Differential equation based on cost for recursive argumentation graph (DECRAG): Continuous framework for recursive bipolar argumentation with similarity

In this paper, we present a new method to get the value of claims and relations of an argumentation graph. By the method, the values of claims and relations are determined by the average of the solutions of the differential equation derived from the argumentation graph. We show that the overhead of obtaining multiple solutions can be reduced by parallel execution and the values obtained by the method agree with our intuition. We can get a confidence interval of each value and a rough grouping of claims and relations by the method. We show the results for toy graphs, large artificial graphs, and a natural language argumentation graph.

Research on argumentation in Artificial Intelligence recently investigates new methods that contribute to the vision of developing robust argumentation machines. One line of research explores ways of reasoning with natural language arguments coming from information sources on the web as a foundation for the deliberation and synthesis of arguments in specific domains. This paper builds upon arguments represented as argument graphs in the standardized Argument Interchange Format. While previous work was focused on the development of semantic similarity measures used for the case-based retrieval of argument graphs, this paper addresses the problem of clustering argument graphs to explore structures that facilitate argumentation interpretation. We propose a k-medoid and an agglomerative clustering approach based on semantic similarity measures. We compare the clustering results based on a graph-based semantic measure that takes the structure of the argument into account with a semantic word2vec measure on the pure textual argument representation. Experiments based on the Microtext corpus show that the graph-based similarity is best on internal evaluation measures, while the pure textual measure performs very well for identifying topic-specific clusters.

Argumentation Machines search for arguments in natural language from information sources on the Web and reason with them on the knowledge level to actively support the deliberation and synthesis of arguments for a particular user query. The recap project is part of the Priority Program ratio and aims at novel contributions to and confluence of methods from information retrieval, knowledge representation, as well as case-based reasoning for the development of future argumentation machines. In this paper we summarise recent research results from the project. In particular, a new German corpus of 100 semantically annotated argument graphs from the domain of education politics has been created and is made available to the argumentation research community. Further, we discuss a comprehensive investigation in finding arguments and argument graphs. We introduce a probabilistic ranking framework for argument retrieval, i.e. for finding good premises for a designated claim. For finding argument graphs, we developed methods for case-based argument retrieval considering the graph structure of an argument together with textual and ontology-based similarity measures applied to claims, premises, and argument schemes.

Rule-based methods are often used to learn surrogates of black-box models within Explainable Artificial Intelligence. Decision trees, among others, are routinely used for such purposes and inherently possess more explainability. Unfortunately, they might be convoluted in large-scale scenarios, with large sizes and many branches, thus hampering such inherent property. They also fail at modelling contrastive information and conflictuality among rules. This research proposes a novel method based on computational argumentation that aims to solve such shortcomings of decision trees. In particular, it proposes a mechanism for automatically extracting rules from trained dense neural networks, the arguments. It then describes a procedure for automatically extracting their conflicts using the notion of attacks. Arguments and attacks are integrated into argumentation frameworks, which are directed graphs that can be used as surrogate models for explaining black boxes. The dialectical status of the arguments in such graphs can be evaluated with formal semantics and then aggregated toward a rational outcome corresponding to the target classes of the black-box models. Such graphs are empirically evaluated against eight objective metrics, including completeness, correctness, fidelity, robustness, number of rules, average rule length, fraction of classes and fraction overlap. They are also compared with the corresponding surrogate decision trees. Findings show how argumentation graphs are highly comparable to decision trees regarding explainability across selected objective metrics. However, it is potentially more appealing given that argumentation graphs offer richer justification and explanations by modelling rules’ conflictuality.

In many domains of public discourse such as arguments about public policy, there is an abundance of knowledge to store, query, and reason with. To use this knowledge, we must address two key general problems: first, the problem of the knowledge acquisition bottleneck between forms in which the knowledge is usually expressed, e.g., natural language, and forms which can be automatically processed; second, reasoning with the uncertainties and inconsistencies of the knowledge. Given such complexities, it is labour and knowledge intensive to conduct policy consultations, where participants contribute statements to the policy discourse. Yet, from such a consultation, we want to derive policy positions, where each position is a set of consistent statements, but where positions may be mutually inconsistent. To address these problems and support policy-making consultations, we consider recent automated techniques in natural language processing, instantiating arguments, and reasoning with the arguments in argumentation frameworks. We discuss application and “bridge” issues between these techniques, outlining a pipeline of technologies whereby: expressions in a controlled natural language are parsed and translated into a logic (a literals and rules knowledge base), from which we generate instantiated arguments and their relationships using a logic-based formalism (an argument knowledge base), which is then input to an implemented argumentation framework that calculates extensions of arguments (an argument extensions knowledge base), and finally, we extract consistent sets of expressions (policy positions). The paper reports progress towards reasoning with web-based, distributed, collaborative, incomplete, and inconsistent knowledge bases expressed in natural language.

This paper combines three computational argumentation systems to model the sequence of argumentation in a famous murder trial and the appeal procedure that followed. The paper shows how the argumentation scheme for argument from expert opinion can be built into a testing procedure whereby an argument graph is used to interpret, analyze and evaluate evidence-based natural language argumentation of the kind found in a trial. It is shown how a computational argumentation system can do this by combining argument schemes with argumentation graphs. Frighteningly, it is also shown by this example that when there are potentially confusing conflicting arguments from expert opinion, a jury can only too easily accept a conclusion prematurely before considering critical questions that need to be asked.

When Expert Opinion Evidence Goes Wrong

In the symbolic artificial intelligence community, abstract argumentation with its semantics, i.e. approaches for defining sets of valid conclusions (extensions) that can be derived from argumentation graphs, is considered a promising method for non-monotonic reasoning. However, from a sequential perspective, abstract argumentation-based decision-making processes typically do not guarantee an alignment with common formal notions to assess consistency; in particular, abstract argumentation can, in itself, not enforce the satisfaction of relational principles such as reference independence (based on a key principle of microeconomic theory) and cautious monotony. In this paper, we address this issue by introducing different approaches to ensuring reference independence and cautious monotony in sequential argumentation: a reductionist, an expansionist, and an extension-selecting approach. The first two approaches are generically applicable, but may require comprehensive changes to the corresponding argumentation framework. In contrast, the latter approach guarantees that an extension of the corresponding argumentation framework can be selected to satisfy the relational principle by requiring that the used argumentation semantics is weakly reference independent or weakly cautiously monotonous, respectively, and also satisfies some additional straightforward principles. To highlight the relevance of the approach, we illustrate how the extension-selecting approach to reference independent argumentation can be applied to model (boundedly) rational economic decision-making.

In abstract argumentation, multiple argumentation semantics have been proposed that allow to select sets of jointly acceptable arguments from a given argumentation framework, i.e. based only on the attack relation between arguments. The existence of multiple argumentation semantics raises the question which of these semantics predicts best how humans evaluate arguments. Previous empirical cognitive studies that have tested how humans evaluate sets of arguments depending on the attack relation between them have been limited to a small set of very simple argumentation frameworks, so that some semantics studied in the literature could not be meaningfully distinguished by these studies. In this paper we report on an empirical cognitive study that overcomes these limitations by taking into consideration twelve argumentation frameworks of three to eight arguments each. These argumentation frameworks were mostly more complex than the argumentation frameworks considered in previous studies. All twelve argumentation framework were systematically instantiated with natural language arguments based on a certain fictional scenario, and participants were shown both the natural language arguments and a graphical depiction of the attack relation between them. Our data shows that grounded and CF2 semantics were the best predictors of human argument evaluation. A detailed analysis revealed that part of the participants chose a cognitively simpler strategy that is predicted very well by grounded semantics, while another part of the participants chose a cognitively more demanding strategy that is mostly predicted well by CF2 semantics.

We develop a notion of explanations for acceptance of arguments in an abstract argumentation framework. To this end we show that extensions returned by Dung's standard semantics can be decomposed into i) non-deterministic choices made on even cycles of the given argumentation graph and then ii) deterministic iteration of the so-called characteristic function. Naturally, the choice made in i) can be viewed as an explanation for the corresponding extension and thus the arguments it contains. We proceed to propose desirable criteria a reasonable notion of an explanation should satisfy. We present an exhaustive study of the newly introduced notion w.r.t. these criteria. Finally some interesting decision problems arise from our analysis and we examine their computational complexity, obtaining some surprising tractability results.

We introduce DAGGER: a generator for logic based argumentation frameworks instantiated from inconsistent knowledge bases expressed using Datalog+-. The tool allows to import a knowledge base in DLGP format and the generation and visualisation of the corresponding argumentation graph. Furthermore, the argumentation framework can also be exported in the Aspartix format.

Bipolarity in argumentation graphs: Towards a better understanding

The automatic extraction of arguments from natural language texts is a highly researched area and more important than ever today, as it is nearly impossible to manually capture all arguments on a controversial topic in a reasonable amount of time. For testing different algorithms such as the retrieval of the best arguments, which are still in their infancy, gold standards must exist. An argument consists of a claim or standpoint that is supported or opposed by at least one premise. The generic term for a claim or premise is Argumentative Discourse Unit (ADU). The relationships between ADUs can be specified by argument schemes and can lead to large graphs. This paper presents a corpus of 100 argument graphs with about 2,500 ADUs in German, which is unique in its size and the utilisation of argument schemes. The corpus is built from natural language texts like party press releases and parliamentary motions on education policies in the German federal states. Each high-quality text is presented by an argument graph and created by the use of a modified version of the annotation tool OVA. The final argument graphs resulted by merging two previously independently annotated graphs based on detailed discussions.

Transparency is a critical concern to democratic societies. As software permeates our social lives, Software Transparency is becoming a quality criterion that demands more attention from software developers. We present in this paper our approach to capture transparency-related requirements patterns through argumentation. We represented initial transparency knowledge as requirements patterns. These initial patterns stimulated stakeholders' arguments on an open discussion about transparency of a given software. We apply an argumentation framework to capture the stakeholders' arguments on argumentation graphs. Transparency-related requirements patterns are identified by formally analyzing consensus on the argumentation graph. Basically, the identified patterns are possible operationalizations to transparency-related softgoals.

In order to demonstrate its understanding of an editorial, OpEd must be able to answer questions about the issues addressed in that editorial. Chapter 1 established that question answering in OpEd is characterized in terms of three processes: (1) understanding questions from the perspective of the editorial’s memory representation, or argument graph; (2) retrieving conceptual answers from the argument graph; and (3) generating conceptual answers in natural language. Furthermore, Chapter 1 indicated that question comprehension is performed by the same conceptual parser used for editorial comprehension, and answer generation is performed by a recursive-descent, English generator. This chapter examines the techniques used in the process of retrieving information from conceptual representations of editorials.KeywordsSteel IndustryMemory SearchNormal ProfitCausal BeliefREAGAN AdministrationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.