Learning Words in Linguistic Contexts: Using Language to Fast-Map and Acquire Partial Knowledge

• In Event: Poster Session 03
  In Poster Session: PS 03 Topic - Language, Communication

Abstract

Researchers have long debated whether children’s rapid word learning involves fast-mapping, a form of one-shot learning (e.g. Woodward, Markman, & Fitzsimmons, 1994), or accumulating partial knowledge of word meanings gradually, across many situations (e.g. Yurovsky, Fricker, Yu, & Smith, 2014). Most attempts to resolve this debate have focused on acquiring words in isolation, despite children’s input consisting of words embedded in rich linguistic contexts (Hoff-Ginsberg, 1990). In the present work, we argue that children can both fast-map and accumulate partial knowledge during word learning, and propose two experiments to demonstrate the important role that the surrounding language plays in each.

In Experiment 1, we ask whether fast-mapping is mediated by the linguistic context in which a word occurs. During learning trials, children will see two pictures and be asked to choose the familiar one; the other serving as an available referent for a novel word (“Mary feeds the bananas to the wug. Click on the banana.”). Crucially, the available referent will either align or conflict in animacy with the linguistic context (Figure 1). After two filler trials, we will test whether children have mapped the novel word to the available referent by asking them to choose between it and another novel picture of the same animacy (“Mary sees the wug again! Click on the wug”). Following Woodard et al. (2016), we will use mixed-effect logistic regression to predict children’s test-trial accuracy by condition. If fast-mapping behavior is mediated by linguistic context, children should map the novel word to the available referent only when its animacy aligns with the verb during learning. We piloted this experiment with 102 adults (51 per condition) and confirmed this prediction: adults were significantly more likely to select the correct referent when it aligned with the verb during learning (𝛽=0.46, SE=0.20, 𝝌2(1)=5.37, p=0.02, Figure 2).

To test whether children use linguistic contexts to acquire partial knowledge about word meanings, we will change the animacy of the test items in Experiment 2, allowing us to ask whether children acquire a feature of the novel word (e.g.[+animate]), rather than mapping the novel word to a specific referent (Figure 1). We piloted on another 100 adults (50 in each condition) and found that, as predicted, adults were significantly more likely to select the correct referent when it aligned with the verb during learning (𝛽=1.65, SE=0.35, 𝝌2(1)=22.59, p<0.001). Further, a significant number of adults in the conflict condition selected the animate referent (consistent with the verb information), even though the available referent during learning was inanimate.

We will run 80 2-3 year olds (20 per condition) via Lookit (Scott & Schultz, 2017), a feasible task because we’ve already created the study, piloted on adults, and arranged the logistics (e.g IRB approval). Our findings will not only have important implications for theories of word learning, but will also emphasize the important role that language itself plays in children’s early vocabulary development.

Authors

• Anna Alberski, University of Pennsylvania
  Presenting Author

• Kathryn Schuler, University of Pennsylvania
  Non-Presenting Author