The Role of Co-Occurrence Statistics in Lexical-Semantic Development

• In Event: 2-160 - The More You Know: How Prior Knowledge Scaffolds Cognitive Development

Integrative Statement

Fundamental to many cognitive processes, such as memory, reasoning, and learning, is the organization of knowledge about the world into a lexico-semantic network comprised of concepts linked by relations. For example, our knowledge of chickens is not isolated, but can instead be related to knowledge of other members of the same “taxonomic” category (e.g., birds), or other associated concepts, such as eggs. Where does relational knowledge come from?

Prior research on the origins of relational knowledge has focused on the emergence of taxonomic knowledge. In contrast, many theoretical accounts omit a role for association, or identify association as only contributing to relational knowledge early in development. Moreover, we know little about how association is learned. Here, we studied how, starting early in development, association may be learned from environmental statistical regularities with which labels denoting different concepts co-occur. Across two sets of studies, we investigated the influence of long-term, naturalistic and empirically manipulated co-occurrence regularities on relational knowledge.

In the first set of studies, we assessed children and adults’ relational knowledge for concepts whose labels reliably co-occur in environmental input (measured from CHILDES corpora), such as “sock” and “foot”, versus taxonomically-related concepts, such as “car” and “boat”. In one paradigm, participants judged whether a word and subsequent picture were “of the same thing”, and we measured the degree to which relations between non-identical items interfered with “no” responses. In another, participants studied pairs of words, and we measured the degree to which relations between words improved recall accuracy for the second word when prompted with the first. Both experiments revealed a substantial influence of co-occurrence in children and adults. Taxonomic relatedness only influenced relational knowledge in adults, but remained weaker than co-occurrence (Fig1).

The taxonomic relations in the first studies may in principle be extracted from co-occurrence, because taxonomically-related words often share patterns of co-occurrence with other words (e.g., “car” and “boat” both co-occur with “ride” and “fast”). Taxonomic-relational knowledge may emerge later than co-occurrence because shared co-occurrence patterns are harder to track/learn. The second set of studies therefore exposed children and adults to two “triads” of words, embedded within familiar sentence frames, which each consisted of a novel adjective (e.g., “dodish”) that separately co-occurred with a familiar noun (e.g., “horse”) and a novel noun (e.g., “geck”) (Fig2, top). Accordingly, within a triad, the adjective and nouns co-occurred, and the nouns shared co-occurrence patterns. In a subsequent free-association task, both children and adults exclusively relied on adjective-noun co-occurrence. When asked to map the words onto objects, adults reliably used “taxonomic” shared co-occurrence, such that they used novel nouns to label objects from the same category as familiar nouns from the same triad. Only one-third of children reliably produced these responses (Fig2, bottom).

These studies show that: 1) Co-occurrence contributes earlier and more strongly to knowledge organization than taxonomic relations, and 2) This asynchrony may occur because taxonomic relations are learned (in part) from shared co-occurrence, which may be harder to track/learn in childhood.

Authors

• Layla Unger, The Ohio State University
  Presenting Author

• Olivera Savic, The Ohio State University
  Non-Presenting Author

• Vladimir M. Sloutsky, The Ohio State University
  Non-Presenting Author