Program

Introduction to argumentation theory across disciplines: AI and law

The subject of the tutorial is the modeling of legal argumentation by means of formal and computational tools. We will investigate to what extent legal argumentation may be represented in an algorithmic manner and what are the current possibilities in this respect. The point of departure will be given by classical legal syllogism and then we will investigate how tools developed in the field of AI and law such as rule-based systems, argumentation frameworks, and argumentation schemes may be fruitfully applied to cover different aspects of legal reasoning. In the final part, we will consider the application of Machine Learning tools to the subject in question.

Introduction to argumentation theory across disciplines: Computer science and computational linguistics

In this tutorial, I will present recently rapidly growing area of argumentation in computer science and computational linguistics – argument mining. It is built upon text mining which provides techniques and methods for automated extraction of information from texts in natural language. The expansion of text mining is a response to the problem of Big Data, i.e. the problem of data being produced faster than we can process it.

More specifically – what can we mine? In sentiment analysis, we mine attitudes (positive, neutral, negative) towards something, e.g. we try to identify on Internet fora a number of people who like a new Mercedes vs. a number of people who does not like new Mercedes (application: stock market). In opinion mining, we mine people's opinions about products, e.g. people can think new Mercedes is too expensive or people can think new Mercedes is reliable (application: media analysis). The new area of argument mining allows for recognising not only what opinions people hold, but also why they hold them, e.g. people can think that new Mercedes is too expensive, because other cars from the similar class cost significantly less.

Introduction to argumentation theory across disciplines: Linguistics and psychology

This tutorial focuses on the psycholinguistic aspect of the studies on persuasion. The focal point of the talk is the concept of attitude defined as a global favourable, unfavourable or neutral evaluation of an attitude object (e.g. a social issue). The role of pathos (i.e. emotional reactions) in the generation and modification of an attitude is unravelled on the basis of the symbolic and ideological approach. The tutorial explains why an attitude system comprises beliefs, emotions, values, intentions to behave and describes the links between those elements. Persuasion models are presented that clarify how a conflict between internal attitude elements might enhance or hamper persuasive attempts. External aspects of an attitude change such as ethos (i.e. persuader’s credibility) and pathos (i.e. emotional message appeals) and their relation to internal cognitive elements of an attitude are also pursued. Direct and indirect attitude measurement techniques in psycholinguistic research are discussed. Basic principles of research design on attitude change are explained (e.g. controlling the confounding variables) and the implementation of a computerised experiment that measures attitudes in the E-Prime software (i.e. constructing an experiment, pilot testing, formal data collection) is shown.

Machine arguing: theories, systems and applications

Within AI, argumentation is about empowering machines with the capability to argue, so as to resolve conflicts, fill gaps in incomplete information and provide explanations for any outcomes obtained by machine arguing. In this tutorial I will overview a variety of existing theories for modeling argumentation in AI, including abstract, (some forms of) bipolar and (some forms of) structured argumentation frameworks, under a variety of so-called semantics, ranging from extension-based, labelling-based, and gradual semantics; I will also touch upon extensions of these frameworks with preferences and probabilities. I will then provide an overview and hands-on session on a number of argumentation systems, including abaplus, Arg&Dec and proxdd and abagraph (all available here). I will conclude with an illustration of a number of concrete applications of machine arguing, in medical, legal and social settings.

Realistic models of beliefs in a paraconsistent and paracomplete setting

In the presence of dynamically changing real-world data, humans, robots and software agents have to cope with imperfect information. This issue manifests itself in incomplete and/or inconsistent information which, in turn, seriously affects a variety of beliefs different types of dialogues and argumentation are based on. Heuristic techniques for completing missing beliefs or disambiguating inconsistencies are inherently nonmonotonic. Traditional approaches to beliefs and non-monotonic reasoning suffer from high complexity. Moreover, belief changes/revisions, often needed in the course of dialogues, make the complexity even worse.

During the lecture we shall discuss an alternative approach where we make three important shifts: (a) using paraconsistent, paracomplete and modular rather than classical reasoning; (b) querying belief bases rather than applying logical entailment; (c) using belief shadowing rather than belief updates/revisions. We will present a tractable framework for belief bases and belief shadowing, based on a doxastic extension of a four-valued, rule-based language 4QL. The language allows for paraconsistent, paracomplete and modular reasoning, and provides a simple to use tools for lightweight forms of nonmonotonic as reasoning with incomplete and/or inconsistent beliefs. The framework enjoys an open-source implementation and is ready for experimental and educational use.

The presented techniques will be illustrated by applications to speech acts specifications and their use in selected forms of dialogues. We will also indicate research directions which may be interesting for the Argdiap community.

Realistic models of beliefs in a paraconsistent and paracomplete setting

In the presence of dynamically changing real-world data, humans, robots and software agents have to cope with imperfect information. This issue manifests itself in incomplete and/or inconsistent information which, in turn, seriously affects a variety of beliefs different types of dialogues and argumentation are based on. Heuristic techniques for completing missing beliefs or disambiguating inconsistencies are inherently nonmonotonic. Traditional approaches to beliefs and non-monotonic reasoning suffer from high complexity. Moreover, belief changes/revisions, often needed in the course of dialogues, make the complexity even worse.

During the lecture we shall discuss an alternative approach where we make three important shifts: (a) using paraconsistent, paracomplete and modular rather than classical reasoning; (b) querying belief bases rather than applying logical entailment; (c) using belief shadowing rather than belief updates/revisions. We will present a tractable framework for belief bases and belief shadowing, based on a doxastic extension of a four-valued, rule-based language 4QL. The language allows for paraconsistent, paracomplete and modular reasoning, and provides a simple to use tools for lightweight forms of nonmonotonic as reasoning with incomplete and/or inconsistent beliefs. The framework enjoys an open-source implementation and is ready for experimental and educational use.

The presented techniques will be illustrated by applications to speech acts specifications and their use in selected forms of dialogues. We will also indicate research directions which may be interesting for the Argdiap community.

Modelling semantic negotiation

This tutorial will focus on several approaches to computational modelling of semantic negotiation. The main goal of such models is to provide a high-level, computational or algorithmic description of hypothetical mechanisms by which communicating agents can arrive at similar semantic representations. Apart from that, such models help us understand how various pressures (social, cognitive, environmental) might affect semantic negotiation and shape emerging meanings. We will look at models of meaning coordination between dyads (pairs of agents) and at population-level dynamics. Our toolbox will include methods and algorithms from diverse domains: Markov chains, computer simulations, reinforcement learning, evolutionary game theory, signalling games etc.

Argumentation in practice: negotiating the legal conditions of real estate contracts

Can the art of argumentation be useful in the practice of negotiating contracts? During the tutorial there will be presented two cases regarding the course of negotiations of business and legal conditions of two real estate transactions. In the first case, one party of the negotiations has a stronger position, in the other parties will have equal negotiating positions. Relations between business assumptions and the proposed legal conditions and their justification will be presented. During the tutorial, I will also show argumentative tactics in the selection of arguments and their types, presenting them in specific phases of negotiations, as well as examples of the use of defence tactics.

Argumentative Microtexts: A multi-layer and multi-purpose corpus for theoretical and applied studies of argumentation

In this tutorial, we first introduce an annotation scheme for the structure of argumentation in texts, and students will practice with annotating sample texts, which offer different kinds of difficulties. We then turn to the task of building these structures automatically and present two different technical approaches for doing so. Finally, we look "beyond" the bare tree structures that represent the argumentative relations (different kinds of support and attack) between units, and we consider additional layers of annotation that enrich the explanatory power: argumentation schemes, implicit assumptions, and certain semantic aspects of the textual units.

Argument in Human-Agent Teams

A key aim underpinning the development of computational models of argument is to build systems that are effective in support of human interaction and decision-making. Achieving this certainly depends upon an understanding of how people argue and how computational models reflect good argumentative reasoning. Equally, however, we must understand how tools based on these insights have impact on individual and group behaviour. In this tutorial, I will explore the potential for argumentation-based models in support of human-agent teams. Support both for dialogue and for reasoning will be considered, and we will have a hands-on session with a tool developed for intelligence analysis (CISpaces).

Evidential and Legal Reasoning in AI - the role of argumentation

There are many techniques in the broad field of Artificial Intelligence that focus on legal decision-making, legal reasoning and reasoning with legal evidence. For example, statistical algorithms have been proposed to predict decisions by judges in the European Court of Human Rights, and Bayesian Networks have been applied to reasoning with legal evidence. The question we will ask ourselves during the tutorial is: what is the role of argumentation? Are arguments necessary, and are computational techniques based on argumentation useful? We will explore these questions through hands-on exercises in which we will study and model the arguments in different legal cases.

Argumentative dialogues without dialogue types

Pragmatic theories model argumentation as dialogues among agents in which standpoints are advanced and challenged. These theories encourage us to shift attention from formalizations of (informal) logics to formalizations of social practices. One common approach assumes that dialogues come in types determined by goals, having rules that ensure the goals will be reached. This presentation lays out an alternative approach—one that does not invoke dialogue types, common goals or pre-established rules. According to the normative pragmatic program, the orderly exchange of arguments is an achievement of the agents themselves; it is designed. In particular, agents create local norms governing their interaction by taking on carefully tailored responsibilities for what they are saying. We will consider two paradigm cases: how experts give their utterances authoritative force, and how advocates create circumstances in which their proposals will receive serious consideration. In addition, we will test out the the normative pragmatic approach by applying it to ordinary argumentative discourse.

Introduction to R for argument mining

R is a language programming that has been gaining importance in the last years, especially in those fields different from computer science, such as linguistics, humanities or social sciences, because it is simple, powerful and free. In this tutorial, we first introduce the basic data types, programming structures and graphical resources in R oriented to natural language processing and argument mining. Next, we will practice with sample texts some basic tasks in argument mining (such as segmentation, searching of discourse indicators, etc.) and how the results can be presented using different kinds of visualisations.