2018 Friday Session B 1715

Friday, November 2, 2018 | Session B, Conference Auditorium | 5:15pm

Incremental strategies in children’s language production: Evidence from eye-gaze
J. Brough, H. Rabagliati, H. Branigan, C. Gambi

Adults produce language incrementally, and this incrementality can manifest in two different ways. Lexical incrementality refers to sentence planning on a concept-by-concept basis, with formulation and articulation beginning with the most accessible entity, which can be determined by low-level attentional cues (e.g. first seen = first mentioned). In structural incrementality, sentence planning begins with a general understanding of the structural relationships between entities (e.g., agent, patient, with e.g., animate entities more likely to be considered as agents, and agents more likely to be sentence subject), and that structure guides retrieval of lexical items. Eye-tracking experiments, in which speakers’ gaze is analysed as they describe pictures, have provided extensive evidence for incremental production in adults (Kuchinsky et al., 2011) but little is known about online sentence production from a developmental perspective: Do children also produce language incrementally and, if so, do they show both lexical and structural incrementality, or use only one strategy?

40 adults and 48 3- and 4-year-olds described events interpretable from two perspectives (e.g., looking for/hiding, see Figure 1, c.f., Gleitman et al., 2007), while their eye movements were tracked. To measure the effect of animacy, scenes showed actions executed by one human and one anthropomorphic shape. To measure effects of perceptual salience, one character was colourful while the other was in greyscale. We tested how these factors affected whose perspective participants took when describing the scene, i.e., which of the two perspective verbs they chose. This required them to choose who to mention first (as the subject) and second (as the object), e.g. the woman is looking for the star vs the star is hiding from the woman. We then assessed the relationship between what participants said, and where they gazed.

We found both children and adults produced speech incrementally: Figure 2 shows that when both groups mentioned the first character, they had already started to shift gaze to the second character. Next, we analyzed incrementality type. If speakers were lexically incremental, word order would reflect fixation order. If they were structurally incremental, word order would reflect the linguistic bias to map sentence subject onto the most conceptually accessible animate entity (the human), i.e. the woman is looking for the star, independent of fixation order.

Adults provided evidence for lexical incrementality. The first character they gazed at was consistently used as the sentence subject (p < .05). Moreover, on 27% of trials, adults did not even fixate the second character before beginning utterances. These patterns were not present in the child data. Children provided stronger evidence for structural incrementality. More than adults, they consistently mentioned the human first (p < .01), suggesting they were using the most animate and conceptually accessible entity as the starting point of their sentence structures. Interestingly, perceptual salience did not affect either group’s descriptions.

The current work provides strong evidence that children produce language incrementally, and can do so based on structured relations between entities. However, children did not provide evidence for lexical incrementality as a strategy.