Contextual features

Since version 2.18, a system of contextual features has been introduced for some morphosyntactic features that are not marked on a word but can be deduced from the context.

Related publication:

Status of morphosyntactic features Illustration with written and spoken French UD treebanks (Kahane et al., TLT-SyntaxFest 2025)

The following lexical features are used (in French treebanks only for now):

• Gender[ctxt]
• Number[ctxt]

In French, many adjectives do not mark the gender (autre, utile, … as well as noir, joli, … in Spoken French). In Spoken French, most adjectives and nouns do not mark the number (un petit chat vs des petits chats, both pronounced /pǝtiʃa/). Using Gender[ctxt] and Number[ctxt] to keep the information and to have a better parallelism between written and spoken data. It is also useful for the maintenance, to be sure that Gender and Number have not been forgotten.

Note

These lexical feature are not kept in conversion to UD. Gender[ctxt]=Femis replaced in UD by Gender=Fem and Exponence[Gender]=Absent.

See also UD issue 985.