Are phonological phrases exploited on-line for the syntactic
analysis of spoken sentences?

Severine Millotte and Anne Christophe
Laboratoire de Sciences Cognitives et Psycholinguistique, CNRS-EHESS

Prosody has often been proposed as being useful for early syntactic acquisition (e.g. Morgan, 1986). There is evidence that intonational phrases constrain on-line syntactic analysis in adults (e.g. Warren et al. 1995), and are perceived by infants. Here we focus on a smaller unit, the phonological phrase. Recent evidence suggests that both adults and
1-year-old infants interpret phonological phrase boundaries as word boundaries on-line (Christophe et al. 2001). Since phonological phrases are available early enough in processing to constrain lexical access, they might also constrain early syntactic analysis. This is what we tested here with adults.

We constructed pairs of sentences with a local syntactic ambiguity, in which one word was either a verb or an adjective: "[Le petit chien] [mord la laisse] [qui le retient]." "[Le petit chien mort] [sera enterré demain].." ("[the little dog] [bites the leash] [which ties him]" "[the little dead dog] [will be buried tomorrow]"; square brackets mark phonological phrases). Both sentences contain words which are pronounced identically until the ambiguous word; however, they differ in syntactic structure, and as a result in prosodic structure. The ambiguous word is preceded by a phonological phrase boundary when it is a verb, but followed by one when it is an adjective. In experiment 1, subjects completed
auditorily presented sentence fragments cut just after the ambiguous word. They gave significantly more verb responses for verb sentences than for adjective sentences (and vice-versa). This off-line experiment tells us that phonological phrase boundaries are perceived and exploited by adult listeners, but not when they are exploited: It could either be during the
generation of syntactic parses or during selection (e.g. Boland et al. 2001). In experiment 2, subjects detected words defined with their syntactic category (e.g. "mordre", "to bite" for the verb target). They had to respond to the verb target in the verb sentence, but refrain from responding in the adjective sentence. In addition, non-ambiguous control
sentences contained the same target words. If prosodic boundary cues were a perfect indicator of syntactic structure, we would observe no effect of ambiguity whatsoever. We did observe slower responses for ambiguous sentences. However, subjects false-alarmed more to adjective sentences, where the prosodic boundary follows the target word, than to verb
sentences, where the prosodic information comes earlier. This suggest that prosody did play a role, even if it was not sufficient to entirely disambiguate the sentences. A further experiment will use sentences produced with a neutral prosody, in which prosodic disambiguating cues are less reliable; We expect a greater effect of ambiguity in this condition.