1 Phonetically driven Optimality Theory

Bruce Hayes and Donca Steriade begin the introductory chapter (‘Introduction: the phonetic bases of phonological markedness’) by raising a number of questions about sources of phonological markedness: ‘Where do markedness laws come from? Why are sound systems governed by these laws and not by some conceivable others? What is the source of individual's knowledge of markedness-based constraints?’ (p. 1). The answer to these questions, Hayes & Steriade argue, lies in understanding the facts of phonetic difficulty – articulatory effort and perceptual confusability (cf. Lindblom Reference Lindblom, Hardcastle and Marchal1990). Facts of phonetic difficulty are internalised by speakers as ‘phonetic knowledge’, and this knowledge serves as the basis for grammatical generalisations: constraints and their rankings. The activity of constraint rankings in language-particular grammars is reflected in cross-linguistic sound patterns. The goal of the outlined research programme, therefore, is to identify grammatical markedness generalisations by deducing them from facts of phonetic difficulty (observed through experimentation) and cross-linguistically common sound patterns (observed through typological surveys). Hayes & Steriade illustrate this deductive approach by examining the phonetics and phonology of voiced obstruent geminates. Phonetic studies show that it is difficult to sustain vocal fold vibration when producing long obstruents, more so for some places than others. This knowledge of phonetic difficulty, a markedness scale for [+voice] referring to gemination and place, serves as a basis for hierarchies of OT markedness constraints. Different rankings of these constraints with Ident[voice], as the authors show, correspond to cross-linguistically attested patterns of voicing gemination. They note that although the phonetically based analysis of the phenomenon is more complex than a traditional modular analysis, it is arguably more empirically adequate than the former. Hayes & Steriade conclude by highlighting the differences between their framework and the approach presented further by Blevins & Garrett.

All other contributions in this category apply the general programme of research outlined in Hayes & Steriade to various phonological phenomena. These works are relatively similar structurally, combining theoretical analysis with typological surveys, reviews of phonetic literature and, in some cases, phonetic experimentation or computational modelling.

Robert Kirchner, in ‘Consonant lenition’, argues that the key factor behind lenition is minimisation of articulatory effort; moreover, patterns of lenition are directly derived through an interaction of effort-based markedness constraints (Lazy) and perceptually based faithfulness constraints. Articulatory effort is assumed to depend on a number of phonetic parameters and factors, including constriction degree, adjacent segments and speech rate and style, and these are directly incorporated into constraint hierarchies. The approach is further tested in great detail on data from Florentine Italian, a language representative of the typology of consonant lenition. The language exhibits both obligatory/categorical and optional/gradient lenition, which applies depending on context, consonant and speech rate and style. The latter effect is modelled in the paper using register-specific Lazy constraints, thus capturing the optionality of the process. Gradient phonetic effects are expected to be achieved in the framework through the introduction of features that are relatively more fine-grained than traditional phonological features (see e.g. p. 338 on the use of the feature [±more strident]). As Kirchner notes, the presented analysis is tentative and based on certain assumptions about articulatory effort cost that await future confirmation.

Edward Flemming, in ‘Contrast and perceptual distinctiveness’, argues that the functional imperative to avoid perceptual confusion plays a key role in phonological grammar. Unlike articulatory difficulty, perceptual difficulty has to refer to contrasts rather than to individual segments, since perception is based on categorisation of sounds, often involving discrimination of pairs of sounds. Thus, the relatively uncommon, marked, vowels /y/ and /ɯ/ are not difficult per se; rather, the contrast between these vowels is less perceptually distinct than the contrast between the relatively common vowels /i/ and /u/ that have more extreme second formant values. Building on the insights of Lindblom's (Reference Lindblom, Ohala and Jaeger1986) Adaptive Dispersion Theory, Flemming develops a grammatical theory of phonological contrasts – Dispersion Theory (Flemming Reference Flemming1995). The key component of this theory is a set of scalar OT constraints requiring maximisation of the distinctiveness of contrasts referring to auditory features (e.g. Minimal Distance=F2:2 – two sounds have to differ at least by two predefined units of distance along the second formant dimension). Flemming shows that interactions of these constraints with those requiring maximisation of the number of contrasts and minimisation of articulatory effort produce various attested patterns of dispersion and neutralisation. The processes discussed in more detail include enhancement of vowel ‘colour’ and consonant laryngeal contrasts in inventories, vowel reduction in unstressed syllables and the role of contrast in allophonic vowel variation and patterns of nasal harmony. A particularly interesting case illustrating the role of perceptual contrasts is pre- and post-stopping of nasals in Kaingang, an articulatory strategy presumably aimed at maintaining the contrast between nasalised and non-nasalised vowels.

Jongho Jun, in ‘Place assimilation’, notes that standard OT and previous phonological theories provide a satisfactory formal descriptive account of place assimilation, but fail to adequately explain a number of asymmetries observed in its cross-linguistic patterns. The patterns of place assimilation in clusters, identified in a language survey, include the regressive direction of assimilation and greater likelihood of certain segments as targets (e.g. nasals>stops>fricatives; coronals>labials>dorsals) and triggers of assimilation (e.g. non-coronals>coronals). An explanatory account of these patterns, Jun argues, has to incorporate articulatory and perceptual teleologies. His analysis of place assimilation crucially involves a set of faithfulness constraints requiring preservation of perceptual cues to consonant place in various contexts, ranked against the articulatory effort-based Weakening constraint. Thus, for example, the factors underlying the susceptibility of nasals to assimilation are the relatively poor cues to nasal place, compared to the cues to stop place, grammaticalised as the constraint hierarchy in (1):

1. (1)

   

Jun also addresses the issue of optionality and gradiency in assimilation, proposing to derive these effects using a set of gradient, speech rate/style-sensitive Preserve and Weakening constraints.

Abigail R. Kaun, in ‘The typology of rounding harmony’, presents results of a cross-linguistic survey of rounding harmony systems, showing that there are systematic preferences in terms of targets and triggers of harmony. In particular, she finds that non-high front vowels are the most likely triggers of rounding harmony, while high back vowels are the most likely targets. In addition, harmony is more likely if triggers and targets agree in height (e.g. high /y–ɨ/ or non-high /ø–ɑ/). Kaun attributes these asymmetries to a number of articulatory and perceptual factors, based on a survey of phonetic studies of lip rounding. First, high vowels tend to be produced with greater lip rounding than non-high counterparts, as do back vowels compared to their non-back counterparts. These differences have perceptual consequences, so that high and back vowels are perceived as relatively more rounded than non-high and front vowels. Second, implementation of two consecutive lip-rounding gestures is apparently facilitated when the corresponding lingual gestures have the same height targets (i.e. vowels agree in height). Kaun argues that rounding harmony is an optimisation strategy, as it improves recoverability of less salient vowels by extending the span of rounding to more salient vowels. An OT account of the phenomenon uses alignment and markedness constraints grounded in these phonetic factors. The author further presents results of an experimental study of rounding harmony in Turkish loanwords. The patterns exhibited by speakers in the experiment are notably different from Turkish rounding harmony in suffixes, yet apparently consistent with the cross-linguistic patterns. Additional support for the perceptual account of target/trigger asymmetries, Kaun notes, comes from the finding of short/long vowel asymmetry: in some languages, the less salient short vowels trigger rounding harmony, while the more salient long vowels do not.

Katherine M. Crosswhite, in ‘Vowel reduction’, argues that there are two different types of reduction, each involving distinct phonetic mechanisms. Reduction that eliminates non-corner vowels (as in Belarusian and Luiseño) is ‘contrast-enhancing’, presumably resulting from licensing perceptually demanding non-corner vowels to stressed syllables. Phonetic evidence for non-coronal corner vowels as sub-optimal is based primarily on Lindblom's (Reference Lindblom, Ohala and Jaeger1986) Adaptive Dispersion Theory and Stevens (Reference Stevens1989) Quantal Theory. The second type of vowel reduction – ‘prominence reduction’ – is driven by the imperative to avoid phonetically long or salient vowel qualities in unstressed position (as in Bulgarian and Sri Lankan Portuguese Creole). The ultimate outcome of such reduction is a schwa (as in English). Formally, this process is implemented as a hierarchy of constraints referring to scales of accentual and vocalic prominence, which is assumed to be grounded in differences in vowel intensity and amplitude, among other possible factors. Crosswhite notes that prominence reduction should not be equated with vowel-target undershoot (as suggested by Flemming in his paper), and considers the two phenomena to be phonological and phonetic respectively. Cases of languages having more than one type or degree of reduction (e.g. Brazilian Portuguese) are particularly revealing; however, the author does not elaborate on whether these cases are consistent with the proposed analysis. Another interesting observation made by Crosswhite is that all cases of prominence reduction in the survey are limited to stress-timed languages, thus suggesting a link between reduction types and stress- or syllable-timing.

Jie Zhang, in ‘The role of contrast-specific and language-specific phonetics in contour tone distribution’, notes that contour tones in many languages are restricted to either phonemically long vowels or stressed syllables. His review of the literature on the production and perception of tone suggests that unlike level tones, contour tones require longer sonorous rhyme duration, a property shared by both long vowels and stressed syllables. He takes a ‘direct approach’, arguing that licensing of contour tones is implemented with direct reference to phonetic categories of duration and sonorancy, and not with reference to abstract structural categories (such as syllable position). A grammar of tone licensing, he proposes, refers to phonetic scales of contour-bearing ability and tone complexity. Moreover, the grammar is argued to be sensitive to gradient language-particular phonetic differences in duration. Zhang further presents experimental results that appear to support his analysis: an acoustic examination of sonorant rhyme duration in three languages showed a close correlation between duration measurements and licensing of contour tones. Another interesting typological finding of the study is that word-final syllables and monosyllabic words are also among the preferred licensing sites of contour tones. Sonorant rhyme duration is a factor here as well, Zhang argues, as utterance-final syllables and shorter words tend to undergo phonetic lengthening.

Matthew Gordon, in ‘Syllable weight’, challenges the traditional assumption of ‘weight uniformity’ – that light/heavy syllable weight criteria are uniform for various processes in a given language, while possibly being different across languages. His survey of over 400 languages exhibiting six weight-sensitive phenomena – stress, tone, poetic metrics, minimal word requirements and syllable-template restrictions – reveals that weight is not always uniform within a language, being strongly process-specific. In other words, a given language may have different weight criteria for different weight-sensitive phenomena; across languages, however, weight-sensitive phenomena are largely uniform in choosing some weight criteria over others. For example, tone is often sensitive to CVV and CV[+sonorant] heavy syllables, while stress is sensitive to CVV and CVC heavy syllables (regardless of coda sonorancy). Gordon proposes that these phonological differences are rooted in phonetic factors specific to particular weight-sensitive phenomena. He further explores the phonetics of weight-sensitive stress in more detail, showing that the syllable-weight hierarchy relevant to stress correlates closely with the phonetic ‘effectiveness’ (measured as energy of the sonorant portion of the rhyme) of the corresponding syllable types. While phonetic effectiveness is clearly a factor, Gordon argues that the principle of phonological simplicity is equally important in selecting language-particular syllable weight criteria. He further presents the results of an acoustic study of different syllable types in six languages with weight-sensitive stress. The findings show that in all of the languages examined phonological weight distinctions are both structurally simple and phonetically effective, being tuned to language-specific phonetic properties. He interprets the findings as evidence for an ‘integrated model’ of phonology and phonetics where ‘one aspect of phonological structure influences phonetic properties which in turn influence another phonological phenomenon’ (p. 300). Further evidence for this view comes from the intriguing finding of a close correlation between coda inventories (ratios of sonorants to obstruents and voiceless to voiced consonants) and the choice of weight criteria in a given language.

It is interesting to note that while these papers share many theoretical assumptions, there are some clear differences in the details of how phonetic influence on phonology is formalised. In many papers, phonological constraint hierarchies refer directly to scales of phonetic difficulty (Jun, Flemming, Kirchner and Zhang; cf. Steriade Reference Steriade, Hume and Johnson2001); other papers take a more traditional view, assuming that phonological constraints are ‘grounded’ in phonetics (Crosswhite and Kaun); still others espouse an ‘integrated’ phonology–phonetics approach (Gordon). Interestingly, these different assumptions in some cases result in conflicting accounts of what appears to be the same phenomenon, e.g. vowel reduction (Crosswhite and Flemming) and the distribution of contour tones (Zhang and Gordon). Given the considerable proliferation of OT constraints in many phonetically driven accounts, a question that often comes up is whether markedness constraints of this type can or need to be innate, specified in the traditionally assumed Universal Grammar (UG). This question, although not raised in most contributions, is taken up by Hayes & Steriade, who view UG as ‘a set of abstract analytical predispositions that allow learners to induce grammars from the raw facts of speech, and not as a – dauntingly large – collection of a priori constraints’ (p. 6). This represents an intriguing departure from the standard generative view of UG adhered to since Chomsky & Halle (Reference Chomsky and Halle1968) (see also Mielke's Reference Mielke2004 work on emergent features).

While all papers are concerned with cross-linguistic typology of phonological patterns, they differ considerably in their descriptive typological coverage. Typological generalisations presented in some of the contributions are based on impressive surveys of about 200 to 400 languages (Gordon, Zhang and Kirchner). Generalisations presented in some others papers are based on less extensive surveys, likely convenience samples of languages (e.g. 23 languages in Kaun's paper and 17 languages in Jun's). Understandably, the sample size of a survey is often constrained by the frequency of a phonological phenomenon examined and by the availability of descriptive data. Still, generalisations based on smaller and genetically biased samples may not always reflect true universal patterns, being possibly skewed by genetic or areal linguistic developments. One case in point is Kaun's typology of rounding harmony, where eight out of 13 types of rounding harmony are represented exclusively by Turkic languages. Further, the main typological generalisation that the non-high front rounded /ø/ is the best trigger and the high back unrounded /ɨ/ is the best target (although it may indeed be phonetically motivated) is based largely, if not exclusively, on Turkic languages, since no other languages in the sample have both vowels in their inventories (with the exception of Shuluh Höh, a Mongolian language, from Kaun Reference Kaun1995). This highlights another issue that seems to be underplayed in this and some other contributions: phonemic contrast is highly important, since language-particular inventories crucially delimit the set of triggers and targets of many phonological processes (cf. Flemming and Gordon). Further, Kaun's typology does not include some relatively well-known cases of rounding harmony, for example, Hungarian and Cheremis (Mari) (e.g. Odden Reference Odden1994, Ringen & Vago Reference Ringen and Vago1998), whose patterns do not seem to fit into Kaun's factorial typology. In these languages, the triggers of rounding harmony are high and non-high front rounded vowels, and the targets are non-high back unrounded vowels.

Another apparent limitation of some of the papers is that universal scales of perceptual difficulty are mainly deduced from acoustic studies, rather than being established in perceptual experiments. As Wright (discussed further below) notes, ‘not all acoustic features that can be discerned in a spectrogram or waveform will necessarily have an equal impact on the listener’ (p. 42). Jun's paper is particularly relevant in this respect. He attributes the lack of susceptibility of unreleased velars in assimilation (compared to labials and coronals) to the characteristic conversion of F2 and F3 prior to the closure. A number of recent perceptual studies of place, however, have shown that unreleased velars are identified more poorly than unreleased labials, and as badly as or even more poorly than unreleased coronals (Kochetov & So Reference Kochetov and So2007, among others). Similarly, other studies have found that place in nasals is identified as well as, if not better than, place in stops (e.g. Winters Reference Winters2003), thus contradicting Jun's manner scale of perceptibility, intended to explain susceptibility of nasals to assimilation. All of this underscores the need for careful experimental testing of claims about the scales of phonetic difficulty.

Perceptual cues, language processing and sound change

Richard Wright's paper ‘A review of perceptual cues and cue robustness’ complements the contributions reviewed above by providing a detailed survey of phonetic studies, examining the auditorily/perceptually important acoustic properties of consonants and vowels. The survey includes a discussion of various cues to place contrasts (formant transitions, fricative noise, release bursts, nasal cues), manner and voicing contrasts, and vowel quality. He further discusses the concept of ‘cue robustness’ – the ability of linguistically important acoustic information to ‘survive’ in external noise and to overcome inherent distortions during auditory processing. One apparent limitation of Wright's survey is that it lacks the cross-linguistic focus found in other papers, as it concentrates mostly on English sounds (which is, however, hardly to be avoided, given that most of the research on the topic has been done on English). As Wright notes in the introduction, the survey is intended to provide background for a revision of the Sonority Sequencing Constraint, which is envisioned as a ‘scalar constraint in which an optimal ordering of segments is one that maximises robustness of encoding of perceptual cues to the segmental makeup of the utterance’ (p. 35). While this proposal is promising, the section dealing with it is rather brief (slightly over two pages), leaving questions about the exact formal implementation of the constraint and its use in phonological grammar unanswered.

Stefan A. Frisch, in ‘Language processing and segmental OCP effects’, explores the role of higher-level phonological processing in phonotactic patterns of similarity avoidance. He shows that earlier documented gradient OCP effects in Arabic and English roots (Frisch Reference Frisch1996, Frisch et al. Reference Frisch, Pierrehumbert and Broe2004) reflect ‘the functional pressure to make language processing as easy as possible’ (p. 346), particularly by avoiding difficult-to-process sequences of similar sounds. Frisch develops this argument by presenting an extensive and highly informative overview of current theories of language processing – spoken word recognition, lexical neighbourhood effects and phonological encoding in speech production. He highlights the ‘problem of serial encoding’ – as psycholinguistic research shows, sequences with repeated or similar items present problems for speech production and perception, leading to errors and confusion. While arguing for the relation between the phonetic/phonological processing and OCP effects in the lexicon (which may be part of speakers' phonological knowledge), Frisch views it as essentially diachronic: ‘over time, functional pressures on the language have shaped the lexicon by influencing borrowings, the creation of nonce forms, and the loss of lexical items’. He further notes that patterns of reduplication and harmony present an apparent problem for the serial encoding-based processing account, and suggests that these phenomena have to involve more specialised cognitive processes or result from other functional pressures. With respect to harmony, he adds that ‘it may merely be the case that some languages place functional value of repetition avoidance above the value of harmony, while others do the reverse’ (p. 361). This line of reasoning, however, seems problematic, as similarity avoidance and harmony patterns may coexist in a given language (for example, the OCP place effects in Ngbaka, mentioned by Frisch, are accompanied by consonant harmony involving laryngeal features and nasality; Hansson Reference Hansson2001).

In stark contrast to most of the above mentioned contributions, Juliette Blevins & Andrew Garrett, in ‘The evolution of metathesis’, explicitly argue for an indirect, diachronic influence of phonetics on phonological patterns: ‘certain sound patterns are cross-linguistically frequent as a consequence of convergent evolution: the intrinsic properties of speech perception and production result in certain frequent sound changes; these in turn yield common sound patterns’ (p. 118). The goal of this approach, also known as Evolutionary Phonology (Blevins Reference Blevins2004), is to develop ‘phonetically plausible diachronic explanations for phonological patterns’, thus simplifying synchronic models of grammar. Instrumental to this goal is understanding the mechanism of sound change and sources of its regularity. Building on Ohala's work (e.g. 1981), Blevins & Garrett view sound change as resulting mainly from listener-based misperception and reinterpretation of phonological input. The source of regularity in sound change, according to them, is rooted in categorical perception and phonological acquisition. In this paper, the authors concentrate on various kinds of consonant and vowel metathesis, a phenomenon that has been traditionally problematic for phonological theories. The typology of metathesis they present (illustrated by a wealth of examples) distinguishes four main types, for expository reasons referred to as ‘perceptual’, ‘compensatory’, ‘coarticulatory’ and ‘auditory’ metathesis. To provide a background for the analysis of these types, Blevins & Garrett review relevant articulatory and perceptual factors. First, ‘elongated phonetic cues’ to some features such as pharyngealisation, rhoticity and laterality tend to span multiple segments, thus creating a considerable segmentation problem and ultimately resulting in cases of perceptual metathesis (e.g. CˁVC→CVCˁ). Second, V-to-V coarticulation under certain prosodic conditions can give rise to an ‘extreme’ anticipatory coarticulation, resulting in a C-V shift, a case of compensatory metathesis (ˈV₁CV₂→ˈV₁V₂C). Interestingly, this type of metathesis is almost exclusively attested in two language families (Austronesian and Pama-Nyungan), whose languages are known for their small and highly variable vowel inventories (and thus presumably greater degree of coarticulation). Third, extreme cases of C-to-C coarticulation can obscure consonant place cues (more for some places than others), producing patterns of coarticulatory metathesis (e.g. [pk]→[kp]). And finally, sibilant noise in sequences of sibilants and stops tends to be ‘decoupled’ from the auditory stream, leading to confusions in segment sequencing – cases of auditory metathesis (e.g. [sk]→[ks]). Blevins & Garrett conclude the paper by making specific predictions about the types of metathesis that are impossible under the outlined approach (while being predicted by the functionalist approach), as well as predicted, but not yet unattested types.

In the context of this volume, Blevins & Garrett's paper raises intriguing questions about possible alternative accounts of the phenomena examined in many of the papers reviewed above. For example, place assimilation in clusters can be alternatively analysed as resulting from C-to-C coarticulation and subsequent listener-based misperception/phonologisation (Blevins Reference Blevins2004, Reference Blevins, Solé, Beddor and Ohala2007; but see Steriade Reference Steriade, Hume and Johnson2001). To what extent such reanalysis of this and other cross-linguistic patterns is possible is an issue of particular importance for the field of theoretical phonology, as it has implications for the synchronic architecture of phonological grammar and models of phonology–phonetics interface. It is expected that future progress in this direction will be achieved through rigorous experimentation and modelling, combined with extensive typological studies.

In conclusion, the papers showcased in Phonetically based phonology truly represent some of the highlights of the recent research on the role of phonetics in phonology (together with work by Paul Boersma, Elizabeth Hume, Jaye Padgett, and Daniel Silverman, among others). Given the overall diversity of the topics, fascinating typological findings and ground-breaking theoretical analyses, the current volume is likely to appeal to a wide range of scholars working in the areas of theoretical phonology, phonetics, the phonetics–phonology interface, linguistic typology and sound change.