Prosodic cues to word order: what level of representation?

Within language, systematic correlations exist between syntactic structure and prosody. Prosodic prominence, for instance, falls on the complement and not the head of syntactic phrases, and its realization depends on the phrasal position of the prominent element. Thus, in Japanese, a functor-final language, prominence is phrase-initial and realized as increased pitch (^Tōkyō ni ‘Tokyo to’), whereas in French, English or Italian, functor-initial languages, it manifests itself as phrase-final lengthening (to Rome). Prosody is readily available in the linguistic signal even to the youngest infants. It has, therefore, been proposed that young learners might be able to exploit its correlations with syntax to bootstrap language structure. In this study, we tested this hypothesis, investigating how 8-month-old monolingual French infants processed an artificial grammar manipulating the relative position of prosodic prominence and word frequency. In Condition 1, we created a speech stream in which the two cues, prosody and frequency, were aligned, frequent words being prosodically non-prominent and infrequent ones being prominent, as is the case in natural language (functors are prosodically minimal compared to content words). In Condition 2, the two cues were misaligned, with frequent words carrying prosodic prominence, unlike in natural language. After familiarization with the aligned or the misaligned stream in a headturn preference procedure, we tested infants’ preference for test items having a frequent word initial or a frequent word final word order. We found that infants’ familiarized with the aligned stream showed the expected preference for the frequent word initial test items, mimicking the functor-initial word order of French. Infants in the misaligned condition showed no preference. These results suggest that infants are able to use word frequency and prosody as early cues to word order and they integrate them into a coherent representation

Abstract licence: CC BY

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.