Verbal Polysemy in Automatic Annotation.
Maryvonne Abraham · The Florida AI Research Society · 2007
The linguistic theory of Applicative and Cognitive Grammar analyses the language in three levels as follows: the linguistic level, the predicative level, and the semanticocognitive level. The meaning of the words is described at the semanticocognitive level. Here we give the description of the verbal semantics in order to build a lemmatized lexicon of polysemic words. On the one hand, the contextual exploration of texts is based on a linguistic context made up of muted segments of sentences without grammatical analysis. In this paper, we attempt to justify this way of processing by describing the characteristics of the arguments of the parameters of the verb. The problem: how do we read a text ? When we read a text, how do we gain access to the meaning? What are the processes which allow us to read a text and understand its meaning? In order to describe the meaning of a text, we break it up into smaller elements. Text has to be distinguished from the words of the text. The text is available as a syntactic structure that we can analyse from a (formal, categorial) grammar. The words are often polysemic. That is the reason why we try to break them up into smaller elements that we call primitives. We want to ensure a cognitive foundation to these primitives. It is interesting to describe the polysemy of a word. To do this, we try to organize the different significations of a word into a net, then we search for an invariant, using an abductive method [Figure 1]. This invariant does not belong to the language, but to our mental organization, and is seen as a mental representation of the word. 1 [Chomsky, 1979, 1981], [Jackendoff, R., 1978, 1983,1987], [Jacobson, R.]. 2 For [Descles, 1990], [Abraham, 1995], [this invariant is a « cognitive archetype ». For Picoche, 1986], it is a « signifie de puissance ». However, how do we build the meaning of a text from its components? Contextual exploration processes using textual indices. It does not use a lemmatized lexicon, but seeks flexed and conjugated words at the observable level of a text, voices of verbs, expressions dedicated to a domain, etc... The reading processing does not work by decomposing and recomposing the text and the words, but it builds the meaning of a text from the conjugated words seen as compiled knowledge: the transformations applied to the words indicate a part of the role of the words in a sentence. Decomposing and recomposing the meaning are not symmetrical processes. The problem of polysemy Figure 1 : From the form to the formal description of polysemy We focus on the description of the verbal lexicon, at a semantico-cognitive level. Many significations can correspond to a single entry form of the lexicon. We describe them using a semantico-cognitive scheme 3 The method to describe the lexicon is given in [Abraham 1995], [Descles, 187,1990], [Djioua, 2000], [Bogacki, 1983]. This method can be compared to [Pustejovky, 1991,1995]. Signification 1