Phonetics & Phonology Forum is a weekly talk and discussion series featuring presentations on all aspects of phonology and phonetics.
Jump directly to: |
January |
February |
March |
April |
May |
The Niger-Congo languages of Africa are well-known for the complexity of their verb extension systems. In subgroups as far apart as Bantu and Atlantic, a verb root may be extended by several derivational suffixes marking such notions as causative, applicative, reciprocal, and passive. There may also be suffixes marking middle voice, pluractionality, or various inflectional categories (tense, aspect, mood, polarity). A rather complex example comes from the Eastern Bantu language, Ciyao (Ngunga 2000):
taam-uk-ul-igw-aasy-an-il-a 'cause each other to be unseated for/at' (cf. /taam-/ 'sit')The Ciyao example shows that multiple suffixes can combine to form derived stems with no apparent principled upper limit, many Northwest Bantu (NWB) and other more westerly Niger-Congo languages impose prosodic constraints on the stem. The most common such constraints involves an upper limit of syllables, e.g. four in Yaka, Punu (NWB); three in Koyo, Basaa (NWB); two in Mankon (Grassfields Bantu). In addition, whereas all consonants can appear in all positions in stem-unrestricted languages such as Ciyao, the same westerly languages tend to restrict the inventory and combinatorics of consonants in non-stem-initial position.
In this talk I will be concerned with the morphological consequences of such phonological constraints. Not surprisingly, if there is an upper limit on the number of syllables permitted, there may not be enough room for a suffix or suffixes to be added to certain verb bases. Much more surprising, however, is what happens when the sequence of non-initial consonants isconstrained by place of articulation. In Tiene (NWB), for example, stems are maximally C1VC2VC3V (Ellington 1977). In addition, C2 must be coronal (alveolar or palatal), while C3 must be noncoronal (labial or velar). This works out fine in the following example, where the C2of the root is alveolar and the C3 of the stative extension is velar:
faasa 'drive through' --> fas-ak-a 'be driven through'However, in the following example, the C2 of the root is velar and the C3 of the causative extension is alveolar:
lóka 'vomit' --> lósek-E 'cause to vomit' (expected: *lók-es-E)As a result of the place restrictions, the /s/ of the causative occurs as C2, i.e. preceding root /k/, which appears as C3. In both examples, the stem has the shape CVsVk-V, but two outputs are obtained in different ways.
What this means is that an extension may be suffixed vs. infixed depending on its place of articulation and/or that of the base to which is is affixed.
This is highly unusual--and mysterious--and yet I have found such phenomena in two separate Niger-Congo language clusters: (i) the Teke languages, including Tiene, spoken in the Democratic Republic of the Congo, Congo-Brazzaville and Gabon; (ii) the Central Plateau group of Nigeria, e.g. Izere (Wolff & Meyer-Bahlburg 1979, Gerhardt 1984, Blench 2000) and Birom (Bouquiaux 1970, Blench 2005). Since these groups have independently innovated very similar distribution patterns and infixation by place of articulation, the questions that naturally arise are how did the coronal-noncoronal sequential constraint come into being-and why?
There seem to be two logical possibilities: (i) The sequential constraint is an historical "accident": Earlier statistically skewed distributions by place of articulation, which may have had to do with noncoronal consonants being more prevalent in outlying suffixes, have been regularized through analogy. (ii) The sequential constraint is principled: There is a phonological motivation for coronals to precede noncoronals in prosodic constituents.
In my talk I will show that the first hypothesis cannot be correct for several reasons. This leads me therefore to consider different explanations as to why coronals might tend to precede noncoronals in the way described.
This talk will be an informal presentation of a strategy that I'm developing for automatically extracting acoustic phonetic measurements from labeled speech corpora. In this work, I suggest a new approach to automating analysis in acoustic phonetics, where measurements often must be taken by hand. In truth, you CAN run the formant tracker and extract formant measurements automatically in a script, but a substantial proportion of the measurements will be erroneous. Measurements of fricative spectra (loudest spectral peak, spectral moments, etc) and of fundamental frequency are also highly problematic. For corpus phonetics, where we would like to extract hundreds of thousands of acoustic measurements, the status quo is not acceptable. In a nutshell, the strategy that I will present (1) collects a set of good examples of spectra illustrating a number of speech sounds, and (2) scores the entire corpus by matching against these clear examples.
In one study I used the resulting "acoustic attribute scores" to code fricatives with temporal trajectories of <s>ness and <sh>ness to evaluate the coarticulatory effects of preceding rhotics. If time permits over the weekend I may also show results on vowel to vowel coarticulation.
In generative phonology, phonological relationships have conventionally understood in terms of binary or privative features that correspond directly to some phonetic parameter. From the beginning, though, there have been arguments that certain features are multi-valued (a possibility that is entertained for vowel-height in SPE, but rejected on the grounds that it seemed possible to capture all other relevant classes in terms of binary features). For example, it has been argued that vowel-raising chain shifts are best captured through scalar (or quasi-scalar) representations. The current paper draws on evidence from a novel source--phonologically-driven ordering effects in coordinate compounds--to argue that human phonologies make reference to n-ary (multi-valued) relationships. Perhaps more radically, it argues that these scales are not necessarily grounded directly in a phonetic dimensions. That is to say, scales are best understood in terms of their logical structure, rather than their phonetic substance.
In most languages of mainland Southeast Asia, there is a class of symmetrical, binary coordinate compounds (compounds lacking a single morphological head) where the conjoined roots--either nominal orverbal--are in a special semantic relationship. They are either synonyms, antonyms, or representative members of some larger class of entities or predicates). There is no strictly morphological principle which can constrain the ordering of elements of such compounds; however, in some languages this order can be predicted on phonological grounds. In Jingpho (a Tibeto-Burman), the ordering of conjunct in a coordinate compound may be predicted on the basis of vowel height. If the tonic vowels of the two stems differ in height, the stem with the higher vowel will be ordered first (1).
(1) a. tip ‘press’, sep ‘exploit’, tip-sep ‘exploit’, *sep-tip
b. ləko ‘foot’, lətaʔ ‘hand’, ləko-lətaʔ ‘hands and feet’, *lətaʔ-ləko
This ordering effect can be expressed in terms of the binary features [high] and [low] (order [+high] before [-high], order [-low] before [+low]). However, I show that by doing this, we model the generalization in a way that is less insightful that a direct statement in terms of a vowel-height scale (2):
(2) {i, u} >> {o, e} >> {a}
As I show, the same type of scale is needed independently to properly account for vowel-raising chain-shifts like the one seen in Zongdi Hmong.
Jingpho is by no means alone is displaying ordering effects of this kind. An analogous generalization, where tone is the relevant phonological feature, is found in various languages, including most languages of the Hmongic family. In analyzing the case of the phonologically-motivated ordering in one of these languages (Mong Leng), I show that this effect is best understood as the result of competition between two scales over the tones of the language, employing the same constraints needed for Jingpho. In contrast to the Jingpho case, though, the scales motivated by this effect in Mong Lengdo not follow transparent phonetic continua. I further show that an attempt to treat these relationships as purely morphological effects result in a loss of generality, increase the power of the morphology unnecessarily, and fail to account for cases where phonological alternations (in the form of tone sandhi) precipitate alternations in the sequence of conjuncts.
This study suggests that language-learners are capable of incorporating into their grammars phonological generalizations that are quite different from those which conventionally been assumed to exist and, further, that these generalizations may manifest themselves not just through distributional patterns and morphophonological alternations, but also through effects upon the sequence of morphological constituents. Thus it not only raises questions about the relationship between phonological structure and phonetic substance, but also about how, and to what extent, linear sequence in morphology is determined by phonology.
Chong is a typologically interesting Mon-Khmer language, having a 4-way voice-register contrast (Thongkum, 1987, 1991). Thongkum (1991) states that languages which have voice-register utilize a set of cues including phonation type, pitch, and length, to signal differences in word meaning. This contrasts with many tone languages which utilize primarily differences in F0 to indicate tone. Of the phonation types which can exist in a particular language, only 2 or 3 types usually contrast. However, in Chong, words occur with one of four phonation types. In this talk I will present both acoustic and laryngographic fieldwork data from the dialect of Chong in Takhian Thong demonstrating the differences in the voice registers. Data from 4 speakers (2 male, 2 female) are examined.
Similar to previous phonetic descriptions of other Chong dialects, the voice-register contrast involves modal, tense, breathy, and growl phonation types. This last phonation type has been debated in work on Chong phonology, with some researchers (Huffman, 1985; Thongkum, 1987, 1991) suggesting that it involves increased glottal adduction on a vowel before the coda; the phonological shape /CVʔC/. Data from both F0 analysis and the laryngograph show that there is some evidence for this perspective. The intensity of the glottalization on two of the voice registers is contrastive, where final creak associated with a tense phonation across the vowel is distinct from the harsh and more intense glottalization associated with the growl register. Differences in pitch are also associated with voice register. High pitch occurs with tense phonation while lower pitch occurs with other phonation types. The phonation cues for this contrast are more robust than the pitch differences however. For the non-tense registers, pitch differences are minimal.
In the languages of the world, phonation type may or may not cross-classify with a tonal system. In Chong, it cross-classifies with some pitch differences to create a complex set of phonological contrasts that are tonally and laryngeally distinct. While it is rare for languages to contrast phonation type, those languages which have such contrasts usually limit them to a 2-way or 3-way contrast. A larger set of contrasts is extremely rare, occurring only in Chong and perhaps !Xóõ. Laryngeal complexity is not uncommon among Mon-Khmer languages (Huffman, 1976) and Otomanguean languages (Silverman, 1997). Research on languages involving such complex laryngeal patterns informs our knowledge of the behavior of vocal fold vibration, the phonetics of tone, and the range of possible laryngeal contrasts that may exist in a language.
This paper introduces a new type of ideophone based on the author's field research on an undocumented variety of Totonaco (TFM). TFM exhibits semantically linked morpho-phonological templates of a kind not previously described in the literature on ideophones. Ideophones in TFM fall into three main semantic areas of color terms, odor/flavor terms, and manner adverbials, They conform to strict CV templates, which secondarily involve both sound symbolism and reduplication, which vary by semantic field, and are found nowhere else in the lexicon. The paper will also relate properties of the TFM templates to properties of ideophonic templates in other languages.
This paper presents an analysis of downstep in Tiriki, a Bantu language of the Luyia subgroup, spoken in western Kenya. Downstep is mandatory between two high tones when they are non-adjacent in the input and meet via tone spreading rules, but either optional or prohibited (depending on morphological context) when the high tones are adjacent in the input and remain so on the surface. We present an analysis where downstep is caused by floating low tones, and show how the adjacency conditions fall out rather than having to be stipulated, as they are in analyses (e.g. Bickmore 2000) where downstep vs. non-downstep is represented as two separate H tones vs. a single fused H tone.
Optimality Theory (OT) focuses on generalizations found in the character of surface structures that languages strive to attain or avoid. A radically surface-oriented theory like OT denies the significance of generalizations about input-output mappings and about environments of processes. In this talk I argue that signficant generalizations of this type exist.
In pre-OT thinking, vowel syncope was treated as the end-stage of vowel reduction, and thus applied in the same kinds of environments as reduction: to unstressed vowels, vowels in unfooted syllables, posttonic vowels, etc. On the other hand, I show that in OT a variety of constraints can lead to vowel deletion, not all of them directly related to the notion of 'weakness' or lack of stress. Crucially, because constraints are statements about outputs, there is no control over the environment of syncope: it can apply to strsesed vowels as well as to unstressed ones.
I argue, first, that proposed analyses of several languages (Hopi, Southeastern Tepehuan) that rely on stressed vowel syncope are in fact incorrect, and in all such cases alternative and better analyses are available. Second, I show that the absence of reference to the environment of syncope leads to uninsigtful analyses of some other languages (e.g. Lebanese Arabic), where syncope is treated as the result of a variety of surface pressures, but a simple input-based generalization that it applies to unstressed vowels is missed.
I will suggest a solution to these problems that keeps the output-based results of OT intact but affords input-based generalizations their rightful place in the theory.
Repeated experimental results have shown that listeners of a language can have extreme difficulty in perceiving non-native contrasts, especially when a contrast is subsumed within the variation of a single native category (e.g. Best et al. 1991). Perceptual learning and category formation, the mechanisms by which listeners learn to focus on only relevant variation and ignore irrelevant details, is thus a key issue in studying the acquisition of a mature perceptual system. Specifically of interest is how subjects' selective attention to relevant dimensions of categorization develops.
In this talk I present pilot work and preliminary data from ongoing dissertation research into categorization. Specifically, this data is from a training study in which adult native speakers of English are trained to perceive a contrast in alveopalatal and retroflex voiceless sibilants (found e.g. in Polish and Mandarin). This study is unique in examining how subjects learn the different cues to the contrast, e.g. fricative noise and formant transition, and how learning along one dimension affects the other. This is possible through use of a stimuli set that varies along two dimensions representing the cues to the contrast.
Preliminary results indicate that English listeners are highly sensitive to transition information and have considerable difficulty in using fricative noise to categorize these stimuli. This contrasts with a native speaker of Polish who is able bring both cues to bear for categorization. There is also tantalizing evidence that use of the transition dimension is a language specific effect and not simply a result of raw psychoacoustic robustness.
In spite of the phonetic grounding that has characterized much of phonological theory since the 1990s (see e.g. Archangeli and Pulleyblank 1994; Hayes and Steriade 2004),there is a growing body of evidence that language users are aware of relationships between the sounds of their languages which cannot be translated transparently into phonetic terms. This paper argues for the existence of language-specific, phonetically arbitrary scales over phonological categories.The argument is based primarily upon phonologically-conditioned ordering effects in Mong Leng and QeNao coordinate compounds.
Like most languages of mainland Southeast Asia, Mong Leng and QeNao have in their lexicons numerous coordinate compounds (compounds lacking a single morphological head). In both Mong Leng (“Green Hmong”) and QeNao, it is possible to predict the ordering of constituents in such compounds based solely upon the tones of the conjoined stems (specifically the tone of the final syllable of each stem). For Mong Leng, coordinate compounds are always arranged according to the set of precedence relationships given in (1). Thus,a conjunct with the falling tone (e.g. ntsê 'face’) will always be ordered before a conjunct with the breathy tone (e.g. mṳa ‘eye’) so that ntsê-mṳa ‘face’ is a licit cocompound but sequences like *mṳa-ntsê are not. An analogous constraint holds in Qe Nao, but it makes reference to a different scale (2).
I propose that there is a string-internal correspondence relationship (Walker 2000; Rose and Walker 2004) between the stem-final tones of each conjunct in this construction. Iargue, further, that one class of constraints that can hold over entities in such a relationship are “directional antifaithfulness” constraints (WAX). WAX[S] requires that, given two corresponding entities of some type, the preceding is lower along some scaleSthan the following. I demonstrate that the Mong Leng scale given in (1) is actually best decomposed into two interacting scales (a binary “register” R scale independently motivated by tone sandhi behavior and a second scale T). Ordering effect result from a competition between a higher ranked WAX[T] and a lower ranked WAX[R] such that register-motivated orderings only emerge when the two tones under comparison are identical relative to T.
The most striking facet of this analysis is the high level of abstraction necessary in the definition of the relevant scales, as shown in (3). These scales clearly do not divide some phonetic dimension in the same way that other proposed phonological scales like tonal prominence (de Lacy 2002), vowel height, sonority, and “inherent voicing” (Gnanadesikan 1997) do. However, the logical structure of these relations is fundamentally the same as that of other, substantially grounded, scales. Learners have constructed phonological scales without the help of substantial cues.
Against this claim, it may be argued that these ordering effects are purely morphological and therefore have no relevance to our understanding of the organization of phonology. However, I show that a purely morphological analysis would require the specification of morphological features that are wholly predictable from phonological contrasts (and which are, therefore, redundant), that such a hypothesis would save a restricted view of the phonology by unprincipled expansions of the power of the morphology,and that a purely morphological analysis cannot account for data like (4), where a purely phonological alternation (tone sandhi) triggers a reversal in the ordering of the same two morphemes in order to satisfy WAX[T]. Just as significantly, I argue, a phonological account is necessary to account for the type of productivity that these constructions and generalizations display.
These cases are part of a wider class of phonological phenomena characterized by logical coherence but phonetic arbitrariness. Such phenomena,as a class,provide important evidence that phonological knowledge, defined broadly as knowledge about the sound system of one’s language,is composed largely of rich structural relationships which need not be grounded synchronic phonetics.
The Cupeño habilitative aspect provides us with an interesting problem, as its formation depends both on stress placement in the verb's base, and on whether or not the verb root ends in a vowel. Verb roots ending in vowels do not exhibit any morphological changes, whereas verb roots ending in consonants have variably sized allomorphs depending on where stress falls in the word. Furthermore, verb roots ending in epenthesized consonants pattern with vowel-final verbs (e):
| base | Hab | gloss | |
| a) | ngáng | ngá'a'ang | cry |
| tewásh | tewá'a'ash | lose | |
| b) | yúymuk | yúymu'uk | be cold |
| néneng | néne'eng | play pion | |
| c) | pína'wex | pína'wex | sing enemy songs |
| épe'-yax | épe'yax | ease a pack | |
| d) | séyki | séyki | gather seyily |
| yewáywe | yewáywe | talk | |
| e) | kwá | kwá' | eat |
| hiqsá | hiqsá' | sigh |
Previous attempts to characterize the habilitative morpheme in Cupeño by phonological means have fallen short in that they have stipulated structures not otherwise found in the language. Based on work by Alan Yu (2003), I propose that the Cupeño habilitative morpheme is a "true" infix (i.e. an infix with obligatory non-peripherality), requiring a trochaic foot to its left and a final consonant to its right. Words that cannot meet these requirements exhibit 0-morphology. I address the 0-morphology of verbs ending in epenthesized consonants by using Subcategorization frames, which occur prior to phonological processes. Subcategorization theory predicts that the conditioning environment for some allomorphs will be rendered opaque by successive phonological processes, a prediction that is borne out in Cupeño.
This is a work in progress, and I certainly appreciate feedback on the analysis.
SourcesVowel harmony typically targets an unbounded sequence of vowels in some domain (e.g. Finnish). But in Lango, [ATR] harmony holds only between suffixes and root-final vowels:
Common analyses of vowel harmony (based on Agree or Align constraints, e.g.) are built for Finnish-style whole-word harmony, so the Lango facts seem to require a radical departure from this sort of constraint. However, an analysis in which Licensing (Zoll 1998, Crosswhite 2000) drives harmony is available. With Licensing constraints requiring [ATR] features to be linked to root vowels, minimal spreading becomes optimal: As long as the root-final vowel and suffix vowel have the same [ATR] feature, Licensing is satisfied, regardless of the root-initial vowel’s [ATR] feature. Potential alternatives fail: The [ATR] feature of the root-initial vowel cannot be preferentially preserved by positional faithfulness because it harmonizes in monosyllabic roots:
 |  |
There is also little independent evidence for a word-final disyllabic foot, so a foot-based analysis is not well supported. Although Lango’s harmony seems to require a mechanism for spreading features by exactly one syllable, the Licensing analysis shows that this is not necessary.