McMahon, April. (2000) Change, Chance and Optimality, Oxford University Press, paperback, x+201 pp., $19.95, ISBN: 0-19-824125-9 (hardback, $60.00, ISBN: 0-19-824124-0).
Peter Norquest, Joint Program in Linguistics and Anthropology, University of Arizona.
Summary 'Change, Chance and Optimality' centers around, and delivers an intelligent critique of, Optimality Theory (OT: Prince & Smolensky 1993), a constraint-based theory of phonology which has had a tremendous impact on the world of formal linguistics throughout the past decade. This book is comprised of six chapters altogether, summarized in the following paragraphs.
Chapter one provides a brief history of generative phonology before OT, in which the primary machinery used in phonological theory were rules, which were language-specific, descriptive tools and therefore lacked direct explanatory power. One of the main problems with a rule-based framework is that rules may be overused, and a way to counter this problem was to propose constraints on their application. This was a logical step in light of the fact that outputs, as well as inputs, play a controlling function in language, as in the case of 'rule conspiracies' where a number of rules interact in complicated ways to derive what are ultimately the same surface patterns. Constraints were ultimately recognized as necessary phonological tools because of their ability to capture generalizations which were neglected under strictly rule-based approaches.
OT stands in sharp contrast with rule-based theories due to the fact that it does away with rules altogether and relies solely on a single formal object, that of constraints. These constraints are assumed to be part of an innate Universal Grammar (UG), and their interaction should therefore be able to result in as well as explain the set of possible human languages, no more and no less. It is in this way that OT is arguably superior to a rule-based theory, since it is purely constraint-based (and therefore more elegant) and appeals to universal principles of human grammar, nullifying the problem of arbitrary and language-specific parts of a grammar. However, language-external factors such as the abilities and limitations of speakers, which can indeed lead to language-specific developments, are not reconcilable with a deeply universalist model such as OT, and an attempt to do so may lead to such attention on the machinery of the theory (such as expanding the constraint set with constraints of dubious universality) that its real-world application becomes overlooked. McMahon's central concern in this book then is that "OT, in attempting to confront the universal component of phonological behavior, is in danger of failing to cope with the language-specific part." (p. 10).
The focus of chapter two is upon evidence that many allegedly universal constraints are in fact language-specific, and that OT cannot function without the addition of parochial rules or their equivalent. The chapter begins with an outline of those processes which seem particularly amenable to OT analyses (prosodic and metrical processes), and those where OT faces greater challenges and which are better-suited to rule-based analyses (morpho-phonological interactions, alternation, opacity, language-specific processes with no clear rationale, variation, and change). The central question of the chapter is then raised: are all of the constraints which have been proposed to deal with these phenomena really universal, violable, and innate? It is first suggested that there are implicitly two classes of constraints used in the literature. The first is comprised of the kind which are indeed violable and rankable. The second is made up of a smaller number of constraints which are universally undominated, such as NUC (Prince & Smolensky 1993: 87) which requires syllables to have a nucleus. McMahon asserts that we must consider accepting the fact that certain constraints "form a universally undominated class, characterized precisely by the property of inviolability, which then itself requires explanation, probably from outside of phonology ... research on which constraints fall into the inviolable class, and therefore progress on any unifying factors, is unlikely to take place so long as OT denies the existence of inviolable constraints" (p. 17).
Attention is then turned to system-specific strategies, and constraints which are of dubious universality. One constraint of this type is Prince and Smolensky's 'FREE-V', which is proposed for Lardil and states that word-final vowels must not be parsed (in the nominative) (Prince & Smolensky 1993: 101). If all constraints of this kind are considered to be universal , then the list of constraints will be extremely long and its organization a Hurculean task. It is noted that parochial rules outside the constraint system have been reintroduced in a number of analyses, an example of which is the r-insertion rule of McCarthy (1993).
Correspondence Theory is examined in light of the fact that it has been used in a rule-like fashion in some analyses of data which a serial derivation otherwise renders transparent. Opacity facts in two languages in particular are examined: Turkish vowel epenthesis and velar deletion (Kager 1999), and Yawelmani Yokuts vowel lowering (Archangeli & Suzuki 1997). The latter is examined critically in terms of the specific kinds of constraints which are proposed to deal with the data; a disparate correspondence constraint which allows non- identical elements to correspond (input length and output non- highness), and an input markedness constraint which determines the shape of possible inputs, thereby conflicting with Richness of the Base. The first disparate correspondence constraint is argued to not only be extremely language-specific, but (in agreement with a statement in Kager 1999: 381) to stipulate the opaque pattern rather than explaining it.
McMahon criticizes sympathy theory for similar weaknesses, and argues that both sympathy theory and Output-Output (OO) correspondence are representative of an unwanted proliferation of theoretical machinery within OT, which results from its inability to deal with phenomena such as opacity which are more easily explicable in a derivational analysis. In the closing section of chapter two, she points out that a child acquiring a language will have no direct access to candidates in tableaux except for the winner, and this poses a problem for sympathy theory. Focusing upon acquisition under OT, two problems are pointed out for OT: (1) "[L]exicon optimization is hard to implement for alternating forms, where two or more outputs correspond to a single input in classical generative terms ... Irregularities, alternations, and misacquisition are all problematic for learning theory under OT", and (2) "successful constraint ranking depends on access to the inputs: we now discover that postulation of inputs, via lexicon optimization, depends on a ranked set of constraints to provide the necessary lists of violations ... the child needs a ranking to get URs and needs URs to get a ranking" (p. 55). McMahon concedes that acquisition under OT is difficult to evaluate at present given the indecision about types of constraints and their interaction.
The main concern of chapter three is with the difficulties that OT faces in accounting for sound change and residue, as well as variation. McMahon argues that change should not be ignored by a phonological theory for the following reasons: (1) if there are meaningful universal principles, then these should define a space delimiting possible developments; (2) knowing the history of a system can explain otherwise opaque aspects of its current state; and (3) it is not always possible to draw a decisive line between synchrony and diachrony, since there are process types which operate both as sound changes and in synchronic phonological alternations.
A chicken-and-egg question is raised which asks which comes first: a constraint reranking which then motivates a new grammar, or a change in the grammar which results in the acquisition of a novel constraint ranking. McMahon favors the latter view, whereby system-external factors bring about a sound change which is then encoded within a new ranking; the opposite view leaves us wondering what the motivation would be for a constraint reranking which only left visible effects in the grammar after it had already occurred. There is an ensuing discussion of epenthesis and deletion, metathesis, and chain shifts, during which various issues are discussed including the possibility that constraints might be acquired instead of innate, and the fact that local conjunction represents one more example in the proliferation of theoretical machinery which is coming to be more of a bane than a boon to the overall theory.
McMahon argues in the following section that in accounting for a sound change, universal constraints and constraint ranking alone are not enough. Although OT accounts of sound change are meant to be explanatory and not merely descriptive, reference to non-phonological factors (i.e. phonetic and sociolinguistic) seems necessary, and while a reranking of constraints certainly results from a sound change, calling the reranking itself is extremely problematic. Two cases in particular are used illustratively, that of the Middle English Great Vowel Shift, and historical segment loss within Korean. The same types of problems occur as well for cases of variation, where there is an equal ambiguity of the roles within OT between external and internal factors which result in synchronic variation.
In the final part of chapter three, McMahon compares OT with two approaches to historical syntax (Lehmann & Vennemann, and Lightfoot) and poses the larger question of whether or not change and variation can be adequately treated in formal models at all. Four different problems for these kinds of models are posited: (a) the non- universality of universals, (b) the strength paradox, (c) the chicken and egg problem, and (d) the external evidence problem. Upon evaluation, OT fails on the last two, but passes on the first two given the stipulation that there is a distinction made between universal constraints and language-specific constraints (or rules). McMahon warns at last that 'OT seems to be falling into the old Standard Generative trap of seeing description and explanation as interchangeable concepts' (p. 128). Chapter four compares OT with two other systems, one also linguistic (Natural Morphology) and one not (evolutionary biology). McMahon's goal is to show that when other models are examined, they do not manage with only one kind of formal statement; rather, both universal and system-specific mechanisms are needed. It is therefore potentially more natural and ultimately more fruitful to expect explanatory power of a theory which uses and appropriately divides both rules and constraints than one which relies either only on rules or only on constraints.
In Natural Morphology, explanation is not intended to be theory- internal, but rather relies on three major principles (iconicity, uniformity, and transparency) which are motivated by neurobiological concerns such as perception, processing, or memory limitations. These three principles are seen as violable, and while language change is generally expected to take place in the direction of naturalness (defined by typological considerations), 'unnatural' changes may occur because of conflicts between the three principles or between different components of the grammar such as morphology and phonology. Natural Morphology does not restrict itself to these three universal principles, however, but also includes system-dependent or language- specific markedness measures which may come into conflict with the universal principles. Natural Morphology ultimately recognizes three levels of statements: (a) a universal set of principles (much like the putatively inviolable constraints of OT), (b) system-specific structural properties, and (c) historical accidents which shape systems but which are outside the remit of the theory.
McMahon discusses a similar situation in Neo-Darwinian evolutionary biology, where there are also three levels of explanation: (a) a superordinate set of absolute constraints which arise from the laws of physics and mathematics, (b) a class of violable constraints, and (c) a final series of species-specific descriptions. An example of the first level is the relation of volume and surface area, from which we can draw exceptionless generalizations and specify certain limitations, i.e. on skeletal scaling or gas exchange. An example of the second level is the cross-species correlation between brain-size and body- weight, for which a generalization can be made to which exceptions may nevertheless be found. The final level is exemplified by the small front limbs of a Tyrannosaurus Rex or the trunk of an elephant. In McMahon's words (p. 147-8), "It would be possible, but essentially pointless, to propose violable constraints requiring four-leggedness, or marsupialness, or a constraint against trunks which is incredibly high-ranking except in elephants; what we really need is just a recognition that there are some facts we can perfectly adequately describe, but not explain except in terms of historical contingency ... Each species will require a degree of independent description in isolation from the constraints, since it represents the current stage in a long chain of changes, some of which refine the results of originally random events."
Chapter five gives the challenge to proponents of an innatist approach to explain how universal constraints developed in our species by the mechanisms usually thought to be responsible for the development of complex systems. McMahon argues that constraints (for example ONSET) cannot be directly genetically encoded, but there must rather be a level of genetic instructions which allow or favor the learning of language systems which conform to particular criteria.
In biological evolution, gradual accumulations of complexity are much more likely to lead to favorable outcomes than a single great leap, although that is exactly the idea which has been popularized by Chomsky and others. One piece of evidence that the evolution of language was a gradual process is that the parts of the brain which are involved in perception and production have homologues in other species. It is more likely that language evolved as a mosaic with individual parts of UG being introduced over time, although historically 'intermediate' stages may be irrecoverable.
In terms of how the evolution of language relates specifically to OT, the question is posed as to whether the constraint set is fixed or flexible; the former case would have most likely resulted from a macromutation which is unlikely from an evolutionary point of view, but on the latter view novel constraints may have the option of developing and becoming innate via genetic assimilation. The only other alternative to these two possibilities is that there is a subset of universal constraints which have developed under selective pressure, and an additional set of rules or language-specific (and therefore learned) constraints which are not innate. '[I]n animals, as in languages, we face the conclusion that things could be otherwise: we must describe them as they are; explain them when we can; and try to develop a nose for the difference' (p. 176).
The final chapter takes up the theme that 'things could be otherwise.' McMahon argues that the properties of language or of organisms in general could have evolved differently, and constraints hypothesized by the theorist to circumscribe only the variation currently attested are themselves shaped by contingency. Explanation for phonological phenomena must therefore ultimately include theory-external explanation from such areas as phonetics and neurolinguistics.
In discussing the future of OT on the final page of her book, McMahon argues that the first and most important task for the model must be a full assessment of the universality of constraints and a decision as to whether there is an initial subdivision between truly inviolable and violable constraints, and the question posed whether the latter category includes language-specific constraints. It must then be decided if some of these should be rather seen as rules instead of constraints, and if so, what the nature is of the interaction between constraints and rules. If a model including both constraints and rules indeed ends up being superior to one which only uses constraints, then it will have to be decided if it is better to place the constraints on well-formedness, or otherwise on rule-application.
'Change, Chance and Optimality' is an extremely welcome evaluation and critique of OT. It deals with this extremely important and influential theory in a very fair and even-handed way, and does a good job of addressing both internal and external issues which are directly relevant to the efficacy of OT. The treatment of what are widely recognized problem spots for OT, such as opacity and degrees of constraint universality, is clear and straightforward.
The most important part of the book is perhaps the contextualization of OT in relation to other complex systems within the natural world, and the poignant observation that scientific treatment of these systems does not procede in the most profitable way by the postulation of only a single formal object, but rather through the recognition that there are various systems interacting with each other and that what we observe today is a result of both the history of and the current manifestations of these interactions. The research agenda which is offered at the end of the book is one which must certainly be confronted by phonologists, as the failure to do so will leave huge gaps in the attempt to work toward a phonological theory which is maximally explanatory and predictive.
The only disappointing part of her book is the fact that there is not more concrete examples or case-studies of how constraints and rules might be integrated into the kind of model which McMahon ultimately argues for throughout and especially at the end. This is not an extreme fault, since such an undertaking could very easily be a monograph-sized endeavor; however, I hope that we have not heard the last word from McMahon on the topic of OT and that the future will see more from her on the instantiation of rules and constraints in a unified phonological theory. In the meantime, the reader has been given much to consider and to guide them in the world of phonological theory.
Archangeli, Diana & Keiichiro Suzuki (1997). 'The Yokuts challenge', in Iggy Roca (ed.), Derivations and Constraints in Phonology. Oxford: Clarendon Press, 197-226.
Kager, Rene (1999). Optimality Theory. Cambridge: Cambridge University Press.
McCarthy, John (1993). 'A case of surface constraint violation.' Canadian Journal of Linguistics, 38: 169-95.
Prince, Alan & Paul Smolensky (1993). Optimality Theory: Constraint Interaction in Generative Grammar. Manuscript, Rutgers University/University of Colorado at Boulder.
Reviewer's Biographical Sketch I am a PhD student in the Joint Program in Linguistics and Anthropology at the University of Arizona. My research interests include historical phonology with an emphasis in East Asia (Altaic, Sino-Tibetan and Kadai) and on Athabaskan, and the phonetic groundedness of sound change. My ultimate goal is to establish a more explanatory and predictive model of sound change which integrates typological data with phonetic facts. I hope to follow a career path which allows me to continue research in these areas as well be an active part of the linguistics community.