2018 Friday Session B 1615
Friday, November 2, 2018 | Session B, Conference Auditorium | 4:15pm
Preschoolers fail to revise their syntactic analysis, even when their initial interpretation is driven by prosodic cues
A. de Carvalho, J. Trueswell, A. Christophe
Because the prosodic structure of sentences correlates with syntactic structure, listeners exploit prosodic boundaries to detect syntactic boundaries and constrain their parsing (e.g. de Carvalho et al., 2016). However, since not every syntactic boundary is marked prosodically, listeners’ reliance on prosodic information to constrain parsing may be low, in comparison to more reliable cues, such as lexical content. Here we investigated how children/adults interpret an ambiguous noun-verb homophone when prosodic structure and lexical content provide conflicting information.
Homophones belonging to different syntactic categories were used to create temporary ambiguities (e.g., ferme, meaning ‘a farm’ or ‘to close’ in French). Sentences began with prosody supporting one interpretation, but ended with lexical content supporting the opposite interpretation. Sentences were created by cross-splicing two sentences which featured the homophone as a verb (e.g. [la petite]NP [fermeV la boîte] – ‘the little girl closes the box’), and as a noun (e.g. [la petite fermeN] [est jolie] – ‘the little farm is beautiful’; brackets indicate prosodic boundaries). Thus a Verb+Noun sentence was [la petite]NP [fermeV [estV jolie]VP – where listeners first heard ([The little girl] [closes… ), and interpreted ferme as a verb, then had to revise it, when only the noun interpretation was consistent with the last words ‘is beautiful’. In Experiment 1, 4-to-6-year-olds and adults listened to sentences while watching two images side-by-side: one illustrating the noun interpretation (e.g., a farm) and the other the verb interpretation (e.g., a girl closing something). Eye-gaze and pointing were recorded. Both adults and children initially looked toward the picture supported by sentence-initial prosody (Figs.1A-1B), but only adults switched to look at/select the ‘correct’ picture upon hearing disambiguating lexical evidence. Children looked/pointed at the picture consistent with their initial interpretation, supported by prosody (Fig.1B-1D). Experiment 2 asked (a) whether children would revise their initial interpretation if they repeated the sentence aloud before responding and (b) whether they would repeat the prosody they heard, or change it to accommodate the later-arriving lexical information. Children’s repetitions reveal that they most often repeat the sentence ‘as is’ and choose the initial interpretation, but when they do revise, they tend to correct the prosody (Table1). In Exp. 3, children heard the sentences without seeing the pictures, then walked across the room to select one. Selection patterns were similar to Experiments 1 and 2 (Fig.2), suggesting that image presence doesn’t play a role in children’s inability to revise.
Altogether, both adults and preschoolers rely heavily on the initial information provided by prosody to constrain their parsing, even in a situation where prosodic information is misleading. Contrary to adults, children do not recover from their initial commitment (even though they remember and repeat the whole sentence, even when visual information is delayed). This behavior replicates the so-called kindergarten-path effect (Trueswell et al., 1999): children give more weight to the information that arrives first, even though that information is prosodic, and supposedly less reliable than lexical content. These results suggest that prosody is not treated as an inferior source of information, to be ignored in case of conflict.