The mental representation of derived words

Deadjectival nominals with –sa and –mi in Japanese are both phonologically transparent and morphologically decomposable. However, whilst –sa essentially serves to form nouns out of adjectives, –mi forms function as semantic labels with specific meanings. We examined –sa and –mi nominals in three experiments, an eye-movement experiment presenting –sa and –mi forms in sentence contexts and in two word recognition experiments using (primed and unprimed) lexical decision, to investigate the nature of their form-level representations. Whilst the word recognition experiments produced the same pattern of results for –sa and –mi forms, the eye-movement experiment demonstrated clear differences: –mi forms elicited longer reading times compared to –sa forms, except when the particular meanings of –mi forms were contextually licensed. These results show how different semantic properties affect the performance of derived words that have the same type of word level representation.