Two current lines of research in phonology lead in seemingly opposite directions. Substance-based approaches, exemplified by Boersma (Reference Boersma1998) and Steriade (Reference Steriade, Hanson and Inkelas2009), hold that the phonological component of the grammar makes use of detailed phonetic information, and that phonological constraints directly express functional phonetic desiderata such as minimizing articulatory effort and maximizing perceptual distinctness. Bridget D. Samuels, alongside others such as Hale & Reiss (Reference Hale and 2000) and Blaho (Reference Blaho2008), takes the opposite course, positing that phonology is ‘substance-free’ – that is, that the formal mechanisms of phonology are wholly unconcerned with the phonetic content of the representations they manipulate. The particular context in which Samuels pursues this line is the framework of biolinguistics, which considers language as a biological property of the human organism. The scope of Phonological architecture is therefore quite broad, encompassing not only the cognitive phenomenon of phonology itself, but also its interactions and parallels with other components of the grammar, and the extent to which similar phenomena can be identified in other aspects of human and animal cognition.
The book consists of seven chapters. Chapter 1, ‘Introduction’, briefly sets out the aims of the book. Chapter 2, ‘A Minimalist program for phonology’, outlines the rationale for the substance-free approach, and suggests that phonetically substantive phonological patterns are best understood as diachronically emergent. Chapter 3, ‘Phonology in evolutionary perspective’, discusses various facts about animal communication and cognition that may be relevant to how the phonological component of the human language faculty came to be. Chapter 4 deals with ‘The syntax–phonology interface’, and argues that syntactic phases offer the best account for a wide range of cyclic and prosodic effects in phonology. Chapter 5, ‘Representations and primitive operations’, sets out key elements of Samuels's view of the computational apparatus of phonology proper; the main points here are an argument against the syllable and the introduction of a very general mechanism for string-based searching and copying that forms the basis for phonological rules. Chapter 6, ‘Linguistic variation’, returns to the question of the diachronic emergence of sound patterns, discussing possible learning biases that may account for how children deal with ambiguous data in acquisition. A short final chapter, ‘Conclusions’, sums up what has come before.
The picture of phonology that emerges from this book is a distributed one. The formal computational mechanism that instantiates phonological rules in the grammar is fairly simple and very powerful, and resembles a Turing machine more than it does many other models of phonology. Rules take the form of instruction sets that tell the processor to start at a particular position, proceed along the string in a specified direction until it finds a segment matching a given description, and then perform a specified change (copy a feature, or delete something, or insert a new precedence relation). Although an operation called copy is one of the basic components of the system, there is no computational bias in favour of assimilation (as there is in standard autosegmental phonology), because the material to be copied need not already be present in the string; it can be supplied by the rule itself. Representations are also minimal, consisting of linearly ordered feature bundles; there is no geometric organization of features into nodes, and no hierarchical organization of segments into larger prosodic units.
The work that such devices are traditionally understood to do has been shifted in two directions. The effects of the prosodic hierarchy are attributed to syntax, as are the cyclic and level-ordering effects familiar from Lexical Phonology. Both arise from derivation by phase: as the syntactic derivation proceeds, certain types of constituents are shipped off to the phonology for interpretation. Material that has already been interpreted cannot be altered. Two affixes introduced at different levels of syntactic structure may thus exhibit different morpho-phonological behaviour if one is separated from the root by a phase boundary and another is not. At higher levels of structure, phase boundaries delimit multi-word domains for the application of phonological rules. If this picture of the syntax–phonology interface is correct, then the hierarchical structure of syntax does not need to be replicated in a prosodic hierarchy at all; from the point of view of phonology, all the relevant constituents are circumscribed procedurally rather than structurally.
While the application domains of phonological rules are determined by syntax, their content is dictated by phonetics. The rule template itself can express processes that resemble nothing in any natural human language, and Universal Grammar (UG) supplies no constraints that would prevent it from doing so. Rather, the fact that some kinds of rules (such as final obstruent devoicing) are common, while others are rare or unattested, is attributed to the fact that phonological rules develop diachronically from the incorporation into the grammar of naturally occurring phonetic patterns. Coarticulations, reductions, misperceptions, and other such phonetic phenomena become phonological when they are interpreted by learners as rules. Phonetically unnatural sound patterns can arise over time as subsequent changes obscure the origins of older rules, but phonetically natural ones emerge much more quickly and more straightforwardly.
In one sense, this approach is, as mentioned at the beginning of this review, diametrically opposed to theories in which phonology is directly driven by phonetically motivated constraints. Here, the phonology proper is a wholly abstract, purely formal computational device. However, the substance-free and the substance-based views are alike in that they both posit functional phonetic explanations for substantive phonological patterns. In the substance-based approach, the phonetic motivations are expressed synchronically in the grammar itself; in Samuels's substance-free approach, their influence on grammar is diachronic and indirect. But the two lines of thought, in their different ways, both turn away from the practice of constructing formal explanations for substantive patterns. Under either approach, languages that devoice final obstruents do so because voicing is (or historically was) difficult to sustain or to discern in final position, and not because codas tolerate less representational complexity than onsets.
What does it mean to pursue a biolinguistic approach to phonology, or to language in general? In one sense, it means that the field of inquiry becomes broader. Some phonological patterns find an explanation in the interface conditions of syntax, others in the neurophysiological underpinnings of mammalian auditory perception. A successful biophonologist must know enough about both of these things, and many others, to be able to make reasonable inferences about what each of them contributes. In another sense, the central object of study becomes much smaller. If some of phonology is really syntax, and some of it is really phonetics, then relatively little of phonology is phonology. If, as Samuels states (186), citing Mesgarani et al. (Reference Mesgarani, David, Fritz and Shamma2008), ferrets have neurons that respond to the formant pattern characteristic of high vowels, then this suggests that human beings probably are biologically predisposed to make contrastive use of vowel height, but at the same time it casts doubt on the idea that this predisposition is attributable to the presence of [±high] in a set of uniquely human, uniquely linguistic phonological features.
Because biolinguistics is a relatively new endeavour, some of the connections drawn between linguistics and biology seem rather tenuous, particularly those involving the core of the formal linguistic system. For example, in Section 5.8 (‘Operations in evolutionary perspective’), Samuels suggests a parallel between the search operation in phonology and the searching involved in foraging for food. Foraging entails the sensory exploration of three-dimensional space for anything edible; phonological search, on the other hand, traverses a one-dimensional formal object that exists only in the mind, attempting to match a given feature specification. Samuels acknowledges the differences between the two, and suggests that these differences are themselves a fruitful area of investigation. In the current state of the field, though, the connection between foraging and search is largely speculative. It is not obvious that foraging is any more closely allied to search than any other cognitive process that involves pattern matching, and it may be a long time before the study of foraging offers any concrete insights into the study of phonological rules, or vice versa.
Toward the goal of constructing the interdisciplinary bridges that will make such insights attainable, the book is designed to be accessible to biologists as well as to linguists. A glossary of linguistic terms is included in the back matter. (There is no glossary of biological terms, but the biological terminology used in the book does not extend beyond what can be found in a general-purpose dictionary.) A few of the ostensibly phonological representations of English words are presented in orthographic form rather than in transcription, although not in any cases where the phonetic properties of the segments are crucial to the discussion. Finally, and probably as much because of the breadth of the topic as because of the breadth of the audience, no individual phonological case study is discussed in as much depth as it might be in a more specialized work.
Samuels's writing in this book is breezy and confident, and has a conversational tone that is usually engaging, although it can occasionally smack of condescension. An example is her discussion of the question of overgeneration, which includes the following passage:
McCarthy (Reference McCarthy2002: 39) asks a rhetorical question: what should we do if we have a phonological theory that predicts a particular sound pattern to be possible, but “diligent research at Harvard's Widener Library failed to uncover any languages of the predicted type”? Well, although Widener Library is quite large indeed, I know from personal experience that grammars of many potentially relevant languages are housed in Tozzer Library a few blocks to the north; others are only accessible with a special appointment to visit the Archives housed in an underground warren beneath Harvard Yard. I digress, but I hope you see the point: we would be remiss in revising our phonological theory to rule out any sound patterns that are not found in Widener Library's collection. Nevertheless, revision of the theory is precisely what McCarthy suggests. (14)
McCarthy (Reference McCarthy2002) is, fairly clearly, using Widener Library metonymically, to stand for the store of descriptive linguistic knowledge available to scholars in general. (It is also perhaps worth pointing out, as Samuels does not, that McCarthy makes this statement in a discussion of a schematic hypothetical example in the first chapter of an introductory text on Optimality Theory; given the context, a certain degree of simplification is to be expected.) Samuels's excursus on the libraries of Harvard University is probably best read as a high-spirited riff on McCarthy's metonymy: there are always further sources of data to be found, some of which may be deeply buried. It is, however, a little too easy to interpret it as a literal-minded refusal to take McCarthy figuratively at all, or simply as a gratuitous reminder of where the author went to university.
The digression also detracts somewhat from the clarity of the argument in which it appears. Samuels is arguing, following Hale & Reiss (Reference Hale and Reiss2008), that UG defines a set of ‘humanly computable’ languages that is a proper superset of the ‘attestable’ languages – that is, of the whole range of natural human languages past, present, and future, and in all their idiolectal variety. Some humanly computable languages, Samuels claims, are not attestable because they contain phonetically unnatural rules that are vanishingly unlikely to arise diachronically. If factors external to grammar can explain the absence of these rules, then we do not need to complicate our model of UG in order to exclude them. This is a point worth pursuing, but in the Widener argument, Samuels switches the focus to a much easier target, a straw McCarthy who seems to say that UG defines the set of attested languages rather than the set of attestable ones.
The argument against syllable structure in Section 5.3 (‘Towards a flat phonology’) ends with a similarly low blow. In explaining that the theory presented in the rest of the book does not crucially depend on the rejection of the syllable, Samuels writes that ‘if the reader finds the idea of doing phonology without structural syllables to be distressing, he or she need not be discouraged from reading on’ (141). The wording of this statement seems to imply that the only possible objection a reader could have to flat structure is an emotional one. It would have been much more gracious to allow that there might be some rational basis for remaining skeptical even after twenty pages of argumentation. (The syllable, by the way, reappears without comment in Section 6.5, ‘Underspecification and underdetermination’, as a target of metathesis in language games.)
Despite these occasional infelicities, Phonological architecture is a valuable contribution to the study of phonology and related disciplines, and it suggests many promising lines of inquiry for future work. For researchers interested in the relation between form and content in phonological rules, the ideas about acquisition that are touched on in Chapter 6 contain a wealth of possibilities that deserve to be explored in greater depth. If the computational machinery of phonology is as powerful, abstract, and substance-free as Samuels proposes, then a great deal of what we know about sound patterns in language must be attributable to the process by which small children educe formal regularities from the tangled mess of phonetic input they receive.