This paper presents a new word alignment method which incorporates knowledge about Bilingual Multi-Word Expressions (BMWEs). Our method of word alignment first extracts such BMWEs in a bidirectional way for a given corpus and then starts conventional word alignment, considering the properties of BMWEs in their grouping as well as their alignment links. We give partial annotation of alignment links as prior knowledge to the word alignment process; by replacing the maximum likelihood estimate in the M-step of the IBM Models with the Maximum A Posteriori (MAP) estimate, prior knowledge about BMWEs is embedded in the prior in this MAP estimate. In our experiments, we saw an improvement of 0.77 Bleu points absolute in JP–EN. Except for one case, our method gave better results than the method using only BMWEs grouping. Even though this paper does not directly address the issues in Cross-Lingual Information Retrieval (CLIR), it discusses an approach of direct relevance to the field. This approach could be viewed as the opposite of current trends in CLIR on semantic space that incorporate a notion of order in the bag-of-words model (e.g. co-occurences).