Adam Albright (MIT)
Friday March 31st 2017, from 10:00 to 12:00
(followed by a light lunch from 12:00 to 13:00)
ENS (29 rue d'Ulm, 75005 Paris), room 235A
This presentation is co-organized with Maria Giavazzi
For a companion presentation on Thursday March 30th, see here
A tenet of "Classical" Optimality Theory is that constraint domination is strict: violations of a higher-ranked constraint do not become more tolerable when lower-ranked constraints are violated multiple times. (That is, multiple lower-ranked violations cannot "gang up" to overcome a higher-ranked violation.) Pater (2009) shows that even in weighted constraint models such as Harmonic Grammar or Maximum Entropy models, ganging up is also limited by the fact that markedness and faithfulness violations trade off against one another. This result helps to preclude patterns such as "devoice a final obstruent, if there is another obstruent in the word". In this presentation, I show that although such cumulative effects are apparently unattested in processes that give rise to alternations, they are indeed observed in gradient phonotactic restrictions. I discuss two cases in detail: complexity effects across the root in Lakhota, and gradient cluster cooccurrence restrictions in English. In both cases, structures that are rare on their own become even rarer (underattested relative to marginal probabilities) in combination. I show that such effects can, in fact, be modeled in a Maximum Entropy model, as long as the faithfulness violations of Correspondence Theory are replaced with a single cost, modeled here as MParse. Crucially, such effects are expected to arise only under very limited conditions, such as when both structures are independently highly marked, or during acquisition (Albright, Magri, and Michaels 2008).
Suggested reading
Pater, Joe. "Weighted constraints in generative linguistics." Cognitive Science 33.6 (2009): 999-1035.