Using finite-state transducers to build lexical resources for Unitex Arabic package

International audience This paper addresses the issue of generating Arabic verbal inflectional paradigms using the FSA. In the process of the proposed approach the tokens drawn from the corpus are manually lemmatized and then finite state transducers are applied to the lemmas for producing all possible word forms with their full morphological features. The first strength of the approach lie in the algorithm of automatic generation of 184 transitions transducers, which is very cumbersome, if manually, designed. The second strength is the new classification of Arabic verbs; this classification is based on our new suggested inflection scheme that specifies the verb inflection paradigms. All resulting resources are publicly available and currently used as an open package in the Unitex framework under the LGPL license.