Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploration of Marker-Based Approaches in Argument Mining through Augmented Natural Language

Nilmadhab Das, Vishal Choudhary, V. Vijaya Saradhi, Ashish Anand

TL;DR

The paper reframes End-to-End Argument Mining as a generation task by encoding ACs and ARs into label-augmented text using Augmented Natural Language (ANL). It introduces argTANL, a TANL-based framework, and augments it with Marker-Based Fine-Tuning (MKT) strategies and ME-argTANL, which embeds marker information directly in the output to jointly identify markers and argumentative structures. Across three AM benchmarks, ME-argTANL achieves strong component classification performance, with marker-driven fine-tuning offering notable gains in several datasets, though some datasets show marker signals less aligned with their argumentative structure. The study also provides ablation analyses (including Abbreviated argTANL) and error analyses, highlighting practical trade-offs between output length, interpretability, and accuracy, and outlining directions for incorporating domain knowledge into generative AM models.

Abstract

Argument Mining (AM) involves identifying and extracting Argumentative Components (ACs) and their corresponding Argumentative Relations (ARs). Most of the prior works have broken down these tasks into multiple sub-tasks. Existing end-to-end setups primarily use the dependency parsing approach. This work introduces a generative paradigm-based end-to-end framework argTANL. argTANL frames the argumentative structures into label-augmented text, called Augmented Natural Language (ANL). This framework jointly extracts both ACs and ARs from a given argumentative text. Additionally, this study explores the impact of Argumentative and Discourse markers on enhancing the model's performance within the proposed framework. Two distinct frameworks, Marker-Enhanced argTANL (ME-argTANL) and argTANL with specialized Marker-Based Fine-Tuning, are proposed to achieve this. Extensive experiments are conducted on three standard AM benchmarks to demonstrate the superior performance of the ME-argTANL.

Exploration of Marker-Based Approaches in Argument Mining through Augmented Natural Language

TL;DR

The paper reframes End-to-End Argument Mining as a generation task by encoding ACs and ARs into label-augmented text using Augmented Natural Language (ANL). It introduces argTANL, a TANL-based framework, and augments it with Marker-Based Fine-Tuning (MKT) strategies and ME-argTANL, which embeds marker information directly in the output to jointly identify markers and argumentative structures. Across three AM benchmarks, ME-argTANL achieves strong component classification performance, with marker-driven fine-tuning offering notable gains in several datasets, though some datasets show marker signals less aligned with their argumentative structure. The study also provides ablation analyses (including Abbreviated argTANL) and error analyses, highlighting practical trade-offs between output length, interpretability, and accuracy, and outlining directions for incorporating domain knowledge into generative AM models.

Abstract

Argument Mining (AM) involves identifying and extracting Argumentative Components (ACs) and their corresponding Argumentative Relations (ARs). Most of the prior works have broken down these tasks into multiple sub-tasks. Existing end-to-end setups primarily use the dependency parsing approach. This work introduces a generative paradigm-based end-to-end framework argTANL. argTANL frames the argumentative structures into label-augmented text, called Augmented Natural Language (ANL). This framework jointly extracts both ACs and ARs from a given argumentative text. Additionally, this study explores the impact of Argumentative and Discourse markers on enhancing the model's performance within the proposed framework. Two distinct frameworks, Marker-Enhanced argTANL (ME-argTANL) and argTANL with specialized Marker-Based Fine-Tuning, are proposed to achieve this. Extensive experiments are conducted on three standard AM benchmarks to demonstrate the superior performance of the ME-argTANL.
Paper Structure (25 sections, 5 figures, 5 tables)

This paper contains 25 sections, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Task Formulation
  4. ACE task: Component-only Variant
  5. ACRE task: End-To-End Argument Mining
  6. Methodology
  7. The argTANL Framework
  8. Argumentative Marker Extraction
  9. Marker-Based Fine-Tuning of argTANL
  10. Augmented Marker Knowledge Transfer (A-MKT)
  11. Span-Masked Marker Knowledge Transfer (SM-MKT)
  12. Marker Knowledge Transfer through Encoding (E-MKT)
  13. Discourse Marker Knowledge Transfer (D-MKT)
  14. The ME-argTANL Framework
  15. Experimental Setup
  16. ...and 10 more sections

Figures (5)

  • Figure 1: An overview of the proposed generative end-to-end argument mining framework argTANL. Claims are highlighted in italics and marked in red. Premises are underlined and marked in blue. Augmented labels are represented with bold font and marked in green.
  • Figure 2: An example sentence from AAE corpus describing both ways of marker extraction. Here, extracted marker candidates are in bold italics, and ACs are highlighted.
  • Figure 3: Description of ME-argTANL Formulation. Markers are enclosed with the "(( ))" symbol in violet color. Claims are highlighted in red italics. Premises are underlined in blue. Augmented labels are represented in green with bold font.
  • Figure 4: Performance on the ACE task of the End-to-End variant with the varying number of ADUs (left) and of sentences (right) in a paragraph.
  • Figure 5: Representation of Abbreviated argTANL with "ID" tokens replacing repeating spans for a shorter ANL output format.