Résumé : This paper presents the two instruction generation systems submitted by the team TALARIS at INRIA-Nancy to the GIVE- 2 challenge 2010. The first system (NM) aims to provide high-level instructions by giving freedom to the user and relies on an extension of the Dale and Reiter's in- cremental algorithm to generate the refer- ring expressions. The second system (NA) uses a more constrained directive naviga- tional strategy and relies on Reference Do- main Theory (Denis, 2010) to generate a wide range of referring expressions in- cluding pronouns, ellipsis and alternative phrases. The results show that despite NA is more successful than NM, NM is consid- ered a bit more enjoyable by the subjective assessment.