In this handsomely produced volume, a number of renowned linguists analyze various issues relating to the choice between a rule- and a constraint-based approach to the description of sound structure. Rule-based phonology (RBP) has its modern roots above all in classical generative phonology (the SPE model of Chomsky & Halle Reference Chomsky and Halle1968, etc.), whereas the major locus of constraint-based alternatives is currently Optimality Theory (OT) (Prince & Smolensky Reference Prince and Smolensky1993, Kager Reference Kager1999). The chief focus of attention in the collection is naturally the choice between processes and constraints themselves, which is inextricably linked to the issue of the nature of phonological representations. Furthermore, since, ordinarily, RBP involves serial derivation whereas classic OT computations take place in parallel, the question of seriality vs. parallelism is intertwined with the rule/constraint issue. In addition, the book touches upon an abundance of other topics, including issues of general theoretical concern and those related to specific details of analysis. Apart from the frontispiece and concluding matter, the book contains eight chapters. For reasons of space, I will discuss only selected points from each of the papers.
In chapter 1, ‘Introduction: The division of labor between rules, representations, and constraints in phonological theory’, Andrew Nevins & Bert Vaux outline the goals and central themes of the volume and briefly survey the contents of the contributions that follow. A key phrase in the description of the objective of the book is ‘to conduct informed comparison’ (2).
The essay in the book that yields the most broadly conceived comparison of the two competing frameworks is Bert Vaux's own, ‘Why the phonological component must be serial and rule-based’ (chapter 2). Vaux compiles a catalogue of arguments put forth by OT advocates in support of their model, identifies a set of fundamental problems with OT, and points to essential considerations favoring RBP. In his view, major shortcomings of classic OT include: (i) the failure to handle satisfactorily certain instances of opacity (such as the interaction of e-insertion and ʔ-deletion in Tiberian Hebrew /deʃʔ/→[deʃe] ‘tender grass’; 32); (ii) problems in treating optionality (like iterative and variable French schwa deletion; 43); (iii) unnaturalness (OT presupposes that all phonology is natural, disregarding the influence of accumulated historical change); (iv) a lack of theory-coherent and empirically adequate means to account for ineffability or want of surface realization (as when Swedish root morphemes in -dd cannot be inflected by means of suffixes beginning with a dental stop; cf. Vaux's example *rätt (48), which is the expected, but ungrammatical neuter singular of rädd ‘frightened’); (v) the prediction of unattested phenomena (e.g. huge phoneme inventories with minute phonetic distinctions or, conversely, tiny inventories with inordinately broad distinctions; 53f.); (vi) an inappropriate recourse to the questionable notion of conspiracy in criticisms of RBP (55–58); and (vii) the failure to account economically and insightfully for language acquisition, which involves ‘grammar construction … driven by the extraction of generalizations from the data to which the learner is exposed’ (29). On these and other counts, RBP fares better according to Vaux. In view of ‘the empirical, formal, and computational superiority of RBP vis-à-vis OT’, he concludes that ‘the parallel constraint-based architecture that currently dominates phonological theory should be abandoned in favor of a serial rule-based architecture’ (60).
David Odden's rewarding, sixty-page study on ‘Ordering’ (chapter 3) scrutinizes the multifarious mechanisms employed by OT in order to describe the linguistic facts that RBP handles by rule ordering and related derivational devices. Basing himself on data from Bantu languages, Odden determines that OT can account for most of the cases he analyzes – frequently, however, only at the cost of bending its initial theoretical premises. Interestingly, he establishes that, contrary to assumptions in OT, Duke-of-York derivations, which change a property A of a given linguistic form into B and later on restore it to A (i.e. A→B→A) do in fact exist in languages (118). Moreover, they can be formally represented not only in RBP but also in OT, ‘as long as “restored” A is token-wise distinct from underlying A’ (119). Hence, the alleged ability to exclude them is not an advantage of OT over RBP. Furthermore, Odden holds that Sympathy Theory, one of the OT means for simulating the effects of serial derivation, is just ‘a reconstruction of the intermediate step in derivational theory, not an independent concept’ (118).
In her lucid and stringent analysis of ‘Stress-epenthesis interactions’ (chapter 4), Ellen Broselow analyzes four empirical cases from Selarayese, North Kyungsang Korean, Winnebago, and Iraqi Arabic. She determines that in the first three cases the disruption of normal stress patterns is due to the insertion of epenthetic vowels, while in the last case, morphological factors, viz. the need to maintain paradigmatic differences, account for the irregular stress. Furthermore, the author argues that these cases furnish evidence for a parallel rather than serial analysis: a constraint-based model like OT can easily handle them, whereas a rule-based description will lead to complications.
Chapter 5, ‘Reduplicative economy’, by William Idsardi & Eric Raimy, compares two approaches to reduplication, (i) OT as supplemented by Correspondence Theory (CT), the sub-theory within OT that subsumes, for example, input–output faithfulness and base-reduplicant identity; and (ii) generative phonology with added non-linear temporal relations, as proposed in Raimy (Reference Raimy2000). A simple illustration of such a non-linear relation is the English word banana (154), which, given the begin-specification #→ and the end-specification →%, might be rendered underlyingly as illustrated in (1).
(1) #→b→a→n→a→%
In the speech linearization process, the loop in the abstract input form attaches a new output syllable na to the input na already present. The authors argue that whereas CT erroneously predicts many reduplicative patterns that do not in fact occur, the new mechanism introduced by Raimy neatly allows the generation of attested reduplications but disallows unattested ones. Their overall conclusion is that, in order to describe reduplication, the amendment of OT by CT is less commendable than the emendation of RBP by means of Raimy's reduplication model.
Quantity systems in the languages of northern Europe are often complex, sometimes, as in Saami or Estonian, even perplexingly so. In chapter 6, ‘Fenno-Swedish quantity: Contrast in Stratal OT’, Paul Kiparsky works out a novel analysis in the much debated realm of Scandinavian quantity. ‘Fenno-Swedish’, in his parlance, stands for ‘Finland Swedish’, as opposed to Swedish in Sweden, which he christens ‘West Swedish’, a term usually reserved for the linguistic varieties along the Swedish west coast. The theoretical basis of his discussion is a ‘marriage’ (202) of OT with his own Lexical Phonology, which he calls Lexical Phonology and Morphology in OT (LPM-OT) (cf. 8f., 11) or Stratal OT (185, 202). In contrast to classic OT, Stratal OT incorporates several strata or levels:
(2) The levels of Stratal OT
Lexical phonology
Stem phonology (level 1)
Word phonology (level 2)
Postlexical phonology
Kiparsky ‘view[s] each of these phonological subsystems as a parallel OT constraint system’, suggesting that constraints may be ranked differently at the different levels and that ‘[a]ll seriality lies in the interface between the levels’ (201). Radically departing from most earlier analyses of Swedish quantity, this intriguing article constitutes first and foremost a penetrating analysis of quantity differences in Fenno-Swedish dialects, where dialect variation is seen, inter alia, as arising from differential promotion of markedness constraints in the postlexical phonology (188–190). Kiparsky does not directly compare rule- and constraint-based approaches as such, but chooses instead to demonstrate the explanatory advantages of his emendation of classic OT. All in all, it is clear that his Stratal OT constitutes a significant and necessary improvement of ‘standard’ monostratal OT, and, by the same token, brings the latter closer to ordinary RBP.
Taking Chomsky & Halle's (Reference Chomsky and Halle1968) SPE model as his point of departure, but making ample use of derivational conditions or constraints, John Frampton explores in his innovative paper, ‘SPE extensions: Conditions on representations and defect-driven rules’ (chapter 7), processes of iterative footing and syllabification triggered by flaws in the structures to which they apply. Like Kiparsky's study, Frampton's paper does not expressly contrast properties of rule- and constraint-based models (cf. 247, though), but illustrates what a portion of a constraint-rich variant of SPE phonology might look like.
In the final, thought-provoking chapter of the collection, ‘Constraining the learning path without constraints, or The Ocp and NoBanana’, Charles Reiss compares (i) classic constraints-only, markedness-based OT, (ii) a generic model with both rules and constraints, and (iii) ‘a purely derivational theory’ that employs only rules (253). He argues that markedness constraints, being established on grammar-external grounds, do not belong in the grammar itself and, at any rate, often simply reflect the linguist's intuitions or guesses about sound patterns rather than robust statistical tendencies. As an example of the problems with constraints, Reiss picks the Obligatory Contour Principle (OCP), which prohibits identical feature specifications on adjacent autosegments. Drawing especially on work by David Odden (e.g. Odden Reference Odden1988), he elegantly demonstrates the empirical inadequacy of the OCP. Moreover, given that constraints constitute statements about what does not occur in language, he argues that constraints in general are as superfluous as a constraint ruling out bananas (the fruit, not the word) as elements of natural languages. The descriptive coverage of constraints and a satisfactory account of learnability are better produced by a more precise formulation of rules. For Reiss, ‘the goal of phonological theory should be to define the set of computationally possible human languages’ (279). Reacting to the standard idea that phonology, OT or other, is phonetically-based, he maintains that phonological computation per se is not grounded in phonetics: ‘phonetic substance cannot be encoded in the phonology’ (259) and ‘never directly determines how the phonological entities are treated by the computational system’ (258; emphasis in original). For these and other reasons, ‘phonology should not return to the rules-and-constraints models that predate Optimality Theory, but to a pure rule-based formalism’ (301; see also 256).
In sum, the volume under review provides a much-needed critical and reasonably comprehensive scrutiny of general as well as more specific issues in OT and RBP. In addition, Kiparsky illustrates the modification of OT in the direction of Lexical Phonology, and Frampton the expansion of the rule-based SPE model in the direction of constraints. According to the editors, the individual papers ‘present us with strong arguments for each position [i.e. OT or RBP] that should enable us to emerge with a new consensus’ (19). However, they leave it up to the reader to figure out exactly what this consensus is to be. What some readers might miss is an extensive attempt to clearly demonstrate the cognitive plausibility of the numerous devices, mechanisms and principles advocated, and of the overall architecture of the respective frameworks favored by the different authors. While the book contains occasional references to psycholinguistic and experimental work, it might have been illuminating to include at least one or two full-length papers evaluating psycholinguistically some of the specific claims advanced by OT and RBP. For example, the unidirectional character of most modern RBP models, which operate exclusively with derivations from input to output forms, is not as unproblematic from a processing and acquisitional point of view as the usual silence on this fundamental tenet of RBP might suggest.
As concerns style and presentation, the book mostly jumps straight into medias res, and its contents could have been made more accessible to a general readership by defining more technical terms and adding more clarifying examples. Neither the unalphabetized list of technical ‘Abbreviations’ (viii) – of which there are many in the text – nor the list of ‘Symbols’ (ix) is in any way complete. Typographical errors and inconsistencies are few, occuring mainly in the editors' ‘Introduction’ and Vaux's own chapter. Only rarely do less transparent oversights crop up, as when the voiceless bilabial fricative [Φ] and the Danish/Norwegian letter <ø> are both rendered as <ɸ> (297f., 307) or when half a dozen Hungarian names are graphically garbled (e.g. Andr's for András, L'aszlo for László, etc., 324).
While it is obvious that the book's contents confront, above all, OT with serious challenges, which OT theoretists will need to answer convincingly, it will also leave traditional RBP advocates with cause for concern. Taken as a whole, this timely collection is a significant contribution to current phonological theorizing that deserves the attention of a wide group of phonologists and theoretical linguists in both the rule- and constraint-oriented camps.
