Date: Sat, 1 Jun 2002 22:31:42 -0500 From: Mike Cahill Subject: Hall (2001) Distinctive Feature Theory
Hall, T. Alan, ed. (2001) Distinctive Feature Theory. Mouton de Gruyter, vii+372 pp., hardback ISBN 3-11-017033-7, $78.00
Mike Cahill, SIL International
DESCRIPTION OF CONTENTS As Hall in his Introduction chapter points out, the rise of Optimality Theory in recent years has unfortunately meant a shift away from examination of featural issues. But the lack of discussion does not mean all featural problems have been solved. This book contains a selection of papers from the Conference on Distinctive Feature Theory held in Berlin in 1999. Since the number of papers is limited, I will summarize and comment on each of them individually. Then I will comment on some issues that are common to most of these.
SUMMARY AND COMMENTATY ON INDIVIDUAL PAPERS 1) T. A. Hall opens the book with "Introduction: Phonological representations and phonetic implementation of distinctive features." Hall (H) describes the main purpose of the book as twofold. The first is to address the nature of featural representations in phonology, in particular what features are required, i.e. underspecification, and the organization of features, i.e. feature geometry. The second is to address the implementation of phonological features in the phonetic component.
H presents an excellent summary of the history of phonological representations of features. He sketches the view of features as binary, non-ordered bundles of features, examines the concept of binarity and how these phonological features have been regarded as distinct from phonetic features, though he notes some recent researchers who have questioned this distinction. He traces the rise of Autosegmental Theory, illustrating with nasal harmony processes, and culminating in the more richly articulated structure of Feature Geometry, for which he goes into some detail.
H then discusses the various views which have been held on underspecification, including privativity. Inkelas (1995) has proposed a theory of Archiphonemic Underspecification specifically in the context of Optimality Theory, which H summarizes. H briefly mentions that some features once thought to be segmental ones, such as [syllabic], are now captured by other entities ? prosodic features.
The phonology-phonetics connection is the last major section of H's discussion. In traditional generative phonology, there are stages of derivation and representation between underlying phonological representation and the final surface phonetic representation, but all this has been called into question in recent years.
In all these sections, H discusses how the papers in the volume relate to the various topics, with some papers addressing more than one issue. All in all, this lays a very good foundation for the rest of the book, and is an excellent source of references for the overall topic of what issues about features have been in the forefront of the generative phonology enterprise in the last several decades.
2) "Laryngeal dimensions, completion and enhancement," by Peter Avery and William J. Idsardi. In this contribution, Avery & Idsardi (A&I) lay out a new schema of Feature Geometry, for which they offer evidence in the area of laryngeal features. This not only adds another layer of structure to the entire feature tree, but also offers a novel interpretation to terminal nodes. Rather than the Laryngeal node having daughter features [spread], [constricted], and [voiced], as in Clements & Hume (1995), for example, A&I interpose a "dimensions" set of nodes between the individual features and the Laryngeal node. The terminal features are to be interpreted as motor instructions, and, having quite a lot in common with Browman and Goldstein's "gestures," this is what A&I call them. There are three "dimensions" under Laryngeal, each having its two opposing feature gestures. This model has its basis in motor control; muscle groups form antagonistic pairs, and each dimension also has antagonistic gestures. The Glottal Width dimension has [spread] and [constricted] gesture dependents, the Glottal Tension dimension has [stiff] and [slack] gesture dependents, and the Larynx Height dimension has [raised] and [lowered] gesture dependents. All the dimensions are daughters of the Laryngeal articulator node. A&I limit the discussion to the two dimensions of Glottal Width and Glottal Tension.
In A&I's model, contrast in a language is on the level of the dimension, which is monovalent; any particular dimension node is either there or not. The phonetic details of gestures are added in a process they term "completion." Besides this completion, A&I's model includes "enhancement" of contrast, adding another contrastive dimension with its daughter gesture to yield a phonetic equipollent contrast.
A&I provide exemplification and argumentation of their model from English, Japanese, and Korean. In their analysis, English contrasts the presence vs. absence of Glottal Width in obstruents. In the unmarked segments (traditionally "voiced" obstruents), there is variability in voicing depending on context, while the marked Glottal Width dimension ones (traditionally "unvoiced") are more consistent across contexts. In the brief discussion of Japanese, A&I take the well-known Rendaku process blocked by Lyman's Law as evidence that the voiced series is marked, so the contrast is between unmarked and Glottal Tension, and that this contrast is enhanced by the insertion of the Glottal Width dimension, with the completion being accomplished by the introduction of [spread]. A&I's most detailed analysis is of Korean. Korean has a three-way distinction between stops: plain voiceless, voiceless aspirated, and what has been called tense or fortis. A&I's theory prevents them from adopting a variant of Lombardi's (1991) [constricted] analysis of the fortis consonants, and they present several arguments in favor of a length analysis of these. The basic laryngeal contrast in Korean then becomes one of the presence vs. absence of Glottal Width.
3) "Representational economy in constraint-based phonology," by G. N. Clements. This is the longest paper of the volume, and in it, Clements (C) deals with the nature of lexical representations, in particular, what features are specified in the lexicon and phonology. He proposes a theory of Active Feature Specification, in which all features are potentially available to the language-learner, but the only ones which actually occur in lexical and phonological representations are those features which are "active," required for expression of lexical contrast or phonological regularities. C assumes a serial approach to a constraint-based phonology here, as advocated in Clements (2000).
C sketches several problems with underspecification theory in the past, e.g. that previous accounts of underspecification have been based on such considerations as the supposed limitations of the human computational faculty (see Odden 1992 for a specific argument against this notion), and that there are many ways to potentially underspecify a system. C proposes that any underspecification ("representational economy") is a result of the language learner's empirical observations of the language.
C proposes a universal scale of "accessibility" for features, starting with [coronal], then [sonorant], [labial], non-sonorant [dorsal], [strident], [nasal], etc., from which he can build consonant inventories with a high, though not perfect, degree of success. The first 4 features yield /p t k m n/, which are the commonest sounds in languages, which supports his ranking. (He does not discuss how [nasal], 6th on the list, would add any more segments than the [sonorant] already did.) He uses this ranking of features in an algorithm developed by Dresher (2000 and earlier works) to specify what features are present in the lexicon. For Hawaiian, with consonants /p m w n l k h ?/, C first inserts a [+] value for [sonorant], which distinguishes the rest from /p k/. Then [+labial] distinguishes /p m w/ from the others. The [dorsal] feature, next in order, is totally redundant and is not entered. In turn, [nasal], [spread], and [constricted] are applied, resulting in a table of [+] features and many unspecified places. In this table, /k/ is totally unspecified, and this is supported by several lines of evidence of the unmarkedness of /k/ in Hawaiian, such as borrowings into Hawaiian, e.g. Peter -> [pika].
C maintains that features are present in the lexicon only when necessary to distinguish phonemes. In the phonological level, however, features may be added which are redundant but are required to express phonological generalizations. C proposes an Activation Criterion: "redundant feature values are specified in all and only the segments in which they are active." If there is a constraint spreading [nasal], than [nasal] will be activated even if it has not been lexically specified. He illustrates this with Zoque, first going through the algorithm to specify lexical features for each consonant, then inserting [voice], which is redundant and not lexical, to account for postnasal voicing of stops.
Carrying the flag of representational economy further, C next examines feature organization, often represented in a feature tree as in Feature Geometry. He notes that while the structure of the tree is usually held to be the same in all languages, many tiers and features do not play any role at all in a particular language. These are superfluous. In order to eliminate this superfluity, C proposes that for a particular language, "all and only prominent features and nodes are projected onto separate autosegmental tiers." "Prominent" is defined as meeting at least one of 4 conditions: a) it is the argument in constraints SPREAD(X), AGREE(X), or OCP(X), b) X is a floating feature, c) X forms part of a monosegmental contour, or d) X constitutes a morpheme. C still maintains there is a universal geometry; what varies from language to language is what portion of this is actually projected from the root node. One significant consequence of this is that languages may vary in their representation of the same segment, since one or more features of that segment may not be active in both languages. A particular segment thus may vary crosslinguistically in both which features are autosegmental and whether those features are necessary at all to distinguish phonemes. C applies this schema to Zoque nasal place assimilation.
In his main ending sections, C applies his system to well- known cases of coronal transparency such as Tahltan and Basque. Transparent coronals are typically unmarked, and so the tree structure is not projected for them; thus they will not be available to block long-distance assimilation. He also discusses transparency of voicing in Japanese.
This is a fairly thorough view of a new approach to underspecification, and deserves attention and testing against other language data. The main reservation I have on it is not the main point, but the nature of the constraints presented. Often C's "constraints" specify a process and so sound identical to an informal formulation of a rule, e.g. "INSERT ([+voice])" (p.92), "PL-ASSIM: Given a nonhomorganic [+nasal] + [-continuant] sequence occurring in an input, spread the Place node of the stop to the nasal in the output" (p. 104) and "Delete [y] in a place-linked coronal cluster" (p.107). What is the difference between these and a rule? C calls the first of these both a constraint and an operation in different paragraphs.
4) "Place of articulation first," by Mirco Ghini. "First" here relates to an algorithm for assigning underlying vowel features which presumably derives from language acquisition: place features are assigned before height features. Ghini (Gh) deals with vowels of Miogliola Ligurian (from Italy), and assumes a derivational model of phonology, with underlying representations being free of redundancy. In his Feature Geometry representation, [labial], [coronal], and [dorsal] are all monovalent and daughters of an Articulator node under place. However, [labial] for vowels is (functionally but not structurally) a secondary feature providing rounding, thus only being possible to assign after another articulation feature for frontness or backness is assigned.
Miogliola Ligurian has short vowels in stressed position as follows:
Front front rounded back back rounded i y u e œ o æ a
Vowels in unstressed position are limited to: i y u E (epsilon) a
Using the Place of Articulation First principle, Gh assigns features to the 8 short vowels, starting with Dorsal (since Coronal is unmarked and Labial is a secondary articulation). Then he assigns Labial to /u,o/. At this point a tripartite division has been made: a back vowel, a back rounded vowel, and a front vowel. He then inserts Low to differentiate contrasts within each subsystem, thus assigning Low to /æ/ and /o/. All back vowels are now distinct, so no further specification is needed for them. For front vowels, a height distinction is needed, so High is inserted for /i,y/. Though Coronal is unmarked, he needs to insert it at this point on /y, œ/ to give a platform for the Labial feature. The featural result is:
[Coronal] [Dorsal] [Dorsal] [Labial] [Labial] [High] i y e œ a u [Low] æ o
In Gh's view, phonological features arise from relationships and contrasts, but not necessarily from absolute phonetic values. Thus /o/ can be phonologically [Low] though phonetically mid. Gh goes through a similar exercise for the long vowels of Miogliola Ligurian as well.
The use of this approach comes in considering the neutralization of some vowels in unstressed position, e.g. /e, æ/ -> /E/ and /u, o, œ/ -> [u]. There is an asymmetry here in that the neutralized vowel in the first case is phonetically mid, but phonetically high in the second case. However, if we consider the featural specifications above (and assume [E] is essentially equivalent to [e], something Gh glosses over), we see that there is a unity: the /u o/ pair is the height counterpart of the /e æ/ pair. The process delinks Low in unstressed position for both. This also correctly leaves the pair /y œ/ unaffected, since there is no Low assigned to either of these.
The motivation for the assignment of Low before High as a feature is unclear to me. Gh says there will always be a phonologically Low vowel, but why should this be? Unlike Clements, who takes a similar approach, Gh does not lay the groundwork for prioritizing some features over others. One is left with a feeling, justified or not, that this entire algorithm, especially prioritizing Low over High, is posited mainly to make the admittedly complex reduction system of Miogliola Ligurian work. If High were prioritized, for example, the entire system would not work, since /u/ would be High and /o/ unmarked. It would be interesting to apply Gh's algorithm to a completely unrelated language and set of processes. Such would be needed to plausibly claim universality for this approach.
5) "Representing nasality in consonants," by Janet Grijzenhout. In this paper, Grijzenhout (Gr) proposes two types of representation for nasal consonants. The first, which she terms "plain nasal stops," have features [+son, +cons], together with [nasal], for which she assumes privativity. The second type, which she terms "light nasal stops," lack the [nasal] specification, but are phonetically interpreted as nasals, since the absence of the privative [continuant] feature implies a lack of oral airflow. This typology is meant to account for the different behavior of nasals with regard to spreading ("nasal harmony"); the first type of nasal initiates nasal harmony, but the second type does not.
Gr presents phonetic and phonological arguments for the existence of both types of nasals in Acehnese: nasals with a relatively large velic opening trigger rightward nasal harmony (adjacent vowels are nasalized), while those with a smaller velic opening do not. On the other hand, Brazilian and European Portuguese differ in that Brazilian has nasal harmony, while European does not, and G attributes this to the two types of representation mentioned above. Furthermore, in language typologies of nasalization, she attributes the different nasal segment inventories and processes to whether a [nasal] specification is present or not.
To further account for nasality patterns in Navaho, Southern Paiute, and Terena, Gr invokes a SV ("sonorant voicing") feature, following Rice (1993), though holding that the traditional [voice] and [sonorant] are still required for some languages. The more sonorous a segment is, the more compatible it is with SV, but the SV varies crosslinguistically as to what it can attach to. From Gr's brief presentation, it is difficult to see what advantage SV has over [voice] in these cases.
Gr also mentions languages with nasals as the epenthetic consonant, and predicts on the basis of her representations that such languages will not have nasal harmony, since the epenthetic nasal would be minimally specified, not including the [nasal] specification. Conversely, in languages with nasal harmony, it is predicted that a nasal is not a possible epenthetic consonant.
Gr does not formalize any rules or constraints, although her prose indicates a predominance of constraint-based thinking in her analysis.
One distinction that might have been helpful is a distinction between contrastive and noncontrastive nasalization. In many languages, such as English, vowels are allophonically nasalized before or after a nasal consonant. Is this nasal harmony? It is not generally regarded as such. But if this is not nasal harmony, Gr would presumably say that such languages, including English, are not specified for [nasal]. But if there is no [nasal] on a nasal consonant, how does such a vowel get its nasality?
6) "Pattern, pervasive patterns and feature specification," by K. David Harrison and Abigail Kaun.
The focus of this paper is underspecification within Optimality Theory. Harrison and Kaun (H&K) claim, counter to traditional interpretation, that predictably alternating features are not necessarily underspecified, and conversely, that segments that never alternate may be underspecified.
In spite of Lexicon Optimization, OT leaves room for partially unspecified lexical structures. H&K discuss the model of Archiphonemic Underspecification (Inkelas 1995). This model predicts that underspecification should only occur when there is a predictable alternation. In this view, Turkish suffix vowels would be underspecified for backness, since they alternate, but stem vowels would be specified for backness.
Hungarian, however, poses a challenge for this view. Hungarian has a vowel length contrast. For two vowels, the length contrast also carries a quality difference. In one of these, long [e:] corresponds to short [æ] ('ash'). In normal words, short [e] does not occur. In the word game "Veve," however, a sequence of /Vv-/ is inserted before the rhyme (e.g. itt -> iv-itt, ti:z -> t-iv-i:z, etc.). For words with /e:/, short [e] occurs in the reduplicant. H&K present tableaus with fully specified /e:r/ input which gives the correct [eve:r] output, then present a tableau with underspecified input /E:r/ which, with the same set of constraints as the first tableau, yields two equally optimal outputs, [eve:r] and [ævæ:r]. H&K interpret this as evidence that Archiphonemic Underspecification is inadequate, and that even predictable alternations must sometimes be fully specified.
However, I would think that an additional constraint favoring faithfulness of quality in the base over faithfulness of quality in the reduplicant would suffice to rule out the unattested form in their second tableau; they do not consider this option. In addition, one might question a broad assertion whose main evidence is based not on the core phonology of a language, but on the fringe elements of a word game. The position they present, that predictably alternating segments are not necessarily underspecified, I would judge has not been well established here.
H&K propose one constraint for Hungarian, Input-Output Identity for length, abbreviated IDENT-I/O[LG], which raises all sorts of issues on featural specification and representation. This formulation assumes "length" is a feature. But much effort has gone into designating length representationally, as two timing units on a CV tier, for example. Since so much rides on length in H&K's paper, would have been good for them to have expounded a bit on their view of what length is and how it should be represented. In H&K's other assertion, that non-alternating segments may nonetheless be underspecified, they examine evidence from Tuvan and, briefly, from Finnish and Turkish. In Tuvan reduplicated forms, the reduplicant replaces the first vowel of the base with [a] in most cases (the exception being when the base already has [a], in which case the reduplicant has [u]). Crucially, inputs which are disharmonic also have disharmonic reduplicants. But harmonic inputs have harmonic reduplicants, even if they need to re-harmonize because of the reduplicant [a]. For example, /idik/ -> [idik-adIk] (where I = barred i here). In H&K's system of constraints, /idik/ must specify [-back] for only the first vowel, not both, in order to produce the correct winning candidate. A word game in Finnish works in a similar manner. Turkish lacks similar reduplication, but H&K, in a pilot experiment, taught Turkish speakers the Tuvan-style reduplication and asked them to produce the most Turkish-sounding word from the process. The result was similar to the Tuvan.
As a result of these, H&K propose that predictable feature values are in a continuum between "systematic" and "idiosyncratic," rather than a strict binary bifurcation of "alternating" vs. "non-alternating," and that features which can legitimately be underspecified are in the area of "systematic" predictable patterns. The exact boundaries are fuzzy. Languages may be located on various points on the continuum.
7) "Phonetic implementation of the distinctive auditory features [voice] and [tense] in stop consonants," by Michael Jessen. In this paper, Jessen (J) presents a literature overview of the features [voice] and [tense] for obstruents, as well as some new proposals of his own. As the title promises, this is a paper which focuses on phonetic details associated with the two features [voice] and [tense], in particular auditory perspectives. In his overview of [voice], he notes that while this feature is generally interpreted as vocal fold vibration, such vibration is not a single phenomenon, but several gestures that need to be coordinated. Thus languages may be analyzed with [voice] even if the actual presence of closure voicing is not known or measurable. In using the feature [tense], J contrasts with other recent views of aspirated stops as having a [spread glottis] feature or the like, and explicitly seeks to reintroduce discussion of [tense] in current circles.
In J's view, a language may have contrasts between /p t k/ and /b d g/, and depending on phonological patterns of the language, these contrasts might be captured by either [voice] or [tense]. If there is long aspiration in /p t k/ and unstable voicing in /b d g/, a [tense] analysis is most plausible; if /b d g/ has full voicing and /p t k/ is basically unaspirated, a [voice] analysis is preferred. Languages like Hindi which have a three-way distinction need both.
J distinguishes between "basic correlates" and "non-basic correlates" of these features. Basic correlates are those phonetic properties which are unique to a feature, and so are often very stable across various contexts. Aspiration duration is the basic correlate of [tense] in a tense-lax system, while voicing during closure is the basic correlate of [voice] in a voiced-voiceless system. Non-basic correlates are those which are shared by other phonological features, and may be less stable across various contexts, include F0 onset, F1 onset, differences between amplitudes of first and second harmonics, closure duration, preceding vowel duration, following vowel duration.
J also briefly discusses what this type of analysis would look like with the feature [checked], used with clicks, ejectives, and implosives, this feature having a basic correlate of high burst amplitude (which must be interpreted as the spectral sharpness of vowel onsets for implosives). He presents a table with [+] and [-] values of all three features discussed, and proposes that all types of stops are covered with the resulting 8 combinations of features. Some, such as grouping Korean fortis stops, ejectives, and plain clicks together as [-tense, -voice, and +checked], might be regarded as controversial.
While references to quite a few languages are made, no actual language data is presented in this paper. Clarity would have been helped if it had, though it would have lengthened the paper considerably.
8) "Distinctive [voice] implies regressive voicing assimilation," by Bertus van Rooy and Daan Wissing. In this paper, van Rooy and Wissing (vR&W) argue for a narrow interpretation of the feature [voice], that is, the feature is used only when the distinction is between actual vocal fold vibration and its absence. (A broader interpretation of [voice] uses the feature to mark the contrast when any two of these three phonetic situations contrast in a language: negative voice onset time (VOT), short-lag VOT, or long-lag VOT.) VR&W, like Jessen in this volume, contend that voicing distinctions in English are marked by the feature [tense], so that variations in voicing such as "cat[s]" and "dog[z]" are not to be considered when discussing [voice].
VR&W present some data on word-final devoicing, postnasal voicing, and intervocalic plosive voicing, but the main data emphasis is regressive voicing assimilation, for which they give data from Dutch and Polish. They cite several scholars who note that it is languages with prevoicing that exhibit this regressive voicing assimilation, and this seems likely to be a universal. (They do discuss some counterexamples from Dutch and Afrikaans of progressive devoicing which often occurs in preference to regressive voicing, concluding that there is much variation in these languages, and in particular, that the feature [voice] is not well established for these.)
The authors conducted an experiment to test this hypothesis. English, in their view, does not use [voice] distinctively; it has no prevoicing. Tswana does. English does not have regressive voicing assimilation. But when Tswana speakers, also speakers of English, were given sentences containing voiced-voiceless obstruents, as in the name "Dick Dean" /--kd--/, they read these with regressive voicing [--gd--]. The phonological pattern of their native language has carried over into their second language.
VR&W contend that the observation leads to a challenge in OT terms of incorporating in a natural way, a statement of the form "if X, then Y," in particular, "if [voice] is distinctive in a particular language, then regressive voicing assimilation automatically applies." Specifically, the ranking
which yields regressive voicing assimilation must be obligatory if [voice] is distinctive, as shown by the Tswana speakers.
One could wish that vR&W had been more explicit about the privativity of [voice], which seems to be assumed, but is not explicit. They deliberately chose a more general FAITH constraint schema, rather than the more specific IDENT family, or even the possible MAX and DEP constraints, which would be possible when dealing with the presence or absence of a particular feature, and one might also wish they had taken a reasoned stand on a more specific approach to faithfulness.
Konni, a language of northern Ghana, offers a radically different output from the cases vR&W examine (Cahill 1999). Konni has prevoiced stops, but when voiceless and voiced obstruents adjoin, a vowel is inserted, e.g. /biis-bu/ -> [biisibu] 'the breast' (cf. [biis-a] 'breasts'). Regressive devoicing occurs, as in /tig-ka/ -> [tikka] 'the village'. Quite interestingly, the constraints needed for Konni are virtually identical in formulation and ranking to the one above, with the exception of *ObsVoice, which is not used for Konni, and in fact makes no significant contribution to the tableaus vR&W present.
9) "The phonology of /r/" by Richard Wiese. In this paper, Wiese (W) attempts to capture what unites all the different sounds called some sort of "r", or rhotic, when they have no consistent phonetic unity. He concludes that they must be defined prosodically, by their position in the sonority hierarchy, rather than featurally.
W starts with a survey of various phonetic sounds that can be called rhotics. While there is no consistent phonetic basis for unifying rhotics as a class, there is a phonological basis for unifying them (see Walsh-Dickey 1997 and Ladefoged & Maddieson 1996). W lists several criteria, for example, that rhotics often alternate with other rhotics, both synchronically and diachronically.
Can we simplify the list of rhotics? W discusses excluding uvulars or fricatives but rejects these. Languages which have uvular rhotics, such as French, lack coronal ones, so one can still talk about "r" in these languages without ambiguity. Fricative rhotics still participate in the alternations mentioned above.
W then considers, but rejects, possibilities for defining rhotics in terms of featural specifications. The trilled alveolar /r/, termed the prototypical /r/ by W, is fully specified by [+cons, +cont, +son, -lateral, +coronal, -nasal], but this is not constant across rhotics, for which only [+cont] and [-lateral] always hold. Hall (1997), as well as others, has proposed a feature [rhotic] to capture the generalizations of what seems to be a natural class, but W maintains this covers the problem rather than illuminates it. Walsh-Dickey (1997) proposes that rhotics are sounds with a non-primary Laminal node. But W notes that rhotics are generally not laminal, that Walsh-Dickey's system vastly overgenerates possible geometries, and other objections. An underspecification approach would look for what feature/s all rhotics have in common, specify only them, and build from there. [+cont] might seem a good one to start with, but this is not enough to distinguish rhotics from other sounds, and for languages with more than one rhotic, this approach fails.
All of these featural-based approaches proving inadequate in one way or another, W proceeds to propose that /r/ is actually a prosody, more precisely, "/r/ is that point on the sonority scale between laterals and vowels." This embodies two claims: first, that /r/ occupies a unique place on the sonority scale, and second, that this is a plausible way of actually defining rhotics.
W presents evidence that /r/ in both French and German acts as a sonorant even when phonetically a fricative, not grouping with other fricatives in its phonotactics. Given that a rhotic can be a voiceless fricative, it seems inconceivable that this would occupy the same position on the sonority scale as say, an voiced approximant. W asserts that the sonority scale which defines /r/ is phonologically abstract, incorporating neither direct phonetics nor a specific list of segmental features. Though /r/ in a specific language would need specific features, that is not its basic identity.
There is recent corroborating evidence for at least some of W's claims. Parker (2002) in a groundbreaking phonetic study of sonority, finds no distinction between /r/ and /l/ in the phonetic sonority hierarchy in English. However, he finds several lines of phonological evidence that /r/ is more sonorous than /l/, so in the phonological hierarchy, he agrees with W, and claims this ranking is a universal.
There is an index of languages and one of subjects as well in this volume.
COMMENTS ON THE BOOK AS A WHOLE The book is fairly free of typos, though the word "Feature" was omitted in the first footnote, there is a gratuitous "ist" on p. 147, 'my brother' has differing transcriptions on p. 199, there appears to be a missing [-B] in the input of Tableau G on p.227, and a few others. A personal preference of mine would have been footnotes rather than endnotes.
The authors have evidently done some substantial revising of the papers as originally given at the conference under the guidance of various referees, and though the conference was in 1999, there are many updated references, even to 2001.
Below I very briefly summarize how the authors intersected with the main purposes of the volume.
Regarding Underspecification: Avery & Idsardi have a view of phonology in which nothing redundant is specified in the phonological representation, similar in spirit to Archangeli's Radical Underspecifi- cation (1984). Grijzenhout crucially relies on the presence or lack of underlying [nasal] to explain different nasality patterns. Harrison & Kaun develop a whole schem of when underspecification is needed. Van Rooy & Wissing's discussion of [voice] seems to assume privativity. Both Clements and Ghini explicitly assume models in which no redundancy is present in underlying specification, but with significant differences from to Radical and Contrastive Underspecification. Jessen and Wiese do not address the issue.
Regarding structure of representation (Feature Geometry): Avery & Idsardi, Clements, Ghini, and Grijzenhout basically accept Feature Geometry, though A&I have their own extension of it. Harrison & Kaun use autosegmental notation in their OT inputs, but basically ignore it elsewhere, not commenting on any Fature Geometry issues. Both Jessen's and Van Rooy & Wissing's discussions of [voice] does not mention connection to Feature Geometry structure, and an autosegmental representation of [voice] is not assumed either, though Jessen does call for research into whether Feature Geometry can resolve some of the issues he addresses. Wiese seems to assume Feature Geometry, but it does not play a major part in his presentation.
The role of phonetics: Avery & Idsardi encode phonetic gestures into the Feature Geometry, but limit the role of these gestures to enhancement, not contrast. Phonetics is directly represented by their "gestures." Grijzenhout's schema relates the presence vs. absence of [nasal] to degrees of velic opening in Acehnese, but presumably two languages with different specifications of [nasal] could still have phonetically identical [n], for example. The connection of phonological specification and phonetics is thus variable. Van Rooy & Wissing's paper explicitly ties the feature [voice] to a specific phonetic interpretation, i.e. negative VOT. Jessen does the same, also tying [tense] to aspiration, and majors on auditory correlates of features. In Ghini's approach, there is not a direct mapping of phonetics with phonology, in that a high phonetic vowel may not get a High feature, likewise with a Low. Wiese has a crucial disconnect between phonetics and phonology, when he proposes his phonologically abstract sonority scale which has rhotics occupying a single place, though their phonetics may vary considerably. Clements and Harrison & Kaun do not discuss phonetic details.
In a book titled "Distinctive Feature Theory," it is disappointing to not find anything on some major topics. Given the provenance of the book, where all the papers came from a conference, this is not surprising, of course; one is at the mercy of what the participants propose as far as topics go. So the following is a kind of a wish list of topics I would have liked to see covered in a more systematic treatment of features.
One such topic is tone. With perhaps half the world's languages tonal to one degree or another, this would have been a relevant topic. It is true that Avery & Idsardi's paper alludes to their enriched Laryngeal node proposal as intersecting with tone, specifically Glottal Tension and Larynx Height. They say "We also believe that laryngeal features capture both phonation type and tone", which is likely putting quite a burden on these nodes: to include ejectives, implosives, and tone all on the Larynx Height node. They refer the reader to an unpublished paper of theirs for more details.
Another wished-for topic is vowel height and place. A piece of writing that has not gotten the attention it deserves is Parkinson 1996, who shows that the traditional [high] and [low] are inadequate to account for languages with at least five vowel heights and certain phenomena. The place of ATR would be another topic.
Another topic is whether the whole concept of Feature Geometry has any place in current Optimality Theory. Writers in this volume tended to either assume FG (as Grijzenhout did), develop details of it, as Avery & Idsardi did, or assume Optimality Theory and ignore FG, as van Rooy & Wissing did. Padgett (2001, and earlier formulations going back to 1995) discusses this issue directly and claims OT basically does away with the need for FG, while Cahill & Parkinson (1997) claim Padgett's schema is basically a notational variant of FG. More discussion on this would be helpful (though Clements does raise the question "What is the trade-off between constraints and representations in understanding phonological regularities?") As Rennison (2000) notes, the fact that OT does not speak to the area of what features are allowed multiplies the number of constraints that must be considered, since the constraints must refer to structures, which vary according to the theory of representation adopted.
REFERENCES Cahill, Michael. 1999. Aspects of Morphology and Phonology of Konni. Ph.D. dissertation, Ohio State University.
Cahill, Michael, and Frederick Parkinson. 1997. Partial Class Behavior and Feature Geometry: Remarks on Feature Class Theory. Proceedings of NELS 27, pp. 79-91.
Clements, George N. 2000. In defense of serialism. The Linguistic Review 17.2-4: 181-197.
Clements, George N., and Elizabeth V. Hume. 1995. The Internal Organization of Speech Sounds. In Goldsmith, John (ed.). The Handbook of Phonological Theory. Cambridge, MA: Basil Blackwell, pp. 245-306.
Dresher, Elan. 2000. Contrastive features: Manchu to Miogliola. Paper presented at the Workshop on Features, Schloss Freudantal (Konstanz), December 7-9, 2000.
Hall, T. Alan. 1997. The Phonology of Coronals. Amsterdam & Philadelphia: John Benjamins.
Inkelas, Sharon. 1995. The consequences of optimizationfor underspecification. Proceedings of NELS 25.
Ladefoged, Peter, and Ian Maddieson. 1996. The Sounds of the World's Languages. Cambridge, MA: Blackwell Publishers.
Odden, David. 1992. Simplicity of underlying representation as motivation for underspecification. In Elizabeth Hume (ed.) Papers in Phonology (OSU Working Papers in Linguistics 41), 85-100.
Padgett, Jaye. 2002. Feature classes in phonology. Language 78: 81-110.
Parker, Steve. 2002. Quantifying the Sonority Hierarchy. Ph.D. dissertation, University of Massachusetts Amherst.
Parkinson, Frederick. 1996. The Representation of Vowel Height in Phonology. Ph.D. dissertation, Ohio State University.
Rennison, John. 2000. OT and TO ? On the status of OT and a theory and as a formalism. The Linguistic Review 17.2- 4: 135-142.
Rice, Keren. 1993. A reexamination of the feature [sonorant]: the status of 'sonorant obstruents'. Language 69: 308-344.
Walsh-Dickey, Laura. 1997. The Phonology of Liquids. Ph.D. dissertation, University of Massachusetts, Amherst.
ABOUT THE REVIEWER:
BIOGRAPHICAL SKETCH Mike Cahill did on-site linguistic investigation in the Konni language of northern Ghana for several years, including application to literacy and translation work. He received his Ph.D. from Ohio State University in 1999, and is presently primarily interested in African phonology, cross-linguistic patterns in tone, and labial-velar stops and nasals. He currently is SIL's International Linguistics Coordinator.