I conducted a pair of VOT experiments to learn more about how pre-existing contrasts and categories can limit accommodation. Experiment 1 was run with native speakers of English in English (an aspirating language), and Experiment 2 other with native speakers of Hungarian in Hungarian (a prevoicing language). Participants had to shadow artificially manipulated p-initial and b-initial words in their native language. The model talker was female in both studies. Her speech pattern either embodied an exaggerated version of how the participants' native language expresses the /p b/ contrast (an aspirating contrast for English-, a prevoicing contrast for Hungarian-speaking participants) or was the opposite (a prevoicing contrast for English speakers and an aspirating contrast for Hungarians). In addition, pre- and post-exposure reading data were also collected from each participant, as well as pre- and post-exposure labeling results and ratings about the model talker along 9 semantic differential scales (e.g. organized–unorganized, friendly–unfriendly).

I defined two hypotheses for how contrasts could be preserved. Maintain contrasts is a requirement for maintaining a distinction between any pair of contrastive sounds. This is a relatively flexible pressure, which can be satisfied if the contrast is simply shifted to a different range of an already used spectrum. This hypothesis predicts accommodation to the un-native-like realization of the /p b/ contrast (prevoicing contrast for English and aspirating contrast for Hungarian speakers). The alternative hypothesis, Maintain categories is a pressure to adhere to certain phonetic details of sound categories. While this requirement is not directly concerned with the preservation of contrasts, it leads to a similar effect by guaranteeing consistency in the phonetic realization of sound categories. If individual categories stay consistent, contrasts will also be preserved passively. This hypothesis predicts that English speakers will not converge with a prevoicing contrast (since the plain /p/ occupies the acoustic space where /b/ typically is in English) nor will Hungarian speakers with an aspirating contrast (for analogous reasons).

The conclusions from the two studies are quite similar, and present evidence for the second hypothesis (Maintain categories). In both experiments, participants accommodated to exaggerated versions of their native contrasts, but not to realizations of the contrast that were un-native-like. These results were found both in performances from the reading task and during the shadowing task itself. An inspection of the labeling data did not show any large-scale perceptual shift in the un-native-like condition either. Therefore, results from both experiments support the Maintain categories hypothesis, i.e. a token will only be accommodated if it is not phonetically atypical for its category.

These data also have consequences for other topics. The lack of convergence in the unnative- like conditions indicates that neither English nor Hungarian speakers perceive aspiration and prevoicing as two equally salient ends of a unified VOT spectrum. This raises the issue of when modeling VOT as a unified continuum—ranging from negative (prevoicing) to positive (long-lag or aspiration)—is justified. Moreover, this study presents evidence for likeability effects being due to mostly Solidarity- and Superiority-related measures, while the model talker's perceived Dynamism is a less important component for the purposes of accommodation (at least among the examined populations). While this study found no across the board gender effects, in certain cases males showed more sensitivity to these likeability factors (with the female model talker) than females did. In addition, while the English experiment found effects of ethnicity, these could be reduced to distance effects. That is, this was likely behavior that compensated for Black / African American participants tending to have more prevoiced /b/'s in their baseline reading results.

Aside from the empirical question of how phonological representations limit accommodation, this dissertation also discusses how pre-existing representations are in turn updated and changed over the course of accommodation. This is explored through a computationally explicit model. I first outline a basic exemplar model, which is not only capable of simulating distance effects but allows for even larger-scale longitudinal changes. This model is then extended to incorporate the empirical findings of this dissertation (i.e. a mechanism implementing Maintain categories). This reflects a system where phonological representations are in constant interaction and are constantly changing.

Eventually, this can all be incorporated into the language change literature at large. Since broader sound changes have been theorized to be propagated as a result of small, incremental instances of accommodation, any viable theory and model for accommodation has to be compatible with sound change. I argue that Maintain categories is in fact compatible with different kinds of long-term sound change, and outline some topics for future research in this direction.