Informal Workshop on Argumentation/Argument Mining
Date: November 14, 2018
Venue: National Institute of Informatics, Room No. 1901-1903
10:00-11:00 Douglas Walton (Distinguished Research Fellow), University of Windsor, Canada
Title: Tutorial on Argumentation
Argumentation is an interdisciplinary subject that is studied in such fields as artificial intelligence, multi-agent systems, computational linguistics, formal logic, informal logic, education, speech communication, medical communication and law. This tutorial surveys some argumentation theories and tools that can be applied to common kinds of tasks encountered in solving problems in these and other fields. The following theories and tools are included: types of dialogues, argumentation schemes, argument diagrams (maps), argument mining, computational systems for automated argument invention, and hybrid systems combining argument and explanation.
11:00-11:20 Coffee break
11:20-12:00 Oliver Ray, University of Bristol, UK
Title: Using Agreement Statements to Identify Majority Opinion in UKHL Case Law
This paper is concerned with the task of finding majority opinion (MO) in UK House of Lords (UKHL) case law by analysing agreement statements (AS) that explicitly express the appointed judges' acceptance of each other's reasoning. We introduce a corpus of 300 UKHL cases in which the relevant AS and MO have been annotated by three legal experts; and we introduce an AI system that automatically identifies this AS and MO with a performance comparable to humans. (Joint work with Josef Valvoda and Ken Satoh)
12:00-12:40 Gaku Morio, Ryo Egawa, Katsuhide Fujita, Tokyo Univ of Agriculture and Technology, Japan
Title: Annotating Inner- and Inter- Post Argumentation in ChangeMyView
We present our work-in-progress annotation project for ChangeMyView (https://www.reddit.com/r/changemyview/) in Reddit.
In the study, we focus on either inter- and inter- post perspectives in discussion threads.
First, in the inner-post perspective, we annotate token-level component boundaries, types and support/attack relations.
The component types contains "policy", "value", "rhetorical statement", "testimony" and "fact", where the policy and value propositions can be a claim and the others can be premises which support the claim.
Second, in the inter-post perspective, we annotate inter-post relationships between the components.
This annotation could capture the interaction of post-to-post argument.
In the presentation, we will talk about how our annotation is going on, and the experimental plan of future analysis of the annotated corpus.
12:40-13:40 Lunch Break
13:40-14:20 Paul Reisert, RIKEN Center for Advanced Intelligence Project, Japan
Title: Towards Improving Current Automatic Essay Scoring and Constructive Feedback Systems
Automatic essay scoring (AES) and constructive feedback (CF) are two important applications in the argumentation mining community. AES allows for an essay to instantly be assessed, ultimately reducing the time spent on manually grading the essay. Simultaneously, CF enables writers to instantly learn about the weaknesses in their argument and improve it accordingly. In this talk, we will discuss our ongoing efforts for improving current AES and CF systems. The talk will be spread out amongst three main efforts which include i) counter-argument generation, ii) improved modeling of student essay organization score, and iii) incorporating background knowledge for warrant identification.
14:20-15:00 Jan Wira Gotama Putra, Tokyo Institute of Technology, Japan
Title: Annotating Argumentative Relations for Mining Coherence Patterns
Coherence is a crucial aspect of text as it governs how to arrange sentences properly. Past researchers have indicated there are regularities between coherent and incoherent text. This presentation highlights our ongoing research towards a deeper understanding of text coherence by annotating argumentative relations in argumentative essays. On top of annotating relations, sentences in each essay are also rearranged such that it results in a well-formed text. Since student essays are imperfect, they fit our purpose well because we can compare the relational patterns between incoherent and coherent (rearranged) texts. This research benefits natural language generation systems, e.g., as multi-document summarization, and may be applied in intelligent language tutoring systems.
15:00-15:40 Hiroaki Yamada, Tokyo Institute of Technology, Japan
Issue Topic based Argumentative Structure Extraction from Japanese Judgment Documents
The legal professionals rely heavily on the judgment document which is a direct output from court trials. However, there are far too many relevant documents, and each document is long and linguistically complex. Therefore, there is a pressing need to generate well-formed summaries of judgment documents.
Our main observation is that the structure of the legal argument can guide summarization. In the case of Japanese judgment documents, the structure demonstrably exists and is exceptionally well- formed since the writers of the documents, who are judges, consistently follow the principle of legal arguments. The structure is based around the so-called “Issue Topic,” a legal concept corresponding to the pre-defined main points to be discussed in a particular court case. Each Issue Topic has a conclusion part, which contains supporting arguments concerning this Issue Topic. We present an annotation scheme capturing this structure, and a corpus of Japanese civil law judgment documents manually annotated with this scheme. The corpus contains 89 documents and their summaries. We show with an agreement study that our scheme is stable, and we also designed and implemented the first two stages of an algorithm for the automatic extraction of argument structure.
15:40-16:00 Coffee Break
16:00-16:40 Juliano Rabelo, University of Alberta, Canada
Title: A discussion on textual entailment techniques applied to argument mining
Argument mining is the area of natural language processing concerned with the automatic recognition of argumentative structures from natural language text. COLIEE 2018
did not include a direct argument mining task, but included the related task of case law entailment identification. That task consists in automatically determining, given a base case b and a related case r, what paragraphs from r entail the decision of b. In this talk, I will present details of the model which was ranked first place among all competitors in COLIEE 2018 Case Law Entailment Task, and discuss the possible application of similar principles to an argument mining scenario.
16:40-17:20 Cesare Bartolini, University of Luxembourg, Luxembourg
Title: Pitfalls in legal argumentation: where interpretation overrides the legal text
Modern philosophy of law argues that a legal text is not a law, but it must first pass through the eyes of the interpreter. Normally, the various interpretations of a legal text, and thus the legal meanings given to it, do not stray too far off the text itself. Occasionally, however, interpreters give a legal text a meaning strongly out of the ordinary. These may be glaring interpretation errors, but sometimes they are the result of a profound knowledge of the legal system, and grounded on a solid reasoning basis, and as such they are eligible to become "the law". As these meanings can be drastically different from the apparent ones, they can prove a challenge for tools designed to extract the meaning from the legal text. Such tools would need to rely on additional sources, such as court decisions or doctrinal studies, or simply be corrected by human hands. This talk discusses a few notable examples, taken from the Italian civil code, where legal interpretation has given an unexpected meaning to the provision, either altering its content, or shifting a duty to a bearer not mentioned in the legal text, or endowing subjects with non-apparent rights or obligations.
17:20-18:00 Tiago Oliveira, NII, Japan
Title: A Computational Argumentation Framework for Decision-Making in Multimorbidity
Multimorbidity is defined as the presence of two or more chronic medical conditions in an individual, presenting problems in care, particularly when the number of existing conditions is high and there are treatment conflicts. It poses challenges, namely in the use of multiple drugs to treat all existing conditions (drug-drug interactions) and the effects their combination may have on the patient's body or by influencing the evolution of other health conditions (drug-disease interactions). The formalization of multimorbidity decisions serves two purposes: to support the stakeholders in choosing which treatment to apply and to identify the reasons behind decisions. We investigate the use of computational argumentation to both analyze and generate decisions in multimorbidity about consistent recommendations, according to the different goals of stakeholders. Decision-making in this setting carries a complexity related with the multiple variables involved. These variables reflect the concomitant health conditions that should be considered when defining a proper therapy. However, current Clinical Decision Support Systems (CDSSs) are not equipped to deal with such a situation. They do not go beyond the straightforward application of the rules that build their knowledge base and simple interpretation of Computer-Interpretable Guidelines (CIGs). We provide a computational argumentation system equipped with goal seeking mechanisms to combine independently generated recommendations, then identify and discuss its advantages over multiple-criteria decision analysis in this setting.
18:30 Informal Dinner