Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Negation Triplet Extraction with Syntactic Dependency and Semantic Consistency

Yuchen Shi, Deqing Yang, Jingping Liu, Yanghua Xiao, Zongyu Wang, Huimin Xu

TL;DR

This work introduces Negation Triplet Extraction (NTE), a task to extract <subject, cue, scope> triplets from sentences, addressing the missing negation subject in prior negation understanding. It proposes SSENE, an Encoder-Decoder PLM with a syntax-aware encoder and a semantic-consistency auxiliary task, optimizing with a combined loss $\mathcal{L}=\mathcal{L}_1+\alpha\mathcal{L}_2$ where $0<\alpha<1$, and encodes syntactic structure via a dependency-attention mechanism and matrix $\mathbf{M}$. A high-quality Chinese dataset NegComment is constructed from Meituan reviews to evaluate NTE, and extensive experiments show SSENE outperforms baselines in NTE and remains competitive on CDSR, with ablations confirming the benefits of both syntactic information and semantic alignment. The model's ability to leverage distant dependencies and enforce semantic consistency has practical implications for downstream tasks like search, sentiment analysis, and reasoning over negated attributes. The work also demonstrates generalizability to CDSR and contributes a new dataset to advance negation understanding in real-world Chinese text.

Abstract

Previous works of negation understanding mainly focus on negation cue detection and scope resolution, without identifying negation subject which is also significant to the downstream tasks. In this paper, we propose a new negation triplet extraction (NTE) task which aims to extract negation subject along with negation cue and scope. To achieve NTE, we devise a novel Syntax&Semantic-Enhanced Negation Extraction model, namely SSENE, which is built based on a generative pretrained language model (PLM) {of Encoder-Decoder architecture} with a multi-task learning framework. Specifically, the given sentence's syntactic dependency tree is incorporated into the PLM's encoder to discover the correlations between the negation subject, cue and scope. Moreover, the semantic consistency between the sentence and the extracted triplet is ensured by an auxiliary task learning. Furthermore, we have constructed a high-quality Chinese dataset NegComment based on the users' reviews from the real-world platform of Meituan, upon which our evaluations show that SSENE achieves the best NTE performance compared to the baselines. Our ablation and case studies also demonstrate that incorporating the syntactic information helps the PLM's recognize the distant dependency between the subject and cue, and the auxiliary task learning is helpful to extract the negation triplets with more semantic consistency.

Negation Triplet Extraction with Syntactic Dependency and Semantic Consistency

TL;DR

This work introduces Negation Triplet Extraction (NTE), a task to extract <subject, cue, scope> triplets from sentences, addressing the missing negation subject in prior negation understanding. It proposes SSENE, an Encoder-Decoder PLM with a syntax-aware encoder and a semantic-consistency auxiliary task, optimizing with a combined loss where , and encodes syntactic structure via a dependency-attention mechanism and matrix . A high-quality Chinese dataset NegComment is constructed from Meituan reviews to evaluate NTE, and extensive experiments show SSENE outperforms baselines in NTE and remains competitive on CDSR, with ablations confirming the benefits of both syntactic information and semantic alignment. The model's ability to leverage distant dependencies and enforce semantic consistency has practical implications for downstream tasks like search, sentiment analysis, and reasoning over negated attributes. The work also demonstrates generalizability to CDSR and contributes a new dataset to advance negation understanding in real-world Chinese text.

Abstract

Previous works of negation understanding mainly focus on negation cue detection and scope resolution, without identifying negation subject which is also significant to the downstream tasks. In this paper, we propose a new negation triplet extraction (NTE) task which aims to extract negation subject along with negation cue and scope. To achieve NTE, we devise a novel Syntax&Semantic-Enhanced Negation Extraction model, namely SSENE, which is built based on a generative pretrained language model (PLM) {of Encoder-Decoder architecture} with a multi-task learning framework. Specifically, the given sentence's syntactic dependency tree is incorporated into the PLM's encoder to discover the correlations between the negation subject, cue and scope. Moreover, the semantic consistency between the sentence and the extracted triplet is ensured by an auxiliary task learning. Furthermore, we have constructed a high-quality Chinese dataset NegComment based on the users' reviews from the real-world platform of Meituan, upon which our evaluations show that SSENE achieves the best NTE performance compared to the baselines. Our ablation and case studies also demonstrate that incorporating the syntactic information helps the PLM's recognize the distant dependency between the subject and cue, and the auxiliary task learning is helpful to extract the negation triplets with more semantic consistency.
Paper Structure (21 sections, 6 equations, 3 figures, 8 tables)

This paper contains 21 sections, 6 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Task Formalization
  4. Model Overview
  5. Multi-task Learning Framework
  6. Syntax-aware Encoder
  7. Experiments
  8. Dataset Construction
  9. Baselines
  10. Implementation Setup
  11. Overall Performance Comparison
  12. Ablation Study
  13. Justifying the Way of Incorporating Syntactic Information
  14. Justifying the Impact of Syntactic Dependency Attention
  15. Case Study
  16. ...and 6 more sections

Figures (3)

  • Figure 1: An examples of negation triplet extraction with syntactic dependency tree (better viewed in color). The negation subject, cue and scope are marked with blue, orange and green, respectively.
  • Figure 2: Subfigure (b) displays the overall framework of SSENE, in which the encoder $\mathbf{E}_{dep}$ is depicted in Subfigure (a).
  • Figure 3: For the sentence "酒店有高档的配套设施，然而却不能多提供一个枕头。|The hotel has luxurious facilities, yet cannot offer one more pillow.", Subfigure (a) and (b) display the average attention distribution of the first layer in SSENE's Encoder before and after incorporating the syntactic association matrix $\mathbf{M}$, respectively.