Top-down Discourse Parsing via Sequence Labelling

EACL 2021  ·  Fajri Koto, Jey Han Lau, Timothy Baldwin ·

We introduce a top-down approach to discourse parsing that is conceptually simpler than its predecessors (Kobayashi et al., 2020; Zhang et al., 2020). By framing the task as a sequence labelling problem where the goal is to iteratively segment a document into individual discourse units, we are able to eliminate the decoder and reduce the search space for splitting points. We explore both traditional recurrent models and modern pre-trained transformer models for the task, and additionally introduce a novel dynamic oracle for top-down parsing. Based on the Full metric, our proposed LSTM model sets a new state-of-the-art for RST parsing.

PDF Abstract EACL 2021 PDF EACL 2021 Abstract

Datasets


Results from the Paper


Ranked #7 on Discourse Parsing on RST-DT (Standard Parseval (Span) metric)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Discourse Parsing RST-DT LSTM Dynamic Standard Parseval (Span) 73.1 # 7
Standard Parseval (Nuclearity) 62.3 # 10
Standard Parseval (Relation) 51.5 # 10
Standard Parseval (Full) 50.3 # 9
Discourse Parsing RST-DT LSTM Static Standard Parseval (Span) 72.7 # 9
Standard Parseval (Nuclearity) 61.7 # 12
Standard Parseval (Relation) 50.5 # 12
Standard Parseval (Full) 49.4 # 11
Discourse Parsing RST-DT Transformer (dynamic) Standard Parseval (Span) 70.2 # 14
Standard Parseval (Nuclearity) 60.1 # 14
Standard Parseval (Full) 49.2 # 12
Discourse Parsing RST-DT Transformer (static) Standard Parseval (Span) 70.6 # 13
Standard Parseval (Nuclearity) 59.9 # 15
Standard Parseval (Relation) 50.6 # 11
Standard Parseval (Full) 49.0 # 13

Methods