Predicting Above-Sentence Discourse Structure Using Distant Supervision from Topic Segmentation

Abstract

RST-style discourse parsing plays a vital role in many NLP tasks, revealing the underlying semantic/pragmatic structure of potentially complex and diverse documents. Despite its importance, one of the most prevailing limitations in modern day discourse parsing is the lack of large-scale datasets. To overcome the data sparsity issue, distantly supervised approaches from tasks like sentiment analysis and summarization have been recently proposed. Here, we extend this line of research by exploiting distant supervision from topic segmentation, which can arguably provide a strong and oftentimes complementary signal for high-level discourse structures. Experiments on two human-annotated discourse treebanks confirm that our proposal generates accurate tree structures on sentence and paragraph level, consistently outperforming previous distantly supervised models on the sentence-to-document task and occasionally reaching even higher scores on the sentence-to-paragraph level.

Improving Topic Segmentation by Injecting Discourse Dependencies

“…discourse analysis and discourse parsing provide the means to understand and infer the semantic and pragmatic relationships underlying complete documents, well aligned with the local text coherence and highly correlated to the inter-sentential topical consistency, as shown in Louis and Nenkova (2012) and Muangkammuen et al (2020). With a variety of linguistic theories proposed in the past, such as the Rhetorical Structure Theory (RST) (Mann and Thompson, 1988), the lexicalized discourse framework (Webber et al, 2003a) (underlying PDTB), and the Segmented Discourse Representation Theory (SDRT) (Asher, 1993;Asher et al, 2003), we follow the RST framework in this work (1) as we focus on monologue text (as compared to dialogue frameworks, such as SDRT) and (2) since RST postulates complete discourse trees spanning whole documents, directly aligned with the topical structure of complete documents (Huber et al, 2021).…”