A corpus - based , pilot study of lexical stress variation in American English

Phonological free variation describes the phenomenon of there being more than one pronunciation for a word without any change in meaning (e.g. because, schedule, vehicle). The term also applies to words that exhibit different stress patterns (e.g. academic, resources, comparable) with no change in meaning or grammatical category. A corpus-based analysis of lexical stress variation is one way of testing the validity of surveys of speakers " pronunciation preferences for certain variants. Such surveys include Wells " surveys of British English (1999, 2008) and Shitara " s survey of American English (1993). The present paper presents the results of a pilot study of American English, replicating part of Mompéan " s corpus-based study of British English (2010). In the current paper, the corpus consists of talks from the TED website The corpus includes approximately 11.5 hours of transcribed speech (92,750 words) produced by 34 educated speakers (17 men and 17 women with an American accent of standardized variety and some traces of regional pronunciations). To guarantee a minimum of representativeness for this pilot study, the items analyzed were found at least ten times in the corpus. The preliminary list of search terms showing lexical stress variation was based on the 261 items in the 2008 Longman Pronunciation Dictionary for which survey data was provided, lists found in a textbook on American pronunciation (Celce-Murcia et al., 1997/2007), lists in two previous studies (Shitara, 1993: Mompéan, 2010) and from anecdotal knowledge of frequent variants. Detailed results for lexical stress are provided for seven items. The TED corpus results do not always concur with LPD data and raise interesting issues concerning the use of authentic spoken corpus data. The paper also discusses designing and carrying out corpus-based pronunciation studies.