Summary

A Universal Dependencies corpus for spoken French.

Introduction

The corpus was automatically converted from the Rhapsodie treebank and then underwent many manual corrections and improvements.

Development

The corpus is maintained in the SUD format and is available in the SUD_French-Rhapsodie repository.

Prosodic annotations from the original project were imported into the SUD data in 2025. This work is described in the TLT paper:

Maria Paz Botero-Garcia, Emmett Strickland, Bruno Guillaume, Sylvain Kahane, and Anne Lacheret-Dujour. 2025. An intonosyntactic treebank for spoken French: What is new with Rhapsodie?. In Proceedings of the 23rd International Workshop on Treebanks and Linguistic Theories (TLT, SyntaxFest 2025), pages 111–118, Ljubljana, Slovenia. Association for Computational Linguistics.

The richest annotations are available in the prosody_pauses folder in the SUD repository. Several other versions are automatically built from it.

The table below outlines the various available formats and their production methods.

Treebanks	Desc	Files	Production
SUD_French-Rhapsodie-prosody_pauses	SUD Syntax + Prosody (including pauses)	prosody_pauses/*.conllu	Source data
SUD_French-Rhapsodie-prosody	SUD Syntax + Prosody	prosody/*.conllu	grs/remove_pauses.grs
SUD_French-Rhapsodie@p_words	SUD Syntax (on phonological words) ${}^1$	p_words/*.conllu	grs/remove_syllables.grs
SUD_French-Rhapsodie@latest	SUD Syntax	*.conllu	grs/split_amalgam.grs
UD_French-Rhapsodie@conv	UD Syntax	*.conllu in UD repo	fr_SUD_to_UD.grs in converter

$^1$ We called phonogical words, the version in which the almagams (like au or du) are not split into syntactic words (à+le or de+le) as expected in UD.