Children’s Generalization of Novel Relational Nouns in Comparison Contexts: An Eye Tracking Analysis

• In Event: Poster Session 08
  In Poster Session: PS 08 Topic - Cognitive Processes

Abstract

Comparison situations (i.e., being introduced to several exemplars, rather than a single one) are known to favor word generalization performances for nouns (Gentner & Namy, 1999; Graham et al., 2010) and for relational nouns (Gentner et al., 2011) because comparing generates better stimulus encoding and highlights important category related properties. Previous studies have shown that semantic distance between items and the type of distractors influence generalization performances for relational nouns in comparison settings (Thibaut & Witt, 2015). Little is known of the solving strategies children follow when they compare and generalize novel relational nouns, or of the steps that lead to a generalization decision.
In the present study, we use eye-tracking data to identify these strategies that lead to efficient learning and generalization of relational nouns in comparison context.
We contrasted two main hypotheses that predict different search profiles, namely the projection first and the alignment first strategies.
We tested six-year-olds on a classic relational noun comparison and generalization task (Figure 1), in which children had to compare two pairs of objects (made of an entity E and an operator O) illustrating the targeted relational noun and generalize this relational noun by building a third pair of objects with a given test entity (TE) and an operator they had to choose from a set of three objects (Ch) including a relational choice (R), and two distractors (a thematic and a taxonomic distractor, Dis). The relational nouns given were pseudo-words. Children’s generalization was considered correct if they chose the relational choice. We also manipulated the semantic distance between learning pairs (Learning Distance: Close or Far) and between the standard learning pair and the generalization pair (Generalization Distance: Near or Distant). More data is still being collected.
Results analyzed the proportion of fixation times on AOIs and the proportion of transitions between items, sorted into three time slices (i.e., beginning, middle or end of a trial), for correct answers only.
The projection first hypothesis predicts early comparisons between items from the learning and generalization domains (i.e., between E and TE, E-Ch, O-TE, O-Dis, O-R). Alignment first predicts early comparisons between items in the learning domain (i.e., between entities or operators from different pairs EE/OO or between items from a same pair E-O) followed by comparisons between items in the generalization domain (i.e., between TE and R, TE-Dis, between choices Ch) .
We ran a four way ANOVA on the proportion of fixation times, with Learning Distance (Close, Far) as a between factor, Generalization Distance (Near, Distant), slice (Beginning, Middle, End) and Transition type as within factors. The main result was an interaction between slice and transition type F(20,340) = 12.95 p < .001 η^2= 0.432, (Figure 2), that revealed that transitions predicted by the projection first hypothesis have scores close to 0% indicating that they are not used during the task whereas transitions predicted by the alignment first hypothesis are used throughout the task.

Authors

• Eleanor Stansbury, Université de Bourgogne (uB)/Université de Dijon
  Presenting Author

• Arnaud Witt, Université de Bourgogne
  Non-Presenting Author

• Jean-Pierre Thibaut, Université de Bourgogne Franche-Comte
  Non-Presenting Author