Résumé
This document presents two bilingual phrase-based alignment methods handling syntactic constituents (sub-sentential components) of parallel sentences. The methods rely on an asymmetrical parsing of both languages: Light part-of-speech tagging for the target language, syntactic tree building for the 'source' language and the complexity of each is studied. One of their benefits is that they do not require lexical knowledge for granting alignment. Another is that they align constituents of variable length and structure, thus providing information about divergent translations. Their originality rely on the fact that parsing of the supposed source language is reused both in resource building and alignment process. The models and methods can be seen as a subclass of Example Based Machine Translation. They built what can be called as an Alignment Mem- ory, the alignment correspondent of Translation Memories. The information acquisition process, partly supervised, is embedded in an online graphical human interface which accelerates the construction of golden corpora by one or many users.