Colloquium: Kevin McMullin (UOttawa).  Distance, blocking, and optionality on phonological tiers

Many phonological processes apply across large amounts of intervening material, including, e.g., vowel harmony, consonant harmony, and long-distance patterns of dissimilation. These phenomena have long posed a challenge for phonologists, and have served as motivation for incorporating tiers or projections into various theoretical frameworks. In this talk, I will argue that when treating long-distance dependencies as tier-based patterns, there is an inherent relationship between the properties of distance, blocking, and optionality. I will present typological and experimental evidence in support of this claim, further outlining several testable predictions that are the focus of research in progress.

I will first review the findings of McMullin (2016), which argues that patterns of long-distance consonant agreement and disagreement are best treated as members of the Tier-based Strictly Local (TSL) class of formal languages (Heinz et al. 2011). In particular, characterizing them as such allows for a straightforward account of unbounded vs. transvocalic patterns. This distance-based dichotomy is robustly attested cross-linguistically, and a series of artificial grammar learning experiments (McMullin & Hansson, submitted) suggests that it is reflective of a human learning bias. Moreover, the TSL class includes patterns that can be blocked by intervening segments, but excludes a number of computationally complex patterns that are predicted to be possible within the Agreement by Correspondence framework (Rose & Walker 2004, Hansson 2010), but that are seemingly unattested (McMullin & Hansson 2014).

Second, I will argue that it is not necessary to draw a strict distinction between the notions of ‘distance’ and ‘blocking’. Within this distance-free approach, transvocalic distance emerges when consonant-consonant interactions are blocked by other intervening consonants. Many patterns also exhibit ‘distance-based decay’ (Zymet 2014), where the pattern is less and less likely to apply as distance increases. I will demonstrate not only that this is a predicted property for any dependency that can be blocked, but that allowing for blocking to be optional makes it possible to derive an expected rate of decay equivalent to distance-based options offered in the literature (McMullin, submitted). While this approach is amenable to many theoretical frameworks, I will illustrate specifically how it can be understood in TSL terms.

Finally, I will provide a short overview of ongoing research focused on long-distance learning that supports (or at least unintentionally exploits) this connection between distance, blocking, and optionality, and present preliminary results from an experiment designed to reveal how humans interpret optional dependencies and whether they can learn patterns with blocking.