Abstract

In distributional semantics, words acquire their meaning by exploiting the statistical information that is inherent in their linguistic environment. A common criticism towards such representations is that they do not explicitly encode semantic features unlike more traditional models of semantic memory, questioning the cognitive relevance of such statistical mechanisms during language learning and processing. Here we show that distributional semantic models provide a good fit to data obtained from implicit language learning experiments on adults. In these experiments participants are introduced to novel non-words, which co-occur with already known words conditioned on underlying semantic regularities, such as concrete/abstract, animate/inanimate. Participants can implicitly learn such underlying semantic regularities, although whether they do so depends upon the nature of the conceptual distinction involved and their first language. Using datasets provided from four behavioural experiments, which employed different semantic manipulations, we obtained generalisation gradients that matched closely those of humans, capturing the effects of various conceptual distinctions and cross-linguistic differences.