The Weekly Newsletter of MIT Linguistics

Colloquium 2/27 - Gunnar Hansson

Speaker: Gunnar Hansson (UBC)
Title: Learning Long-Distance Phonotactics: Dissimilation, Assimilation and Correspondence
Date: Friday, February 27th
Time: 3:30-5:00p
Place: 32-141

In current phonological theory, the leading constraint-based approach to modelling long-distance consonant assimilation is Agreement by Correspondence (ABC; Walker 2000, Hansson 2001/2010, Rose & Walker 2004). In the ABC model, long-distance agreement is dependent on an abstract, and in principle covert, correspondence relation that carves up the surface string into sets of corresponding segments (equivalence classes) in ways negotiated by ranked and violable constraints. Some of these (CORR constraints) demand that similar segments stand in correspondence, while others (CC-Limiter constraints) can curtail correspondence by demanding that pairs of correspondents obey certain criteria. However, there is reason for skepticism as to whether the formal machinery of the ABC model has adequate empirical support, in particular its fundamental notion that similarity-based agreement is a mediated effect (similarity-based correspondence; correspondence-based agreement). The factorial typology of ABC generates many unattested patterns that are highly suspect, while other attested and formally simple patterns become hard to capture; various proposed amendments merely highlight underlying flaws in the overall architecture. Recently the implications of the ABC model for consonant dissimilation have been explored in detail by Bennett (2013, 2014), who identifies a number of “mismatch predictions” whereby the typologies of dissimilation and harmony should be opposites of one another. I will report on a series of artificial grammar learning experiments (in collaboration with Kevin McMullin) which investigate how inductive biases and heuristics affect the learning of nonadjacent phonotactic dependencies between consonants. The central question is what implicational relationships, if any, learners assume to hold between “transvocalic” consonant pairs (…CVC…) and pairs separated by greater distance. Some of these experiments follow a poverty-of-stimulus paradigm, testing how learners generalize in the absence of overt evidence (cf. Wilson 2006, Finley & Badecker 2009, Finley 2011, 2012), whereas others are designed to provide explicit evidence for patterns of questionable status (cf. Lai 2012, White 2014). The findings suggest that, with respect to the hypothesis space, learners treat dissimilatory dependencies and assimilatory dependencies (harmony) equivalently, in ways that run counter to predictions of the ABC model. I will outline future studies that aim to clarify the formal nature of “transvocalic” locality: whether the relevant criterion is syllable-adjacency (Odden 1994, Rose & Walker 2004, Bennett 2013) or adjacency on a consonantal tier (Hansson 2010, McMullin & Hansson 2014).