2018 Friday Session B 0900

Friday, November 2, 2018 | Session B, Conference Auditorium | 9am

Children’s N400 is sensitive to both predictability and frequency: Evidence from natural listening

T. Levari, J. Snedeker

In adults, the N400 response, classically detected in cases of semantic violation, has been shown to sensitively reflect the ease of lexical activation and semantic integration (Kutas & Federmeier, 2011). Adults show smaller N400s to words that are frequent (Dambacher, Kliegl, Hofmann, & Jacobs, 2006; Van Petten & Kutas, 1990) and predictable (Dambacher et al., 2006; Davenport & Coulson, 2011). However, studies of child processing have focused solely on responses to overt semantic violations (e.g. Atchley et al., 2006; Holcomb, Coffey, & Neville, 1992). Adult-like n400s are present as early as 19 months (Friedrich & Friederici, 2004), although children’s responses are more distributed and delayed (e.g. Atchley et al., 2006; Holcomb et al., 1992). What we don’t know is whether child N400s are sensitive to frequency or sentence and discourse level constraints. These are central questions for understanding children’s lexical processing. Is it guided by top-down expectations generated from prior semantic, syntactic, and/or pragmatic contextual constrains or does it largely reflect bottom-up facilitation, reflecting a word’s frequency or local semantic associations? To test this we use a natural listening paradigm, based on an adult procedure implemented by Brennan, Cantor, Eby, and Hale (2016). Adults and children, ages 5-10, listen to a story as ERPs time-locked to the onset of every word are recorded. We gather over 700 responses in a 15-minute session, while children perform a task that is natural and familiar. Each content word in the story is coded for frequency and three different measures of predictability. First, words are coded depending on their association to the preceding context, as measured by Latent Semantic Analysis (LSA) (Wolfe & Goldman, 2003). Second, two measures of cloze probability were calculated using Amazon Mechanical Turk. Subjects were asked to guess every word in a sentence while either being presented with sentences out of the blue (sentence cloze) or within the context of the full story (discourse cloze). All measures were then used to calculate surprisal, defined as the negative log-probability of a word in a given context (Hale, 2001; Levy, 2008). Baseline concerns related to item-by-item ERP analyses were addressed by controlling for relevant features of the words directly before and after the target word. Preliminary analyses (n=18 children, n=18 adults) show that, unlike adults, children’s N400 responses are sensitive to frequency (b=-0.42, t=-2.27, p < .05). However, above and beyond effects of frequency, both child and adult responses are sensitive to surprisal as calculated by sentence level (Children: b=-0.55, t=-2.97, p < 0.01; Adults: b=-0.25, t=-2.2, p < 0.05) and discourse level cloze (Children: b=-0.45, t=-2.35, p < 0.05, Adults: b=-0.27, t=-2.3, p < 0.05), but not LSA. If we assume cloze probability reflects higher level processing of top-down constraints from semantic/syntactic/pragmatic levels, our findings suggest that both adults and children use these representation during lexical processing. However, in children, lower level information, like frequency, continues to play a critical role. This finding is consistent with current theories sentence comprehension development in which children rely more on bottom-up cues (Snedeker et al., 2013).