Event News

Lecture on informal workshop on Argument Mining


January 12 (Friday)




National Institute of Informatics
Room 1208


13:00-14:00, Paolo Torroni

Context-Independent Argument Mining

Argument mining is a booming area at the intersection of artificial intelligence, machine learning and computational linguistics. The wide-scale availability of natural language resources, together with recent technological advances, promise to unlock unforeseen applications and open new research perspectives. However, the majority of current approaches are highly domain-specific. The challenge here is to develop methods that can be smoothly adapted to several domains and different applications. In this talk, we discuss recent trends in argument mining, with a special emphasis on the corpora that are being developed for this task. After a brief review of the state of the art, we present our recent work on context-independent argument mining, where we show that the rhetorical structure of argumentation in natural language can provide effective cues to mining systems.

14:15-15:00, Naoya Inoue, Paul Reisert

Towards Capturing Implicit Reasoning of Argumentative Texts

Automatic understanding of implicit reasoning underlying argumentative texts has potential for applications such as automated essay grading and multi-document summarization. In this talk, we introduce our recent efforts towards implicit reasoning identification. First, we discuss our annotated corpus of argumentation templates, where we augment an existing, reliable corpus of argumentative texts with such implicit reasoning via predefined templates. We also discuss our first attempt of an automated model for implicit reasoning identification. In the second part of the talk, we discuss our ongoing work for extending abductive reasoning with distributed representations for implicit reasoning identification.

15:15-16:00, Vitor Castro

Detection of Argumentative Texts in the User-Generated Web

Argument mining is a relatively new, rapidly developing area inside of natural language processing. The automatic extraction of argument structure can be applied to fields as varied as legal precedent discovery, summarization, opinion mining and writing assistance. In this context, the user-generated web poses unique challenges, such as a relative lack of well-formed arguments, high linguistic variability, and the sheer amount of available data. One aspect of this research field that has been largely overlooked up until now is determining which texts in this huge quantity of documents are adequate for the mining of argument microstructure and subsequent processing, i.e. which texts are argumentative in nature. In this presentation, we will first show how this problem has been described in previous work, but not, to the best of our knowledge, tackled effectively just yet. Then, we will talk about how we intend to approach the problem by using claim detection information and other machine learning techniques to automatically classify documents as topic-relevant argumentative or not. We plan to evaluate our method by performing additional annotations on top of an existing corpus.

16:15-17:00, Gaku Morio, Katsuhide Fujita

Supporting Large-scale Online Discussions by Argument Mining

Online discussions have been attracting much attention in recent years. They allow people to hold discussions online, whereas these have traditionally been conducted face-to-face at town meetings, etc. However, in large-scale online discussions, some original problems such as groupthink and flaming happen. Therefore, supporting such problems with argument mining techniques is important research topics in large-scale online discussions. In our presentation, we present supporting applications of the large-scale discussions based on argument mining: competence estimation and automatic type classification of posts. In addition, our ongoing research progress to support the large-scale discussions by argumentation mining.

17:15-18:00, Katsumi Nitta

Toward Automated Scoring of Argumentation Skills

Every year, intercollegiate negotiation competition is held in Japan, and more than 40 teams participate in this competition. The score sheets consists of 30 criteria to evaluate the argumentation skills such as strategy of negotiation, constructive proposal of alternatives, effective discussion and so on. The tentative result of estimating the scores by argumentation records is presented.


Ken Satoh (ksatoh(at)nii.ac.jp)