Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DMON: A Simple yet Effective Approach for Argument Structure Learning

Wei Sun, Mingxiao Li, Jingyuan Sun, Jesse Davis, Marie-Francine Moens

TL;DR

This work tackles argument structure learning (ASL) by modeling contextual relations among arguments with a relationship tensor. The proposed Dual-tower Multi-scale cOnvolution network (DMON) introduces a four-part design: pairwise argument encoding into a tensor ${\mathbf{H}}$, a cropping strategy to augment scarce labeled data, a bidirectional MSRM-based learning mechanism to capture contextual head/tail relations, and a confidence-based label fusion strategy. Empirical results on AbstRCT, SciDTB, and CDCP show consistent state-of-the-art gains across medical, scientific, and legal domains, with ablations confirming the importance of each component. The approach demonstrates that explicitly modeling contextual argument relations and bidirectional cues yields robust ASL performance and improves interpretability for discourse analysis tasks.

Abstract

Argument structure learning~(ASL) entails predicting relations between arguments. Because it can structure a document to facilitate its understanding, it has been widely applied in many fields~(medical, commercial, and scientific domains). Despite its broad utilization, ASL remains a challenging task because it involves examining the complex relationships between the sentences in a potentially unstructured discourse. To resolve this problem, we have developed a simple yet effective approach called Dual-tower Multi-scale cOnvolution neural Network~(DMON) for the ASL task. Specifically, we organize arguments into a relationship matrix that together with the argument embeddings forms a relationship tensor and design a mechanism to capture relations with contextual arguments. Experimental results on three different-domain argument mining datasets demonstrate that our framework outperforms state-of-the-art models. The code is available at https://github.com/VRCMF/DMON.git .

DMON: A Simple yet Effective Approach for Argument Structure Learning

TL;DR

This work tackles argument structure learning (ASL) by modeling contextual relations among arguments with a relationship tensor. The proposed Dual-tower Multi-scale cOnvolution network (DMON) introduces a four-part design: pairwise argument encoding into a tensor , a cropping strategy to augment scarce labeled data, a bidirectional MSRM-based learning mechanism to capture contextual head/tail relations, and a confidence-based label fusion strategy. Empirical results on AbstRCT, SciDTB, and CDCP show consistent state-of-the-art gains across medical, scientific, and legal domains, with ablations confirming the importance of each component. The approach demonstrates that explicitly modeling contextual argument relations and bidirectional cues yields robust ASL performance and improves interpretability for discourse analysis tasks.

Abstract

Argument structure learning~(ASL) entails predicting relations between arguments. Because it can structure a document to facilitate its understanding, it has been widely applied in many fields~(medical, commercial, and scientific domains). Despite its broad utilization, ASL remains a challenging task because it involves examining the complex relationships between the sentences in a potentially unstructured discourse. To resolve this problem, we have developed a simple yet effective approach called Dual-tower Multi-scale cOnvolution neural Network~(DMON) for the ASL task. Specifically, we organize arguments into a relationship matrix that together with the argument embeddings forms a relationship tensor and design a mechanism to capture relations with contextual arguments. Experimental results on three different-domain argument mining datasets demonstrate that our framework outperforms state-of-the-art models. The code is available at https://github.com/VRCMF/DMON.git .
Paper Structure (20 sections, 5 equations, 6 figures, 6 tables)

This paper contains 20 sections, 5 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Argument Structure Learning
  4. Structured Learning in NLP
  5. Method
  6. Pairwise Arguments Representation
  7. Cropping Strategy
  8. Bidirectional Learning Mechanism
  9. Label Fusion
  10. Experiments
  11. Datasets
  12. Experimental Set-up
  13. Baselines
  14. Results and Discussion
  15. Ablation Study and Analysis of Results
  16. ...and 5 more sections

Figures (6)

  • Figure 1: A simple example of an argumentative structure showing attack (orange arrow) and support (green arrow) relationships.
  • Figure 2: We select argument $C$ as the observation object. This example shows the correlation between the head (red) and tail (blue) relationship information and a relationship tensor. Each element in this relationship tensor is the concatenation of two arguments.
  • Figure 3: The structure of Multi-scale Residual Convolution Neural Network (DMON) during training. During testing, no cropping mechanism is used. In this example, the yellow and green cells in a prediction matrix correspond to attack and support relations.)
  • Figure 4: The structure of the multi-scale residual module (MSRM). k1, k2, k3 and k4 represent the kernel size of the respective 1D convolution layer.
  • Figure 5: Macro-F1 scores when changing window size of contextual arguments.
  • ...and 1 more figures