Longitudinal phonetic variation in a closed system

1 Introduction We present a progress report on a large-scale corpus study of the trajectories of different phonetic parameters over the course of three months in a diverse group of speakers situated in a linguistically and socially closed system. The data analyzed so far, and presented in this paper, consist of 806 Voice Onset Time (VOT) measurements from four contestants in Season 9 of the reality-television show Big Brother (Channel 4, United Kingdom). This show offers a unique opportunity to study " medium term " phonetic change in individuals: longer than studies of accommodation in the laboratory or in conversation (hours to days), but shorter than real-time sociolinguistic studies (years). We show that VOT measurements in this dataset, which consists of conversational speech over the course of three months, show several effects consistent with previous work (e.g. the dependence of VOT on place of articulation). We further show that change in the mean VOT of different speakers is non-linear over time, with speakers' VOT trajectories neither fluctuating around a single mean, nor drifting in a single direction throughout the observed period; in addition, a significant portion of this time dependence can be explained in terms of a perturbation to the social organization of the community of speakers, rather than time per se. While this last finding is post hoc in the sense that the hypothesis was formalized only after building other models of the data, and is therefore subject to verification with a larger dataset, it presents preliminary evidence that social factors can drive phonetic change in individuals over extended periods of time.