Combining Linguistics with statistics for multiword term extraction: a fruitfull association?