Prosodic constraints on allophonic distribution in adult L2 acquisition

Abstract

Data from three American teenagers acquiring German during an exchange year show uneven acquisition patterns with respect to allophonic variation. Acquisition of [ ç ] / [ x ], as in [ dıç ] dich 'you' and [bax] Bach 'stream', is early, but acquisition of final obstruent devoicing is not acquired at all during the data collection period. In pairs of words exhibiting alternations, such as [ kınt ] ~ [ kındɐ ] Kind / Kinder 'child / children' the final /d/ is either realized as *[ d ] or the obstruent is voiceless in both positions: [ kınt ] ~ *[ kıntɐ ]. Factors such as markedness (Eckman, 1977) and similarity (Flege, e.g., 1995) fail to account for these acquisition patterns. An account based on learners' transfer of their American English intervocalic flapping (of /t/ and /d /) captures these facts and tentatively provides evidence in support of the Asymmetry Hypothesis (Young-Scholten, 1994). Flapping itself is not necessarily transferred in a direct way, but the domain of application larger than that of final devoicing is; transfer is thus taken to have the indirect effect of leading learners to erroneously surmise that a voiced alvealor stop in the input could be derived from underlying /t/. That the [ kınt ] ~ *[ k*ıntɐ] pattern is found only for alvealor stops supports this conclusion.