<<

versus

in *

Dominique Rodier McGill University

o. Introduction Some of the most recent studies in have been concerned with the 'Principles and Parameters' approach to the study of the language faculty, as illustrated in work by Bouchard (1982), Kaye and Lowenstamm (1981, 1982, 1983), Piggott and Singh (1984), etc. Certain phonological processes are found in numerous languages. To describe them simply in terms of language-specific rules would suggest that very important generalizations are being missed. The alternative to describing those processes by means of phonological rules would be to appeal to the principles and parameters of Universal . As Singh (1980, 1981a,,c) has argued, some phonological processes can be characterized as repair strategies triggered by well-formedness conditions on syllabic structure. Each of these strategies (e.g. Deletion, , etc.) would be the result of the setting of some parameter, an option that, in principle, is given to every grammar. Each grammar would have to decide which strategy would be used to repair which are not well formed. The choice of a particular parameter may be linked to universal or language-specific constraints on structure. The setting of one parameter may also be prompted by the prior setting of another parameter. Compensatory Lengthening is one of those phonological processes that are found in language after language. In this analysis, I will try to explain this particular process in terms of properties of syllable structure and argue that it can be seen as a repair strategy which is triggered by the need to preserve a certain kind of syllable structure. The determination of syllable structure assumed in this paper will be that of Piggott and Singh (1984). In fact, this conception of syllabification is

50 Compensatory Lengthening and Gemination 51

crucial to my analysis of Compensatory Lengthening.

1. The Framework

Following Piggott and Singh (1984), I will assume that every syllable (~) has the two constituents: the onset (0) and the rime (R). In addition, the rime has an obligatory constituent, the nucleus (N) and it may also contain a second (optional) constituent, the coda (C). I will also assume that the terminal nodes of syllable trees are not segments themselves but a set of skeletal slots or points to which segments are linked. Hence the structures in (1) are allowed (the XIS represent skeletal point ) : la. ------r--- b. ------if --- ° R ° R I ------N N C I I I X x X x x

Of these, Cl a l is the mi n imal st ruc t ur e that a syllable can have. Each of the nodes labeled 0, Nand C may branch, allowing for syllable structures of greater complexity than those shown in (1). According to Piggott and Singh, there are obviously likely to be universal and language-specific constraints on what may be associated with the terminal elements of the onset or coda. There are also likely to be constraints on the complexity of structures with branching onsets, nuclei and codas. Kaye and Lowenstamm (1981), argue that the terminal elements of the nucleus must all be specified [+vocalic]. Since vowels, glides, liquids and nasals are all possible constituents of the nucleus, they are all specified as [+vocalic] segments by Kaye and Lowenstamm. Piggott and Singh (1984) take a more conventional approach and regard these segments as . However, they propose that the nucleus may dominate [+] , at least in some languages, among them being English. The qualification of the last statement is intended to allow for the possibility that languages may vary in terms of what segments may be linked to the terminal elements of the nodes labeled 0, Nand C. Piggott and Singh (1984) further propose that nasals and liquids may appear as the second (i.e. the right-most) constituent of a nucleus. Thus, a branching nucleus may characterize either a long vowel, a diphthong or a sequence of a vowel followed by a nasal or liquid. They also propose that the [s] may also appear as a constituent of the nucleus.

Piggott and Singh argue that their analysis of Epenthesis depends crucially on their conception of syllabification, i.e. of the set of strategies by which all the segments in a string 52 Rodier are (exhaustively) assigned to well-formed syllables. Since my own analysis also depends on their conception of syllabification, let us look at their assumptions: first, there is a principle which applies relatively freely, constructing an initial set of syllable structures. This initial syllabification principle (ISP) is stated roughly in (2). The ISP is mediated by a set of principles which include those cited in (3) (Piggott and Singh, 1984; 91-92).

2. Every skeletal slot is assigned to a position in a syllable, maximizing onsets (ISP).

3a. Every node of syllable structure must dominate at least one skeletal slot. b. A vowel is associated with at least one nucleus.

c. A segment must be associated with a slot in the nucleus if it is specified [+vocalic]. d. A slot which is linked to a segment unspecified for the feature [vocalic] must be associated with a nucleus unless it precedes a slot linked to a [+vocalic] segment in which case the former is assigned to the onset. e. A [-vocalic] segment cannot be assigned to the nucleus

Resyllabification is conceived of as a set of strategies by which segments are reassigned to different positions if the ISP creates structures that violate some substantive constraint, either universal or language-specific. One example given by Piggott and Singh of resyllabification triggered by a language-specific constraint is that of the glide [y] in French. Principle (3d) will assign the glide to the nucleus. However, according to Kaye and Lowenstamm (1981), the glide in the final syllable of a word like 'travail' [travay] should be in the coda. The structure given by the ISP would then be as in (4a) and the structure after resyllabification would be as in (4b). Language-specific constraints or principles are regarded as marked options; thus the need for resyllabification.

4a. rr b. tJ 0 ----R 0 --- --R I ~--..... 6. N N c I /""--. /\ I I tra v a y tra v a y Compensatory Lengthening and Gemination 53

According to Piggott and Singh, the principles of syllabification presented in (2) and (3) do not initially permit sonorant consonants in the nucleus. It would seem that the appearance of such consonants in the nucleus is the manifestation of a marked option which is specifically permitted by a language-particular rule or condition. According to Piggott and Singh, sonorant consonants may be assigned to the nucleus through one or two devices. A given language may include among the set of resyllabification strategies a condition on syllabification such as the one cited here in (5) below (Piggott and Singh, 1984; 94). 5. A [-vocalic] segment is associated with the nucleus iff it is also [+sonorant].

Alternatively, a language may undergo the rule which moves a sonorant into the nucleus. Crucially, application of the rule Move Sonorant follows resyllabification. Both the condition in ~and the rule Move Sonorant are constrained by restrictions on which position in the nucleus sonorant consonants may occupy. It is assumed that in the unmarked case the sonorant appears as the righthand constituent of a nucleus. Also, the rule Move Sonorant and the condition cited in (5) are parametrically determined. 2. Compensatory Lengthening and Gemination Let us now look at the phonological process of Compensatory Lengthening, comparing it to another quite similar process, the formation of Gemination. In their paper on Compensatory Lengthening in Tiberian Hebrew, Kaye and Lowenstamm (1983) claim that the representation of Semitic and phonology specifies the prosodic information involved in a particular structure, i.e. in most cases, the number and shape of syllables that characterize a particular morphological class. Along with this prosodic information, segmental information is incorporated in the form of a consonantal and a vocalic melody. Kaye and Lowenstamm argue that Compensatory Lengthening and Gemination arise when there is a discrepancy between the number of Level 2 positions (i.e. the number of points available for mapping by the melodic tier) and the number of terminal nodes on ·Level 1 trees in some morphological classes, as shown in (6) where the rightmost branch of the medial syllable does not dominate a position at Level 2. This sort of configuration determines the presence of a null element (a). 54 Rodier

6. Levell

Level 2

Level 3 hit / g, d, 1 / / a /

The null element position has, then, to be filled. If the (ff) is a C, it will serve as the support for the spreading of £ in (6); if it is interpreted as a V, the position will be filled by a, thus generating either hitgaddeel or the ill-formed *hitgaadeel. What is necessary here is that the right branch of the rime of the unlabeled medial syllable of the representation be identified as a C, not a V. This amounts to saying that the derived label dominating the branching rime should be R, not N. This is what Kaye and Lowenstamm expect from their principle (10)<1> which restricts the distribution of null elements to non-branching constituents. In fact, in Tiberian Hebrew, Gemination is the normal case. Compensatory Lengthening appears only when in a configuration like (6) the medial cannot geminate (i.e. in TH, the resonants / 7,8 , h, r, h / which never geminate in any context). From this analysis of the facts in Tiberian Hebrew, Kaye and Lowenstamm conclude that Gemination is the unmarked or 'default' way of resolving the problem of the identity of the null element and that one should expect that a language in which every consonant can geminate will never display Compensatory Lengthening (CL). Specifically their solution expresses the fact that Gemination takes precedence over Vowel Lengthening. 3. The Ancient Greek data The analysis of Compensatory Lengthening and Gemination in Ancient Greek, however, illustrates a type of situation quite different from that of Tiberian Hebrew. The Ancient Greek data are from Ingria (1980).

The most striking fact concerning Compensatory Lengthening in Ancient Greek is the divergence between two Aeolic dialects (Lesbian and Thessalian) and all the other Greek dialects. In general in Ancient Greek, when a segment is deleted the vowel preceding that segment is lengthened, even when the segment is not adjacent to the vowel underlyingly. In Lesbian and Thessalian (henceforth L/T), however, the result is a lengthened adjacent consonant, whether this consonant precedes or follows the deleted segment in the underlying representation. The following examples may be taken as illustrative. Compensatory Lengthening and Gemination 55

7a. /s/ following sonorant in the sigmatic <2> Underlying Form Surface Output

ekrinsa L/T -ekrinna ,judge' Elsewhere ekri:na a:lJgelsa L/T _a:I)gella 'announce' Elsewhere _a:l)ge:la ep" t'"ersa L/T ephth erra 'destroy' Elsewhere =eph t" e: ra

b. /s/ following sonorant elsewhere

bolsa: L/T -bolla: 'council' Elsewhere bo:la:

aw s c r s L/T aWWJ:s 'dawn' Doric a rw o r s Ionic - :>:..:>:S Attic -he o e s

c. /s/ preceding sonorant

esmi L/T emmi ,I am' Elsewhere e:mi

selasna: L/T selanna: 'moon' Attic-Ionic sel.£:na: Elsewhere -sela:na: kneslioy L/T khell ioy 'thousand' Attic. =kh i :1ioy Elsewhere _khe: 1ioy

kh. esr L/T kh erras 'hand' Elsewhere -ele: ro

naswos L/T nawwos 'temple' Doric -na:wos Homeric -n!'):os Attic ne o s 5 56 Rodier

8. /y/ following sonorant

phany J: All dialects _ph.aynJ: 'snow' (present)

ankonya All dialects _ankoyna 'halyard' but: kriny::>: L/T krinn::>: ,judge' (present) Elsewhere k r i en o s kh.ary::>: All dialects _kh.ayrJ: 'rejoice' (present) morya All dialects _moyra 'fate' but:

phe' e r y o i L/T ph.e·err j : 'destroy' Elswhere =ph.thoe :r::> : However:

ba Ly o : All dialects ball::>: "throw' (present)

aI)gelyo: All dialects _aI)gell:l: 'announce'

In contrast with Kaye and Lowenstamm's analysis of Tiberian Hebrew, in Ancient Greek the null element which triggers Compensatory Lengthening or Gemination does not arise from a discrepancy between Levelland Level 2, but rather is left in the position of a deleted segment. To be more precise, I argue that, in Ancient Greek, Compensatory Lengthening and Gemination apply after syllabification has been completed. The phonological rules which apply to the data in (7) and (8) before Compensatory Lengthening or Gemination takes place are the following (Ingria's formulation):

9. V +sonorantJ s V [-syllabic

1 2 3 4 -----+)0 1 3 2 4 Compensatory Lengthening and Gemination 57

10. 'hI I +sonorant] s >%J/ _____ [-syllabic

11. V 1 Y V

1 2 3 4 ___ --?>>- 1 2 2 4

12. V [+sonorant ] y +consonantal

1 2 3 -----7-~1 3 2

13. 'hI I V +sonorant ] y >- .ff / [-back] [+consonantal

4. A Parametric Analysis In my analysis of Compensatory Lengthening and Gemination in Ancient Greek dialects, I will argue that the choice between those two processes will be determined by the principles of syllabification and the way in which a particular parameter is set. I will also argue that in fact those phonological processes are structure preserving repair strategies. One very interesting characteristic of the phonological rules whose interaction seems to trigger Compensatory Lengthening, is that they all involve [+sonorant] segments and/or [s], those being the only consonants which, according to Piggott and Singh, can be part of the nucleus. In fact, the glides, the nasals, the liquids and [s] appear to be the only segments whose deletion triggers Compensatory Lengthening. Let us then assume that the dialects of Ancient Greek which have Compensatory Lengthening have also the condition in (5) (repeated here for convenience) as part of their grammar and that L/T do not have this condition.

5. A [-vocalic] segment is associated with the nucleus iff it is also [+sonorant]. That condition should be expanded to indicate that [sl can also be assigned to the nucleus. The Empty Node Condition proposed by Piggott and Singh (1984) will also play an important role in the process of Compensatory Lengthening. 58 Rodier

14. Empty Node Condition

A well-formed syllable cannot contain an empty position unless this position is exhaustively and exclusively dominated by 0 (onset).

Let us now look at the derivation of some of the data given in (7) and (8) and see how the principles of syllabification and the phonological rules apply in L/T and in the other dialects. First, the segmental representation is mapped onto Level 2 according to the initial syllabification principle (ISP) (plus some language-specific constraints on syllable structure) and the universal principles in (2) and (3), as is shown in (15) and (16) • l5a. Elsewhere b. Lesbian s Thessalian if' ( 11 ()

~ I I I 1\I hI V V k r i n y ::> k r i n y ~

l6a. Elsewhere b. Lesbian & Thessalian (J tJ (J 0- ff tJ /\ /\ I /\ /\ I I I I I I I I g I ,0 I e k r 1 n s a e 1 n s a

Because the dialects of Ancient Greek, apart from L/T, have condition (5) as part of their grammar, (15a) and (16a) are resyllabified . l5a. t:f IX 1\ /~\ I I I V I I I I r i ny, f!f e . r 1 n s a Compensatory Lengthening and Gemination 59

The phonological rules (12) and (13) then apply to the syllable structures in (15a-b) and the phonological rules (9) and (10) apply to the syllable structures in (17a-b), giving the structures shown in (17) and (18).

17a. Elsewhere b. Lesbian & Thessalian IJ {J ~ D'

I V I IA I I I V .i k r 1 £J n k r i g n J

18a. Elsewhere b. Lesbian & Thessalian ff () 0 f'J' If {f 1\ /\ 1\ !\ I I I 1 I I t I I I I I I XI e k r i 0 n a ~ e k r 1 n n a

The Empty Node Condition then comes into play and triggers a repair strategy. In the case of Lesbian and Thessalian, it is Gemination since the null element is dominated by the right branch of the rime. In the case of the other dialects, Compensatory Lengthening applies since the null element is dominated by the right branch of the nucleus. This gives the following structures:

19a. Elsewhere b. Lesbian & Thessalian (f IA(J N .c.!r .Ix~ I I ~ I 'V/' I I I V ""-./ k r 1 n :J k r i n ~ 60 Rodier

20a. Elsewhere b. Lesbian & Thessalian rr /\ AhAI I I I <:> I I I I ff' e k r i n a £1 e k r i n a

Thus, the choice between C and V to fill up the null element is made as a consequence of its position on the CV skeleton, without any need for a language-specific rule of Compensatory Lengthening or Gemination. The only thing we need to know is if condition (5) is part of the grammar or not. So a parameter of Universal Grammar (i.e. the possibility of having the condition in (5) in the grammar or not) will in turn determine a choice in another parameter (Compensatory Lengthening or Gemination). The correlation appears to hold in language after language displaying CL. Each time there is, in a language, lengthening of a preceding vowel after the deletion of a segment, one can expect that the deleted segment will be a [+sonorant] or [s]. In fact, [s] seems to be very much part of the process which triggers Compensatory Lengthening. In French, for example, Compensatory Lengthening was triggered at some point in the history of the language by the loss of [s] before a [t] (cf. beste > bete, fenestre > fenetre). In , there was a rule deleting-TS] before [+sonorant, +anterior] segments, which would also trigger CL.

Singh (1980, 1981a,b,c) has argued that certain phonological processes are most appropriately seen as repair strategies. They are, according to him, invoked to eliminate violations of well-formedness conditions that are defined over prosodic domains such as the syllable, the and the word. Piggott and Singh (1984) have argued that Epenthesis, both consonantal and vocalic, is most appropriately seen as an automatic consequence of certain types of readjustment of metrical structures necessitated by violations of well-formedness conditions (i.e. in the case of Epenthesis, phonotactic constraints involving sequences of segmen ts ). If Compensatory Lenthening and Gemination (which are, as was shown above, the effects of a unitary phenomenon) are to be seen as repair strategies, what then are they to be considered an automatic consequence of? In other words, why this particular strategy in this particular context? Deletion alone cannot be seen as the reason for that particular strategy to apply, leaving aside all the other possibilities (e.g. pruning). I will argue that Compensatory Lengthening and Gemination Compensatory Lengthening and Gemination 61

are appropriately seen as structure-preserving repair strategies. I will further argue that the structure which has to be preserved is a heavy syllable<3>. This means that deletion in the onset will not trigger either of those two processes. Let us look at some new data in Ancient Greek which give support to my analysis of the two processes.

5. The Attic Dialect In Ancient Greek, there is a rule which metathesizes /y/ and /w/ with a preceding sonorant consonant when any vowel precedes. It also deletes /w/ before a [+sonorant, +consonantal] and deletes /y/ in the context of rule (13). takes place, followed by deletion. The derivation of /ksenwos/ 'foreigner' then gives the expected result [kse:nos]. What is more interesting, however, is what happens in Attic, Lesbian and Thessalian where deletion of the /w/ takes place to the right of the consonant. These dialects must have rule (21). 21. +sonorant ] w ~~ [+consonantal ------

The derivation of the underlying form /ksenwos/ is then as follows in Attic and L/T. In (22) the segmental representation is mapped onto Level 2 according to the initial syllabification principle (ISP) and the universal principles in (2) and (3).

22a. Attic b. Lesbian & Thessalian ! t ff

~ / ~ I I I I l I I I I I I k s e n w 0 s k s e n w 0 s

Since Attic has condition (5) as part of its grammar, (22a) is resyllabified (i.e. the branching rime becomes a branching nucleus). 62 Rodier 22a·h IA

I I I I I I I k s e n w 0 s

The (21) then applies to the syllable structures in (22a-b), giving the structures shown in (23) where there is now a null element in the onset.

23a. Attic b. Lesbian & Thessalian d' IY « a- .: IA I I I I I I I I I I I I k s e n ,fi!f 0 s k s e n J! 0 s

The Empty Node Condition then comes into play but cannot trigger Compensatory Lengthening or Gemination since the heavy· syllable is still whole, the null element being in the onset. To fill up the null element, Attic and L/T have to resort to another repair strategy, Resyllabification, maximizing onsets. This will give structures (24a-b). The argument for the application of Compensatory Lengthening or Gemination before Resyllabification is that a language strives to preserve structures (i.e. as Singh pointed out, given a choice between a structure-preserving and a structure-changing repair strategy, the former is preferred). As was argued above, those two processes are used to preserve heavy syllables. In the case of (23a-b), then, they cannot be used to any profit since the the null element is in the onset. Resyllabification is used instead. 24a. Attic b. Lesbian & Thessalian tf

I I I I I I 0k s e n 0 s e n o s Compensatory Lengthening and Gemination 63

As one can see, Resyllabification results in the loss of a heavy syllable for a light one. But since Resyllabification does not leave behind a null element it is impossible for Compensatory Lengthening or Gemination to apply to save the structure.

This analysis of the word /ksenwos/ in Attic and L/T, in fact, supports my claim that Compensatory Lengthening and Gemination are used to preserve heavy syllables in more ways than one. One could argue that the reason why Gemination occurs only in the dialects that do not have condition (5) (i.e. L/T) is because a segment cannot spread from an onset to a nucleus, while the spreading of a segment from an onset to a coda is allowed. However, the fact that in (23b) and (24b) Gemination is not used although nothing should block the spreading of a segment from the coda to the onset shows that to consider Gemination and Compensatory Lengthening as repair strategies to preserve heavy syllables is the right solution. Another very interesting example is found in Cretan. In a word like /tans/ which ends with /ns/ two phonetic forms may be found: [tas] when followed by __ # C and [tans] when followed by __ # V. The structures in (25) are shown after resyllabification because of the condition in (5). 25. Cretan

a. b. ~ t

I I I AI I I I I AI I I t a n s # C V t a n s # ~ V

In (25b) the Empty Node Condition comes into play triggering Resyllabification, maximizing onsets. The result is a branching nucleus followed by an onset.

26. ff ~

AA, I I I I tan s # V

In (25a), however, the situation seems more complex. In Cretan, there is a rule which deletes /n/ before /s/, but only 64 Rodier

when the two segments inv?lved are tautosyllabic. One mu~t.assume then that this rule applles after the Empty Node Condltlon has triggered Resyllabification, otherwise /n/ would also have been deleted in (25b). The structure of (25a) is as in (27) after the Deletion rule has applied.

27. b' !f /~ 1\ I I I I I I t a ,Ej s # C V

There are two possible strategies here: Resyllabification or Compensatory Lengthening. However, since there is no change in syllable quantity, Compensatory Lengthening is not needed. So the Empty Node Condition must come into play, triggering Resyllabification, erasing the right branch of the nucleus. Following the condition in (5), /s/ is then assigned to the nucleus, making the branching rime a branching nucleus •

28. lJ

I I I t a S # CAV

Once more, Compensatory Lengthening does not apply, since the heavy syllable was perfectly safe throughout the derivation. 6. Conclusion I have argued that Compensatory Lengthening and Gemination are both effects of the same phenomenon: a structure-preserving repair strategy. I have argued that the choice between these two phonological processes is determined by the presence or absence of a parameter of Universal Grammar, the condition in (5). I have also argued that the repair strategy discussed here is a consequence of the need to preserve heavy syllables. Some of the most obvious implications of this analysis of Compensatory Lengthening and Gemination are the following: CL allows the retention of a branching nucleus when a heavy syllable of the type VV has lost its rightmost branch. Gemination allows the retention of a branching rime when a heavy syllable of the type VC has lost its rightmost branch. As predicted, onsets are not Compensatory Lengthening and Gemination 65

subject to that repair strategy; Resyllabification must apply instead. The most important implication of this analysis, however, comes from the assumption that [+sonorantj consonants and [sJ can be part of the nucleus. Since Compensatory Lengthening is seen as a repair strategy to maintain a heavy syllable, it follows that a language which does not consider CVV syllables to be heavy will not permit sonorants into the nucleus. We have, then, a grammar of Ancient Greek which, in setting one parameter (that cvv is a heavy syllable), has determined the choice of another parameter, the condition in (5) (or rather the possibility to have that choice) and, by having condition (5), the choice of a phonological process, Compensatory Lengthening. The grammar of Lesbian and Thessalian, on the other hand, does not consider Cvv to be a heavy syllable<4>. In choosing this parameter, the grammar has eliminated the possibility of condition (5) being a part of it. The grammar has also, with that prior decision, determined the choice of Gemination as the repair strategy needed to keep a syllable heavy.

Notes

* Versions of this paper have been presented at the April, 1985 GLOW conference in Brussels and at the June, 1985 Canadian Linguistic Association meeting in Montreal. I received a McGill University travel grant to attend the GLOW conference, for which I am grateful. <1> Principle (10) Null elements may not appear in branching constituents, where constituent refers to the prosodic constituent immediately dominating the null element (Kaye & Lowenstamm, 1983: 9).

<2> Greek possesses two morphologically distinct . The more common is the sigmatic aorist, which is formed by the suffixation of , /s/, to the stem, followed by the suffixation of a set of person and number markers. The other, called the second aorist, involves a vowel mutation as well as affixation.

<3> CVV in the case of Compensatory Lengthening; eve in the case of Gemination. 66 Rodier

<4> See Ingria (1980).

References

Bouchard, D. (1982) ' without a rule of Nasalization'. Unpublished MS. MIT.

Halle, M. and J.R. Vergnaud (1980) 'Three Dimensional Phonology'. Journal of Linguistic Research I, 83-105.

Ingria, R. (1980) 'Compensatory Lengthening as a Metrical Phenomenon'. Linguistic Inquiry 11, 465-495.

Kaye, J.D. and J. Lowenstamm (1981) 'De la Syllabicite'. Unpublished MS. Universite du Quebec a Montreal.

Kaye, J.D. and J. Lowenstamm (1982) 'Syllable Structure and Markedness Theory'. Proceedings of the 4th GLOW Conference. Pisa, Italy.

Kaye, J.D. and J. Lowenstamm (1983) 'Compensatory Lengthening in Tiberian Hebrew'. Unpublished MS. Universite du Quebec a Montreal.

McCarthy, J. (1979) 'On and Syllabification'. Linguistic Inquiry 8, 752-763.

McCarthy, J. (1981) 'A Prosodic Theory of Nonconcatenative Morphology', Linguistic Inquiry 12, 373-418.

Piggott, G.L. (1981) 'On the Status of Rules of Insertion and Deletion'. Presented to Canadian Linguistic Association, Dalhousie University, Halifax, Nova Scotia.

Piggott, G.L. and R. Singh (1984) 'The Empty Node in Phonology: An analysis of Epenthesis'. McGill Working Papers in Linguistics, Vol. 1, No.2, 64-109.

Singh, R. (1980) 'Old French Epenthesis and Syllabic Structure'. Canadian Journal of Linguistics 25, 226-230.

Singh, R. (1981a) 'Free at Last'. Paper presented at the 1981 Annual Meeting of the Canadian Linguistic Association, Halifax, N.S., May 1981. Compensatory Lengthening and Gemination 67

Singh, R. (198lb) 'The English Negative Prefix in-'. Recherches Linguistiques a Montreal 17, 139-143.

l Singh, R. (1981c) 'The English • Recherches Linguistigues a Montreal 17, 145-148.

Dans certaines langues, les procedes phonologiques de I'allongement compensatoire et de la gemination peuvent etre consideres comme des strategies de reparation. Dans certains cas, lorsqu'une consonne est effacee deux resultats s~nt possibles: ou bien la voyelle precedant Ie segment qui a ete efface est allongee (allongement compensatoire) ou bien Ie resultat est une consonne adjacente allongee (gemination). Le but de cette analyse du grec ancien est de discuter ces deux procedes apparament differents et de demontrer qu'ils s~nt, en fait, un meme phenomene; les differents effects sur la syllabe venant de l'action de certains principes de syllabification et de 1a mise en position du parametre indiquant quel type de syllabe est considere 1ourd.