• 1 Laboratoire de Recherche informatique - Badji Mokhtar University

In this paper, we present two methods for correcting Arabic words generated by text and/or speech recognizers. These techniques operate as post-processors and they are conceived to be adaptable. They correct rejection and substitution word errors. The former one is very linked to the dictionary and is called 'lexicon driven', when the orther is very general exploiting contextual information and called 'context driven'. Arabic language properties are very useful in morpho-lexical analysis and so they were strongly exploited in the development of the second method. Substitution errors are rewritten in rules for being used by a rule based system. The extensions to the other levels of language analysis are considered in perspectives.