri A en H van der Hulst On Foot Typ ology NELS
Lahiri A and J Koreman Syllable Weight and QuantityinDutch
Proceedings of the West Coast ConferenceonFormal Linguistics
Langeweg S The Stress System of Dutch PhD dissertation University
of Leiden
Lass R English Phonology and PhonologicalTheory Synchronic and
Diachronic Studies Cambridge University Press Cambridge Cambridge
studies in linguistics
Lev elt C On the Acquisition of Place PhD Dissertation Leiden
Levin J A Metrical Theory of Syl labicity PhD Dissertation MIT
Lieb er R An IntegratedTheoryofAutosegmental Processes SUNY
Press Albany SUNY Series in Linguistics Mark Arono Editor
Lindau M The Feature Expanded Journal of Phonetics
Lit J van A Structural Condition on Autosegmental Licensing in P
Branigan J Gaulding M Kub o and K Murasuji eds SCIL
MIT Working Pap ers in Linguistics Cambridge Mass
References
Lo dge K French Phonology Again Deletion Phenomena Syllable
Structure and Other Matters in J Anderson and J Durand eds Ex
plorations in Dependency Phonology Foris Dordrecht pp
Lombardi L Laryngeal Features and Laryngeal Neutralization PhD Dis
sertation University of Massachussetts at Amherst
Martin A Klinker reductie Een Casus Manuscript Universiteit
Utrecht
Mascar o J Vowel Reduction as Deletion Handout Going Romance
Conference Utrecht
McCarthy J Formal Problems in Semitic Phonology and Morphology
Do ctoral Dissertation MIT Published by Garland Press New York
McCarthy J Feature Geometry and Dep endency A Review Phonetica
McCarthy a L In xation R eduplicative dans les Langages Secrets Lan
gages Sp ecial issue on Les javanais edited byM Pl enat
McCarthy J b The Phonology of Pharyngeal Articulations in Semitic
Pap er presented at the rd Conference on Lab oratory Phonology UCLA
Los Angeles
McCarthy J c Synchronic Rule Inversion BLS
McCarthy J and A Prince Proso dic Morphology Manuscript Amherst
Brandeis
McCarthy J and A Prince QuantitativeTransfer in Reduplicati ve and
Templatic Morphology in Seuk Dik Kim ed Linguistics in the morning
calm Hanshin Seoul pp
McCarthy J and A Prince Foot and Word in Proso dic Morphology
The Case of the Arabic Broken Plural NLLT
McCarthy J and A Prince a Proso dic Morphology I Manuscript
Amherst Rutgers
McCarthy J and A Prince b Generalized Alignment in G Bo oij and
J van Marle eds Yearbook of Morphology Kluwer Dordrecht
McCarthy J and A Prince a The Emergence of the Unmarked Opti
mality in Proso dic Morphology NELS
McCarthy J and A Prince b Proso dic Morphology An Overview Talk
presented at Proso dic Morphology Workshop
McCarthy J and A Taub Review of C Paradis and J F Prunet eds
The Sp ecial Status of Coronals Phonology
Melvold J Structure and Stress in the Phonology of Russian PhD Dis
sertation MIT
Mester A and J It o Feature Predictability and Undersp eci cation
Palatal Proso dy in Japanese Mimetics Language
References
Michaels D Natural and Unnatural Phonology in W Dresher H
Lusch utzky O Pfei er and J Rennison eds Phonologica Cambridge
University Press Cambridge pp
Mohanan K P The TheoryofLexical Phonology Reidel Dordrecht
Mohanan K P On the Bases of Radical Undersp eci catio n NLLT
Mohanan K P Fields of Attraction in Phonology in J Goldsmith
Mohanan T Syllable Structure in Malayalam Linguistic Inquiry
Mo ortgat M and H van der Hulst Leeft het Nederlands op Grote Vo et
Manuscript Leiden University
Morin Y The Status of Mute e Studies in French Linguistics
Morin Y Cross syllabic Constraints and the French e muet Journal
of Linguistic Research
Morin Y De l AjustementduChevaenSyllabeFerm e dans la Phonolo
gie du Fran cais in Verluyten
Mors C ter Een Vo caalinsertieparameter Of Waarom Nederlanders
Geen Indianen Zijn TABU
Moulton W The Vowels of Dutch Phonetic and Distributiona l Classes
Lingua
ctions Foris Muysken P and H van Riemsdijk eds Features and Proje
Dordrecht Studies in Generative Grammar
Myers S Vowel Shortening in English NLLT
Nesp or M and I Vogel Prosodic Phonology Foris Dordrecht
Neyt A and W Zonneveld Metrische Fonologie De Representatie van
Klemto on in Nederlandse Monomorfematische Wo orden De Nieuwe Taal
gids
Nijen Twilhaar J Generatieve Fonologie en de Studie van Oostneder
landse Dialecten Generative Phonology and the Study of Eastern Dutch
Dialects P J Meertens Instituut Amsterdam
Nijen Twilhaar J Genus en Fonologisch Gedrag van Nomina op Schwa
in G Bo oij and J van Marle eds Dialectfonologie P J Meertensinstituut
Amsterdam
No oteb o om S Some Regularities in Phonemic Sp eech Errors IPO An
nual progress report
No oteb o om S Production and Perception of Vowel Duration A Study
of Durational Properties of Vowels in Dutch PhD Dissertation Rijksuni
versiteit Utrecht
References
Noske R Syllabi ca tion and Syllabl e Changing Rules in French in H
van der Hulst and N Smith eds The Structure of Phonological Represen
tations Part II Foris Dordrecht
Noske R La Syllabi catio n et les R egles de Changement de Syllab e en
Fran cais in Verluyten
Noske R ATheory of Syl labi cation and Segmental Alternation With
Studies on the Phonology of French German Tonkawa and Yawelmani
PhD Dissertation Tilburg University
Nouveau D Language Acquisition Metrical Theory and Optimality A
Study of Dutch WordStress PhD Dissertation OTS Utrecht
Noyer R Mobile A xes in Huave Optimality and Morphologica l Well
formedness in E Duncan M Hart and P Spaelti eds Proceedings of
WCCFL University of California Santa Cruz
Noyer R Optimal Words Towards a Declarative Theory of Word
Formation Manuscript Princeton
Nuyts J Het Antwerpse Vokaalsysteem Een Synchronische en Diac hro
nische Schets Taal en Tongval pp
Odden D Vowel Geometry Phonology
Odden D Tone African Languages in J Goldsmith pp
Odden D and M Odden Ordered Reduplicati on in K hehe Linguistic
Inquiry pp
Oostendorp M van Formal Prop erties of Metrical Structure in S
Krauwer M Mo ortgat and L des Tomb e eds Proceedings of the Sixth
ConferenceoftheEuropean Chapter of the Association for Computational
Linguistics OTS Utrecht pp
Oostendorp M van A xation and Syllable Integrity in Dutch in R
Bok Bennema and C Cremers eds Linguistics in the Netherlands
John Benjamins Amsterdam
Oudenaarden J Wat Zeggie Azzie Val dan Leggie Een Speurtocht
naar het Dialect van Rotterdam Contact Utrecht
Overdiep G Moderne NederlandscheGrammatica Tjeenk Willink
Zwolle
Padgett J StrictureinFeatureGeometry PhD Dissertation University
of Massachussetts at Amherst
adgett J Stricture and Nasal Place Assimilation Natural Language P
and Linguistic Theory
Padgett J Feature Classes Manuscript to app ear in University of Mas
sachussetts Working Papers
References
Painter C Cineradiographi c Data on the Feature Covered in Twi
Vowel Harmony Phonetica
Paradis C and J F Prunet eds The Special Status of Coronals Aca
demic Press San Diego
Paradis C and J F Prunet A Note on Velar Nasals The Case of
Uradhi Canadian Journal of Linguistics R evue canadienne de linguistique
Peterson K A Comparative Lo ok at Nguni Verbal Tone in I Ha k
and L Tuller eds Current Approaches to African Linguistics Foris
Dordrecht pp
Philips C Are Feature Hierarchies Autosegmental Hierarchies in A
Carnie H Harley T Bures eds Papers on Phonology and Morphology
MIT Working Papers in Linguistics MITWP Cambridge Mass
Polg ardi K Derived Environment E ects in Optimality Theory
Manuscript Leiden University
Pollo ck J Y Verb Movement Universal Grammar and the Structure
of IP Linguistic Inquiry
Prince A Improving Tree Theory Proce edings of the Eleventh Annual
Meeting of the Berkeley Linguistcs Society Berkeley BLS Universityof
California pp
Prince A Quantitative Consequences of Rhythmic Organisation in K
Deaton M Noske and M Ziolkowsky eds CLS Volume I I Papers from
the Parasession on the Syl lable in Phonetics and Phonology CLS Chicago
Prince A and P Smolensky Optimality Theory Manuscript Rutgers and
Colorado
Prokosch E AComparative Germanic Grammar Linguistic so cietyof
America UniversityofPennsylvania Philadel phi a
Prunet J F Spreading and Locality Domains in Phonology PhD Dis
sertation McGill University
Pulgram E Syl lable Word Nexus Cursus Mouton The Hague Paris
Janua Linguarum Studia Memoria Nicolai van Wijk Dedicata Series
Minor
Pulleyblank D Vo calic Undersp eci cation in Yoruba Linguistic In
quiry
Ramers K H Vokalquantit at and qualit at im Deutschen Niemeyer
T ubingen
Redenb erger W Lusitian Portuguese A is ATR and CP in
Studies in Romance Linguistics Proceedings of the M P Hawigara ed
Fifth Symposium on Romance Linguistics Newbury House Rowley Mass
Not consulted cited by Donegan
References
Riad Tomas Structures in Germanic Prosody A Diachronic Study with
Special Reference to the Nordic Languages PhD Dissertation Sto ckholm
University
Rialland A Schwa et Syllab es en Fran cais in L Wetzels and E Sezer
eds Studies in Compensatory Lengthening Foris Dordrecht
Rice C Stress AssignmentintheChugach Dialect of Alutiiq CLS
Papers from the General Session pp
Rice C Proso dic Trapping and Sievers Law Manuscript Universityof
Troms
Rice K On Deriving Sonority A Structural Account of SonorityRela
tionships Phonology
Rijen H van M en Tilb orgs W oordeboek Gianotten Tilburg
Rijk R de Aprop os of the DutchVowel System Manuscript MIT
Robinson O Synchronic Re exes of Diachronic PhonologicalRules PhD
dissertation Cornell University
Ro ca I Generative Phonology Routledge London
Rogier L J Aantekeningen en Correcties op de Rotterdamse Dialect
pro even Tijdschrift vor TaalenLetteren
Rosenthall S Vowel Glide Interaction in a Theory of Constraint Inter
action PhD Dissertation University of Massachussetts at Amherst
Rotenb erg J The Syntax of Phonology PhD Dissertation MIT
Rowicka G Polish Palatal Assimilation in Pre xed Words in R Bok
Bennema and C Cremers eds Linguistics in the Netherlands John
Benjamins Amsterdam
Rowicka G and J van de Weijer Proso dic Minimality Constraints in
the Lexicon of Polish The Case of Derived Imp erfectives The Linguistic
Review
Rubach J Abstract Vowels in Three dimensional Phonology the Yers
The Linguistic Review
Sagey E The Representation of Features and Relations in Non linear
Phonology Garland New York
Schane S The Phonological and Morphological StructureofFrench PhD
Dissertation MIT
Schane S On the Abstract Character of French E muet Glossa
Schane S There is no FrenchTruncation Rule in R Campb ell et al
eds Linguistic Studies in R omanceLanguages Georgetown University
Press Washington D C pp
Schlindwein D The PhonologicalGeometry of Morpheme Concatenation
PhD dissertation University of Southern California
References
Schultink H Boundaries Word Classes and the Accentuation of Derived
Words in Dutch in W Zonneveld et al eds Studies in Dutch Phonology
Martinus Nijho The Hague pp
Schutter G de Asp ekten van de Nederlandse Klankstruktuur Antwerp
Papers in Linguistics
Selkirk E The Phrase Phonology of English and French PhD Disserta
tion MIT Published by Garland Press
Selkirk E Comments on Morin s Pap er The FrenchFo ot On the
Status of Mute E Studies in French Linguistics
Selkirk E The Role of Proso dic Categories in English Word Stress
Linguistic Inquiry
Selkirk E a The Syntax of Words MIT Press Cambridge Mass
Selkirk E b The Syllable in H van der Hulst and N Smith eds The
Structure of Phonological Representations Foris Dordrecht
Selkirk E Phonology and Syntax MIT Press Cambridge Mass
Selkirk E On Derived Domains in Sentence Phonology Phonology
Yearbook
Selkirk E Dep endency Place and the Notion Tier Manuscript Uni
versity of Massachussetts
Selkirk E On the Inalterabili ty of Geminates in P M Bertinetto M
Kenstowicz and M Loparco eds Certamen Phonologicum II Papers from
the Cortona Phonology Meeting Rosenb erg and Sellier Torino pp
Selkirk E A Two ro ot Theory of Length Manuscript Universityof
Massachussetts at Amherst
Siegel D Topics in English Morphology PhD dissertation MIT
Siegel D The Adjacency Condition and the theory of Morphology
NELS
Sievers E Zur Accent und Lautlehre der Germanischen Sprachen
Beitr age zur Geschichte der Deutschen Sprache und Literatur
Sievers E Grundz uge der Phonetik Breitkopf und K artel Wiesbaden
th edition
Slis I and M van Heugten Voiced voiceless Distinction in DutchFrica
tives in H Bennis and A van Kemenade eds Linguistics in the Nether
lands Foris Dordrecht pp
Slo otweg A M and J Wester Nederlandse Wo ordklemto on Cijfers en
Structuren Gramma
Sluyters W Representing Diphthongs PhD Dissertation Nijmegen Uni
versity
References
Smith N R Bolognesi F van der Leeuw J Rutten and H de Leeuw
Aprop os of the DutchVowel System Years On in H Bennis and A
van Kemenade eds Linguistics in the Netherlands Foris Dordrecht
Spa J GeneratieveFonologie Levende talen Also in Hul
shof pp
Sp encer A A Non linear Analysis of Vowel Zero Alternations in Polish
Journal of Linguistics
Sproat R On Deriving the Lexicon PhD dissertation MIT
Stanley R Redundancy Rules in Phonology Language
Stemb erger J Glottal Transparancy Phonology
Steriade D RedundantValues in A Bosch B Need and E Schiller
eds CLS Parasession on Autosegmental and Metrical Phonology
Chicago CLS pp
Steriade D Reduplication and Syllabl e Transfer in Sanskrit and Else
where Phonology
Steriade D Closure Release and Nasal Contours in M Hu man and
R Krakow eds
Steriade D Undersp eci cation and Markedness in J Goldsmith ed
pp
Stewart J Tongue Ro ot Positions in Akan Vowel Harmony Phonetica
Stro op J Metathesis van s en p Spektator
Stro op J Afgedwongen Nasalisering Tijdschrift voor Nederlandse Taal
en Letterkunde
Szpyra J Ghost Segments in Nonlinear Phonology Polish Yers Lan
guage
Takahashi T ConstraintInteraction in Aranda Stress in J Harris ed
UCL Working Papers in Linguistics
Trager G L and H L Smith An Outline of English Structure Battenburg
Press Norman Oklahoma Studies in linguisti cs Occasional pap ers
Tranel B Concreteness in Generative Phonology University of California
Press Berkeley
Tranel B Floating Schwas and Closed Syllable AdjustmentinFrench
in W Dressler et al eds Phonologica Cambridge University Press
Cambridge
Tranel B FrenchSchwa and Non linear Phonology Linguistics
Tranel B A Prop os de l AjustementdeE enFran cais in Verluyten pp
References
Tranel B French Liaison and Elision Revisited A Uni ed Account
within Optimality Theory Manuscript University of California Irvine
Trigo Ferre R L On the Phonological Derivation and Behavior of Nasal
Glides PhD Dissertation MIT
Trommelen M The Syl lable in Dutch With Special Reference to Diminu
tive Formation Foris Dordrecht
Trommelen M On ie uu and oe DutchVowel Length revisited in
G de Haan and W Zonneveld eds Formal Parameters of Generative
Grammar OTS Yearbook III OTS Utrecht pp
Trommelen M Lexical Word Pro cesses in Dutch Linguistic Review
Trommelen M and W Zonneveld Stress Diphthongs r in Dutch in
H Bennis and A van Kemenade eds Linguistics in the Netherlands
Foris Dordrecht
Trommelen M and W Zonneveld Klemtoon en Metrische Fonologie
Coutinho Muiderb erg
Trub etzkoy N Grundz uge der Phonologie Travaux du Cercle Linguis
tique de Prague Praag
Uriagereka J GovernmentFunctions and the Classical Arabic Verb
Paradigm Handout of the th GLOW Collo quium
Vago R The TreatmentofLongVowels in Word Games Phonology
Yearbook
Vagra M Phonetic Explanations in Phonology the Case for Glottal
Stops in Italian Word nal Stressed Syllables Manuscript Scuola Normale
Sup eriore Pisa Not consulted cited in Burzio
Vater H Zum Silb en Nukl eus im Deutschen in P Eisenberg K
Ramers H Vater eds Silbenphonologie des Deutschen Gunter Narr Ver
lag Tubingen pp
Velde H Van der A Trend Study of a Trendy Change Manuscript
Nijmegen University
Vennemann T The Phonology of Gothic Vowels Language
Vennemann T PreferenceLaws for Syl lable Structure De Gruyter
Berlin
Vennemann T Syllable Structure and Syllable Cut Proso dies in Mo dern
Standard German in P M Bertinetto and M Lop orcaro eds Certamen
Phonologicum Proceedings of the Cortona Phonology Conference Rosen
b erg and Sellier Torino
Verdam J Uit de Geschiedenis der NederlandscheTaal Thieme Zut phen
References
Vergnaud J R On the Foundations of Phonology Pap er presented at
the th GLOW Collo quium Not consulted cited in Charette
Verluyten S La Structure Proso dique des Mots Fran cais Hi erarchique
ou Lin eaire in K van den Eynde M Dominicy and S Verluyten eds
Linguistics in Belgium Univeristeit van Antwerp en Antwerp en Antwerp
Papers in Linguistics
Verluyten S ed La Phonologie du Schwa Fran cais John Benjamins
Amsterdam
Visch E A Metrical Theory of Rhythmic Stress Phenomena PhD Dis
sertation Utrecht University
Visser W Schwa app endixen in het Fries in G Bo oij and J van Marle
eds Dialectfonologie P J Meertensinstituut Amsterdam
Vogel I Phonological interfaces in Italian in M Mazzola ed Proceed
ings of Linguistic Studies on RomanceLanguages XXIII
Vlugt M van der and S No oteb o om Auditory Word Recognition is not
more SensitivetoWord Initial than to Word Final Stimulus Information
IPO Annual Progress R eport
Vries J de et al De Slot t in de Consonantclusters te Leiden een So ci
olingu sti sch Onderzo ek Forum der Letteren
Wennerstrom A Fo cus on the Pre x Evidence for Word Internal
Proso dic Words Phonology
Weijer J van de Basque A ricates and the Manner Place dep endency
Lingua
Weijer J van de Segmental Structure and Complex Segments PhD Dis
sertation Leiden University
Weijnen A Verklaren van Dialectverschijnsel en Text of lecture Ni
jmegen University
Weijnen A Vergelijkende Klankleer der Nederlandse Dialecten Staats
drukkerij uitgeveri j Den Haag Aan het Woord
Wetzels L The Timing of Latent Consonants in Mo dern French in C
Neidle and R Nunez Cedeno eds Studies in Romance Linguistics Foris
Dordrecht pp
Wetzels L Umlaut en Verkleinwo ordvorming in het Schinnens Gramma
W etzels L Possible and Imp ossible Assimilation Pro cesses The Place of
the Feature continuant in the Feature Tree Manuscript Free University
of Amsterdam
Wiese R Schwa and the Structure of Words in German Linguistics
References
Wiese R Silbische und Lexikalische Phonologie Studien zum Chinesis
chen und Deutschen Niemeyer Tubingen
Wijk N van De Nederlandsche Taal Tjeenk Willin k Zwolle
Wijk N van Phonologie Een Hoofdstuk uit de Structurele Taalweten
schap Nijho Den Haag
Wilde Van Buul G de Het Enclitisch Pronomen Personale van de e en
e Perso on Singulari s in het Rotterdams Tijdschrift voor Nederlandsche
taal en letterkunde
Williams E On the Notions Lexically Related and Head of a Word
Linguistic Inquiry
Winteler J C Die Kerenzer Mundart des Kantons Glarus in ihren
Grundz ugen Dargestel lt Carl Winter Heidelb erg
Yip M English Vowel Ep enthesis Natural Language and Linguistic
Theory
Yip M The Tonal Phonology of Chinese PhD Dissertation MIT
Yip M Tone in East Asian Languages in J Goldsmith pp
Zonneveld R van A x grammatika Een Onderzoek naar Woordvorming
in het Nederlands PhD Dissertation Rijksuniversitei t Groningen
Zonneveld R van Word Rhythm and the Janus Syllable in H van der
Hulst and N Smith eds The Structure of Phonological Representations
II Foris Dordrecht
Zonneveld W Fonologische Polariteit en de Vlaams Brusselse Vokaalver
schuiving Leuvense Bijdragen
Zonneveld W AFormal TheoryofExceptions in Generative Phonology
The Peter de Ridder Press Lisse
Zonneveld W The DescriptivePower of the Dutch Theme Vowel Spek
tator
Zonneveld W Schwa Sup erheavies Stress and Syllabl es in Dutch The
Linguistic Review
Zonneveld W and M Trommelen Inleiding tot de Generatieve Fonologie
Coutinho Muiderb erg
Zubizaretta M L Vowel Harmony in Andalusian Spanish MIT Work
ing Papers in Linguistics
Zwart C J W Dutch Syntax A Minimalist Approach PhD Dissertation
University of Groningen
Summary in Dutch
Klinkerkwali tei t en
lettergreeppro jectie
In dit pro efschrift wordt de ho ofdigheidshyp othese vo or lettergrep e n nader
uitgewerkt Met name wordt b estudeerd in ho everre fonologische ken
merken van prototypische lettergreepho ofden dat wil zeggen klinkers van
invlo ed zijn op de structuur van de proso die rondom het ho ofd Centraal
in de discussie staan het Standaard Nederlands en een aantal Nederlandse
dialecten zoals het Tilburgs en het Rotterdams maar o ok enkele meer ex
otische talen zoals het Indonesisch het Frans het Gotisch en het No ors
worden b esproken Na een algemeen inleidend ho ofdstuk komen de vol
gende onderwerp en aan b o d
In ho ofdstukken en wordt een analyse gemaakt van het contrast
tussen gespannen versus ongespannen of lange versus korte klinkers in het
Standaard Nederlands Er wordt b eargumenteerd dat dit verschil alleen
b evredigend b eschreven kan worden met een kenmerk Retracted Tongue
Ro ot RTR Vervolgens is er een conditie die dit kenmerk relateert aan
het op en versus gesloten karakter van de lettergreep Zeer sterkeevidentie
vo or deze analyse komt uit talen die een op en gesloten contrast combineren
met klinker harmonie op het kenmerk RTR Brabantse dialecten Tilburgs
Antwerps Hofstaads zijn in dit verband o ok van b elang omdat ze de pa
rameters lengte en gespannenheid onafhankelijk com bineren en omdat ze
laten zien dat klinkerlengte over het algemeen corresp ondeert met RTR
niet met RTR zoals tot nu to e op basis van het Standaard Nederlands werd
aangenomen
In ho ofdstukken en wordt de Nederlandse sjwa b esproken De
stomme e mo et worden geanalyseerd als fonetisch en fonologisch nage
no eg leeg Met deze kenmerklo osheid corresp onderen een aantal bijzondere
eigenschapp en van deze klinker hij is de reductiekli nker de ep enthetische
klinker en als onderliggende klinker komt hij alleen vo or in lettergrep en
Summary in Dutch
met een zeer eenvoudige structuur een enkelvoudige onset en een co da
waarin ho oguit een sonorant mag staan Er wordt aangeto ond dat al deze
eigenschapp en o ok inderdaad direct samenhangen met de leegheid van de
sjwa
In ho ofdstuk wordt de analyse over de Nederlandse sjwa to egepast op
een aantal andere talen met name het Frans en het No ors Het Frans staat
anders dan het Nederlands juist wel een complexe onset en geen co da to e in
sjwa ho ofdige lettergrep en Het No ors heeft zowel een Franse sjwa eple
als een Nederlandse sjwa tiger deze sjwa s zijn echter o ok in het No ors
gescheiden wo orden als katrel komen niet vo or De verschillen tussen de
drie talen wordenzoveel mogelijk b eschreven in termen van onafhankelijk
gemotiveerde andere verschillen en b ovendien wordt gewezen op enkele zeer
opvallende overeenkomsten
In ho ofdstuk wordt tenslotte ingegaan op het voorkomen van klinkers
in niet ho ofdige p ositie Dit geb eurt aan de hand van een uitgebreide anal
yse van de Rotterdamse ie je alternantie in monomorfemische wo orden het
clitisch pronomen van de tweede p erso on singularis en het diminutiefsuf
x In al deze vormen zijn de klinkers i en sjwa met elkaar in comp etitie
Deze comp etitie kan uitstekend b eschreven worden in termen van de Opti
maliteitstheorie Het blijkt dat de i zich inderdaad gedraagt als een b eter
lettergreepsho ofd en dat de sjwa alleen aan de opp ervlakte komt als real
isatie van de volle klinker als nucleair ho ofd geblokkeerd wordt Vergelijking
van dit Rotterdamse pro ces met de Wet van Sievers over het Gotisch levert
bovendien evidentie tegen de lengte analyse van het Standaard Nederlands
in talen met echte lengte zoals het Gotisch breekt i tot ji Het blijkt zo te
zijn dat vo oral het kenmerk ho og een voorkeur heeft vo or de extracefalis
che p ositie op allerlei niveaus
Ho ofdstuk geeft een overzichtvan de familie van b ep erkingen die de
kern vormen van de meeste analyses in dit pro efschrift de familie van
pro jectieb ep er kingen Deze verbinden allemaal het voorkomen of niet
vo orkomen van een b epaald kenmerk of van een gro ep kenmerken aan
het voorkomen of niet voorkomen van een b epaald typ e proso dische struc
tuur Het wordt aangeto ond dat het b egrip familie ons een elegante en
b evredigende metatheorie van fonologische pro jectie verscha en
Tenslotte wordt in een app endix een complete klinkergrammatica van
het Nederlands uitgewerkt en er worden enige b eschouwingen gewijd aan
het in het pro efschrift gebruikte formalisme
Summary in English Vowel
quality and phonological
pro jection
In this thesis we try to work out some consequences of the hyp othesis that
syllables are headed constituents In particular we examine to what extent
phonological features of prototypical heads that is to say vowels have
an in uence on the structure of the syllable around the head Central to
the discussion are Standard Dutch and some Dutch dialects likeTilburg
Dutch and Rotterdam Dutch but also other languages like Indonesian
French Gothic and Norwegian receive some treatment After a general
intro ductory chapter the following topics are discussed
In the chapters and we analyse the contrast b etween tense and
lax or long versus short vowels in Standard Dutch It is claimed that
this di erence can only b e describ ed in a satisfactory way when using the
feature Retracted Tongue Ro ot RTR We also p osit a constraintwhich
relates this feature to the op en versus closed nature of a syllable Very
strong evidence for this relation is drawn from languages such as Andalusian
Spanish and Eastern Javanese whichcombine the op en closed contrast with
vowel harmony on the feature RTR Also a dialect like Tilburg Dutchis
of interest b ecause it has b oth a tense lax and a long short contrast In
this dialect we can also see that vowel length correpsonds to RTR not to
RTR as is usually assumed on the basis of Standard Dutch
In chapters and we discuss Dutchschwa This vowel has to b e
analysed as almost empty b oth from a phonetic and phonological p ointof
view This featurelessnes s corresp onds to several sp ecial prop erties of this
vowel it is the reduction vowel the ep enthetic vowel and as an underlying
vowel it only supp orts syllables with a very simple structure viz a simple
onset and a co da in whichwe only nd a sonorant consonant It is shown
that all of these prop erties are indeed directly dep endent on the emptiness
of schwa
Summary in English
In chapter the analysis of Dutchschwaisappliedtoanumber of
other languages esp ecially French and Norwegian Frenchschwaallows a
complex onset but no co da and therefore seems markedly di erent from its
Dutchcounterpart Norwegian has b oth a French eple and a Dutchschwa
tiger These schwas however still have a di erent status in the language
we do not nd words suchas katrel These di erences b etween the three
languages are describ ed in terms of indep endently motivated constraints en
on top of this it is shown that the three languages still show a few very
remarkable similarities
In chapter we discuss the o ccurrence of vowels in extracephalic p osi
tion The main topic of discussion here is a striking alternation b etween the
high vowel i and the syllable j we nd in Rotterdam Dutch monomor
phemic forms second p erson singular clitics and diminutive forms The
comp etition b etween these two forms can b e describ ed in a theory of con
straint con ict such as Optimality Theory It app ears that i acts as a b etter
nuclear head and that the form with schwa only surfaces as a realisation
of the full vowel is blo cked A comparison of the Rotterdam pro cess to
Sievers Law in Gothic additionally provides us with extra evidence against
a length analysis of Standard Dutch
Chapter gives an overview of the family of constraints whichform
the backb one of most analyses in the thesis the family of pro jection con
straints All of these constraint connect the o ccurrence or non o ccurrence
of a certain feature to the o ccurrence or non o ccurrence of certain typ es
of proso dic structure It is shown that the notion of a constraint fam
ily provides us with an elegant and satisfying metatheory of phonological pro jection