Cross-Framework Discourse Relation Classification Though Unifying Dimensions

Anonymous

TL;DR: Unifying dimensions for cross-framework discourse relation classification

Abstract: Existing discourse corpora annotated under different frameworks adopt distinct but somewhat related taxonomies of relations. The integration of these corpora has been an open research question. Previous studies on the interoperability of different discourse formalisms are mainly theoretical, although such research is performed with the hope of benefiting computational applications. In this paper, we show how the unifying dimensions (UDims) that originate from the Cognitive approach to Coherence Relations (CCR) (Sanders et al., 2018) can facilitate cross-framework discourse relation (DR) classification. To address the challenges of using predicted UDims for DR classification in model learning, we adopt the Bayesian learning framework based on Monte Carlo dropout (Gal and Ghahramani, 2016) to obtain more robust predictions. Data augmentation enabled by our proposed method yields strong performance. We compare different possible models and analyze the experimental results from different perspectives.

Paper Type: long

Research Area: Discourse and Pragmatics

Contribution Types: NLP engineering experiment

Languages Studied: English