EOTV¨ OS¨ LORAND´ TUDOMANYEGYETEM´ BOLCS¨ ESZETTUDOM´ ANYI´ KAR
Functional phonological analysis of the Hungarian vowel system
A magyar maganhangz´ orendszer´ funkcionalis´ fonologiai´ alapu´ elemzese´
SZAKDOLGOZAT / MASTERS THESIS
Temavezet´ o/Supervisor:˝ Peter´ Rebrus tud. fomts.˝
Kesz´ ´ıtette/Author: Daniel´ Szeredi
Theoretical Linguistics Major
Elmeleti´ nyelveszet´ szak
Budapest, 2009. NYILATKOZAT
Alul´ırott kijelentem, hogy a jelen dolgozat sajat,´ on¨ all´ o´ munkam´ eredmenye.´ Nem hasznal-´ tam fel elkesz´ ´ıtes´ ehez´ semmifele´ olyan forrast,´ beleertve´ az internetet is, amelyet nem tuntettem¨ fel a megadott bibliografi´ aban.´ Az esetleges forrasmegjel´ ol¨ es´ nelk´ ul¨ valo´ atv´ etelekkel´ kap- csolatban tudomasul´ veszem, hogy azok a szerzoi˝ jog megsert´ es´ et´ jelentik. Budapest, 2009. aprilis´ 15.
CERTIFICATE OF RESEARCH
By my signature below, I certify that my ELTE M.A. thesis, entitled Functional phonological analysis of the Hungarian vowel system is entirely the result of my own work, and that no degree has previously been conferred upon me for this work. In my thesis I have cited all the sources (printed, electronic or oral) I have used faithfully and have always indicated their origin. Date: 15 April 2009 Contents
1 Introduction 1
2 Hungarian vowel system 2
2.1 Earlierresearchoncentralization ...... 5 2.2 Earlier research on frequency effects on vowel reduction ...... 6 2.3 Earlier research on secondary stress in Hungarian ...... 7
3 The experiment on vowel reduction 9
3.1 Settings ...... 9 3.2 Results...... 10
3.2.1 The phoneme /E/ ...... 11 3.2.2 Thephoneme/o/ ...... 12
3.2.3 The phoneme /6/ ...... 13 3.3 Summaryofresults ...... 14
4 The experiment on frequency effects 16 4.1 Settings ...... 16 4.2 Results...... 17
4.2.1 Averagestatistics ...... 17 4.2.2 Vowelreduction ...... 18
4.2.3 Duration and intensity ...... 19 4.3 Summaryofresults ...... 19
5 The experiment on secondary stress 20 5.1 Settings ...... 20
5.2 Results...... 21 5.2.1 Fundamentalfrequency...... 21
5.2.2 Duration ...... 23 5.2.3 Vowelquality...... 25 5.2.4 Intensity ...... 26
5.3 Summaryofresults ...... 28
6 Analysis 30 6.1 Theoretical relevance of subphonemic phenomena ...... 30
6.1.1 StressinHungarian...... 30 6.1.2 Diachronic arguments in phonology ...... 31 6.2 Vowelreduction...... 32
6.2.1 StandardOT ...... 33 6.2.2 Functional grounding – frequency effects ...... 35
6.2.3 Phonetic grounding – Dispersion Theory ...... 37 6.2.4 Phonetic grounding – Dispersion-Focalization Theory ...... 38 6.2.5 Representation of phonetic facts ...... 39
6.2.6 ElementTheory...... 41 6.2.7 Predictions of the ET-DFT framework ...... 43
6.3 Stresspattern ...... 44 6.3.1 Metrical Stress Theory – trochaic analysis ...... 44
6.3.2 Metrical Stress Theory – Szinnyei-type analysis ...... 46 6.3.3 Metrical Stress Theory – unbounded analysis ...... 48 6.3.4 Diachronic evolution ...... 49
6.3.5 Theoreticalconsequences ...... 52
7 Summary 52 7.1 Claimsandconsequences...... 52 7.2 Furtherresearch...... 53 ... Medborgare om etthundra ar˚ finns du ej langre¨ till ... Cornelis Vreeswijk Acknowledgements
I would like to thank for the help and guidance of my supervisor, Peter´ Rebrus, and prac- tically everyone in the Department, particularly for the ‘wise men’ in phonology: Peter´ Siptar´ and Miklos´ Torkenczy¨ for their support and advices through these long years and the ‘young stars’ Zsuzsanna Bark´ anyi,´ Tamas´ B´ıro,´ Zoltan´ Kiss for, well ... pretty much the same, and obviously for Kinga Gardai´ without whom we would be nothing. My international acknowledgements go to Maria Gouskova, John Harris, Patrick Honeybone, Shinichiro Ishi- hara, Jonathan Kaye, Mary Pearce, Anthi Revithiadou, Curt Rice, and Jochen Trommer for every word they exchanged with me and for the conversations that have led me to understand phonology better, and of course the Tromsøites for bringing the laughs along. I thank Barıs¸ Kabak, Katalin E.´ Kiss, Balazs´ Suranyi,´ Istvan´ Szakadat,´ Tamas´ Varadi´ and all the MOKK people for trusting me and giving me the chance to work in linguistic projects that widened my point of view. I want to particularly thank for the organizers of the EGG Summer School, where I had the chance to meet many persons who have inspired me. Coming back home, I have to bow my head for my professional and psychological mentor, Sylvia Blaho, for all the advice, whether they’d helped me or not. I greet my fellow classmate phonologists: Zsofia´
Gyarmathy, Marton´ Soskuthy´ and Peter´ Racz´ for helping create the (just slightly competi- tive) atmosphere among students that phonology, and our way of phonology is the coolest thing out here. Finally, but maybe most importantly I would like to thank my family for their patience and full-scale support and Brigi for being here. 1 Introduction
Hungarian has been a much discussed language in the literature due to its unique features among European languages, but there are several claims that have been assumed as valid for Hungarian without clearly checking the validity of the information.
For example, in Metrical Theory Hungarian is claimed to have a very basic left-to-right trochaic foot assignment, without extrametricality and with the main stress on the leftmost foot (Hayes 1995: 330). The examples of the sources cited by Hayes are however suspicious to a native Hungarian speaker therefore even the principles of Hungarian stress are yet to be made clear. A concurrent experiment on this issue presented below reveals, however, that stress in Hungarian as a syllable-timed language has a lesser role than in stress-timed languages like English, and it depends more on discourse-level factors.
An other assumption made on Hungarian closely related to the controversial stress pattern of the language is that there is scarce centralization or reduction of (unstressed) vowels, as it is usual in syllable-timed languages (Roach 1982). Whereas this seems to be true phonemically, the extent of phonetic centralization has been scarcely researched. A commonplace claim is that there is no, or at most very weak vowel reduction in Hungarian. The situation however seems to be more complicated. In section 2 I will briefly summarize the canonized description of the vowel inventory of
Hungarian, and in section 3 I will describe the experiment conducted to see if vowel reduction exists in Hungarian. In the next two sections I will summarize the other two experiments set to investigate issues that became important during my research into the topic: lexical frequency effects that might, or might not affect the degree of centralization in section 4, and the (non-existence of) secondary stress in Hungarian in section 5.
In section 6 I will show the process of finding a probably adequate phonetically grounded theoretical background to describe subphonemic processes that will have been found in this
1 [-back] [+back] [-round] [+round] [-round] [+round] high /i/ /i:/ /y/ /y:/ /u/ /u:/ mid /e:/ /ø/ /ø:/ /o/ /o:/ low /E/ /a:/ /6/
Table 2.1: Vowel inventory of Standard Hungarian
thesis. In the subsection 6.1 I will argue for the phonological importance of examination of subphonemic phenomena like vowel reduction in Hungarian, and the importance of the theoretical background to discuss the source and the effects of these processes, as many have questioned if it is even worthy studying these kinds of processes and dismissed them as phenomena out of the reach of phonology.
2 Hungarian vowel system
Descriptions of Hungarian vowels in the standard dialect elicit a system that might first look symmetrical as it contains 7 short and 7 long vowels, with the symmetricity also supported by the orthography that clearly sets up pairs of short and long vowels: :
:
2 There are several phonological alternations that are used in morphological processes. The most important ones are:
• vowel harmony: /6/:/E/, /a:/:/e:/, /o/:/E/:/ø/, /o:/:/ø:/, /u(:)/:/y(:)/
• length alternation: /6/:/a:/, /E/:/e:/, /o/:/o:/, /ø/:/ø:/, /u/:/u:/, /i/:/i:/, /y/:/y:/
• some other lexical alternations: /ov/∼/6v/∼/6j/:/o:/, /Ev/∼/øv/∼/Ej/:/ø:/, /o/:/6/ and
/ø/:/E/ (in suffixes), etc.
As it’s been pointed out above, orthography and the phonological length alternation system makes the 7+7 vowel system look perfectly symmetrical. This symmetry can be doubted however if one looks at the greater phonetic distance between phonologically short-long pairs of low vowels, and the fact that /E/ appears in two suffixal sets in vowel harmony. It appears in the ‘low’ series, alternating with back /6/, eg. in the illative (and colloquial inessive) suffix [b6]∼[bE]: [6 ha:zb6] ‘in/to the house’, [6 kErdbE] ‘in/to the garden’, [6 gødørbE] ‘in/to the pothole’. However it also appears in the ‘mid’ series, where it alternates with back /o/ and, after front rounded vowels /ø/, eg. in the allative suffix [hoz]∼[hEz]∼[høz]: [6 ha:shoz] ‘to the house’, [6 kErthEz] ‘to the garden’, [6 gødørhøz] ‘to the pothole’.
The explanation for the special role of the vowel /E/ can be seen when taking non-standard (Kiss 2001) and older varieties of Hungarian (E. Abaffy 2004a: 346) into account, as it can be seen that this phoneme is the merger of an earlier low /æ/ and an earlier mid /e/ phoneme. This yields an unsymmetrical 8+7 system, which is found in many Hungarian dialects. Some archaic varieties and Old Hungarian had a symmetric 8+8 system however, with the long /æ:/ differentiated from the long /e:/, and (showing the most archaic pattern) without the labializa- tion of /6/ (E. Abaffy 2004a). This system, found in the most archaic Paloc´ (Northern) dialects (marked with a P when different) and the Hussite orthography (marked with an H to show the different marking of front low-mid vowels) of the 14th century (Korompay 2004: 295) can be
3 [-back] [+back] [-round] [+round] [-round] [+round] high /i/ /i:/ /y/ /y:/ /u/ /u:/ mid /e/ H
Table 2.2: Vowel inventory of Old Hungarian and archaic Paloc´ dialects
seen in Table 2.2.
It can be seen that the present vowel system is the result of some diachronic processes resulting in a less symmetric vowel system. It must be noted however that as such, it is also the source of diachronic processes that will or even can now be seen in some registers. Length distinction of high vowels is quite vague in colloquial Hungarian (Siptar´ and Torken-¨ czy 2000, Pintzuk et al. 1995) and absent in western dialects (Kiss 2001), thus resulting in a 7+4 or 8+4 system. On the other hand, new long vowels are emerging from lengthening before /l/, /r/, /j/ and loanword non-phonemic [w] in codas, where these consonants may be dropped, a phenomenon originating from eastern dialects. These new long vowels have exactly the same quality as the original short vowel, therefore giving emergence to contrasts between arra [6:r6] ‘that way’ and ara [6r6] ‘bride’ on one hand and hold [hofl:d] ‘Moon’ and hod´ [hofi:d] ‘beaver’ on the other hand. The conclusion is that at present, the vowel system in colloquial standard Hungarian is not a stable system, it is in the middle of diachronical processes and shows the effects of dialectal varieties. This might call for the examination of a process yet scarcely researched: vowel reduction.
4 2.1 Earlier research on centralization
As it can be seen, phonological alternations in Hungarian are in their true sense ‘lexical’ processes as they are structure-preserving. There is no alternation pattern that needs to be defined with a reference to (the lack of) stress, or any kind of reduction or centralization.
Therefore if some kind of phonetic reduction shows up in spoken Hungarian, it must be described as a subphonemic process. Textbooks for foreigners usually also contain the assertion that there is no reduction in
Hungarian, moreover, they usually claim that every Hungarian vowel is pronounced distinctly and clearly, which should mean they are not to be pronounced in a reduced or centralized form, eg. ‘Die Betonung des Wortes liegt immer auf der ersten Silbe. Die Vokale werden aber auch in allen folgenden Silben voll und klar ausgesprochen.’ [The stress of the word always falls on the first syllable. But vowels are pronounced full and clear in every following sylla- ble.] (Ginter and Tarnoi´ 1974/1993: 11). Prescriptivist tradition also holds that distinct clear pronounciation of sounds of speech is typical for Hungarian, contrary to other languages, and
‘blurring’ segments is a maleficent effect of modern life and these other languages (eg. Benko˝ 1992).
However, reduced or centralized vowels do occur on the surface, but their distribution and frequency is still an issue that has been lightly researched (Acs´ and Siptar´ 2001). de Graaf (1987) confirmed in an auditory experiment with two speakers that vowels do centralize to- wards a target of a [@] in Hungarian, and the extent of centralization is dependent on its context: free vowels are less reduced than vowels in isolated words and much less reduced than vowels in context. Gosy´ (1997) has made an experiment where continuous speech of a male speaker with average voice parameters and no speech defects was recorded. Then several vowels were cut from the signal and these short isolated sound segments were played back to 10 participants, who were asked to transcribe the sound they had heard in Hungarian orthography. If they
5 categorized any vowel that is not underlying /ø/ as
2.2 Earlier research on frequency effects on vowel reduction
Recent years have seen a renewed interest in the effect use has on language. Proponents of usage-based functionalism (Bybee 2001; Pierrehumbert 2001) claim that language phenom- ena gain additional explanation if viewed from a perspective of use.
Generative theories regard reduction as a phonetic change, having an even effect on all lexical items (Kiparsky 1985). Usage-based functionalists, however, argue that an item’s representation is a set of the richly detailed exemplars of its use, so since reduction occurs during a word’s use, if it is used more, reduction will be larger. Jurafsky et al. (2001) discuss a dataset of content words ending in t/d in American English. They find that high frequency words are significantly shorter, and final segment deletion also correlates with token fre- quency. Hooper (1976) discusses correlations of frequency and temporal reduction in a study on schwa-reduction and deletion in American English with similar results. The experiment on vowel reduction that will be presented in section 3 have shown that very frequent function words like articles (definite a, az and indefinite egy) and the comple- mentizer hogy are reduced the most as they receive the least stress. The question is whether
6 their reduction follows from solely their lexical class or also from their high frequency.
2.3 Earlier research on secondary stress in Hungarian
While it is generally accepted that Hungarian has initial primary stress, the secondary stress pattern of the language – as noted above – has been subject to considerable debate. This de- bate is summarized shortly in van der Hulst et al. (1999: 452) and in Hayes (1995: 330), who comes to the point that different analyses may be available because of dialectal differences. The most widely cited view, originating from Kerek (1971) and further back to Balassa
(1890), is that the language displays a very basic quantity-insensitive left-to-right trochaic foot assignment without quantity sensitivity and extrametricality, i.e., that secondary stress falls on every odd-numbered syllable.1 Conversely, Szinnyei (1912: 12) argues for a quantity- sensitive pattern: ‘Der Hauptton fallt¨ in einem alleinstehenden Worte auf der ersten Silbe. In langeren¨ Wortern¨ fallt¨ außerdem ein Nebenton auf die dritte und funfte¨ oder (wenn die dritte kurz ist) auf die vierte und sechste [Silben]’ [Main stress falls in isolated words on the first syllable. In longer words, there is a secondary stress on the third and fifth or (if the third one is light) on the fourth and sixth syllables]. This assumption is taken by Lotz (1939) as well, and this pattern and analysis will be called the Szinnyei-type pattern and the Szinnyei analysis from now on. The most complex analysis involves not only two, but three levels of stress in Hungarian. Hammond (1987) uses an intermediate abstract unit called ‘colon’, which consists of two feet, but is below the word layer. In his analysis, primary stress falls on the first syllable of the word, secondary stresses fall on the first syllables of each following colon, and the left syllables of feet not heading a colon get tertiary stress. His examples in his notation where primary stress is marked by an acute, secondary by a circumflex, and tertiary by a grave accent
1Hayes (1995) cites Kerek (1971) calling for a righ-to-left Latin type trochaic analysis for Hungarian. This analysis would be the oddest of all, but it should be noted that Kerek uses arguments from verse, which is much influenced by classical poetry in Hungarian.
7 include [k´ıˇsku:nf`e:lEéhˆa:za:b`On] ‘in Kiskunf´ elegyh´ aza’´ (sic!), [m´EgvEst`EgEthˆEtEtl`EnEknˆEk] ‘to those unbribable’. These views, however, lack both phonetic and phonological evidence to support them: the claims have been made without prior articulatory or acoustic measurements, and they also lack support from phonological processes like stress-sensitive vowel lengthening or shorten- ing, or foot-initial strengthening or medial weakening of consonants. Moreover, an impres- sionistic survey of native speaker judgements raises doubts regarding both accounts: speakers reject secondary patterns as well as both quantity-sensitive and quantity-insensitive tertiary footing. Hammond’s examples are the oddest of all with the tertiary account, where secondary and tertiary stress can fall on word-final syllables, that cannot happen according to the other traditions. The examples above show that secondary stress can fall on inflectional suffixes as well in his account which seems quite implausible. His article contains orthographic errors and examples with words that are close to be nonsensical, and are used mainly in jocular context which may divert their pronunciation quite much like [´Elka:p`osta:ˇsˆı:tott`OlOnˆı:tott`a:tok] ‘you.PL have decabbagised it’. The idea that the secondary stress pattern is described in the literature incorrectly ac- cording to native intuition or even measurements is not unique to Hungarian. Arantes and
Barbosa (2008) have done research on Brasilian Portuguese, which is also described to have (right-to-left) trochaic foot pattern, and have found that there is no secondary stress in their measurements, phonetic cues of stress usually increase until the lexically determined primary stress, and vowels after the primary stress receive much less stress, as it can be seen on Figure
2.1. While presenting this idea in the 6th OCP in Edinburgh with Sylvia Blaho, we received many interesting comments from researchers native in different languages, many of whom claimed that the usual description of secondary stress pattern of their native language is also
8 Figure 2.1: F1 values for the Brasilian Portuguese [a Zarata"kaka] (Arantes and Barbosa 2008)
dubious for their intuition. This might also show that a falsifiable experimental account for secondary stress in languages is very much needed as data on which Metrical Theory and other stress theories have been built upon seems to be unvalidated even for much-researched languages, therefore it is hard to trust data cited in these theories for lesser known languages that are of great importance for them.
3 The experiment on vowel reduction
3.1 Settings
Seven native Hungarian speakers took part in this experiment2. Fourteen (including two filler) sentences were recorded from each participant six times in a row, so for the later sessions the register they used was more casual. The aim of the experiment was to investigate four envi-
2The results of this experiments were presented in the Mokus´ Conference and published as Szeredi (2008).
9 ronments under three different stress conditions. The four environments were (z) [t6z], (a)
[S6k], (o) [hoé] and (e) [kEé]; and the vowels in these sequences were either (1) having pri- mary stress – ie. first syllable of a content word, (2) unstressed – last syllable of a content word (usually in an odd-numbered syllable to avoid possible secondary stress related effects) and (0) unstressed in functional words. Afterwards 12 words were recorded from each par- ticipant, asking them to read these words slowly and carefully to attain the formants for slow careful speech vowels (g). The hypothesis before the experiment was that centralization would be found in casual natural speech, affecting both the F1 and F2 formants, therefore converging all vowels to- wards a target of a [@]-like quality. This seemed to be natural as the vowel reduction was presumed to be non-phonological, therefore any preservation of any phonological features had seemed to be unlikely. Nevertheless, it was expected that the pattern of centralization would resemble the way this kind of process works in other languages, so some kind of the- oretical model could explain the results, and this model could make some prediction on the behavior of other Hungarian vowels. It was also hypothesized that the rate of the centraliza- tion would correlate with stress: the more stressed the vowel, the less centralized it would be.
3.2 Results
The recorded data were analyzed using the Praat software (Boersma and Weenink 2008).
The results have shown that it is clear that Hungarian speakers tend to centralize vowels in certain positions even if they read the sentences from paper. It can be seen however that different environment do not behave in the same way.
10 Figure 3.1: Average positions of vowels in environments with phoneme /E/
3.2.1 The phoneme /E/
Figure 3.1 shows the average position of the careful speech vowels and the average position of the vowels in the (e1), (e2) and (e0) environments in faster, more casual speech.
It can be seen that the second formant of the /E/ phoneme does not show centralization. However, this can be attributed to the effect of the neighboring palatal /é/ consonant, which can hinder the velarization of the preceding vowel.
The first formant of the vowel in this environment shows centralization: its is almost always higher than the F1 of the vowel in careful speech. The only exception is the fifth speaker, who read the (eg) environment as a mid-high [e], either because of emphatic or di- alectal pronounciation, so his (e1), (e2) and (e0) vowels are more open and more centralized.
However, these environments were pronounced in the same way as they were by other speak- ers. Another oddity is that the (e0) variable of the third speaker is very open, pronounced as
[æ], however this vowel is very short, about 20 ms long. In other records of the speech of this speaker, the (e0) is mid-high, like the (e0) of other speakers, so this phenomenon can be attributed to the fact that the formants show the impact of neighboring consonants on a very
11 Figure 3.2: Average positions of vowels in environments with phoneme /o/
short vowel that is almost elided. The hypothesis on the graduality of centralization corresponding with (the lack of) stress doesn’t seem to be correct in the case of /E/, as the realizations of the vowel in this environ- ment are approximately in the same place in the dimensions of the first three formants, so all of them is pronounced in casual speech as [efl]. This result is odd looking at the analysis of de Graaf (1987) as well, who finds the short [E] to reduce its F2 formant.
3.2.2 The phoneme /o/
As it can be seen in Figure 3.2 the realizations of the phoneme /o/ behave in the opposite way than the realizations of /E/. The first formant does not undergo centralization in casual speech. This is quite easily explained as the careful speech (og) environment usually has a mid, or centralized F1 itself, so no more centralization is possible. However, the second formant shows significant centralization, and this process is indeed related to the lack of stress as it was supposed above, so the stressed (o1) environment usually has a low F2 but (o2) is more centralized and (o3) in the function word hogy ’that (complementizer)’ has a quality quite
12 variables F1 F2 F3
a0-z0 71 Hz -15.3 Hz -206.8 Hz
a2-z2 153.8 Hz 29.7 Hz -182.8 Hz
a1-z1 0 Hz 120.3 Hz -191.5 Hz
a-z 75.1 Hz 44.9 Hz -193.7 Hz
ag-zg 26.3 Hz 60.3 Hz -167.7 Hz
Table 3.1: Differences between formants of vowels in [S6k] (a) and [t6z] (z) environments
close to [@]. The environment of this vowel in the experiment is also a palatal /é/, thus it is possible that the fronting seen is affected by this consonant. Then an explanation for the graduality of centralization could be that the less stressed the vowel is, the shorter it is, thus the neigh- boring palatal consonant has a greater effect on its formants. However, as the environments containing underlying /6/ do not stand beside palatal consonants but still show centralization, it seems that the effect of /é/ cannot be the only factor in this fronting process.
3.2.3 The phoneme /6/
The /6/ phoneme shows the most typical schwa-oriented centralization, as both of its for- mants undergo this process in less stressed syllables. This phoneme was examined in two environments to see the extent of centralization in both versions of the definite article (/6/ and
/6z/). The average difference between parameters of the vowels is shown in Table 3.1, and can be seen on Figures 3.3 and 3.4 as well.
It can be seen that there is a significant difference between the value of the third formant of these two vowels, as it is much higher in the case of the [t6z] environment, nearing the
13 Figure 3.3: Average positions of vowels in environments with phoneme /6/ in the dimensions of F1 and F2
value of non-rounded vowels. Whether the effect of the [t z] environment is in fact a loss of rounding is not sure, but if that would be the case it would probably lead to the raising of the second formant as well but rather the opposite fact seems to be true in the data. Lowering of F1 of the vowel (that is raising the vowel) in the (z) environment also seemed to be much more prevalent, and reduction in vowel length is also stronger for the (z) environment. This reduction was so strong that some speakers had only one or two records out of six with an observable vowel segment in the (z0) environment.
3.3 Summary of results
This experiment has shown that subphonemic vowel reduction in Hungarian is an existing phenomenon and it has to be taken into account seriously. Table 3.2 shows the average quality of the vowels in the discussed environments. Having summarized the results the two hypotheses seemed to be right in many circumstances:
• The /o/ and /6/ phonemes were centralized in casual speech towards [@], and realiza-
14 Figure 3.4: Average positions of vowels in environments with phoneme /6/ in the dimensions of F1 and F3 (a) (z) (e) (o) (g) [6fi] [6fi] [E] [o] (1) [5»] [5] [efl] [off] (2) [5] [@] [e] [@] » fi fl ¯ (0) [@»] [@fi] [efl] [@]
Table 3.2: Average vowel quality in examined environments
tions of the /E/ phoneme had a slightly centralized F1 formant.
• The extent of centralization depended on the stress in the case of the /o/ and /6/ phonemes.
This pattern of centralization is shown in the dimensions of F1 and F2 on Figure 3.5.
15 Figure 3.5: Tendencies of centralization for each environment
4 The experiment on frequency effects
4.1 Settings
The experiment conducted in section 3 has shown that function words are reduced the most and it has been also seen that they receive the least stress.3 This fact might be accounted for only by referring to the lack of stress that is the essential property of function words. There may be an other explanation, however, that strengthens the factors for reduction or centralization: the frequency of these words is very high. This experiment was conducted to see if frequency effect shows up for content words in Hungarian.
The usage-based null hypothesis is that frequency effect should be seen for content words as well. Yet, the results seem to go against these predictions. In the study four native speakers were asked to read sentences with two environments ([tok] and [t6k]) that occured inside words of different frequencies according to the Szoszablya´ webcorpus (Halacsy´ et al. 2004). The examined words appeared in the same syntactic position and the investigated vowels
3The core of this chapter is an abstract written with Peter´ Racz´ that has been accepted to the 17th Manchester Phonology Meeting, see Racz´ and Szeredi (2009)
16 Word freq. F1 F2 dur. int. p6t6kot (‘brook.ACC’) 16256 569 1350 48 58 cs6t6kot (‘swelter.ACC’) 102 535 1331 37 57
Table 4.1: Mean phonetic values of the environment [t6k] in word with different frequency
Word freq. F1 F2 dur. int. titok (‘secret’) 55430 519 1147 79 64 ŃolStok (‘folding ruler’) 26 500 1317 39 56
Table 4.2: Mean phonetic values of the environment [tok] in word with different frequency
were all in the second syllable to ensure that vowel reduction has an effect on them. Each word was placed in the same syntactic and prosodic environment to make sure there are no external effects. The Praat software was used again for the phonetic analysis.
4.2 Results
4.2.1 Average statistics
Table 4.1 shows the statistics of the [t6k] environment with the lemma frequencies from the corpus and phonetic reduction cues (the average formant values in Hertz, average duration in milliseconds and average intensity in decibels). As it can be seen, if there is correlation of the frequency of the given words and their reduction in this environment, it is the other way round than expected, as the vowel in the less frequent form is longer. The next set, with the [tok] environment is even more problematic, as it can be seen in the table 4.2.
17 Figure 4.1: Average place of vowels in test words in the F1-F2 plane
Figure 4.2: Place of the vowel [o] in test words in the F1-F2 plane for each elicitation
4.2.2 Vowel reduction
Figure 4.1 shows the average positions of each word for every speaker of every word in the
F1-F2 plane. It can be seen here as well that the [o] of the frequent word titok is remarkably less reduced for every speaker than the vowel of the much less word colstok. There is no significant difference for the [6] set either. To see how consistently less reduced the [o] vowel of titok is, see Figure 4.2 that shows every elicitacion of the [tok] set of test words on the F1-F2 plane.
18 duration (ms.) speaker titok colstok [tok] difference patak csatak [t6k] difference 1 51 46.025 4.975 43.25 37.275 5.975 2 153.125 57.075 96.05 53.25 45.675 7.575 3 62.55 26.325 36.225 57.65 41.3 16.35 4 88.9 24.4 64.5 38.625 26.925 11.7 intensity (dB) speaker titok colstok [tok] difference patak csatak [t6k] difference 1 75.775 71.325 4.45 68.775 66.45 2.325 2 70.625 62.95 7.675 62 59.525 2.475 3 61.275 48.45 12.825 54.975 56.6 −1.625 4 69.225 39.45 29.775 46.725 46 0.725
Table 4.3: Average duration and intensity for test words for each speaker
4.2.3 Duration and intensity
Not only vowel reduction shows the anti-frequency based effect, but vowel length and inten- sity as well, and the unexpected result is again consistent for all speakers as it can be seen in table 4.3, with the difference calculated in the direction contrary to the hypothesis: a positive number means the less frequent word being less prominent. The only speaker and pair where this ‘wrong’ correlation is not shown is speaker 3 with the [t6k] set of words.
4.3 Summary of results
It has been seen that the results of this experiment have totally and consistently contradicted the hypotheses. The reason for this and that it is seen for every speaker and almost every elic- itation, not only for the gross average is definitely going against findings in other languages, and needs to be researched. These results could mean that Hungarian vowel reduction is dif- ferent in nature from the above examples or could suggest that the correlation of frequency of use and sound change is more subtle than it would be expected by usage-based functionalism. The explanation that test words were not well chosen is still left on the table.
19 5 The experiment on secondary stress
5.1 Settings
To test the controversial claims on Hungarian secondary stress, a study examining 4 native Hungarian speakers both in casual and in formal careful speech has been conducted.4
Raphael et al. (2007: 232) elicits three main cues for stress: fundamental frequency (F0), vowel duration and peak intensity. F0 is determined in Hungarian by the information structure of the sentence (Varga 2002), while vowel duration is only influenced by the lexical length contrast (see section 2) and other local factors like the quality and quantity (as geminates and consonant clusters are quite common in Hungarian) of neighboring consonants. That leaves intensity as the only reliable cue in Hungarian for determining stress. Therefore in the experiment, intensity has been investigated the most, but to get more ver- ification for the above claims about the problems with fundamental frequency and duration, these factors were checked out as well. The participants were asked to read five sentences three times in a casual setting, thus acquiring the three casual sets; and afterwards they were instructed to read the list twice as if they were in a formal situation on television, thus getting two careful speech variants for each sentence. Praat (Boersma and Weenink 2008) was used for analysis here as well. Not all words that were included in the test are analysed here, only four of them will be taken into account as they may show the effects of most factors on stress, see Table 5.1. Four speakers were involved in the experiment, but it is hard and methodologically incorrect here to show averages for four speakers in given cases, so for illustration, usually data for one selected speaker will be shown.
4Parts of this chapter have been presented with Sylvia Blaho at the 6th OCP in Edinburgh, United Kingdom, see Szeredi and Blaho (2009)
20 orthography trochaic analysis Szinnyei analysis gloss and characteristics befektetok˝ ["bEfEktEtø:k] ["bEfEktEt()ø:k] ‘investor.PL’, effects of the verbal prefix, lexically long vowel in the end, iambic syllable weight pattern ( ` — ` —) felekezetieke´ ["fElEkEzEtijE()ke:] ["fElEkEzEtijEke:] ‘denomination.ADJ.PL.POSS’, eas- ily comparable as it contains many light syllables with [E] as nuclei, Hammond analysis may be tested: [f´ElEk`EzEtˆıjEk`e:] katolicizmus ["k6toliŃiz()muS] ["k6toliŃizmuS] ‘catolicism’, syllable weight would suggest a Szinnyei-type analysis: ` ` ` — ` kereskedelem ["kErESkEdE()lEm] ["kErESkEdElEm] ‘commerce’, mostly easily compa- rable light syllables with [E] as nu- clei (the word is followed by the word [e:S] ‘and’, therefore the last syllable is light in context as well)
Table 5.1: Words examined in the stress experiment
5.2 Results
5.2.1 Fundamental frequency
As Varga (2002) has claimed before, pitch in Hungarian largely marks information on syn- tactic, pragmatic or contextual levels, and is not very much reliable as a cue for phonological or subphonemic matters. The results show that there might be some correspondance of F0 with the supposed trochaic stress pattern of examined words, but pragmatic matters often overwrite the informa- tion given by the fundamental frequency. It would be quite hard to illustrate this point, and its full presentation would need more space than this paper could provide, but Figures 5.1 and
5.2 show some of the problems. Figure 5.1 shows the pitch of several elicitations (that could be used for illustration) of two speakers uttering the word [fElEkEzEtijEke:] in one picture. It
21 Figure 5.1: Fundamental frequency in elicitations of the word [fElEkEzEtijEke:]
can be seen that some pattern of trochaic footing can be seen as the second syllable of each foot has – by large – a bit lower pitch, but exceptions and alterations are numerous.
For the third supposed foot ([tijE]) of this pattern breaks up more as the pitch is many times raised for the second syllable, giving the word a pitch pattern usually seen for the final two syllables of yes/no questions (Bolinger 1989: 58). This question-like behavior is quite odd given that this word is not clause final and the clause that contains it is conditional: ... ez az iskola azoke´ a felekezetieke´ lehet, ... ‘this school may be a property of those from that denomination’. This intonation may indicate contrast with the next clause: ... mert hogy nem allami´ , az biztos ‘because that’s for sure, that it is not state property’. Speakers’ inter- pretation might have been contrastive once and non-contrastive the other time, thus resulting in unpredictable and chaotic pitch pattern seen for this word.
Looking at the same kind of figure for the casual speech elicitations of the word [k6toliŃizmuS] (Figure 5.2) it can be seen that the stress pattern does not show up in the fundamental fre- quency domain, as it seems to be the function of the pragmatic and syntactic environment of the word again. Usually there is a peak of pitch on the second syllable of the word which cannot be the effect of any described stress pattern, it seems to be a discourse-driven phe- nomenon again.
22 Figure 5.2: Fundamental frequency in casual speech elicitations of the word [k6toliŃizmuS]
5.2.2 Duration
As expected, vowel duration is mainly influenced by the phonological contrast between long and short vowels, and the consonantal environment. Figure 5.3 shows the duration of vow- els in the test word [bEfEktEtø:k] ‘investor.PL’ for a speaker’s all readings, both casual and careful speech with the means. It can be seen that the first vowel is quite long, which might derive from primary stress, but it is a nucleus of a light syllable as well. The second vowel is the shortest, which can be attributed to its voiceless environment and the fact that it is a lexically short vowel in a heavy syllable. The third vowel is longer as it is in an open syllable again and the lexically (obligatorily) long [ø:] is about equally long as the first vowel of the word.
The durational analysis of the word [fElEkEzEtijEke:] shows even more how unreliable vowel duration is in Hungarian to analyze stress or to use it as a cue for any phonetic analysis on vowels. As it can be seen on Figure 5.4, each utterance has different duration patterns, and the mean reveals other problems as well: the only lexically long vowel of the word is the shortest, and there is no effect of primary stress on the first [E] sound, but the fourth one is definitely the longest, which may be an effect of a Szinnyei-type secondary stress or the consonantal environment.
23 Figure 5.3: Duration of vowels for a speaker in the word [bEfEktEtø:k]
Figure 5.4: Duration of vowels for a speaker in the word [fElEkEzEtijEke:]
24 Figure 5.5: F2 of vowels for a speaker in the word [kErESkEdElEm]
5.2.3 Vowel quality
The answer to the question if vowel quality can help determine the place of stress is definitely positive, as vowel reduction is working in syllables receiving lesser prominence. As it was seen in Figure 2.1, Arantes and Barbosa (2008) could use F1 reduction in their study of Brazilian Portuguese stress.
In the experiment above in Section 3, however, it could be seen that stress is only one factor determining the extent of vowel reduction in Hungarian, therefore claims based on vowel quality need to be taken carefully. As we have seen, the vowel /E/ does not reduce its
F2 value, but for the word [kErESkEdElEm] F2 values seem to correlate with the Szinnyei-type proposed stress system, although the (supposed) secondary stress is very weak here as well, as it can be seen on Figure 5.5 (the higher the second formant is, the vowel is less centralized and supposed to be more prominent).
The first four /E/ vowels of the word [fElEkEzEtijEke:] shows, however, no reasonable traces even of the primary stress, suggesting that vowel quality here is affected by the neigh- boring consonantal environment much more than prominence, as it can be seen on Figure 5.6.
25 Figure 5.6: F2 of the first four [E] vowels for a speaker in the word [fElEkEzEtijEke:]
5.2.4 Intensity
As it’s been noted, intensity has been found to be the best indicator of stress in Hungarian.
Having eliminated influences of clause structure, words consisting entirely of light syllables usually do not seem to have a secondary stress at all: the first syllable has the highest intensity and the intensity peaks on the following vowels decrease gradually. This pattern of downstep with no or very little sign of some vowels being more prominent than others can be seen for the four-syllable word [bEfEktEtø:k] in Figure 5.7.
The word [k6toliŃizmuS], which is supposed to be a good example for the Szinnyei-type stress pattern with a light third syllable therefore with a secondary stress on the fourth shows the very same downstep pattern as it can be seen on Figure 5.8. The fourth syllable has actually the lowest intensity in the word, with the final [u] being a bit louder on the average, but it can be seen that there is a reading where the speaker almost dropped the fourth vowel
[i] thus affecting the mean intensity value. In the results it is noticeable that intensity does not seem to correlate with syllable weight.
26 Figure 5.7: Intensity of vowels for a speaker in the word [bEfEktEtø:k]
Figure 5.8: Intensity of vowels for a speaker in the word [k6toliŃizmuS]
27 Figure 5.9: Intensity of vowels for a speaker in the word [kErESkEdElEm]
Where there is a slight increase somewhere in the intesinty downstep pattern, it is in the position where Szinnyei (1912) predicts it, as it can be seen for the word [kErESkEdElEm] in Figure 5.9, where the fourth syllable resists the downstep and is louder somewhat than the former syllable. The word [fElEkEzEtijEke:] shows the same pattern, but somewhat curiously, the intensity of the second syllable is higher than the first syllable, showing some kind of clash for the first ‘foot’, as it can be seen in Figure 5.10.
5.3 Summary of results
This experiment has proved that it is not easy to get unambiguous results for the different cues of stress when investigating a syllable-timed language with no phonological constraints on unstressed syllables like Hungarian. However the experiment has definitely falsified the most prevalent claim (henceforth fallacy) that Hungarian has a left-to-right trochaic footing with secondary stress on each odd-numbered syllable. The only sign of ‘foot’ if this abstract representational notion is meant to have a reflection on the surface at all can be seen on the
28 Figure 5.10: Intensity of vowels for a speaker in the word [fElEkEzEtijEke:]
fundamental frequency pattern, but Figure 5.1 has not been fully convincing at all.
Quantity-sensitivity is a harder issue: there have been no clear-cut signs for long vowels or closed syllables to attract some kind of prominence like secondary stress, but the Szinnyei (1912) analysis have shown to have some ground here: words with light third syllable tend to have some weak extra prominence on the fourth syllable, maybe to compensate for the lapse of the previous sequence of unstressed syllables.
What seems to be clear here is that the use of classical Metrical Theory analysis on Hun- garian has to be revised as the trochaic pattern assumed to appear seems to be a claim forced on Hungarian (and maybe other languages as well) for theoretical purposes. Experimental data have shown that sources used by textbooks in this theory are not checked, led by a combination of intuition and theoretical bias, and these sources and claims might need to be checked for every single language that are the cornerstones of theoretical analyses, even if they are well known and very much researched European languages.
29 6 Analysis
6.1 Theoretical relevance of subphonemic phenomena
Subphonemic processes have not been seen as very much relevant to phonology or phonologi- cal theory. There are reasons however that could give one the idea to work on subphonological data using phonological accounts developed to explain phonological phenomena. Looking at subphonemic vowel processes in Hungarian and vowel reduction or central- ization processes in general it will apparent at first glance that the target and the extent of centralization is not universal but seems to be affected by the quality of the ‘underlying’ vowel. In Hungarian, as it has been seen, front vowels do not centralize their F2 formant, which might bring forth the idea of language-specific factors like vowel harmony affecting the process. Language-specificity means that the issues arising here are phonologically rele- vant. There are more problems, however, that will lead to the same conclusion.
6.1.1 Stress in Hungarian
Discussing concurrent description of stress in Hungarian in Section 5, it could be seen that the different analyses from Szinnyei (1912) through Kerek (1971) to Hammond (1987) lack phonetic facts for their theory. For the later analyses and summaries (like Hayes 1995 and van der Hulst et al. 1999) it can be probably said that their pursue for explanatory adequacy results in a much lesser degree of observational adequacy (Chomsky 1965). This probably happens because the need for the description of data to meet some theoretical need has been more important than the precise description of stress patterns in a given language.
It is quite hard to determine the precise stress pattern in a language without measuring subphonemic data either in a perceptional or in an auditory study like the one in section 5. As it has been seen in subsection 5.2.3, when conducting a study like that, other phonetic issues, mostly non-phonemic have to be taken into account, like changes in vowel quality. This
30 means that if subphonemic vowel reduction in Hungarian exists (and it does, as seen in section
3), its examination could help investigating this issue as reduction is usually related to vowel duration and intensity. This would mean that subphonemic vowel reduction in Hungarian – even if it is ‘non-phonological’ – would help to decide on matters that are part of the core of a given phonological theory, ie. Metrical Theory and other theories of stress in this case.
6.1.2 Diachronic arguments in phonology
The second argument that could confirm that investigating subphonemic phenomena is worth for phonological theory comes from a lineage of research (eg. recently Evolutionary Phonol- ogy, Blevins 2004) that derives universal features of languages from the diachronic nature of language: they share similar traits because the way languages change is similar, so con- straints on synchronic phenomena can be traced back to constraints on diachronic processes. This means that contrary to generative tradition where universals are explained by using the innatist axiom and assuming the existence of a Language Faculty, there is no need for such a (from a certain point of view) extra-linguistic explanation and constraints on synchronic phenomena can be traced back to constraints on diachronic processes, that can, and should be described inside the domain of linguistics. This claim also means that synchronic phonological accounts that are able to account for patterns found in the subphonemical domain like vowel reduction in Hungarian should be preferred over those that are not, because subphonological behavior in a generation is the source of phonological rules/constraints/patterns of a later generation. Ohala (1981) describes the way how the listener, and in an acquisitional point of view, the learner is the source of sound change in different scenarios. These scenarios take subphonemic issues into account, and Ohala describes how these issues can lead to a phonological shift in the next generation. Simulational studies like Oudeyer (2006) and Wedel (2007) (and see Zuraw 2003 for a summary for such studies) examine language change with the use of simulation, in stochastic
31 Target of centralization in Hungarian Catalan SLPC Bulgarian Belarusian Central Italian Southern Italian a ? @ @ @ a a a 6/O @ -o- - o - E/æ e @ e-- e - e ? @ eia e i o @ uoua o u
Table 6.1: Hungarian vowel reduction compared to other reduction patterns
or connectionist models. These studies have shown how small differences in the start of a simulation, or translating this to diachronic linguistics, in an earlier generation can lead to a phonological sound change in one simulation and the lack thereof in an other one. These stu- dies therefore also underline the importance of investigating subphonemic data to understand language change, and therefore language itself according the diachronical, or evolutionary theoretical background seen above.
6.2 Vowel reduction
The vowel reduction pattern seen in the experiment (section 3) fits well into the picture when compared to (phonemic) reduction in other languages used by Crosswhite (2004) and Harris (2005) to illustrate their analyses (SLPC abbreviating Sri Lankan Portuguese Creole), as seen in table 6.1.
Looking at the table it can be surely declared that the processes that underlie subphonemic vowel reduction in Hungarian do not seem to have unique or grammatically unusual features when compared to phonologic reduction processes in other languages. For the reasons men- tioned in section 6.1.2, this is important as this means that subphonemic vowel reduction in Hungarian can be examined using current phonological theories of vowel reduction and it can be used to compare different phonological theories and check their validity on on a
32 subphonemic process.
6.2.1 Standard OT
The standard OT analysis for vowel reduction will be shown as it is described in Crosswhite
(2004). She describes two kinds of reduction patterns found in languages: prominence re- duction and contrast enhancement. The Hungarian data above shows prominence reduction as the low back vowel seems to converge towards a central [@]-like quality rather towards the corner vowel [a]. Prominence reduction is described in Crosswhite (2004) using feature-specified faith- fulness constraints in a language-specific order like MAX[+BACK], MAX[+ROUND] and so forth; and using a universally ranked markedness scale of vowel qualities and their occurrence in unstressed situations in the form of the following hierarchy:
UNSTRESSED/a ≫ ... ≫ UNSTRESSED/high vowels ≫ UNSTRESSED/@
There is no explanation in this theory why [@] is less prominent than high corner vowels, and in which detail it differs from other mid vowels like [e], [o] or lax [I], [E] or [O]. The above ranking is supposed to be universally bound and language-specific reduction patterns are explained using the faithfulness MAX constraints ranked to different places in the above hierarchy. Register-specific reduction can be explained with the gradual demotion of these faithfulness from an initial high position. Analyzing Hungarian data using this account will show some more of its weaknesses.
First, looking at the behavior of unstressed /E/ shows that two faithfulness constrains have to be used: a MAX[+LOW] constraint that can be demoted in casual speech and a MAX[+FRONT] constraint that cannot. To illustrate this, the following tableau shows the evaluation of candi- dates in careful speech where both MAX constraints are ranked high (*U/V being an abbre- viation for *UNSTRESSED/V):
33 /E/[+stress] MAX[+FRONT] MAX[+LOW] *U/E,O *U/e,o *U/i,u *U/@
☞[E] *
[e] * *
[@] * * *
In casual speech the MAX[+LOW] constraint is ranked lower (anywhere under UNSTRESSED/E,O), thus resulting prominence reduction:
/E/[-stress] MAX[+FRONT] *U/E,O MAX[+LOW] *U/e,o *U/i,u *U/@
[E] *
☞[e] * *
[@] * * * A prediction given by this analysis is that faithfulness constraints could be demoted further, thus yielding other centralization targets for /E/ in even more casual speech: [i] or [I] if
MAX[+LOW] is demoted under UNSTRESSED/e,o; and even [@] if MAX[+FRONT] can be demoted as well.
The analysis for the behavior of /o/ needs a MAX[+BACK] faithfulness constraint, as [o] and [@] are contrasted only in this feature. The demotion of the MAX[+LOW] constraint has no effect on /o/ because centralization or raising would not violate this constraint, as [o] itself is already [-low]. Careful speech evaluation is then as follows:
/o/[+stress] MAX[+BACK] *U/E,O *U/e,o *U/i,u *U/@
☞[o] *
[@] * * With the demoted faithfulness constraint in casual speech:
/o/[-stress] *U/E,O *U/e,o MAX[+BACK] *U/i,u *U/@
[o] *
☞[@] * *
34 The constraint selection and their re-ranking patterns above can be tested with the reduction of /6/. This phoneme can be represented as [+back] and [+low]. Its UNSTRESSED constraint should be between UNSTRESSED/a and UNSTRESSED/E,O. First, careful speech:
/6/[+stress] MAX[+BACK] MAX[+LOW] *U/6 *U/E,O *U/e,o *U/@
☞[6] *
[@] * * *
In casual speech MAX[+BACK]andMAX[+LOW] have to be demoted under *UNSTRESSED/e,o as seen above, so in this register the evaluation of candidates is as follows:
/6/[-stress] *U/6 *U/E,O *U/e,o MAX[+BACK] MAX[+LOW] *U/@
[6] *
☞[@] * * *
The problem with this analysis is the lack of its explanatory power and some of its ad hoc nature. One might point out that both MAX[+BACK] and MAX[+FRONT] have to be used in order to get the Hungarian reduction pattern. The actual rank where faithfulness constraints get demoted to seems to be random, and this analysis lacks any type of phonetic or other grounding for this. The question of graduality is also avoided here: if a certain constrant is demoted to a certain point in casual speech, its quality is set to the target phonetic realization instead of the somewhat gradual and stochastic realization pattern that can be seen in the data. Stochastic
OT (Boersma and Hayes 2001) can help here however, but it still does not have the functional grounding that is needed if the approach set in section 6.1.2 is taken seriously.
6.2.2 Functional grounding – frequency effects
The experiment described in section 4 showed that at least in the test words taken into account there, frequency effect on vowel reduction predicted by usage-base functionalist research like
35 the findings of Jurafsky et al. (2001) and Hooper (1976) seemed to work the other way round than expected. It is also quite shocking that the contradictory results are found consistently for every speaker. This finding does not mean, however, that analyses based on frequency or analogy are falsified here. Further research is needed to understand the contradictory nature found in this experiment. The lack of significant difference can be explained somehow by reasoning that syllable-timed systems do not show frequency effects on vowel reduction. But there is significant difference in the [tok] set, which definitely calls for an explanation.
This explanation might be the wrong choice of words, and some phonetic processes that have not been described for Hungarian as of yet. The two [tok] words chosen were /titok/
‘secret’ and /ŃolStok/ ‘folding ruler’. The process that might have influenced the quality, length and intensity of the second syllable of these words is the effect of the length of the first syllable.
For the frequent word /titok/, the [i] in the first syllable was for every speaker and almost every elicitation again extremely short, to the point of deletion in some utterances, and this may have resulted despite the ‘underlying’ position of primary stress on the first syllable in compensatory lengthening and stress shift to the second syllable, resulting in the usual pro-
I nunciation [t "tofl:k] or even [t:ofl:k] for some utterances. Speaker 2 (see 4.3) used extremely short first syllables and consequently, he had the longest second syllables due to this com- pensatory lengthening.
The infrequent word /ŃolStok/ shows converse effects however: as described in section 2, [l] may delete with compensatory lengthening of the previous vowel. The first syllable either with or without l-deletion is heavy: [Ńol] or [Ńofl:], which is followed by the complex [St] onset of the following syllable, therefore the second syllable can be more reduced in the usual pronunciation like [Ńofl:St@k].
36 6.2.3 Phonetic grounding – Dispersion Theory
An important phonetic notion when talking about vowel reduction is the dispersion of phonemes in a vowel inventory. Dispersion Theory portraits vowel reduction as the lesser need of dis- persion in unstressed environment, ie. the lesser need of a wide space for vowels. Flemming
(2004) has two phonologically grounded types of constraints based on this theory: a con- straint that enforces maximal dispersion of vowels in the space available (like MINDIST=F1:3 which is violated if two vowels are closer to each other than 3 measurements on an arbitrary scale) and constraints that prefer to minimize articulatory effort (like *SHORT LOW V for F1 centralization or *HIGH EFFORT for F2 centralization). As it evaluates vowel inventories it also uses a constraint that requires to maximize the numbers of vowel contrasts in a language. This latter property of this analysis is in fact a disadvantage, as well as the use of auditory effort-based constraints, while the listener (or the learner, the child) is practically the source of sound change (Ohala 1981). Using this theory also needs more research in Hungarian, as the effect of vowel length, and its retention in the standard speaken dialect is not very well understood yet. Boersma and Hamann (2008) provide a bidirectional grammar to include both participants of speech in Dispersion Theory. While their solution is elegant and dismisses the problem of exclusively articulatory explanation, their grammar still evaluates phonemic inventories and shows how languages usually arrive to a stable balanced system. The examination of subphonemic vowel reduction in Hungarian and other subphonemic processes needs an ex- planation to peek into an unstable system and the way these unstable inventories develop over time, and needs an explanation that does not evaluate a whole phonemic (or vowel in this case) inventory of a language, but is able to evaluate specific forms, phonemes and real- izations. The analysis of de Graaf (1987) on Hungarian vowel reduction has shown the effect of reduced dispersion in vowels in context when contrasted to vowels in isolation. He measures
37 Speaker nr. Free vowels Vowels in words Vowels in context long short long short long short 1 768 583 695 336 428 255 2 859 582 596 372 430 321
Table 6.2: Acoustic Systems Contrast measurement of Hungarian vowels in de Graaf (1987)
an Acoustic Systems Contrast (ASC) variable that is the mean square distance of a given vowel’s position from the central vowel’s place in the F1-F2 plane. The ASC values in his experiment can be seen in table 6.2.
6.2.4 Phonetic grounding – Dispersion-Focalization Theory
Dispersion-Focalization Theory (DFT, Schwartz et al. 1997) adds the focalization factor to the dispersion-based analyses. Focalization is meant to refer to the observed characteristics about the least ‘marked’ vowels in spoken languages. The assertion is that the most easily perceptable auditory feature of vowels is the convergence of two formants, ie. when they are so close to each other that their energy adds up to one salient frequency peak which is easily perceived by the human ear and mind. Dispersion and focalization work slightly against each other, but their balance can explain the distribution of vowel inventory types in the world (cf.
Kingston 2007). Focalization is easy to see on the spectra of the four corner vowels compared to the central non-focalized [@] vowel as it can be seen on Figure 6.1. It is assumed in this theory that human perception can very clearly hear the difference between the place of the highest peak in the spectrum, which is the salient convergence of two or three formants (high F1 and F2 for [a], high F2 and F3 for [i], low F0, F1 and F2 for [u]). This is contrasted with the lack of salient convergence, which means a more even spectrum seen for [@]. The spectrograms of these vowels can be seen, and the salient peaks can be more easily understood with the eye of a
38 20 20 Hz) Hz) / /
0 0
-20 -20 Sound pressure level (dB Sound pressure level (dB
0 5000 0 5000 Frequency (Hz) Frequency (Hz) Hz) Hz) / / 20 20
0 0 Sound pressure level (dB Sound pressure level (dB -20 -20 0 5000 0 5000 Frequency (Hz) Frequency (Hz)
Figure 6.1: Spectra of the vowels [a] and [i] (top row) and [u] and [@] (lower row)
linguist used to spectrograms on Figure 6.2.
6.2.5 Representation of phonetic facts
The question whether representation or derivation is of greater phonological interest is de- bated since the beginning of generative linguistics. The point of view taken here is definitely not generative because of the support for the diachronic explanation for universal features in languages. It will be argued, however, that representational approaches can help describing subphonemic phenomena and their diachronic behavior.
Diachronicity is sometimes imported to generative analyses either in a form of lexical derivation or in the form of abstract complex representation of underlying forms. Vowel-zero alternations in Slavic for example are still usually described using an underlying abstract set of phonemes called yers that are eliminated either through derivation (from Halle 1959
39 5000 5000 Frequency (Hz) Frequency (Hz)
0 0 1.425 1.838 3.738 4.152 Time (s) Time (s)
5000 5000 Frequency (Hz) Frequency (Hz)
0 0 5.06 5.473 7.704 8.117 Time (s) Time (s)
Figure 6.2: Spectrograms of the vowels [a] and [i] (top row) and [u] and [@] (lower row)
to Rubach 1993) or through running default processes in a representationalist approach (eg.
Scheer 2002 in strict CV phonology). The notion of yers is available from the knowledge of older forms of Slavic, where these (later abstract) phonemes had a phonetic value of a reduced [i] and [1] usually noted in Slavistics with the characters <ž> and <ż> respectively. The use of diachronicity in this paper is very much different. As described in section 6.1.2 the importance of phonological treatment for subphonemic data is explained through diachronicity, and representations should help in understanding language change. A given representational approach in this analysis would not describe phonological, rather phonetic facts in this case. This approach has to be well grounded phonetically and has to be linked to phonetic facts and it must provide falsifiable predictions on universal cross-linguistic level and falsifiable predictions on vowels not yet examined.
Racz´ (2009: 55–72) describes how a certain subphonemic vowel change (Southern En-
40 glish glottalization) can be analyzed using analogy-based functionalism, and illustrates how underlying representation (thus phonology) can shift under the pressure of analogy of fre- quent words. He also points out, however, that this might result in a postulation of a generative- type rule by the final generation, despite the tools and theoretical background he uses is dis- carded by generative theory. I argue for the same point here, that despite generative phonology may discard functional arguments, the combination of these arguments and certain representationalist approaches might even save those theories in the sense that the usually solid base of the given the- ory (Metrical Theory or Element Theory based on Dependency Phonology and Government Phonology), and findings of that theory are not lost, but functional (meaning phonetic here) grounding might give additional understanding to them, and diachronic approaches can get access to these theories.
6.2.6 Element Theory
Harris (2005, 2007) uses the phonetic grounding of focalization (see DFT in section 6.2.4) in his Element Theory (ET) framework for vowels. The abstract elements (A), (I) and (U) used in the phonological representation of vowels therefore have a phonetically grounded meaning as well, representing the perceptionally most salient effects that carry information over the neutral [@]-like carrier signal, cf. Figure 6.2:
• (A) – called a mAss, the salient convergence of high F1 and central F2
• (I) – called a dIp, the salient convergence of high F2 and F3
• (U) – called a rUmp, the salient convergence of F0, low F1 and low F2
The representation of corner vowels is a the simple element that characterizes them. The central vowel is represented by the lack of any segmental information: /@/ = (). Non-corner
41 vowel Catalan reduction Hungarian reduction phoneme representation quality representation quality representation /a/ (A) [@] () /6/∼/O/ (A+U) [u] (U) [@] () /o/ (A+U) [u] (U) [@] () /E/ (A+I) [@] () [I] (@+I) /e/ (A+I) [@] ()
Table 6.3: Comparison of Catalan and Hungarian vowel reduction in ET-DFT following Har- ris (2005)
vowels are treated as complex, therefore their representation is the combination of elements, eg. /o/ = (A+U), /e/ = (A+I). In more detailed systems one element is called the ‘head’ of the representation having a greater impact on the vowel quality, and the other is a ‘dependent’, eg. /O/ = (A+U) and /ofi/ = (A+U). In this ET-DFT framework vowel reduction is then easily handled as the loss of some elements, and different reduction patterns are handled as the question of retention or loss of one of them. Harris (2005) shows how this works in a variety of languages, including Catalan (see the table in the top of section 6.2) which seems to be quite similar to the Hungarian pattern attested above in this analysis: while Hungarian reduces (A) and (U) while keeping (I), Catalan reduces (A) and (I) and retains (U), as it can be seen on Table 6.3.
The result of centralization of Hungarian /E/ in representation in this framework would be a segment with an empty head (represented here by (@)) with a dependent (I), which renders a lax [I] quality (cf. Harris and Lindsey 1995). This quality is actually quite close to the realization seen in the experiment above. The representation of Catalan /O/ and Hungarian /6/ is handled to be the same as the (not reduced) realization of these phonemes are really close to each other and there is no theoretical issue that would necessitate their differenciation.
42 6.2.7 Predictions of the ET-DFT framework
The above analysis has the advantage of having theoretical background, having phonological groundedness and is capable of predictions on the behavior of vowels not examined on one side and maybe lets to take a peek into the tendencies of changes in the vowel inventory of
Hungarian. The main prediction would be that the palatality distinction is strongly retained in Hungarian, as the element (I) is retained, so one would not expect front vowels like /y/, /ø/ or /i/ to centralize their F2 formants, their target of reduction is predicted by the Element Theory analysis to be [I]. This could seem quite odd as the vowel quality of [ø] is the one substituted by Hungarian speakers for a [@]-like sound (Gosy´ 1997), therefore its loss of roundedness is a very strong – and falsifiable prediction of this model. The long term prediction of keeping the (I) element intact is that F2 vowel harmony seems to be well and alive and its loss (like the one that occurred in the genetic relative Estonian, cf. Viitso 1998) is not probable. Should vowel reduction in unstressed syllables get stronger by time and its effects enter the phonological domain, at least two reduced phonemes (back
[@] and front [I]) behaving according to vowel harmony are predicted to survive, if such bold predictions can be made by the models used above.
This prediction seems to be even more plausible looking at the research of Pearce (2009). She has shown that vowel harmony has in fact an effect on subphonemical vowel reduction in the way of conserving the harmonizing feature. She also shows some results on Hungarian, where she finds quite similar effects to those in the experiment above: back vowels centralize while front vowels do not. She has shown this pattern for more F2-based vowel harmony systems (even with genetic relatives of Hungarian like Finnish) as well as for other auditory cues, like [±ATR].
43 6.3 Stress pattern
As we have seen in the experiment described in section 5, the general claim that Hungarian has a left-to-right weight-insensitive bounded stress pattern is untenable. It is worth, however, looking at the formal acconts for the different theories of the stress pattern in Hungarian and to see if they can be useful finding a right account for the lack of secondary stress in Hungarian. Metrical Theory analyses use a limited number of parameters that characterize a given stress pattern (Hayes 1995). The interaction of these parameters, and their analysis as violable constraints led to Metrical Stress Theory to be the first to be described in Optimality Theory (Prince and Smolensky 1993/2004, McCarthy and Prince 1993). I will attempt to formalize different accounts of the Hungarian stress pattern using these theoretical backgrounds below.
6.3.1 Metrical Stress Theory – trochaic analysis
The analysis of Hungarian bieng a weight-insensitive pattern of stress falling on each odd- numbered syllable is parametrized in the fashion of Hayes (1995) like this:
Foot Construction Parse words to syllabic trochees Degenerate Feet Degenerate feet are banned (no secondary stress
on final syllables) Word Layer Construction End Rule Left The above means the parsing of the example word [fElEkEzEtijEke:] to the following layers:
word layer (x ) foot layer (x .) (x .) (x .)
syllable layer σ σ σ σ σ σ σ
fE lE kE zE ti jE ke: Optimality Theory analysis for a system like this following McCarthy and Prince (1993) would mean a low ranked WSP constraint (Weight-to-Stress Principle: heavy syllables must be stressed, unstressed heavy syllables invoke a violation mark), and a high ranked PARSE-σ
44 so as secondary stresses occur and left-heading constraints also ranked high: MAIN-LEFT making the End Rule Left work, ALL-FEET-LEFT (henceforth abbreviated to ALLFT-L) to make sure left-to-right parsing and TROCHAIC to have the first syllable to be the head of a foot. The degenerate foot at the end of the word may be accounted for with a NONFINAL constraint.
The ranking is important here: NONFINAL has to be ranked over PARSE-σ in order to avoid degenerate feet and PARSE-σ has to be ranked over ALLFT-L to get secondary stresses. The following tableau shows the effect of each constraint (right-aligning constraint are ranked below WSP):
/fElEkEzEtijEke:/ NONFINAL PARSE-σ ALLFT-L MAIN-L TROCH WSP
☞"(f´ElE)(k´EzE)(t´ıjE)ke: * ** *
(f´ElE)(k´EzE)"(t´ıjE)ke: * ** *! *
"(fEl´E)(kEz´E)(tij´E)ke: * ** *!** *
"(f´ElE)(k´EzE)(t´ıjE)(ke:) *! ***
(f´ElE)(k´EzE)(t´ıjE)"(ke:) *! *** *
fE(l´EkE)(z´Eti)"(jEke:) *! *** * *
"(f´ElE)kEzEtijEke: **!*** *
45 The analysis of Hammond (1987) is similar to the above, but with the colon as the extra layer, providing the possibility of three grades of stress, and without the ban on degenerate feet and cola: word layer (x ) colon layer (x . ) (x .)
foot layer (x .) (x .) (x .) (x) syllable layer σ σ σ σ σ σ σ
fE lE kE zE ti jE ke:
6.3.2 Metrical Stress Theory – Szinnyei-type analysis
The analysis has to be changed if the weight-sensitive rule that Szinnyei (1912) proposes and the experiment in section 5 shows that it has some plausibility is added to the description.
Stress assignment here seems to be quite strange: a trochaic left-aligned foot is constructed for the first two syllables and the remainder of the word behaves like if a new left-to-right trochaic assignment cycle would have begun, with the possibility of extrametricality for the third syllable if it is light.
This ‘sub-pattern’ of the word following the third syllable resembles the stress alignment in Wargamay (Hayes 1995: 140–142). The stress rules in Wargamay are:
• if the initial syllable is heavy, it receives the main stress, eg. ["mu:ba] ‘stone fish’, ["gi:baóa] ‘fig tree’
• if the initial syllable is light and the syllable count of the word is odd, stress the second
syllable, eg. [gag"ara] ‘dilly bag’, [éuó"agajmiri] ‘from Niagara Vale’
• if the initial syllable is light and the syllable count of the word is even, stress the first
syllable again, eg. ["bada] ‘dog’, ["giéawulu] ‘freshwater jewfish’
The last rule is the difference between the third-to-last Szinnyei pattern of a word, and it is
46 the crucial rule that enables Wargamay to be analysed to be a right-to-left moraic trochee system. This means that if the Szinnyei-type analysis of Hungarian stress still uses feet with two syllables, extrametricality needs to be used on the third syllable, which is not possible given that extrametricality can only occur on the edges of well defined domains, known as the Peripherality Condition (Harris 1983).
This pattern however resembles the stress system of an other language: Estonian as pre- sented in (Hayes 1995: 316–329). Not taking the complication of overlong syllables into account Estonian stress rule is basically the same to the Szinnyei-type stress rule in Hun- garian, with the notable exception that the weight of the third syllable decides only (and optionally for Estonian) for the place of the first secondary stress in the word, and after the second foot, the weight of the following syllable decides again of the position of the next foot. To illustrate this difference, let’s see the following hypothetical example (with ternary feet):
• word shape: — — ` — ` ` — ` `
• Estonian: ( —´ — `)( —´ ` `)( —´ ` `)
• Szinnyei-Hungarian: ( —´ — `)( —´ `)( `´ —)( `´ `)
There are more explanations for this type of non-peripheral extrametricality: Prince (1980) allows ternary feet of the shape ( —´ ` `), which easily solves the Estonian problem, but interferes with the usual notion of feet consisting of two syllables. Hayes provides an alter- native repair for the problem: Weak Local Parsing, which is defined in (Hayes 1995: 308) as the possibility to ‘skip’ over a / ` / after a constructed foot ‘where possible’. This analysis is however quite close to importing the notion of extrametricality to non-peripheral environ- ments, which might mean a blow to the core of this theory again.
47 6.3.3 Metrical Stress Theory – unbounded analysis
If the suggestion in section 5 that there is no secondary stress in Hungarian is right, then the
Hungarian pattern has to be described as unbounded: there is no need for the notion of feet here, or a foot is always equal to the prosodic word. The analysis of the test word above goes like this: word layer (x ) (foot layer (x ...... ))
syllable layer σ σ σ σ σ σ σ
fE lE kE zE ti jE ke: Prince (1985) and Bakovic´ (2004) argue that unbounded systems like the above (which are characterized as consistently leftmost) should be analyzed using the same theoretical appara- tus that has been worked out for bounded systems. This means, that the whole issue can be accounted for by demoting the constraint PARSE-σ anywhere below ALLFT-L, as this will yield a winning candidate which in fact contains a binary foot in its left edge. Since the asso- ciation of stress in the word is the same for a representation of (σσσσσ´ ) and (σσ´ )σσσ, they argue that this representational difference is not important at all. Then the tableau needs just a little change:
/fElEkEzEtijEke:/ NONFINAL ALLFT-L PARSE-σ MAIN-L TROCH WSP
"(f´ElE)(k´Eze)(t´ıjE)ke: *!* * *
☞"(f´ElE)kEzEtijEke: * ***** *
This analysis leads however to the abuse of the notion of foot, as it doesn’t have any theo- retical or any functional let alone phonetic reality. Binary feet are used here just to preserve the unity of the theory, and to incorporate the othervise quite simple unbounded stress pattern into it with its notions, constraints and tools used for much more complicated phenomena. Unbounded systems like the one above do not need ‘feet’ as there is no secondary stress
48 Vulgar Latin Eur. Portuguese Spanish French Italian RP English(non-Romance) [Ńive"tate] ‘city’ [s1"dad1] [Tju"DaD] [si"te] [Ùi"t:a] ["sIti] ["rosa] ‘rose’ ["Ko:z5] ["ro:sa] ["Koz] ["rOza] ["ô@Uz] Table 6.4: Deletion and reduction processes not affecting place of stress in Romance lan- guages
or any type of binary alternation in them, therefore their analysis could be easily handled with a pair of constraints that require primary stress on a word, but prohibit it in non-peripheral positions. Obviously a more thorough analysis of unbounded systems would need more tools, and the principles of Optimality Theory would require for constraints used on complicated bounded systems to be universal and therefore to be available when analyzing unbounded systems. A winning surface candidate with a binary foot that has no functional reasons to be there is, however, definitely not realistic.
6.3.4 Diachronic evolution
There might be an explanation of the ‘weak local parsing’ of Estonian and Hungarian if the
Szinnyei-account is assumed here to be verified. This explanation is again of diachronic nature: stress is a conservative feature in many ways.
This can be seen in Romance languages, where the place of stress is almost never moved from the place wher it stood in their ancestor (Vulgar) Latin. There have been many phono- logical diachronic processes of reduction or deletion going on segmentally, especially in French and Romanian, but final /e/ is deleted in Spanish and often in Italian as well. The clear-cut right-to-left moraic trochee system with word-final extrametricality of Latin (Hayes
1995: 91–92) has therefore been rendered quite opaque, but the place where stress stood never changed. Latin stress of Romance loanwords was kept by the learned` tradition in English, but not as strictly as strictly stress conservativity affected the daughter languages, as it can be seen on table 6.4.
49 This might help to build an account to what happened in Hungarian resulting in the intra- word extrametrical pattern. The investigation of ‘weak local parsing’ in Estonian would exceed the limits of this thesis, but there is a phenomenon seen in Estonian stress, that shows how diachronical explanation can be given to extrametricality, and how this explanation helps it understand why extrametricality is peripheral. In Estonian, as described in (Hayes 1995: 56–58), word final consonants are extramet- rical, ie. word final CVC syllables are counted as light, while word final CV:, CVCC and internal CVC syllables are heavy. The diachronical explanation for this is easy to be seen when Estonian is compared to Finnish, which is more conservative phonologically. The main difference between the two languages here is the deletion of word final vowels in Estonian, and their retention in Finnish. While the structure of the word in this process changes as the final syllable becomes closed, the weight pattern is conservative and does not change in Estonian.
This means that Finnish word final CVCV corresponds to Estonian CVC, eg. Finnish mayr¨ a¨ [mæyræ] ∼ Estonian mager¨ [mæker] ‘badger’. The weight pattern is not different: Finnish ` ` ` (as Finnish is held to be a mora-timed language, see Aoyama 2001) or — ` (in a syllabic analysis) versus Estonian ` ` with the last ` syllable having lost, but the final consonant being extrametrical. The process can be seen as if the ‘extrametrical’ consonant would still have stayed in the syllable with the disappeared vowel: *ma.ge.ra¨ ` ` ` > ma.ge.rø¨ ` ` ø which can be synchronically analyzed as ma.ge¨
Hungarian non-peripheral extrametricality, or its ‘weak local parsing’ can be explained in a similar fashion. Words in Proto-Hungarian had a similar shape to the Finnish dominantly vowel-ending system (E. Abaffy 2004b). This system eroded in two ways: the loss of final vowels, like in Estonian, and the loss of vowels in open syllables word-internally, which was
50 ∗ ∗ ∗ • ( σ´1 `2 ) ( σ´3 `4 ) σ → *( —´ 1 ) ( σ´3 `4 ) σ → (Clash removal) ( —´ 1 σ3 `4 ) σ as a ternary foot, or with `4 as extrametrical as it has been peripheral and footed at the beginning of the process.
∗ ∗ • ( σ´1 `2 ) ( σ´3 —4 ) σ → *( —´ 1 ) ( σ´3 —4 ) σ → (Clash removal) ( —´ 1 σ3 —4), but as —4 could attract stress either by WSP or in proposed metrical trochees, this ∗ could be reanalyzed → ( —´ 1 σ3 ) ( —´ 4 σ5 ) σ .
Table 6.5: Proposed diachronic process of the evolution of a Szinnyei-type stress pattern
a recurrent process stretching from the Proto-Hungarian to the Old and Middle Hungarian period, with Slavic loanwords involved as well, eg. Slavic [malina] ∼ Hungarian [ma:ln6] ‘raspberry’. The third process involved here is the buildup of affixes from distinct words, eg.
Old Hungarian (1195) [timnyŃe: bEleyl] ∼ Modern standard [tømløŃe:bøl].
These processes resulted in a recurrent change in the shape of σ1 `2 σ3 → —1 σ3. Ifa ‘default’ left-to-right syllabic trochaic analysis is proposed for the older stress system and a conservative stress pattern for the vowel change, the diachronic process can be described as seen in table 6.5.
This proposal has many flaws and presuppositions, for example the fact that other Finno- Ugric languages like Finnish and Karelian (Hayes 1995: 329–330) show ‘weak local parsing’ as well was not taken into the discussion here. Despite this it definitely shows a new account, which combines Metrical Theory analysis with explanation of diachronical nature. It can also explain the Peripherality Condition of extrametricality: this phenomenon occurs at edges because segment deletion is frequent there and segment deletion does not necessarily lead to the reanalysis of the stress pattern or syllable weight rules of the previous language stage.
51 6.3.5 Theoretical consequences
This study of Hungarian stress did not lead to a particular theory that may be used to predict falsifiable claims for so far unchecked data or for the future. Running through Metrical Stress analyses above it can be seen that all descriptions of Hungarian stress have their explanation in Metrical Theory, but the lack of functional approaches here really hurts these analyses as by the end the data are again helping out the theory and not theories are modeling the data. It has been shown that diachronical analysis shows that the representational Metrical
Stress Theory account doesn’t need to be dismissed if the solutions of ternary feet and weak local parsing seem to be weak solutions to the problem: diachronical analyses can show that the findings of the theory might be correct by large, but taking diachronic processes into ac- count, the combination of the two can yield more analyses, which can be more plausible and more functionally grounded.
7 Summary
7.1 Claims and consequences
This thesis has had two main claims besides presenting experimental data from areas previ- ously not well researched. The first claim is that subphonemic phenomena that have been dismissed by phonologic theory on the grounds that they are extra-linguistic or part of the
‘performance’ or being merely some kinds of errors in speech are well enough worth to be examined, as they can be described by several theories, and can falsify claims given by some theories that had excluded phonetic facts from their field of description. It has been argued that with the recent re-emergence for diachronic explanations for syn- chronic patterns these subphonemic phenomena need to be researched and functionally well grounded theories are needed to describe them. These descriptions must account for pho-
52 netic facts, the gradient nature of subphonemic processes and must provide enough theo- retical power to provide falsifiable predictions. It has been shown that for vowel reduction phenomena, the combination of the once very much distant Dispersion-Focalization Theory and Element Theory is an account that meets these expectations. For the description of the problematic Hungarian stress process the abstract representational theory of Metrical Theory seemed to be inadequate, but if the (theoretically problematic) Szinnyei-type stress pattern is assumed to have some reality the theory does not need to be dismissed as functional (di- achronic) explanations can help correcting the flaws of the theory.
7.2 Further research
Further research would obviously need to check the validity of the predictions of the ET-DFT framework sketched up in section 6.2.7 and to widen the array of subphonemic phenomena that could prove or falsify claims of this framework. The development of different functional phonological theories also needs to be tracked as newer theories and newer findings may provide more understanding to vowel reduction and other subphonemic patterns.
There will definitely be further research in the question of the influence of frequency effects on vowel reduction, as the surprising results in section 4 lead us to look for more pairs of words that can be tested. If these results are found for the new pairs, the controversiality of our findings would be quite motivating to work on this topic theoretically as well. The secondary stress pattern of Hungarian might need more research as there seem to be some arguments pointing towards a Szinnyei-type pattern to be accepted as an analysis for Hungarian stress (seen in section 5), but more test words need to be taken into account. The arguments for a Szinnyei-type analysis consisted of the lack of downstep in intensity, and the significance of this might also be checked statistically and cross-linguistically.
A further goal might be to get the pieces together that have begun to be collected above and provide a theoretical background for any further experimental study on a lesser studied
53 language, particularly on subphonemical phenomena. The field of functionalist phonology at this time is quite scattered and research is going on a lot of topics, and it might be hard for an experimental researcher to find the relevant theories that can be useful for presenting their idea.
This goal would then be carrying on investigating the interaction between different func- tionalist approaches to phonology as well as keeping an eye open to representationalist ap- proaches that can be used as a model for the description of some phenomena, especially when they can be linked to phonetically or diachronically grounded explanations. It has been shown that fields previously thought to be well researched like vowel systems, intonation and stress in languages previously thought to be well researched like Hungarian much studied Indo- European languages can still come up with a lot of surprises. If descriptions of phenomena in a such well known language are so much mistaken, a plan for reinvestigating intuition-led claims in a more explicit and falsifiable manner seems to be necessary.
54 List of Tables
2.1 Vowel inventory of Standard Hungarian ...... 2
2.2 Vowel inventory of Old Hungarian and archaic Palocdialects´ ...... 4
3.1 Differences between formants of vowels in [S6k] (a) and [t6z] (z) environments 13 3.2 Average vowel quality in examined environments ...... 15
4.1 Mean phonetic values of the environment [t6k] in word with different frequency 17
4.2 Mean phonetic values of the environment [tok] in word with different frequency 17 4.3 Average duration and intensity for test words for each speaker ...... 19 5.1 Words examined in the stress experiment ...... 21
6.1 Hungarian vowel reduction compared to other reduction patterns ...... 32 6.2 Acoustic Systems Contrast measurement of Hungarian vowels in de Graaf
(1987)...... 38 6.3 Comparison of Catalan and Hungarian vowel reduction in ET-DFT following
Harris(2005) ...... 42 6.4 Deletion and reduction processes not affecting place of stress in Romance languages ...... 49
6.5 Proposed diachronic process of the evolution of a Szinnyei-type stress pattern 51
55 List of Figures
2.1 F1 values for the Brasilian Portuguese [a Zarata"kaka] (Arantes and Barbosa 2008) ...... 9
3.1 Average positions of vowels in environments with phoneme /E/...... 11
3.2 Average positions of vowels in environments with phoneme /o/...... 12 3.3 Average positions of vowels in environments with phoneme /6/ in the dimen- sionsofF1andF2...... 14
3.4 Average positions of vowels in environments with phoneme /6/ in the dimen- sionsofF1andF3...... 15
3.5 Tendencies of centralization for each environment ...... 16 4.1 Average place of vowels in test words in the F1-F2 plane ...... 18
4.2 Place of the vowel [o] in test words in the F1-F2 plane for each elicitation . . 18
5.1 Fundamental frequency in elicitations of the word [fElEkEzEtijEke:]...... 22
5.2 Fundamental frequency in casual speech elicitations of the word [k6toliŃizmuS] 23 5.3 Duration of vowels for a speaker in the word [bEfEktEtø:k] ...... 24 5.4 Duration of vowels for a speaker in the word [fElEkEzEtijEke:]...... 24
5.5 F2 of vowels for a speaker in the word [kErESkEdElEm] ...... 25 5.6 F2 of the first four [E] vowels for a speaker in the word [fElEkEzEtijEke:] . . . 26
5.7 Intensity of vowels for a speaker in the word [bEfEktEtø:k] ...... 27 5.8 Intensity of vowels for a speaker in the word [k6toliŃizmuS] ...... 27
5.9 Intensity of vowels for a speaker in the word [kErESkEdElEm] ...... 28 5.10 Intensity of vowels for a speaker in the word [fElEkEzEtijEke:]...... 29 6.1 Spectra of the vowels [a] and [i] (top row) and [u] and [@] (lower row) . . . . 39
6.2 Spectrograms of the vowels [a] and [i] (top row) and [u] and [@] (lower row) . 40
56 References
Acs,´ Peter´ and Peter´ Siptar´ (2001). Tul´ a gondozott beszeden.´ In: Ferenc Kiefer (ed.), Struk-
turalis´ magyar nyelvtan 2. Fonologia´ , Akademiai´ Kiado,´ Budapest, 2nd ed., pp. 550–580.
Aoyama, Katsura (2001). A psycholinguistic perspective on Finnish and Japanese prosody.
Springer.
Arantes, Pablo and Plinio A. Barbosa (2008). F1 and spectral correlates of secondary stress in Brazilian Portuguese. In: Proceedings of the Speech Prosody 2008 Conference, Campinas,
pp. 559–562.
Bakovic,´ Eric (2004). Unbounded stress and factorial typology. In: John McCarthy (ed.),
Optimality Theory in Phonology: A Reader, Blackwell, London. ROA-244, Rutgers Opti- mality Archive, http://roa.rutgers.edu/.
Balassa, Jozsef´ (1890). Hangsuly´ a magyar nyelvben. [Stress in the Hungarian language].
Nyelvtudomanyi´ Kozlem¨ enyek´ 21, pp. 401–431.
Benko,˝ Laszl´ o´ (1992). Hangzoss´ ag´ es´ erthet´ os˝ eg.´ In: Gabor´ Kemeny´ and Jeno˝ Szant´ o´ (eds.),
Mondd es´ ´ırd, Auktor Konyvkiad¨ o,´ pp. 15–17.
Blevins, Juliette (2004). Evolutionary phonology. The emergence of sound patterns. Cam-
bridge University Press, Cambridge.
Boersma, Paul and Silke Hamann (2008). The evolution of auditory dispersion in bidirec- tional constraint grammars. Phonology 25, pp. 217–270.
Boersma, Paul and Bruce Hayes (2001). Empirical tests of the gradual learning algorithm. Linguistic Inquiry 32, pp. 45–86.
57 Boersma, Paul and David Weenink (2008). Praat: doing phonetics by computer. Version
5.0.41. http://www.praat.org.
Bolinger, Dwight (1989). Intonation and its uses: melody in grammar and discourse. Stan- ford University Press.
Bybee, Joan (2001). Phonology and language use. Cambridge University Press, Cambridge.
Bybee, Joan and Paul Hopper (eds.) (2001). Frequency and the emergence of linguistic struc-
ture. John Benjamins.
Chomsky, Noam (1965). Aspects of the Theory of Syntax. MIT Press.
Crosswhite, Katherine (2004). Vowel Reduction. In: Hayes et al. (2004), pp. 191–231.
E. Abaffy, Erzsebet´ (2004a). Hangtort¨ enet.´ (Az omagyar´ kor) [Phonology of the Old Hungar- ian period]. In: Kiss and Pusztai (2004), pp. 301–351.
E. Abaffy, Erzsebet´ (2004b). Hangtort¨ enet.´ (Az osmagyar˝ kor) [Phonology of the Proto- Hungarian period]. In: Kiss and Pusztai (2004), pp. 106–128.
Flemming, Edward (2004). Contrast and Perceptual Distinctiveness. In: Hayes et al. (2004),
pp. 232–276.
Ginter, Karoly´ and Laszl´ o´ Tarnoi´ (1974/1993). Ungarisch fur¨ Auslander¨ . Tankonyvkiad¨ o,´ 8th ed.
Gosy,´ Maria´ (1997). Semleges maganhangz´ ok´ a magyar beszedben´ [Central vowels in Hun-
garian speech]. Magyar Nyelvor˝ 121, pp. 9–19.
Gosy,´ Maria´ (2006). A semleges maganhangz´ o´ nyelvi funkcioi.´ [The linguistic functions of
the central vowel]. In: Maria´ Gosy´ (ed.), Beszedkutat´ as´ 2006, pp. 8–23.
58 de Graaf, Tjeerd (1987). Vowel contrast reduction in Hungarian, Finnish, and other lan-
guages. Studia Uralica 4, pp. 47–58. Akten der dritten Tagung fur¨ uralische Phonologie. Eisenstadt, 28. Juni– 1. Juni 1984.
Halacsy,´ Peter,´ Andras´ Kornai, Laszl´ o´ Nemeth,´ Andras´ Rung, Istvan´ Szakadat´ and Viktor
Tron´ (2004). Creating open language resources for Hungarian. In: Proceedings of Lan- guage Resources and Evaluation Conference (LREC04). European Language Resources
Association, .
Halle, Morris (1959). The Sound Pattern of Russian. Mouton, The Hague.
Hammond, Michael (1987). Hungarian cola. Phonology Yearbook 4, pp. 267–269.
Harris, James W. (1983). Syllable Structure and Stress in Spanish: A Nonlinear Analysis. Linguistic Inquiry Monographs 8, pp. 1–158.
Harris, John (2005). Vowel reduction as information loss. In: Philip Carr, Jacques Durand and Colin J. Ewen (eds.), Headhood, elements, specification and contrastivity, Benjamins,
Amsterdam, pp. 119–132.
Harris, John (2007). Information Flow in Phonology. Handout for a lecture given at the EGG Summer School in Brno.
Harris, John and Geoff Lindsey (1995). The elements of phonological representation. In: Jacques Durand and Francis Katamba (eds.), Frontiers of Phonology: atoms, structures,
derivations, Longman, Harlow, pp. 34–79.
Hayes, Bruce (1995). Metrical stress theory. University of Chicago Press, Chicago & Lon- don.
Hayes, Bruce, Robert Kirchner and Donca Steriade (eds.) (2004). Phonetically Based Phonology. Cambridge University Press, Cambridge.
59 Hooper, Joan B. (1976). Word frequency in lexical diffusion and the source of morphophono-
logical change. In: W. Christie (ed.), Current Progress in Historical Linguistics, North Holland, Amsterdam, pp. 96–105. van der Hulst, Harry, Bernadet Hendriks and Jeroen van de Weijer (1999). A survey of word prosodic systems of European languages. In: Harry van der Hulst (ed.), Word prosodic
systems in the languages of Europe, Walter de Gruyter, chap. 7, pp. 425–476.
Jurafsky, Daniel, Alan Bell, Michelle Gregory and William D. Raymond (2001). Probabilistic
relations between words: Evidence from reduction in lexical production. In: Bybee and Hopper (2001), pp. 229–254.
Kerek, Andrew (1971). Hungarian metric: some linguistic aspects of iambic verse. Mouton, The Hague.
Kingston, John (2007). The Phonetics-Phonology Interface. In: Paul de Lacy (ed.), The Cam- bridge Handbook of Phonology, Cambridge University Press, Cambridge, pp. 353–382.
Kiparsky, Paul (1985). Some consequences of lexical phonology. Phonology yearbook 2, pp. 85–138.
Kiss, Jeno˝ (ed.) (2001). Magyar dialektologia.´ [Hungarian dialectology]. Osiris, Budapest.
Kiss, Jeno˝ and Ferenc Pusztai (eds.) (2004). Magyar nyelvtort¨ enet.´ [The history of Hungar- ian]. Osiris, Budapest.
Korompay, Klara´ (2004). Helyes´ırast´ ort¨ enet.´ (Az omagyar´ kor) [Orthography in the Old Hun- garian period]. In: Kiss and Pusztai (2004), pp. 281–300.
Lotz, John (1939). Das ungarische Schprachsystem. Stockholm.
McCarthy, John and Alan Prince (1993). Generalized Alignment. In: G. Booij and J. van Marle (eds.), Yearbook of Morphology, pp. 79–154.
60 Ohala, John J. (1981). The listener as a source of sound change. In: C. S. Masek, R. A.
Hendrick and M. F. Miller (eds.), Papers from the Parasession on Language and Behavior, Chicago Linguistic Society, Chicago, pp. 178–203.
Oudeyer, Pierre-Yves (2006). Self-Organization in the Evolution of Speech. Studies in the
Evolution of Language, Oxford University Press, Oxford. Translation by James R. Hur- ford.
Pearce, Mary (2009). Vowel harmony effects on vowel reduction. Paper presented at the 6th OCP, Edinburgh, United Kingdom.
Pierrehumbert, Janet B. (2001). Exemplar dynamics: Word frequency, lenition and contrast.
In: Bybee and Hopper (2001), pp. 137–157.
Pintzuk, Susan, Miklos´ Kontra, Klara´ Sandor´ and Anna Borbely´ (1995). The Effect of the
Typewriter on Hungarian Reading Style. In: Working Papers in Hungarian Sociolinguis- tics, Linguistic Institute, Hungarian Academy of Sciences, Budapest, vol. 1.
Prince, Alan (1980). A Metrical Theory for Estonian Quantity. Linguistic Inquiry 11, pp. 511–562.
Prince, Alan (1985). Improving tree theory. In: Proceedings of the Berkeley Linguistics Soci-
ety, vol. 11, pp. 471–490.
Prince, Alan and Paul Smolensky (1993/2004). Optimality Theory: Constraint interaction in
generative grammar. Blackwell.
Racz,´ Peter´ and Daniel´ Szeredi (2009). Testing usage-based predictions on Hungarian vowel reduction. Abstract submitted to the 17th Manchester Phonology Meeting and accepted for
a poster presentation.
61 Racz,´ Peter´ (2009). A case study of an English phonological phenomenon – Glottalization in
Southern English. Master’s thesis, Eotv¨ os¨ Lorand´ University, Budapest.
Raphael, Lawrence J., Gloria J. Borden and Katherine S. Harris (2007). Speech science primer. Lippincott, Williams & Wilkins, Baltimore.
Roach, Peter (1982). On the distinction between ‘stress-timed’ and ‘syllable-timed’ lan- guages. In: David Crystal (ed.), Linguistic controversies, Edward Arnold, London.
Rubach, Jerzy (1993). The Lexical Phonology of Slovak. Clarendon Press, Oxford.
Scheer, Tobias (2002). How yers made Lightner, Gussman, Rubach, Spencer & Co invent CVCV. Ms., University of Nice. http://www.unice.fr/dsl/papers.htm.
Schwartz, J.-L., L.-J. Boe,¨ N. Vallee´ and C. Abry (1997). The Dispersion-Focalization Theory of vowel systems. Journal of Phonetics 25 (3), pp. 255–286.
Siptar,´ Peter´ and Miklos´ Torkenczy¨ (2000). The phonology of Hungarian. Oxford University
Press, Oxford.
Szeredi, Daniel´ (2008). Centralized vowels in Hungarian. In: Laszl´ o´ Kalm´ an´ (ed.), Papers
from the Mokus´ Conference, Tinta Publishing House, Budapest.
Szeredi, Daniel´ and Sylvia Blaho (2009). (The non-existence of) secondary stress in Hungar- ian. Poster presented at the 6th OCP, Edinburgh, United Kingdom.
Szinnyei, Josef (1912). Ungarische Sprachlehre. Goschen,¨ Berlin.
Varga, Laszl´ o´ (2002). Intonation and stress: evidence from Hungarian. Palgrave Macmillan, Basingstoke.
Viitso, Tiit-Rein (1998). Estonian. In: Daniel Abondolo (ed.), The Uralic Languages, Rout- ledge, London & New York, pp. 115–148.
62 Wedel, Andrew B. (2007). Feedback and regularity in the lexicon. Phonology 24, pp. 147–
185.
Zuraw, Kie (2003). Probability in language change. In: Rens Bod, Jennifer Hay and Stefanie Jannedy (eds.), Probabilistic Linguistics, MIT Press, Cambridge, MA, pp. 139–176.
63 Magyar nyelvu˝ absztrakt
Szakdolgozatomban a magyar nyelv ket´ eddig kevess´ e´ vizsgalt´ jelenseg´ et´ mutatom be kortars´ fonologiai´ elmeleti´ hatt´ er´ hasznalat´ aval:´ a maganhangz´ ok´ redukcioj´ at´ (vagy semlegesedes´ et)´ esam´ asodlagos´ hangsuly´ helyet.´
A maganhangz´ oredukci´ o´ vizsgalata´ soran´ elvegzett´ k´ıserletben´ 7 beszel´ o˝ vett reszt.´ A k´ıserlet´ negy´ kornyezetet¨ vizsgalt´ harom´ hangsulyhelyzetben:´ a [t6z] (z), [S6k] (a), [hoé] (o) esa[´ kEé] (e) szekvenciak´ maganhangz´ oj´ anak´ minos˝ eg´ et´ (1) tobbsz¨ otag´ usz´ o´ elsosz˝ otagj´ aban,´ (2) tobbsz¨ otag´ u´ szo´ utolso´ szotagj´ aban,´ illetve (0) funkciosz´ oban.´ Annak ellenorz˝ es´ ere,´ hogy az egyes beszel´ ok˝ gondozott beszed´ eben´ ezek a maganhangz´ ok´ milyen hangsz´ınnel fordulnak elo,˝ ezeket a hangsorokat izolaltan,´ lassan is felolvastak´ (g). Az eredmenyek´ azt mutatjak,´ hogy az (a), (z) es´ (o) kornyezetek¨ valoban´ mutatnak maganhangz´ oredukci´ ot,´ a hangsulyhelyzet´ ukt¨ ol˝ fugg¨ oen,˝ az (e) kornyezet¨ nem: beszelt´ nyelvi kornyezetben¨ minden hangsulyhelyzetben´ azonos mert´ ek´ uz˝ ar´ od´ ast´ mutat.
Egy masik,´ Racz´ Peterrel´ egyutt¨ elvegzett´ k´ıserletben´ 4 beszel´ o˝ altal´ felolvasott mondatok alapjan´ a maganhangz´ oredukci´ o´ esasz´ o´ gyakorisaga´ koz¨ otti,¨ mas´ nyelvek alapjan´ feltetelez´ est´ teszteltuk.¨ Ennek a k´ıserletnek´ az eredmenye´ meglepo˝ lett, mivel a hipotezisekkel´ ellentetben´ a ritkabb´ szavak mutattak nagyobb redukciot.´ Ennek tobb¨ oka is lehet, amelyeket meg´ tovabb´ kell kutatni, am´ elkepzelhet´ o,˝ hogy a k´ıserleti´ szavak kivalaszt´ asa´ a kudarc oka, viszont ´ıgy egy ujabb,´ eddig nem le´ırt jelenseget´ van lehetos˝ eg´ le´ırni: a szotagok´ kozti¨ hosszus´ ag´ ki- egyenl´ıtod˝ es´ et.´
A harmadik, Blaho Sylviaval´ egyutt¨ elvegzett´ k´ıserletben´ ismet´ 4 adatkozl¨ o˝ beszed´ et´ vizsgaltuk,´ hogy a magyar masodlagos´ hangsuly´ helyet´ megallap´ ´ıtsuk. Tobbf¨ ele´ vitatott elmelet´ koz¨ ul¨ ketto˝ tunt˝ a k´ıserlet´ alapjan´ valosz´ ´ınunek:˝ egyreszt´ az elosz˝ or¨ Szinnyei (1912) altal´ le´ırt rendszer, mely szerint a masodlagos´ hangsuly´ a paratlan´ szotagokon´ talalhat´ o,´ kiveve,´ ha a harmadik szotag´ rovid,¨ s ekkor a negyediktol˝ kezdve a paros´ szam´ u´ szotagokra´
64 kerul¨ a masodlagos´ hanguly.´ Masr´ eszt´ pedig elkepzelhet´ onek˝ tartottuk, hogy a magyar nyelvben egyaltal´ an´ nincsen masodlagos´ hangsuly,´ ennek feltetelez´ ese´ mas´ nyelvek elemzese´ soran´ hasznalt´ elmeletek´ magyar nyelvre valo´ kenyszer´ ´ıtese.´ Veg´ ul¨ dolgozatomban megvizsgaltam´ a fenti jelensegeket´ kortars,´ illetve regebbi´ gen- erat´ıv alapu´ fonologiai´ elmeletek´ seg´ıtseg´ evel.´ Ravil´ ag´ ´ıtottam arra, hogy, bar´ a fenti je- lensegek´ nem fonologiai´ szintuek,˝ ezeknek fonetikai vagy mas´ funkcionalis´ elmelet´ altali´ elemzese´ hasznos lehet fonologiai´ elmeletek´ vizsgalat´ ara´ is. A diakron´ alapu´ (Blevins (2004) alapjan´ ,,evoluci´ os”)´ fonologia´ szam´ ara´ is fontos ezeknek a fonetikai jelensegeknek´ a megert´ ese,´ mivel egy generaci´ oban´ tapasztalt nem-fonologiai´ jelensegek´ a kovetkez¨ o˝ generaci´ okban´ a nyelvvaltoz´ as´ soran´ fonologiai´ jelentos˝ eget´ kaphatnak. A maganhangz´ oredukci´ o´ vizsgalata´ soran´ a Harris (2005) alapjan´ felvazolt´ fonetikai hat-´ terq-rel´ (a diszperzios-fokaliz´ aci´ os´ elmelettel,´ vo.¨ Schwartz et al. (1997)) megeros˝ ´ıtett, megis´ reprezentaci´ okon´ dolgozo´ elemelmeletet´ talaltam´ a leghasznosabbnak a magyar jelenseg´ el- emzes´ ere.´ Igy´ egyreszt´ a katalan´ maganhangz´ osemlegesed´ eshez´ konnyen¨ hasonl´ıthatov´ a´ valt´ a magyar minta, masr´ eszt´ pedig az (I) elem tartoss´ aga´ (ez az elem nem tunik˝ el a redukcio´ folyaman)´ a maganhangz´ oharm´ onia´ robusztussag´ ara´ is ramutathat.´ A masodlagos´ hangsuly´ elemzese´ soran´ vegigvettem´ a magyar hangsulyra´ vonatkozo´ elmeletek´ formalis´ elemzes´ et,´ ramutatva,´ hogy a hangsulyrendszerek´ elemzes´ ere´ hasznalt´ elmelet´ csupan´ az elmeletbe´ nehezen illesztheto˝ segedeszk´ oz¨ okkel¨ kepes´ a magyar, illetve a hozza´ hasonlo´ eszt´ es´ finn hangsulyrendszerek´ elemzes´ ere.´ Felvazolok´ ugyanakkor egy diakron´ alapu´ elemzest,´ amely esetleg iranyt´ mutathat az elmelet´ problem´ ainak´ funkcionalis´ alapu´ megoldasa´ fele.´
65