L’alternance actif / passif en français : une étude statistique sur corpus écrit

Abstract

We study the active/passive alternation in written French to discover preference constraints. We have extracted and annotated a sample of 500 clauses (250 active and 250 passive) of the French Treebank (Abeillé et al., 2019) and modeled them with logistic regression (Baayen, 2008). Short passives (without a par-phrase) are more frequent than long passive (77% of passives are short); in 80% of short passives, the omitted argument is present in the context. We show that argument length and information structure are important factors in active/passive alternation. We also show that the passive construction obeys more general cross-linguistic harmonic argument alignment constraints (Bresnan et al., 2001 and 2007).

Keywords

• written corpus
• passive
• preference constraints
• statistical modeling