Comparing Models of Associative Meaning: An Empirical Investigation of Reference in Simple Language Games

  • 2018-10-08 21:51:44
  • Judy Hanwen Shen, Matthias Hofer, Bjarke Felbo, Roger Levy
  • 9

Abstract

Simple reference games are of central theoretical and empirical importance inthe study of situated language use. Although language provides rich,compositional truth-conditional semantics to facilitate reference, speakers andlisteners may sometimes lack the overall lexical and cognitive resources toguarantee successful reference through these means alone. However, languagealso has rich associational structures that can serve as a further resource forachieving successful reference. Here we investigate this use of associationalinformation in a setting where only associational information is available: asimplified version of the popular game Codenames. Using optimal experimentdesign techniques, we compare a range of models varying in the type ofassociative information deployed and in level of pragmatic sophisticationagainst human behavior. In this setting, we find that listeners' behaviorreflects direct bigram collocational associations more strongly thanword-embedding or semantic knowledge graph-based associations and that there islittle evidence for pragmatically sophisticated behavior by either speakers orlisteners of the type that might be predicted by recursive-reasoning modelssuch as the Rational Speech Acts theory. These results shed light on the natureof the lexical resources that speakers and listeners can bring to bear inachieving reference through associative meaning alone.

 

Quick Read (beta)

loading the full paper ...