Contenu du sommaire : Orféo : un corpus et une plateforme pour l'étude du français contemporain

The orféo project enabled the development of a Corpus for the Study of Contemporary French (céfc). I present here the spoken part of the céfc as a collection of corpus which have been standardized in order to facilitate the use of resources. céfc is available for free download or searchable online.

The céfc corpus includes data from several different sources to make observable the diversity of oral French at least partly, solving the problems inherent to the heterogeneity of these data is intrinsic to the constitution of this resource and motivated by its objective. This article will describe, step by step, the methodological approach that enables us to build a homogeneous resource by pooling these different sources in order to provide coherent automatic annotations and to facilitate the analysis of an oral corpus of several million words.

The paper discusses the principles and criteria used in elaborating the POS tagset and the structure of the corresponding lexicon at use for the automatic parsing of the orféo corpus. This architecture is compared with the current Lexique des Formes Fléchies du Français (lefff) dictionary, available under open source license. The linguistic and natural language processing challenges are dwelled on. A specific attention is devoted to the processing of multiword expressions. Some ways of improvement of the system are provided.

Syntactic annotation of Spoken French: orféo's choices
This article presents the syntactic annotation choices for the orféo project. A corpus of Spoken French of more than 180 000 words was manually annotated in dependency syntax, then a 3 M word corpus was automatically parsed. The annotation choices are compared with those of the rhapsodie project, which preceded orféo, with universal dependencies (ud), which started shortly after orféo, and with surface-syntacticud (sud), which synthesizes orféo and ud's choices. orféo is characterized by a consideration of macrosyntax and list phenomena, as well as a restricted tag set that allowed a quick and more easily reproducible annotation.

Automatic syntactic parsing of the spoken part of the céfc
This paper presents the linguistic annotation tools that were developed in the framework of the orféo project and used to annotate the different corpora. Two important points are developed. The first one is sentence segmentation, which is a difficult problem when processing speech transcriptions. We show that taking into account syntax allows to obtain good segmentation performance. The second is the introduction of metadata features in the parsing process in order to adapt the models to the variety of data collected. The results obtained on the orféo corpus validate the proposed approaches and make it possible to estimate the quality of the annotations produced automatically on the orféo corpora which are not validated manually.

Prosodic annotation in orféo
The prosodic annotation described here relates not to the speaker emotions, attitudes or socio-geographical features, but to the prosodic structure, defined as the hierarchical organization of sentence accent phrases. This assemblage in successive levels maintains with the syntactic structure a variable degree of congruence. Prosodic annotation operates in two steps: (i) localization of stressed syllables (excluding emphasis) defining accent phrases by their final position (in French) and (ii) description of the prosodic events occurring on stressed vowels, events instantiated by specific melodic contours assumed by hypothesis to indicate between accent phrases dependency relations which define the prosodic structure.

A contrastive study of shell nouns in written and spoken corpora through automatic extraction
This paper describes a contrastive analysis of shell nouns in three registers: spoken texts, written texts and computer-mediated communications available in the Corpus d'Étude pour le Français Contemporain (céfc). Our method consists in an automatic extraction of lexical-syntactic patterns based on morphosyntactic and syntactic annotations provided by the céfc; followed by a manual evaluation. Results give new insights into the variation and behavior of shell nouns across registers.

The present article focuses on the pragmaticalisation of the French temporal phrase à un moment donné (“at one point”, “at a given moment”), and more particularly on one question (that was left unresolved by a previous study) about the role of the adjective donné (“given”). We hypothesize that it plays in favor of a greater subjective implication (favouring thus the pragmaticalisation of the phrase). We verify this hypothesis by a systematic examination of à un moment vs. à un moment donné, based on the oral part of the Corpus Study for Contemporary French. We also consider other possible variants: variants with il y a ("there is”) (il y a un moment (donné) | “there is a given moment”) or without the preposition à (“at”) (un moment (donné) | “a given moment”).

This article, based upon the Corpus d'Étude pour le Français Contemporain (céfc), aims at providing a mainly qualitative analysis of partial direct interrogatives present in the oral genre of commercial interactions, where goods are exchanged for money. A distinction is made between partial interrogatives occurring in non-transactional parts (such as greetings), on the one hand, and in transactional parts (such as requests for goods or payment), on the other hand. In particular, we highlight the specialisation of combien in direct or (by metonymy) indirect quantification of the product, on a semantic level, as well as the strong presence of a verbless interrogative structure where a noun is followed by comment, on a syntactic level. This article thus contributes to a characterisation, on syntactic, semantic and interactional levels, of partial direct interrogatives, and, more generally, to further reflection on oral genres.