ABSTRACT
BACKGROUND: Given the impact of lexical properties such as valence, arousal, and concreteness in language processing, recent computational methods have been designed to extrapolate these values from different sources, such as word co-occurrence or word association corpora. These methods have been proven to be particularly successful approaches to extract lexical features from word association data. Consequently, valence, arousal and concreteness seem to be represented in word association, and we hypothesize that they might in fact be critical mediators in the process. METHOD: To test our hypothesis, we paired the cue and associate words from three databases in three different languages with their valence, arousal and concreteness values. We then conducted linear regression analyses to see if an associate's score in each dimension could be predicted by the scores of its cue word. RESULTS: The analyses showed that the score of the cue words in each of the three dimensions was a strong predictor of the scores of their associates in the same dimension. Furthermore, words that were more strongly associated tended to have more similar scores. CONCLUSIONS: We showed that across different languages, word association is mediated and can be predicted by concreteness, arousal and valence.
