Here are a some thoughts about the relationship between phonological theory and experimentation that might be useful in starting our discussion of this topic, which we will be pursuing throughout the semester. A few of us found the following quote from the Pierrehumbert et al. paper (p. 5) of interest:
“Given the rapid pace of change in theoretical linguistics, and the great expense and labor of laboratory research, the shrewd experimentalist will not devote an experiment to even the most central claim of any single linguistic framework. Instead, he or she will look for a topic which represents a source of tension across many frameworks, or which has remained unsolved by traditional methods over many decades…One class of topics which lend themselves to advances using experimentation are theoretical issues . In using this term, we do not mean issues which arise as corollaries of the main assumptions of individual frameworks. Rather, we mean the issues which can be formulated after a deep and sustained effort to compare different frameworks.”
We might be less certain than Pierrehumbert et al. about the wastefulness of investigating “even the most central claim of any single linguistic framework”, but I think that the distinction between phonological theories (or frameworks) and theoretical issues is an important one to keep in mind. Most, if not all, the papers that we will be reading are set up as experimental investigations of theoretical issues rather than tests of theories themselves. Here are some of the research questions that the papers in the syllabus are investigating, in an order roughly matching the papers’ order in the syllabus:
- What is the nature of phonotactic knowledge? Does it play a role in speech processing outside of well-formedness judgments?
- What is the relationship between knowledge of phonotactics and knowledge of alternations?
- Do speakers have knowledge of phonological universals?
- Are “natural” alternations/generalizations easier to learn that unnatural ones?
Perhaps the best example of a paper in which an experiment is set up a test of a corollary of a theory is Berent et al.’s recent paper in Phonology, in which they make explicit the assumptions under which OT predicts that universals should play a role in perception. But this is the exception, rather than the rule.
Why are experiments typically set up as investigations of theoretical issues, rather than tests of theories? One reason, as Pierrehumbert et al. imply, is that this gives the results a durability beyond the lifespan of a given theory. Another reason is that while theories of generative phonology often make claims to “psychological reality” (see esp. the Goldrick paper for discussion), they aren’t directly theories of cognition. Think about what an argument for a phonological theory consists of. The usual goal is to construct the maximally elegant theory that captures all and only the attested typological data. Presumably, a theory constructed with that goal in mind will have important things to tell us about how phonological knowledge is cognitively represented, but it’s not directly a theory about cognition. Thus, it is close to a category error to try test a generative phonological theory in a psycholinguistic experiment.
Theoretical issues arise in the course of constructing our theories of phonological typology, and these are well-suited to explore experimentally. To the extent that the results line-up with the structure of a given theory, you can then see that as an indication that the theory is on the right track. Ultimately, we do want our theories of phonological typology to be consistent with what we know about cognition, since that increases the odds that our typological work is in fact going to yield fruit in shedding light on the nature of phonological knowledge.
One last thought: about “framework” and “theory”. I haven’t gone back to explore the roots of this distinction, but a way that I’ve recently found useful to employ these terms is in saying that OT, or HG, is a framework, while OT or HG with a given constraint set is a theory. The theory makes specific typological predictions; the framework does not (though the structure of the framework may of course play a role in shaping what kinds of predictions can be made…)