Brain potentials differentiate compositional and non-compositional processing of Multi-Word Expressions: The case of idioms.

It is widely accepted that the meanings and the forms of Multi Word Expressions (MWEs) are stored in the mental lexicon (e.g., [1,2]). To comprehend these over-learned expressions the combination of single words meanings might not be readers’ most effective strategy. In the present study we used ambiguous idioms (e.g., land on someone’s feet) that have both a conventional meaning and a semantically well-formed literal interpretation, to isolate and compare the cognitive operations at play during compositional and non-compositional processing of the same sequences of words. The strings were inserted in different contexts: an Idiomatic Context Condition (IC) (1); a Literal Context Condition (LC) (2); a Control Condition (CC) where the last word of the idiom string was embedded in a literal sentence (3). The predictability of the last word of the idiom strings was equally very high across conditions. In Experiment 1, we recorded Event-Related Potentials (ERPs). We assumed that processing idiomatic configurations of words [3] involves recognition [4] and whole meaning integration processes. Therefore we expected that the recognition of the idiomatic strings should be facilitated in IC compared to LC and this could be reflected in a larger P300 effect. The integration of the idiomatic meaning in the sentential context might be reflected in N400 effects, [5]) or, alternatively, in P600 effects [6] if additional inferences based on vocabulary information are needed to preserve sentence coherence. The ERP waveforms showed an early positive effect (P300-like) on the penultimate word of the expression in IC. The processing of the last constituent was also influenced by the previous context: a late positive effect was observed on the last idiomatic constituent in IC compared to CC. To test whether the literal meaning of the idioms was computed even in idiomatic contexts [3] we ran a cross-modal lexical decision times experiment (Experiment 2) in which the visual target was either related or unrelated to the literal meaning of the last constituent of the idiom string. The results showed that regardless of context the literal meaning of the last constituent of the expression was still available at the offset of the idiom string. This suggests that the meanings of the constituent words are activated even when the idiom has been recognized as a conventional expression. Overall, our results show that idiom comprehension processes differ from literal processing: idioms need to be first recognized as such and recognition is easier when idioms are inserted in idiomatic contexts. At the end of the expression the conventional meaning is integrated into the sentence and because of the local ambiguity of the string readers might need to draw additional inferences to preserve sentence coherence, retrieving and integrating the conventional meaning of the expression.