Over the spring break, Andries Coetzee gave a colloquium presentation at the University of Southern California. In his talk, he presented an overview of his research on phonological variation. He argued that an adequate model of phonological variation should allow grammatical and non-grammatical factors to co-determine the properties of the variation that speakers produce, and then developed one such a model in noisy Harmonic Grammar. This research attempts to integrate the seemingly incompatible traditions of formal phonological theory, Labovian variationist linguistics, and exemplar models of grammar. The title and abstract of his presentation is included below.

Phonological Variation: Grammar, Usage Frequency, Speech Rate and Lexicality

Significant advances have been made over the past two decades with regard to incorporating variation into generative models of phonology (Anttila 1997, Boersma & Hayes 2001, etc.). However, the models that have been developed in this tradition are exclusively grammatical, implying that grammar is solely responsible for variation. This runs counter to evidence from the variationist sociolinguistic tradition, and the speech production/perception literature. Research in these fields have documented that, in addition to grammar, other social and cognitive factors also influence variation. A more complete model of variation will allow input from multiple factors, and have an explicit integration function for combining the contribution of different factors. In this presentation, I will develop a model of phonological variation, based on Harmonic Grammar, that formally integrates the contributions of both phonological grammar and other not strictly grammatical factors.

The presentation will focus specifically on three non-grammatical factors: (i) Usage frequency. Variable simplification phenomena are more likely to apply to words of higher than lower frequency. I will illustrate this based on a corpus study of word-final t/d-deletion (more deletion in frequent just than infrequent lust), and an ultrasound study of coda l-vocalization (more extreme vocalization in frequent milk than infrequent ilk). (ii) Speech rate. Assimilatory processes, such as cross-word nasal place assimilation (greem box for green box), are more likely to apply at faster than slower speech rates. Based on the results of a perception experiment, I will show that
listeners rely on this correlation between speech rate and assimilation – they are more likely to interpret the acoustic sequence [… m b …] as an instance of /… n b …/ at faster than slower speech rates. (iii) Lexicality. Listeners are more like to apply perceptual accommodation for phenomena such as cross-word nasal place assimilation if the resulting percept will result in a word than if not (i.e. accommodation is more likely in greem box than in um pa).