29th European Summer School in Logic, Language, and Information
University of Toulouse (France), 17-28 July, 2017

Foundations of argumentation for argument mining

Laura Alonso Alemany, Patrick Saint-Dizier

Language and Computation (Foundational)

First week, from 9:00 to 10:30


This course aims to provide an introduction to the fundamental concepts in the area of argumentation mining. This is an area with strong background in linguistics and in artificial intelligence, with clearly applied goals. In this course we aim to provide students with an insight of the phenomena that are found in argumentative texts and dialogues, their complexity and their characteristics. Then, we describe different approaches to formalizing these phenomena and the argumentation process in general. Finally, we present different applications of argumentation mining, in the domains of legal texts, debating or mining scientific literature, and we show different approaches to bridge the gap between theory and application. We devote a special emphasis to the manual annotation of argumentative texts for different purposes, for students to narrow down abstract concepts and to better understand the idiosyncrasy of argumentative phenomena, all of this with applied goals in sight.


Course Outline

1.Linguistic and conceptual foundations of argumentation

The aim of this first class is to introduce and gradually define argumentation, informally, and from a conceptual, pragmatic and linguistic perspective. Formal features are developed in the following class.

1.1.What is argumentation?

- Where do we have argumentation? what purpose does it serve? what is an argument, how to construct an argumentation? from debates to argumentation. Standpoints and argumentation.

- Domains or applications where argumentation is prominent: law, science, debates, politics, user-generated contents such as opinion expression,

- General form of arguments, general concepts and terminology,

- Argumentation vs. explanation, persuasion, rational argumentation, ethos vs. pathos.

1.2.Argumentation in language and communication

- Analytic, dialectic and rhetoric, controversies and discussion

- The language of argumentation: linguistic cues, syntactic aspects of interest for NLP,

- Argumentation and discourse structure (RST and others),

- Argumentation and rhetoric: planning and strategy issues, figures of speech, fallacies, etc.

- Argumentation and the listener, written vs. oral argumentation,

- From Aristotle to Perelman: the most relevant concepts that have been put forward (through time) to capture and instrumentalize the intuition behind argumentation and rhetoric,

- A typology of argumentation schemes in language, inventories of strategies for argumentation (prototypically, Walton, Cicero’s)

2.Formal and computational models for argumentation

The goal of this class is to a basic understanding of the foundational aspects of argumentation.

2.1.Formal aspects

- Informal logic and critical thinking, analysis of refutation,

- The structure of argumentation, simple and complex arguments, unexpressed premises,

- Basic concepts: supports, attacks, counter-attacks and the real situations, bipolar systems, agreement and disagreement,

- Causality and argumentation,

- The soundness of argumentation, forms of evaluations of arguments

- Value systems,

- Theories of argument structure (Toulmin and followers), macro-structure of arguments, Van Eemeren approach, formal dialectics.

2.2.Computational representations

- Reconstructing argumentative discourse: Formal representations of arguments and argumentation.

- Genre specific aspects of argumentation

- Situation of argument mining and the foundational problems raised: argument and relation detection.


It is important to show that argumentation mining should interface between information extraction, discourse parsing and argumentation theory, while argumentation theory assumes that arguments and their relations have somehow been identified.

At the end of this class we will have built a glossary and definitions of the relevant concepts that we will be using for models and formalization of argumentation.

3.Corpus compilation and annotation

The goal of this class is to provide students with an overview of existing efforts to manually annotate naturally occurring data with argumentative analyses, specially to aid in machine learning approaches to argumentation mining. A hands-on exercise will be provided for students to put to practice the difficulties described in theory.

3.1.Methodology for (argumentative) corpus annotation

- The role of guidelines for human annotators, the process of guideline development (annotation-abstraction-putting them in common cycle), overview of some annotation guidelines for argumentative corpora. Typology of methods to ensure coherence between annotations: formal tests, intuition, textual evidence?

- The support from other linguistic analyses for argumentation: discourse structures, illocutionary analyses. Useful aspects (or attributes), interesting aspects.

- Metrics for inter-annotator agreement, metrics ruling out chance agreement.

- An overview of available tools for corpus annotation.

3.2.An overview of annotated corpora

An overview of existing annotated corpora will be provided, describing the kind of corpus, the theoretical framework underlying each annotation effort, and discussing some controversial examples. We will focus in illustrating how theoretical concepts are actually implemented to naturally occurring examples, how these annotated corpora may be useful for a variety of tasks, and possibilities and difficulties to capture some of the phenomena in a machine learning approach. The class in day 3 ends with a presentation of some interesting examples to annotate and to be discussed the following day, giving some time during the class in day 4 to discuss in groups.

4.Argument analysis, argumentation construction

The goal of this class is to provide a high-level overview of argumentation mining systems.

4.1.Functional architecture of an argument mining system

- Preprocessing, to provide some additional information for the argumentative analysis: finding syntactical, semantical, and document structure information.

- Identifying Argumentative Discourse Units (ADUs) in text. Approaches using shallow cues, exploiting syntax and exploiting discourse parsing.

- Identifying Relations between ADUs. Shallow approaches, deeper approaches. The use of cues. Hedging in argumentation. The need to make implicit ADUs explicit and how to overcome it.

- Applying NLP techniques developed for other uses to argumentation mining: clause splitting for ADU identification, sentiment or textual entailment for relation identification.

4.2.Feeding the results of analyses to a human or application

- Construction of an argument synthesis: clusters of related arguments, graphs

- Question answering over argument graphs,

- A set of relevant arguments for a given argumentation scheme in progress…

- Ranking arguments by strength or persuasiveness

- An overview of applications: opinions, debates, scientific corpora (in contrast to social media), teaching argumentation and didactics


We will finish the course by discussing the presented concepts, implementations, applications and difficulties, and outlining interesting areas of work and open questions.

Expected level and prerequisites

As a foundational course, no prerequisites are expected.