Multi-dimensional meanings in lexicon formation

摘要

This chapter introduces a language game experiment for studying the formation of a shared lexicon when word meanings are not restricted to a single domain, but instead consist of any combination of perceptual features from many different domains. The main difficulty for the language users is that upon hearing a novel word they cannot be sure which aspects or properties of the referred object comprise the meaning of the word. We introduce an Adaptive Language Strategy which pays considerable attention to the adaptive nature of individual word meanings and allows the language user to use its linguistic items in a flexible manner, leading to extensive re-use. Using grounded language game experiments we show that this Adaptive Strategy elegantly copes with the problems introduced by using embodied robotic data and allows scaling towards large meaning spaces and population sizes. The strategy is further compared to a second one which lacks some of the adaptive and flexible features of the first strategy, and show that this non-adaptive strategy struggles to keep a high level of performance under taxing conditions.