Corpus Annotation of Macro Discourse Structures

We present our discourse annotation project, ANNODIS, which aims to make available a diversified French corpus annotated with discourse information, along with a set of tools for annotation and corpus exploitation. An original aspect of the project is that it combines two theoretically and methodologically different points of view on discourse: bottom-up and top-down. In the bottom-up perspective, basic constituents are identified and linked via discourse relations. In a complementary manner, the top-down approach starts from the text as a whole and focuses on the identification of configurations of cues signalling higher-level text segments, in an attempt to address the interplay of continuity and discontinuity within discourse. The focus of this paper is the annotation scheme used in the top-down approach, which revolves around enumerative structures. These structures, which are of particular interest to our project because of their ability to occur in nested configurations and at all levels of granularity (from within a sentence to across text sections), are the discourse object chosen to “bootstrap” our approach. We describe the different stages involved: corpus selection, pre-processing and “marking” techniques, and the specific interface facilities, designed to make it possible for coders to navigate and scan the text in order to identify relevant spans at different granularity levels. keywords: Corpus annotation, discourse organisation, text segmentation, macro discourse structures, computational linguistics I. A MACRO APPROACH TO DISCOURSE ORGANISATION The work presented here is part of a larger project, the ANNODIS project, the aims of which is to create an annotated corpus of French written texts for the study of discourse organisation. Discourse organisation is difficult to study with corpus linguistics methods because of the lack of