Mining Argument Components in Essays at Different Levels

EasyChair Preprint 10521

Abstract

The research of arguments in student essays has long been the subject of automatic approaches to argument mining. The task has often been modeled as a sequence tagging problem where the text is ei- ther analyzed in its entirety by a transformer model or split into smaller homogeneous units, such as sentences or paragraphs. However, previous research has highlighted how the various text sections may have different functions, and how the position of specific argument components obeys precise structural dependency criteria. For this reason, we propose an approach exploiting such structural information: in this work we present a hybrid training approach that takes into account the specific structural part of the essays, in order to be able to mine different types of argu- ment components at different levels. Our hybrid approach achieved an improvement over essay-level and paragraph-level training, in particular in the extraction of some specific argument components.

Keyphrases: Natural Language Processing, argument mining, machine learning, transformers

@booklet{EasyChair:10521,
  author    = {Roberto Demaria and Davide Colla and Matteo Delsanto and Enrico Mensa and Enrico Pasini and Daniele P. Radicioni},
  title     = {Mining Argument Components in Essays at Different Levels},
  howpublished = {EasyChair Preprint 10521},
  year      = {EasyChair, 2023}}