The variation of action verbs in multilingual spontaneous speech corpora

Most high frequency verbs referring to Action in our ordinary communication are General; that is, they productively extend to different actions in their own meaning. Moreover, languages can categorize actions differently. Despite its importance the variations of these verbs is largely unknown, and this lack of data prevents us from facing crucial aspects of lexical typology. The range of productive variations of Action verbs can be induced from spoken corpora, since references to actions are frequent in oral communication. This paper presents data derived from multilingual corpora (English and Italian) within the IMAGACT project and illustrates the methodology, the corpus design requirements, and the overall results obtained in this corpus-based research on cross-linguistic lexical semantics. The methodology identifies data that is relevant to semantic competence, separating the contexts in which the verb is used in its own core meaning from metaphors and phraseology. It makes use of visual prototypes rather than definitions in representing Action concepts, so allowing the display of typological variations across languages in a simple and informative manner. In the Italian corpus, among 677 verbs referring to Action, 106 are General, each of them comprising 3 to 15 action types. This subset records the majority of the cases in which there is reference to Physical Action and is for this reason a core area in the semantic knowledge of the language. Data regarding semantic variation can emerge only if a large enough variety of interactive context is recorded. As a whole, the incidence of metaphorical and phraseological usages in the verb occurrences is high (39%), but is higher in formal uses of language. Reference to Action is concentrated in informal, interactive contexts and especially in interactions with children in the early phases of language acquisition, which also testifies the higher variation of verbs across action types.