One of the great achievements of linguistic study in the twentieth century is the awareness and development of the idea that speech sounds are decomposable into smaller properties – variously referred to as features (e.g. Chomsky & Halle Reference Chomsky and Halle1968), elements (e.g. Kaye, Lowenstamm & Vergnaud Reference Kaye, Lowenstamm and Vergnaud1985, Harris Reference Harris1994, Backley Reference Backley2011), components (e.g. Anderson & Ewen Reference Anderson and Ewen1987), particles (e.g. Schane Reference Schane1984) and gestures (e.g. Anderson & Ewen Reference Anderson and Ewen1987, Browman & Goldstein Reference Browman and Goldstein1989). All these terms have been used to identify properties generally considered to be primitives, which form part of the phonological module of the language faculty. Furthermore, recent phonological theories take the position that it is these properties which function as the minimal units of phonological contrast, and regard each primitive as a category associated with its own stable phonetic signature.
What is common to all scientific pursuits, however, is the inevitability of theoretical disagreement. And in the case of phonology, disagreement exists over the number and kinds of primitives that are assumed to be psychologically real. According to San Duanmu in his book A Theory of Phonological Features, this stems from two issues: (i) no existing theory has yet proposed a system of primitives which is based on a sufficient amount of data, such as those which are available from established sound inventory databases; and (ii) the data available in the phonology literature and from databases contain a substantial amount of phonetically vague (and not necessarily phonological) information, together with occasional clerical errors, which has led scholars to make bad choices when building a systematic model of primitives. To address (i), Duanmu closely examines two databases of transcribed sound inventories, UPSID (UCLA Phonological Segment Inventory Database; 451 inventories, 13,966 phoneme tokens; Maddieson Reference Maddieson1984, Maddieson & Precoda Reference Maddieson and Precoda1990) and P-base (628 inventories, 19,959 phoneme tokens; Mielke Reference Mielke2004–2007). With regard to (ii), as a means of determining whether given information is phonological or phonetic, and whether it contains clerical errors or not, Duanmu employs the well-established notion of contrast, making clear what kinds of principles should be used in the research process to arrive at a system of primitives that is, at the very least, capable of distinguishing all sounds in the languages of the world. His arguments are developed within a framework which employs primitives known as distinctive features. Unlike other theories of primitives, the features discussed in Duanmu’s book are conceived under the assumptions that (a) features are bivalent (+/– value specification) with respect to attributive oppositions; (b) a feature cannot be phonetically realized unless it is harnessed to a full set of other (+/– value-specified) features, and these features collectively determine the phonetic identity of a whole segment; and (c) features are defined according to properties of articulation, which itself is supported by the motor theory of speech perception.
The book comprises eight chapters including an introduction (Chapter 1) and some concluding remarks (Chapter 8). After an overview of the field of research and a summary of the book’s main content (in Chapter 1), Chapter 2 discusses the method of identifying the ideal number and kinds of phonological features to use. This is done by appealing to the reasons for using data from the two sound inventory databases (UPSID and P-base), and by referring to the key principles (Principle of Contrast, Maxima First, Known Feature First) employed in Duanmu’s approach to determine the smallest set of features that can succeed in distinguishing between all the sounds in the world’s languages. Using the method motivated in Chapter 2, the vowel contrasts in the two databases are then scrutinized in Chapters 3 and 4, while an analysis of consonant contrasts is provided in Chapter 5. On the basis of the analyses in Chapters 3–5, Duanmu proposes a feature system which employs the minimum number of features required to represent all sound contrasts; this is preceded by a review of seven different feature models found in the literature. Using the proposed feature model, Chapter 7 discusses the representation of complex segments.
Below I discuss several points and arguments which I consider to be important in a review of Duanmu’s work. In my opinion, there are three respects in which A Theory of Phonological Features distinguishes itself from similar books and chapters that discuss the merits of different feature systems for representing the sounds of the world’s languages from a linguistic point of view.
Firstly, in Chapter 2, the author makes a point of casting light on the problems that all phonologists share when they refer to published data from the literature and from databases. He acknowledges that it is almost impossible to find a database that is error-free – partly because some errors are present in the original sources and these are difficult to verify, and partly because errors can also emerge as a result of clerical oversight. He then claims that, to make such databases useful for current research purposes, they should be read and scrutinized in conjunction with three general principles: the Principle of Contrast, Maxima First and Known Feature First (19).
Secondly, by making extensive use of these three principles together with the UPSID and P-base databases, the author investigates the upper bound of the set of cross-linguistically possible contrastive segments and the features that prescribe this set. The book reaches the conclusion (54) that just four binary features ([back], [high], [round], [ATR]) are sufficient for representing the sixteen basic vowels shown in the following table (ART = advanced tongue root):
As a simpler, yet still viable alternative to traditional feature systems in which [high] and [low] are typically used to define three degrees of vowel height, Duanmu proposes to adopt a two-height system to represent basic vowel contrasts by combining the features [high] and [ATR]. It should be noted that, depending on the language being described, the choice of symbol for some cells can vary since the feature system is phonological and therefore not designed to encode precise phonetic values of the kind which are found in IPA transcriptions.
Thirdly, the consonant feature system that Duanmu proposes is, like his vowel feature system, also minimal in size for the purposes of representing all the consonantal contrasts in the world’s languages. Simplex consonants are represented by referring to seven active articulators and the following manner and non-manner features (104). (Note that complex sounds are dealt with separately because they require representations with intra-segmental or inter-segmental hierarchical structure.)
As indicated above, there are restrictions on the combination of articulators and features, these restrictions deriving from how the given articulators move. For example, the articulator Tip, which can be moved backwards and forwards to create various degrees of obstruction against the airflow, can be combined with [location], [narrow], [stop] and [fricative]. On the other hand, the only movement that the articulator Root is capable of involves the advancement dimension: [advanced] or not.
In Duanmu’s system, the traditional manner types are represented primarily by the features [stop] and [fricative]: [+stop, –fricative] and [–stop, +fricative] define stops and fricatives respectively, while [+stop, +fricative] and [–stop, –fricative] represent affricates and approximants respectively. The Oral/Nasal distinction is expressed via the presence of [+/–stop] under Velum, thereby replacing the traditional feature [nasal]: [+stop] and [–stop] for Velum define a given segment as being oral and nasal respectively.
In addition to the articulators and features given above, this system also assumes three values for [location]: [front], [central] and [back]. In order to create the protruded/bilabial/labiodental distinction, all three values are subsumed under the articulator Lips. Likewise, the same three values are given under Tip to distinguish interdental, alveolar and retroflex. Unlike Lips and Tip, the articulator Body has only the values [front] for palatals and [back] for velars; this is because the segment type corresponding to [central] is not articulated under Body.
Due to space limitations, this review cannot account for how to represent the other standard categories of consonants in Duanmu’s system. However, from the brief summary of feature combinations given above it should be apparent that the proposed system is based entirely on the physiological mechanisms involving the vocal organs. It is, of course, a matter of controversy as to whether features should be based on articulation and whether feature combinations should describe the properties of sounds within the phonological module or not. However, if the motor theory of speech is correct, then the type of feature system developed in Duanmu’s book – which employs a smaller number of categories than we find in traditional feature systems – should prove beneficial to theories of sound structure in recognizing the mechanisms of articulation executed in the vocal organs.
Nevertheless, there are several aspects of Duanmu’s approach which should be made clear at this point. First, because some of the proposed features and combined structures are similar to those found in other theories of phonological primitives, these competing approaches should be acknowledged at the relevant points in the discussion. For example, the proposed features [stop] and [fricative] parallel the ‘element’ primitives [stop] (written ǀʔǀ) and [noise] (written ǀHǀ) that have long been employed in Element Theory (Harris Reference Harris1994, Backley Reference Backley2011). And because the definitions of primitives in Element Theory are different from those used in Distinctive Feature theories, the author should refer to the arguments developed in the Element Theory literature as a means of supporting the proposed features and his combined manner representations.
Second, although the author’s attempt to reduce the number of required features deserves a positive evaluation from a minimalist perspective, it may not be judged to have such a significant impact on traditional feature theory as a whole if we choose to interpret certain categories (e.g. the seven articulators, and the [location] values) as being different in kind from other primitive categories. If those categories are regarded as primitives on a par with features, then the extent to which Duanmu has succeeded in reducing the number of features is not as great as it first appears. To avoid giving this impression, the book would benefit from including more discussion on how the articulators differ from nodes/the corresponding units, and how the [location] values +/– are different from the proposed features (e.g. [high] and [round]), for which +/– values are also assigned.
Aside from these minor issues, on the whole Duanmu has produced a book which is well structured and in which the arguments are clearly laid out. In addition, it is clear that his discussion makes the book theoretically distinct from similar publications and contributes to the debate over the nature of an optimal feature system for phonology. Furthermore, the book has proposed not only a feature system that is minimal and sufficient for representing the contrastive sounds observed in all natural languages, but it also contributes to the development and validation of the feature-based model of segment-internal architecture which prevails in current phonological studies.
