Hugo Hernault

A sequential model for discourse segmentation

Identifying discourse relations in a text is essential for various tasks in Natural Language Processing, such as automatic text summarization, question-answering, and dialogue generation. The first step of this process is segmenting a text into elementary units. In this paper, we present a novel model of discourse segmentation based on sequential data labeling. Namely, we use Conditional Random Fields to train a discourse segmenter on the RST Discourse Treebank, using a set of lexical and syntactic features. Our system is compared to other statistical and rule-based segmenters, including one based on Support Vector Machines. Experimental results indicate that our sequential model outperforms current state-of-the-art discourse segmenters, with an F-score of 0.94. This performance level is close to the human agreement F-score of 0.98.

Semi-supervised discourse relation classification with structural learning

The corpora available for training discourse relation classifiers are annotated using a general set of discourse relations. However, for certain applications, custom discourse relations are required. Creating a new annotated corpus with a new relation taxonomy is a timeconsuming and costly process. We address this problem by proposing a semi-supervised approach to discourse relation classification based on Structural Learning. First, we solve a set of auxiliary classification problems using unlabeled data. Second, the learned classifiers are used to extend feature vectors to train a discourse relation classifier. By defining a relevant set of auxiliary classification problems, we show that the proposed method brings improvement of at least 50% in accuracy and F-score on the RST Discourse Treebank and Penn Discourse Treebank, when small training sets of ca. 1000 training instances are employed. This is an attractive perspective for training discourse relation classifiers on domains where little amount of labeled training data is available.

From rhetorical structures to document structure: shallow pragmatic analysis for document engineering

In this paper, we extend previous work on the automatic structuring of medical documents using content analysis. Our long-term objective is to take advantage of specific rhetoric markers encountered in specialized medical documents (clinical guidelines) to automatically structure free text according to its role in the document. This should enable to generate multiple views of the same document depending on the target audience, generate document summaries, as well as facilitating knowledge extraction from text. We have established in previous work that the structure of clinical guidelines could be refined through the identification of a limited set of deontic operators. We now propose to extend this approach by analyzing the text delimited by these operators using Rhetorical Structure Theory. The emphasis on causality and time in RST proves a powerful complement to the recognition of deontic structures while retaining the same philosophy of high-level recognition of sentence structure, which can be converted into application-specific mark-ups. Throughout the paper, we illustrate our findings through results produced by the automatic processing of English guidelines for the management of hypertension and Alzheimer disease.