Integrated Language Science and Technology Seminar

Monday, September 10

1:30pm – 3:30pm

3401 Walnut Street, Room 401B

Ryan Budnick – University of Pennsylvania

An Acquisition-Based Account of the Typology of Non-Rhythmic Stress

Word-level stress systems are highly constrained, compared to the limitless number of possible systems. Previous work has explained this restricted typology through mental representations such as parameters (Hayes 1985; Dresher & Kaye 1990) and constraints (Eisner 1997; Bakovic 1998; etc.). I argue that possible stress systems can be strongly restricted by non-representational properties of language acquisition. Crucially, unattested systems are less stable than attested ones in the presence of under-segmentation errors (perceiving multiple words as one word) and over-segmentation errors (perceiving one word as multiple words), both of which are found in child behavior (Brown 1973; Peters 1983) and computational models of word segmentation (Lignos 2012; Phillips & Pearl 2015).

The model accounts for the edge-oriented behavior of quantity-insensitive stress systems (such as stressing the first or penultimate syllable of each word), excluding other possibilities, such as stressing the middle syllable, or stressing the first syllables of words with odd numbers of syllables and the penultimates syllable of words with even numbers of syllables. In the quantity-sensitive domain, fewer unattested systems are excluded, but attested systems are stable. It is likely that the inclusion of additional phonetic assumptions outside the acquisition system can combine with this model to make tighter predictions here; such an extension may also prove fruitful when considering rhythmic (or iterative) stress systems. The model also makes novel empirical predictions, including a correlation between stress patterns and word minimality which is borne out by typological data.