Location: Royce Hall 362

Phonological Selection in Small Sublexicons

Affixes can impose phonological restrictions on the stems they attach to: only consonant-final stems, or stems no bigger than two syllables. These restrictions remain poorly understood and under-formalized. For example, Prosodic Morphology-era analyses characterize adjectival -er/-est affixation as selecting for a trochee, but it remains a mystery why this trochee can be disyllabic, while the -en suffix ("stiff-en", "redd-en") admits only obstruent-final monosyllables. I
consider the problem from the perspective of sublexical theory, whereby the generalizations about such patterns are statistically extracted rather than being part of the morphological rules governing the affixes. The challenge taken on in this talk is how to extract generalizations from small sublexicons.

I explore size restrictions imposed by two suffixes of Russian: the adjectival "notable body part" -ast suffix (e.g., glaz-ast-ij 'big-eyed') and the nominal "baby diminutive" -onok (e.g., akulʲ-onok 'baby shark'). The -ast suffix selects for disyllabic stems, while the baby diminutive selects for maximally trisyllabic stems. In Sharoff's (2005) list of Russian words that occur at least once per million in the Russian National Corpus, -ast adjectives are represented by just 17 lemmas, and -onok baby diminutives by 29 lemmas. An exploration
of a larger corpus shows that both affixes are productive, and their size selection constraints are projected in a frequency-matching pattern. The frequency-matching even extends to the semantics of the stems, specifically the categories that typically figure in inalienable possession.  

My explanation for this is that the rules contain lexical lists (sublexicons) but no phonological (or semantic) context. Learners form phonological generalizations by calculating disparities between what is attested in the sublexicon defined by each affix vs. what is attested in the set of comparable stems in the language as a whole. The generalizations tend to be coarse (to borrow Pierrehumbert's (2001) term) because a small sublexicon is likely to support only a generalization that is informed by every item in the list: all stems
include syllables, consonants, and vowels, but not all stems supply evidence of finer-grained underattestation at the level of segmental features. I also argue that a Prosodic Morphology-style subcategorization frame cannot capture the 2-syll and 3-syll selection in Russian because these units do not correspond to prosodic constituents in the language.