Discovering Arabic structures from texts: what a formal analysis can tell us

Abstract

We present a new approach to discover Arabic language structures from electronic texts. The method is based on a distributional analysis inspired from Arabic Grammatical Tradition (AGT) and Harris, and uses a minimum knowledge about the Arabic language. The idea underlying this research is that in the absence of a formal model of Arabic language and freely usable Natural Language Processing (NLP) tools for this language, what we can learn about the structures of this language only by a formal analysis of raw corpora written in it. By the word 'formal' we mean, here, that the analysis is based only on the form of the written texts and uses only a minimum knowledge about the language. This knowledge consists in some minimal hypothesis inspired from AGT and that never refers to the meaning of words, utterances and texts. On the contrary, we do not yet know these structures, but we want to discover them formally by automatic algorithms.