Cross-situational inference and meaning space structure

Abstract

Humans use language to direct each other’s attention to complex meanings. Researchers hold differing opinions on the evolutionary origin of these meanings: some hypothesise that human ancestors had innate, pre-linguistic concepts, similar to the referents of present-day nouns and verbs; others that simple meanings were broken down from initially complex meanings associated with specific situations. Experimental investigations of language evolution, meanwhile, have generally assumed a structured, pre-existing meaning space. This dissertation argues that a closer look at the ostensive-inferential nature of human communication supports a different account of the emergence of meanings and challenges the assumption of a pre-existing meaning space. Building on an experimental paradigm from Xu & Tenenbaum (2007), participants were presented with scenes of complex events to test how their meaning inferences were affected by the interaction of suspicious patterns in training input with their world knowledge. The results showed that the complexity of meaning lexicalised depended on this interaction, with substantial variation caused by both differing salience effects of events and differing world knowledge of participants. This result shows that meaning grounding in humans is not simply a matter of matching words to pre-existing cognitive concepts, or of associating a word with a specific recurring situation, but is an intelligent process crucially sensitive to the speaker’s intention to communicate. Further investigation is called for into the role of communication in forming a meaning space, and what determines this meaning space’s structure.