![]()
1
An Introduction to Optimality Theory
1.1 How OT Began
Around 1990, Alan Prince and Paul Smolensky began collaborating on a new theory of human language. This collaboration led in fairly short order to a book-length manuscript, Optimality Theory: Constraint Interaction in Generative Grammar. Photocopies of the manuscript were widely distributed and had a terrific impact on the field of linguistics, even though it wasn’t formally published until more than a decade later (as Prince and Smolensky 1993/2004). OT had and continues to have its greatest effect on phonology, but it has also led to important work in syntax, semantics, sociolinguistics, historical linguistics, and other areas. OT belongs on anyone’s list of the top three developments in the history of generative grammar.
One of Prince and Smolensky’s goals for OT was to solve a longstanding problem in phonology. Phonological theory in the tradition of Chomsky and Halle’s (1968) The Sound Pattern of English (SPE) was based on rewrite rules. The rewrite rule A → B/C___D describes an input configuration CAD and an A → B transformation that applies to it. Rewrite rules can describe lots of phenomena, but they do a poor job of explaining how phonological systems fit together. (For a brief explanation of SPE’s main assumptions, see the boxed text at the end of this section.)
To illustrate, we’ll look at some data from Yawelmani, a nearly extinct dialect of the California Penutian language Yokuts (Newman 1944).
1 In this language, syllables cannot be bigger than CVC (consonant-vowel-consonant). Various phonological processes are involved with this limit on syllable size. For instance, Yawelmani has a process that deletes a vowel at the end of a word, as the data in (a) of (1) show. (The “.” marks the boundary between two syllables.) But the data in (b) show that final vowels do not delete when they are preceded by a consonant cluster. The explanation for the difference between (a) and (b) is that deletion after a cluster would require a syllable that is too big or leave a consonant that cannot be syllabified: *[xatk
].
2Yawelmani also has a process of vowel epenthesis that applies to a cluster of three consonants in the middle of a word. (See (2). The data in (a) show epenthesis into triconsonantal clusters, and the data in (b) show that there is no epenthesis in smaller clusters.) If there were no epenthesis process, then the result would again be a syllable that is too big or a consonant that cannot be syllabified: *[
ilk.hin].
It is certainly possible to state SPE-style rewrite rules to account for these two processes in Yawelmani – V → Ø/VC___# and Ø → i/C___CC will do the job nicely. But, as Kisseberth (1970) first argued, those rewrite rules are missing an important generalization about the special role of surface-structure constraints in both rules. Final vowel deletion cannot create bad syllables in surface forms, and epenthesis exists to eliminate them. Adopting a suggestion from Haj Ross, Kisseberth called this situation a conspiracy.
When two or more rewrite rules are involved in a conspiracy, they directly or indirectly support some constraint on surface forms. In Yawelmani, the relevant constraints are a CVC limit on syllable size and a prohibition on unsyllabified consonants. Final vowel deletion is
blocked from applying when it would produce a surface form like *[xatk
], which cannot be fully parsed into
MAXimally CVC syllables. Epenthesis is
triggered to apply by the need to fix sequences that cannot be parsed into CVC syllables. In every conspiracy there is a constraint on surface forms, which we can refer to more succinctly as an output constraint, since it evaluates the output of the grammar. There is also some mixture of processes that are blocked by that output constraint and/or processes that are triggered by it.
Conspiracies are common in the languages of the world, and so it was a matter of some concern that the SPE theory of rewrite rules couldn’t explain them. A rewrite rule, by its very nature, describes an input configuration and an operation that applies to it. A conspiracy is completely different: it refers to an output configuration, it involves several different operations, and those operations may participate in the conspiracy by applying or failing to apply, depending on the circumstances. When analysts try to describe conspiratorial behavior in terms of rewrite rules, they have to start using counterfactuals, as I did in the preceding paragraph: “blocked from applying when it would produce,” “to fix sequences that could not be parsed.” Statements like these show that the analyst understands what’s really going on in the language, but counterfactual conditions have no place in SPE’s theory of how to apply rules. When a phonologist says something like “The epenthesis rule ensures that the language has only unmarked syllables,” he or she is describing an intuition about how the system works. But that intuition has to be expressed formally, in the theory itself. Otherwise, we are just telling ourselves stories.
At around the same time that phonologists were beginning to grasp the importance of output constraints, syntacticians were having a similar revelation. For example, clitic movement in Spanish is triggered by an output constraint requiring that second person clitics precede first person clitics (Perlmutter 1971). That is why the direct and indirect objects appear in different orders in TeIO meDO present...
