«Formal Phonology* David Odden OSU Abstract Two problematic trends have dominated modern phonological theorizing: over-reliance on machinery of ...»
Two problematic trends have dominated modern phonological theorizing: over-reliance on machinery
of Universal Grammar, and reification of functional properties in grammar. The former trend leads to
arbitrary postulation of grammatical principles because UG “has no cost”, which leads to a welter of
contradictory and unresolvable claims. The latter trend amounts to rejection of phonology and indeed
grammatical computation, as a legitimate independent area of scientific investigation. This paper outlines Formal Phonology, which is a metatheoretical approach rooted in an inductive epistemology, committed to seriously engaging the fundamental logic of the discipline, one which demands justification of claims and an integrated consideration of what is known about phonological grammars, eschewing ad libitum conjectures and isolated positing of novel claims without evaluating how the claim interacts with other aspects of phonology. Debate over the proper mechanism for apparent segment-transparency in harmony, or the binary vs. privative nature of features, is ultimately doomed if we do not have a clear awareness of what a “grammar” and a “phonology” are. Misconstruing the nature of a phonology as being a model of observed behavior negatively affects theoretical choices, leads to confusion over what could motivate a claim about the nature of grammar, and in general, a lack of developed epistemological foundation leads to confusion over how to approach theory-construction.
1. The object of study in Generative Phonology An obvious fact about language is that developmentally-normal adults can produce and comprehend an unbounded set of sentences in their language. What is most striking is that speakers can produce and interpret vast numbers of utterances that they have never heard before and could not have learned. This is only possible if speakers use a stock of primitive units plus a system of rules to create utterances, and children learn the primitives and rules rather than learning actual utterances. This then raises two central
scientific questions. First, what is the nature of the rule system that enables speakers to create utterances:
what does the system do, and how does it do it? Second, how are those rules automatically learned by observation of speech behavior, when the child acquires its language. What is actually learned?
A central feature of the theory of generative grammar is the claim that there are special cognitive properties which are particular to the human language faculty. This means that human language has a particular nature, and its nature does not reduce to general statements about human mental ability. A system of rules – a grammar – operates on stored representations, and the fundamental goal of generative grammatical research has been to discover the nature of grammars and representations. The generative enterprise then logically reduces to positing theoretical conclusions in the form of general propositions * This is a fragment of a draft of a longer work, still in progress. It is the result of numerous influences, and I hope those whose ideas appear here do not object to my co-opting their ideas and not even bothering to give credit where credit is due. I do want to specifically point to the obvious influence of the work of Hale and Reiss. Thanks to Kevin Gabbard, Kati Hout, Martin Krämer, Mike Marlo, Mary Paster, Markus Pöchtrager and two anonymous reviewers for comments on earlier drafts of this paper.
See Chomsky (1965: 4) for a standard characterization of a generative grammar as a perfectly explicit description of the competence of the ideal speaker-hearer.
© 2013 David Odden. Nordlyd 40.1: 249-273, special issue ‘A Festschrift on the Occasion of X Years of CASTL Phonology and Curt Rice’s Lth Birthday’ ed. by Sylvia Blaho, Martin Krämer and Bruce Morén-Duolljá. University of Tromsø.
FORMAL PHONOLOGY(“language sounds are feature bundles”, “features are attributes with two values”...), and empirically evaluating those conclusions as a model of the language faculty.
Generative grammar (and any good science) carries with it an ontological commitment to reality, the claim that its theoretical conclusions describe, or “correspond to” facts about a thing. If a theory does describe the nature of language, it is valid or true as a statement about language; if a theory misdescribes the nature of language, it is false as a statement about language. Evaluation of a theoretical conclusion means determining the extent to which a theory describes a fact about what human language is. Comparison of theories means determining which theory is closest to exactly describing the nature of language.
Thus the theoretical conclusion (accepted by most phonologists) that utterances are composed of rule-governed concatenations of simpler units lacking intrinsic meaning – are formed by combining segments – is accepted as an undeniable fact. If a language were discovered whose utterances did not decompose into reusable units, e.g. if utterances in a language were composed of single units varying continuously in pitch or amplitude, then the claim would be refuted. Numerous additional, more sophisticated and specific conclusions have been posited in the course of the development of Generative Phonology.
This brief summary of the relationship between Generative Phonology and its object of study is hardly controversial and hopefully familiar. A very important question about Generative Phonology has, however, been glossed over, namely how theoretical conclusions should get to the intellectual marketplace and be judged. A proper methodology for phonological theory does not rely on random actions or emotional reactions, it requires a rational i.e. logical method of relating ideas to reality. What then is the nature of that method? While certain implicitly methodological terms are widely used in the course of talking about phonological theories (“simple”, “constrained”), there is little discussion of what these terms refer to, and why or even whether they are being used validly in our scientific investigations. The purpose of this paper is to outline Formal Phonology, which is a metatheoretically-driven approach to phonology, focusing on the proper logic of phonological investigations which leads to true statements about the nature of the human language faculty.
This approach holds that creation of scientific knowledge is a unified process where valid conceptcreation means that existing knowledge implies a concrete theoretical concept, once focus is placed on the relevant facts. Theoretical conceptualization is not a cycle of arbitrary stipulation of isolated claims followed by a brief search for counterexamples, it is the continuous evaluation of the correspondence between ideas and the totality of what the ideas are about. The FP perspective on the relationship between facts and ideas is that facts are primary because they are what exist independently, and ideas are secondary – they are how we understand reality. A theory is a system of hierarchically-related concepts, and FP holds that foundational concepts must first be firmly established as correct, before erecting higher-order concepts on that base. It follows that positive means of justifying a claim are important, and “not being false” is too weak a criterion of proof. Not only must a claim be uncontradicted by the facts, it must also be the conceptually-simplest proposition that describes the facts. Finally, FP demands that a claim be appropriate to the thing that it is about. FP is about grammatical computations in a phonology, therefore claims about broader sound-system behavior are irrelevant unless they directly prove some fact about that computaThe disclaimer “it doesn’t matter”, in answer to a question about the nature of language, is usually false or misleading.
It is false on a literal reading, unless the question is really meaningless. The physical shape of a symbol in a theory does not matter, so whether the predicate “becomes” is notated as “=”, “→”, “⇒”, “→” or “becomes”, the symbols have the same interpretation. It is thus meaningless to ask whether the concept “becomes” is written as “=”, “→”, “⇒”, “→” or “becomes”, because concepts are not physically-written objects, and it does not matter if rules are written as “a = b” or “a → b”. It does matter if features are binary or one thing becomes another, if one cares about the nature of language.
The statement is misleading when intended as “This theory is only a partial theory of the domain, and makes no claims about J – it is an open question what the nature of facts of J are”. An honest and non-dismissive way to express such a claim is to say “This theory is only a partial theory of the domain, and makes no claims about J – it is an open question what the nature of facts of J are”. Very often, the implication is that the only thing that matters is what the system “does”, i.e. the class of string-sets (“languages”) admitted under the theory. But this is false. To repeat: generative grammar is a theory of the language faculty – an ability, which is how the system does what it does.
tion. These last two desiderata lead to the conclusion that concern about overgeneration is not a valid criterion for evaluating a theory.
2. The nature of Formal Phonology qua theory There are two approaches to reaching theoretical conclusions, which may appear to differ only in style.
One, found in much of the earlier research in generative grammar, is the top-down approach. Under this approach, a general hypothesis is proposed, consequences of the hypothesis are deduced, and to the extent that the specific consequences seem to describe facts about language, the general hypothesis is eventually judged true, or rejected if the consequences describe falsehoods. In this approach, great emphasis is placed on hypothesis-testing, the deduction of specific claims that follow from the hypothesis, and matching empirical observation to such claims. An important question arises regarding the hypotheses posited in this approach: how are such hypotheses created? This question has received relatively little attention in the literature. This leaves open the possibility that hypotheses might validly be the result of day-dreams, as Kekulé claimed was the source of his idea about the structure of the benzene ring, though it is unlikely that hypotheses with overtly irrational bases would be given much attention in linguistics. In the top-down approach, the process of postulating a linguistic hypothesis is basically arbitrary, and what matters is whether the hypothesis has been disproven because it makes a false prediction.
An alternative approach, one gaining more attention in current theoretical linguistics, is the bottomup inductive approach, which emphasizes that which the top-down approach ignores, namely hypothesiscreation. The inductive approach methodologically rejects arbitrary postulation of hypotheses, and rejects unjustifiably far-reaching claims. Instead, the approach requires that proposed hypotheses be integrated into a system of factual knowledge. The inductive approach relies on the fact that there already is a substantial empirical foundation, which is a precondition for positing a universal generalization worthy of further consideration. Since greater emphasis is placed on valid hypothesis-formation – the integration of an idea with existing knowledge to yield a concrete proposition – deductive inference plays a relatively minor role in scientific progress, though it remains useful for revealing unappreciated consequence of a hypothesis. A linguistic corollary of the insistence that the scope of a claim should not exceed that for which there is evidence, is that principles cannot be added to Universal Grammar ad libitum: there must be compelling evidence for positing the addition of a new principle or entity in UG. In part, this paper outlines the logic and practical application of this inductive approach to phonology.4 A second important metatheoretical issue about the nature of and arguments about grammars pertains to modularity and the explanatory scope of phonological theory. Approaches to language, since before generative grammar, have tended in two directions regarding the scope of grammar. For some, phonology is very broad and includes all aspects of linguistic and communicative behavior, and for others, the scope of investigation is more narrowly defined. This paper follows the principle that phonology is a specific and narrowly-defined domain, as advocated in numerous works by Hale and Reiss, where a phonological grammar is a formally-statable system of symbolic computations on “sound representations”, and the goal of phonological theory is to discover the nature of those representations and computations.
Autosegmental Phonology, Metrical Phonology and Optimality Theory are concrete claims about the form of phonological grammars, and carry no methodological commitments to what a phonology is or how a theory should be constructed; they also inherit the strong-nativist practice of freely attributing devices to Universal Grammar because UG is not learned (it is known a priori) and is thus considered to be This does not mean that the inductive approach only allows extremely specific hypotheses with a tiny range of empirical application – it misconstrues the nature of theory-building to only state what has been observed. Rather, the approach rejects claims made without proper justification and scrutiny with respect to conceptual simplicity.
Formal Phonology says nothing about how to create an idea which leads to a theoretical concept, rather it addresses the logical relation that such an idea should have to existing knowledge, and how the resulting concept should be evaluated.