The Handbook of Phonological Theory

Chapter 1

Rules v. Constraints

DAVID ODDEN

1 Background

The goal of a theory of phonology is to elucidate the nature of “phonology” at a conceptual and predictive level. The title of this chapter refers to a comparative evaluation of rules and constraints as successful theories of phonology which implies having a standard of evaluation, and adequate clarity as to what “rules” and “constraints” refer to. Neither prerequisite is trivial to satisfy

1.1 The Scope of Inquiry

Certain assumptions about the nature of phonology must be considered, even lacking agreement on which assumptions to make. First and foremost, deciding whether phonology is based on rules or constraints, or a mix of the two, requires having objectively expressible statements of phonologies within different frameworks whose consequences can be compared. Therefore the theories must have a definite form, that is, they must be formalized. The entities which make up a phonological grammar should be expressions, which are finite sequences of elements taken from a specified set, and combined by rules of construction that define well-formed statements of rule or constraint. The value of formalism is its power to make objectively-interpretable statements about the phonology which can be checked against fact. To evaluate rules versus constraints as models, we should then consult the formalisms of the theories, to see whether one theory better passes the test of empirical and aesthetic adequacy1 Problems in this area are not trivial; certain theories of constraints or rules are severely under-formalized so that it is hard to know what predictions the theory makes; and a number of theories are under-applied in the sense that it is impossible to determine from examples how particular phenomena would be analyzed.

Assuming that we are comparing formal theories, we must resolve questions about the scope of phonology, including how much of “phonetics” or “morphology” is phonology, and whether all facts bearing on phonology are the responsibility of the theory Generative phonology traditionally encompasses a broad range of processes which might be considered phonetic (allophonic) or morphological (rules with lexical or morphological conditions), but the edges of phonology may also be contracted for theoretical purposes, viz. restrictiveness. Thus Webb (1974: 127) excludes metathesis from phonology, stating that “synchronic metathesis is not a phonological process. In the residual cases of metathesis, the rule is always morphologically restricted,” enabling the “Weak Metathesis Condition,” a restriction against reordering in phonology. If phonology is deemed to be concerned only with biuniquely recoverable surface-true relations between sounds (e.g. allophonic vowel nasalization in English), and abstract phonological alternations are to be described by the formal methods of morphology, a theory designed to account for just surface phonotactics cannot be meaningfully compared to one designed to account for both phonotactics and abstract morphophonemics.2 A surface-phonotactic view of phonology thus must ignore a substantial portion of research into phonological grammars, on Bedouin Arabic (Al Mozainy 1981), Finnish (McCawley 1963; Harms 1964; Karttunen 1970; Keyser and Kiparsky 1984; Kiparsky 2003a), Chukchi (Krauss 1981), Kimatuumbi (Odden 1995), Klamath (Kisseberth 1973; White 1973), and Ojibwa (Piggott 1980), and numerous other languages.

There are also questions as to the level of explanation demanded of a theory — do we demand formal explanation, or formal and functional explanation? Much of the course of phonological theorizing has involved the increasing absorption of substantive factors into the theory, in an attempt to narrow the gap between prediction and observation. Comparative evaluation of theories implies determining which theory is better at making definite the notion “possible rule” or “possible constraint.” The notion “possible” is used in two ways. One sense is theoretical well-formedness, that is, a rule constructible by free combination of elements, according to a theory of the form of rules. In that sense, “A→B/C___D” would be a possible rule, but “→B___/ACD” would not. McCawley (1973: 53) points to a different sense, the metaphysically possible, claiming “One who takes ‘excessive power’ arguments seriously has as his goal characterizing ‘phonological rule’ so as to include all and only the phonological rules that the phenomena of a natural language could demand….” This notion of “possible rule” seems to mean what does exist, so is attested, or that which we have solid scientific or philosophical reason to conclude must exist now or in the past or future, just waiting to be discovered. The latter kind of “possible” depends on metatheoretical expectations, so McCawley intuits that assimilation of nasal to labials alone is not a possible rule (the present author does believe that such a rule is possible, if unlikely).

Whether such a rule is possible is not central to this discussion: what is essential, is distinguishing the undiscovered from that which is impossible by the nature of language. Expansion of the substantive content of phonological theory narrows the predictive gap, though, complicates the theory and renders it redundant with respect to the extragrammatical physical explanations for the gap. If phonology is only a system of symbolic computations where the syntax of computations defines a broad class of possible rules, and separate aspects of languages referring to substance (perception, acoustics, articulation, language learning, and the transduction between grammar and linguistic behavior) explain why some formally allowed rules have negligible probability of attestation (as argued by Hale and Reiss 2008; Morén 2007), then failure to capture a generalization about substance within the theory of computation is not an argument against the theory of computation. But there is no universal agreement that the object of investigation is the computational apparatus rather than the full and undifferentiated panoply of factors influencing linguistic sound.

A second metatheoretical question affecting a comparison is whether phonology describes abstract string collections, or the mental faculty which generates them. If phonology only models strings, then considerations such as the results of psycho-linguistic tests or problems regarding infinities in the model — infinite sets of candidate or sub-rules — are irrelevant to theory selection.3 An example of how different conclusions are reached depending on whether one considers just the strings, versus the strings plus the mechanisms, is Mohanan (2000: 145–146) versus Calabrese (2005: 34). Mohanan contends that a rule [+nasal] → [+voice] is “logically equivalent” to a negative constraint *[+nasal,-voice], while Calabrese contends that rules and constraints are totally different means of implementing a linguistic action and are ontologically different. Mohanan is correct that the rule and the constraint describe the same string classes — are weakly equivalent; Calabrese is right that the imputed mental mechanisms of rules versus constraints are different — are not strongly equivalent.4

Even if we presume that phonology should be concerned with a mental faculty as well as the sets of strings, we must also determine whether phonology is concerned with all sound-related behavior, or just that behavior which generates the strings. A mentalist view of phonological grammars would care whether insertion of [i] after a word-final obstruent is regulated by a rule or a constraint, and whether this takes place in a single step or many steps; but a mentalist view of phonological grammars does not automatically care about the behavior of speakers of such a language under certain types of psycholinguistic testing, since a mentalist view of grammar does not automatically hold that all aspects of the mind pertaining to language sound are contained in the phonological component of a grammar.

To properly contrast “rules” versus “constraints” in phonology, we must also determine what these terms refer to, because we want our conclusions about differences between rules and constraints to reflect the concepts themselves, and not quirks of particular theories of rules or constraints. Many definitions of “rule” are offered in the Oxford English Dictionary, but the ones that seem closest to its linguistic use are:

A fact (or the statement of one) which holds generally good; that which is normally the case.

A principle regulating the procedure or method necessary to be observed in the pursuit or study of some art or science.

(Grammar). A principle regulating or determining the form or position of words in a sentence. In modern linguistics, usually applied to any one of a system of rules that can be formulated in such a way that together they describe all the features of a language.

The closest applicable definition of constraint is

The exercise of force to determine or confine action; coercion, compulsion.

In addition, the terms “principle,” “condition,” and “convention” are often used in linguistics to describe what often seems to be the same thing as a constraint, perhaps with the implication of...