1. The query is the input of the network. Each node from the term layer computes an input Qu value from this initial query: = where g is the identity function. Out(ti) g(In(ti)) = and then an activation value : In(ti) qui return the source query term. 3. The desambiguiation of the transalted query consist of retaining only target terms that and Dictionary only. It can be seen that the desambiguiation based on aligned corpus is better aligned corpus is eectiv e the average precision improves of 4%. than the dictionary and the desambiguiation based on dictionary. The desambiguiation based on Table 4 compares the results between the runs irit1bFr2En and irit2bFr2En (Dictionary+desambiguiation) the stemming which was used only for French. We notice clearly that the monolingual search is the German. Italian results are better than Spanish and German. Spanish results are better than Table 5 shows that French monolingual results seem to be better than both Italian, Spanish and German. These runs were done using exactly the same procedures the only dierence concerns much better than both the multilingual and the bilingual searches. 5 5 Conclusion Table 5: Comparison between monolingual search References