Gouskova colloquium Friday Feb 18 at 3:30

Maria Gouskova (New York University) will present “Morpheme Structure Constraints Revisited” in the Linguistics colloquium series at 3:30 Friday February 18, by Zoom. An abstract follows. All are welcome!

Abstract:
Most constraint-based frameworks embrace Richness of the Base: the assumption that no interesting generalizations are stated as constraints on the lexicon (a.k.a. Morpheme Structure Constraints, or MSCs). The main argument against MSCs is that they introduce duplication into the theory. When the same constraints define the shapes of morphemes and restrict derived words, the latter, surface-oriented constraints should be sufficient. Unlike MSCs, surface-oriented constraints are less abstract, and are independently necessary. This echoes earlier criticisms of MSCs: they are redundant, abstract, and unlearnable.

In this talk, I revisit MSCs in the context of Russian voicing. Russian voicing was Morris Halle’s (1959) original battleground against structuralism—which he, incidentally, also criticized for having a duplication problem. By treating contrastive oppositions differently from non-contrastive ones, structuralism fails to capture the generalization that Russian voicing assimilation works on all obstruents alike, whether they contrast for voicing phonemically (/b/ vs. /p/) or are obligatorily voiceless (e.g., /tʃ/). My concern is not the undergoers; rather, it is the lack of certain contrasts predicted by the popular Positional Faithfulness account of voicing neutralization in Optimality Theory (Lombardi 1999 and many others). I will show that even though this account captures the phonetics and typology of voicing contrasts, it has a problem with Russian. There are several alternatives, but all encounter some kind of a duplication problem. I will argue for MSCs against consonants such as the affricate /dʒ/ in the lexicon. Another alternative would include a host of markedness constraints covering positions where [dʒ] does not occur, but this move introduces a duplication into the analysis: the phonology of certain consonants must be handled twice. These constraints, moreover, are neither phonetically grounded nor formally sensible; all they do is plug the holes in the analysis.

My account, like everyone else’s, has a duplication problem. But unlike other analyses, it explains facts such as the handling of loanword [dʒ], which is borrowed as a CC cluster in Russian, and which behaves as though it is never represented as an affricate in the system. I conclude with a discussion of a learnability proposal for MSCs within a constraint-based framework, Minimum Description Length (Rasin & Katzir 2016). I discuss some complications that arise in applying Minimum Description Length to learning certain kinds of hidden structure, especially structure that allows words to be shorter and grammars to be simpler