Abstract:Due to its great importance in deep natural language understanding and various down-stream applications, text-level parsing of discourse rhetorical structure (DRS) has been drawing more and more attention in recent years. However, all the previous studies on text-level discourse parsing adopt bottom-up approaches, which much limit the DRS determination on local information and fail to well benefit from global information of the overall discourse. In this paper, we justify from both computational and perceptive points-of-view that the top-down architecture is more suitable for text-level DRS parsing. On the basis, we propose a top-down neural architecture toward text-level DRS parsing. In particular, we cast discourse parsing as a recursive split point ranking task, where a split point is classified to different levels according to its rank and the elementary discourse units (EDUs) associated with it are arranged accordingly. In this way, we can determine the complete DRS as a hierarchical tree structure via an encoder-decoder with an internal stack. Experimentation on both the English RST-DT corpus and the Chinese CDTB corpus shows the great effectiveness of our proposed top-down approach towards text-level DRS parsing.

Improved Discourse Parsing with Two-Step Neural Transition-Based Model

Transition-Based Discourse Parsing with Multilayer Stack Long Short Term Memory

Recursive Deep Models for Discourse Parsing

Modeling Discourse Cohesion for Discourse Parsing Via Memory Network.

Improving Feature-Rich Transition-Based Constituent Parsing Using Recurrent Neural Networks.

RST Discourse Parsing with Tree-Structured Neural Networks

Improve Discourse Dependency Parsing with Contextualized Representations

A Two-Stage Parsing Method for Text-Level Discourse Analysis

A Structure Self-Aware Model for Discourse Parsing on Multi-Party Dialogues

Transition-based Dependency Parsing Using Two Heterogeneous Gated Recursive Neural Networks.

A Neural Transition-Based Approach for Semantic Dependency Graph Parsing

Multi-level Cohesion Information Modeling for Better Written and Dialogue Discourse Parsing

RST Discourse Parsing with Second-Stage EDU-Level Pre-training

Improved Character-Based Chinese Dependency Parsing by Using Stack-Tree LSTM.

Shift-Reduce Constituent Parsing with Neural Lookahead Features.

A Neural Probabilistic Structured-Prediction Model for Transition-Based Dependency Parsing.

A Multi-stage Strategy for Chinese Discourse Tree Construction

A Top-Down Neural Architecture towards Text-Level Parsing of Discourse Rhetorical Structure

Discourse Representation Structure Parsing for Chinese

DRTS Parsing with Structure-Aware Encoding and Decoding

Neural Network Models for Implicit Discourse Relation Classification in English and Chinese without Surface Features