Phrasal prosody disambiguates syntax

Abstract

Two experiments tested whether phonological phrase boundary cues, as produced by naïve speakers, constrain syntactic analysis in French. Pairs of homophones belonging to different syntactic categories (verb and adjective) were inserted within locally ambiguous sentences that differed in their prosodic structure (e.g., [les pommes dures]…– hard apples… – versus [les pommes] [durent…] – apples last …– where brackets indicate phonological phrase boundaries). In Experiment 1 six speakers, unaware of the ambiguities, recorded the sentences. Acoustical analyses showed that they all produced reliable prosodic cues (phrase-final lengthening and pitch rise). Experiment 2 tested whether listeners exploited these prosodic cues to constrain syntactic analysis. They listened to the sentences beginnings (cut after the ambiguous word) and completed them in writing. Their assignments of the target words to their correct syntactic categories were better than chance. We discuss these results in light of the on-going debate about the production of disambiguating prosody by speakers who are unaware of the ambiguities.

Acknowledgements

The work reported in this paper was made possible by a Ph.D. and a post-doctoral fellowship from the Direction Générale de l'Armement, as well as a Marie Curie Intra-European fellowship (n° 024843) to Séverine Millotte. It was also made possible by a grant ‘Action Concertée Incitative Jeunes Chercheurs' to Anne Christophe (n°JC6059) from the French Ministry of Research, as well as a grant from the French Agence Nationale pour la Recherche (‘Early Language Acquisition: Experiments and Computational Approaches’). We also wish to thank Caroline Floccia, Julie Franck, Jeff Lidz, as well as three anonymous reviewers for valuable comments on an earlier version of this manuscript.

Notes

¹Ten French adults were asked to estimate the frequency and the plausibility of each target word in ambiguous sentences, using a scale from 1 (not frequent or not plausible at all) to 7 (very frequent or very plausible). The average frequency and plausibility of verb and adjective targets were balanced: [frequency: 4.8 for verbs vs. 5.1 for adjectives, t(21) < 1; plausibility: 5.0 for verbs vs. 5.3 for adjectives, t(21) = 1.4, p=.1].

²Two items were excluded from the analysis because they led to a large response bias (88% of adjective responses, whatever the sentence they heard, for item 21 ‘violet - violait’/purple - violated ; 97% of verb responses for item 22 ‘cool - coulent’/cool - sank). The analyses were run on 20 ambiguous items (the same results were obtained when these two items were kept in the analysis).

³Since adjective and verb responses were complementary (with the exception of the discarded responses), we used only the mean number of adjective responses in the statistical analyses.