Mapping a corpus-induced ontology of action verbs on ItalWordNet

Action verbs are the least predictable linguistic type for bilingual dictionaries and they cause major problems for NLP technologies. This is not only because of language specific phraseology, but it is rather a consequence of the peculiar way each language categorizes events. In ordinary languages the most frequent action verbs are “general”, since they extend productively to actions belonging to different ontological types. Moreover, each language categorizes actions in its own way and therefore the cross-linguistic reference to everyday activities is puzzling. A cross-linguistic stable ontology of actions is difficult to achieve because our knowledge on the actual variation of verbs across types of actions is largely unknown. This paper briefly presents the problems and the building strategies of the IMAGACT Ontology, which aims at filling this gap, and compares some early results on a set of Italian verbs with the information contained in ItalWordNet. 1 The Semantic Variation of Action verbs within and across Languages Action verbs refer to at least an eventuality where an agent/causer theta role fills its argument structure. In all language modalities, action verbs bear the basic information that should be processed in order to make sense of a sentence. Especially in speech, they are the most frequent structuring elements (Moneglia and Panunzi, 2007), but unfortunately no one-to-one correspondence can be established between an action verb, conceived as a lexical entry, and an action type, conceived as an ontological entity. For instance, the English verb to take can refer to qualitatively different actions. In some uses the agent assumes the control of an object and changes its location (1); in some other uses the agent receives the object (2); in other cases, the agent takes the object away from somebody else (3): (1) John takes the umbrella