Information-theoretic variables in Spanish-English bilingual speech

Abstract

Languages lenite similar segments to different extents—a finding that can be accounted for with information-theoretic variables like frequency, predictability, and informativity [Cohen Priva, 2017, Language 93: 569–597]. Prior research addresses the role of segment information content across languages, but assumes that information-theoretic variables operate on an in-language basis. While appropriate for monolingual speech, this assumption is problematic for bilingual speech, as an individual’s languages are known to influence one another (e.g. [Fricke et al., 2016, J. Mem. Lang. 89: 110–137]). In this paper, I report on a study using the Bangor Miami corpus [Deuchar et al., 2014, Advances in the Study of Bilingualism: 93–110], addressing how well information-theoretic variables predict lenition in Spanish-English bilingual speech. Specifically, do they operate on an in-language or cross-language basis? To address this, I focus on the duration of word-medial intervocalic fricatives shared by both languages—/f/ and /s/. Accounting for variables known to affect segment duration (e.g. speech rate), I use linear mixed-effect models to assess the contribution of information-theoretic variables in accounting for duration variation in bilingual speech. Four models are evaluated, compared, and discussed: (i) English in-language, (ii) English cross-language, (iii) Spanish in-language, and (iv) Spanish cross-language.