An experimental and computational study of thick terms

Thick terms like ‘courageous’, ‘smart’ and ‘tasty’ combine description and evaluation. These terms are contrasted with thin evaluative terms like ‘good’ and ‘bad’ and descriptive terms like ‘Italian’ and ‘green’. While this contrast raises several questions about the distinction between facts and values and the objectivity of evaluative language, it is unclear how thick terms combine description and evaluation. Here we contribute to address this issue with two experiments involving a cancellability task and a Cloze task, coupled with computational modelling. We found that words’ affective valence and co-occurrence patterns extracted from large corpora of natural language both reliably predict thick terms’ cancellability and cloze effects. This finding highlights the key role of automatic valenced associations, presumably acquired through experience of co-occurrent words within a shared cultural milieu, in explaining how thick terms combine evaluation and description.