Syntactic and Semantic Effects on the Production of Ordering Errors

96. Speakers are sensitive to prediction mismatches between two cues to grammatical category in spontaneous speech Esteban Buz, Thomas Farmer, T. Florian Jaeger University of Rochester Speakers and listeners have a shared goal: trying to successfully communicate. Speakers produce an encoded message and listeners attempt to decode the message. Often, speaker messages are temporarily ambiguous, and listeners use informative cues to generate expectations about the interpretation of ambiguous speech. One cue that comprehenders are able to use is a word’’s phonological typicality (PT), or, the degree to which the sound properties of an individual word are typical of other words in the same grammatical category [1, 2]. Prior work has demonstrated that PT can influence reaction times to words in a sentential context that is highly predictive of lexical category, and that it can bias the category to which a noun/verb homonym is assigned during on-line comprehension tasks [3, 4]. We ask whether or not speakers are sensitive to PT, and specifically, how speakers behave when the information about lexical category membership that is provided by PT conflicts with that of syntactic probability. In a conversational speech corpus, we investigated the duration of function words (FNs) in sequences of FN NOUN. We find an interaction between the bigram probability of a noun and the phonologically typical (‘‘nouniness’’) of the NOUN. When a noun was unexpected, the FN was produced with shorter duration as typicality increased. As noun expectedness increased, this negative effect of typicality on FN duration decreased. The results suggest that grammatical information cascades down during the lemma selection, phonological encoding, and articulation processes, consistent with cascading activation models of production [5].