AF-CBA uses case-based argumentation to justify classifier predictions by arguing about differences between cases. We extend the mechanism by modelling which differences can compensate for each other by constructing arguments using domain knowledge. This involves a secondary argumentation framework. To assist experts in defining the appropriate domain knowledge, we use a rule-based classifier for semi-automated knowledge induction. We use the resulting rule set to derive arguments and demonstrate this with an evaluation procedure.

In formal argumentation theory, multiple argumentation-based explanation methods have been formulated based on ideas from social and cognitive science. However, these have not yet been empirically validated. One such idea is that information in an explanation needs to be related; in argumentation-based explanations, this has been captured as there being an attack path between arguments. This study describes and empirically validates two types of relatedness, related admissibility and directly related admissibility. This was done by instructing participants to select arguments from an argumentation framework to explain another argument in this framework. These explanations selected by the participants were compared to argumentation-based explanations that use relatedness. We found that both forms of relatedness are cognitively plausible. This gives insight into how argumentation theory can be applied in the real world to provide explanations.

We discuss explanations for formal (abstract and structured) argumentation – the question whether and why a certain argument or claim can be accepted (or not) under various extension-based semantics. We introduce a flexible framework, which can act as the basis for many different types of explanations. For example, we can have simple or comprehensive explanations in terms of arguments for or against a claim, arguments that (indirectly) defend a claim, the evidence (knowledge base) that supports or is incompatible with a claim, and so on. We show how selection based on necessity and sufficiency can be captured in our basic framework and discuss a real-life application.

We define and study the notions of stability and relevance for precedent-based reasoning, focusing on Horty's result model of precedential constraint. According to this model, precedents constrain the possible outcomes for a focus case, which is a yet undecided case, where precedents and the focus case are compared on their characteristics (called dimensions). In this paper, we refer to the enforced outcome for the focus case as its justification status. In contrast to earlier work, we do not assume that all dimension values of the focus case have been established with certainty: rather, each dimension is assigned a set of possible values. We define a focus case as stable if its justification status is the same for every choice of the possible values. For focus cases that are not stable, we study the task of identifying relevance: which possible values should be excluded to make the focus case stable? We show how the tasks of identifying justification, stability and relevance can be exploited for human-in-the-loop decision support. Finally, we discuss the computational complexity of these tasks and provide efficient algorithms.

AF-CBA is an example-based approach to XAI that draws on the case-based argumentation tradition in AI & Law. It means to explain binary classifications made by an opaque machine-learning model by presenting an argument graph to the user, which represents an argument game about the classification of a case on the basis of precedents derived from labelled data used in the training phase of the classifier. We improve the robustness of this method by modifying it to better handle inconsistent labelling and evaluate an alternative setup that does not require access to the labelled data by using earlier predictions instead.

We extend the result model for precedent-based reasoning with incomplete case bases. In contrast to regular case bases, these consist of incomplete cases for which not all dimension values need to be specified, but rather each dimension is assigned a set of possible values. The outcome of cases then applies for each (combination of) the possible dimension values. Building on earlier proposed notions of justification and stability for incomplete focus cases, we introduce the notion of possible justification statuses, which are required to maintain consistency of the incomplete case base. We demonstrate how these theoretic notions can be applied in practice for human-in-the-loop decision support, discuss their computational complexity and provide efficient algorithms.

This paper presents ORLA (Online Reinforcement Learning Argumentation), a new approach for learning explainable symbolic argumentation models through direct exploration of the world. ORLA takes a set of expert arguments that promote some action in the world, and uses reinforcement learning to determine which of those arguments are the most effective for performing a task by maximizing a performance score. Thus, ORLA learns a preference ranking over the expert arguments such that the resulting value-based argumentation framework (VAF) can be used as a reasoning engine to select actions for performing the task. Although model-extraction methods exist that extract a VAF by mimicking the behavior of some non-symbolic model (e.g., a neural network), these extracted models are only approximations to their non-symbolic counterparts, which may result in both a performance loss and non-faithful explanations. Conversely, ORLA learns a VAF through direct interaction with the world (online learning), thus producing faithful explanations without sacrificing performance. This paper uses the Keepaway world as a case study and shows that models trained using ORLA not only perform better than those extracted from non-symbolic models but are also more robust. Moreover, ORLA is evaluated as a strategy discovery tool, finding a better solution than the expert strategy proposed by a related study.

In this paper, we present a learning-based approach to the symbolic reasoning problem of dynamic argumentation, where the knowledge about attacks between arguments is incomplete or evolving. Specifically, we employ deep reinforcement learning to learn which attack relations between arguments should be added or deleted in order to enforce the acceptability of (a set of) arguments. We show that our Graph Neural Network (GNN) architecture EGNN can learn a near optimal enforcement heuristic for all common argument-fixed enforcement problems, including problems for which no other (symbolic) solvers exist. We demonstrate that EGNN outperforms other GNN baselines and on enforcement problems with high computational complexity performs better than state-of-the-art symbolic solvers with respect to efficiency. Thus, we show our neuro-symbolic approach is able to learn heuristics without the expert knowledge of a human designer and offers a valid alternative to symbolic solvers. We publish our code at https://github.com/DennisCraandijk/DL-Abstract-Argumentation.

We explore the computational complexity of stability and relevance in incomplete argumentation frameworks (IAFs), abstract argumentation frameworks that encode qualitative uncertainty by distinguishing between certain and uncertain arguments and attacks. IAFs can be specified by, e.g., making uncertain arguments or attacks certain; the justification status of arguments in an IAF is determined on the basis of the certain arguments and attacks. An argument is stable if its justification status is the same in all specifications of the IAF. For arguments that are not stable in an IAF, the relevance problem is of interest: which uncertain arguments or attacks should be investigated for the argument to become stable? We redefine stability and define relevance for IAFs and study their complexity.

In this article, we discuss case decision predictors, algorithms which, given some features of a legal case predict the outcome of the case (i.e. the decision of the judge). We discuss whether, and if so how, such prediction algorithms can be used to support judges in their decision making process. We conclude that case decision predictors can only be useful in individual cases if they can give legal justifications for their predictions, and that only these legal justifications are what should matter for a judge.

There has recently been talk of algorithms that predict decisions in legal cases being used by the judiciary to improve the predictability and consistency of judicial decision making. We argue that their use may minimise the error rate of decisions in the long run, but that this would require not only major technical advances but also major changes in legal thinking about what is the most important objective of judicial decision-making: optimising individual justice in a particular case or reducing errors in the long run. We further argue that if algorithmic decision predictors give any useful information in individual cases to judges at all, this is not in its predictions but in its explanations.

Enforcement, adjusting an argumentation framework such that a certain set of arguments becomes acceptable, is an important research topic within the study of dynamic argumentation, but one that has been little studied for structured argumentation. In this paper we study enforcement in a general structured argumentation setting. In particular, we study conditions on the argumentation setting and the knowledge base that ensure (or prevent) the acceptability of sets of formulas for structured argumentation frameworks.

In this paper, we present the information graph (IG) formalism, which provides a precise account of the interplay between deductive and abductive inference and causal and evidential information, where ‘deduction’ is used for defeasible ‘forward’ inference. IGs formalise analyses performed by domain experts in the informal reasoning tools they are familiar with, such as mind maps used in crime analysis. Based on principles for reasoning with causal and evidential information given the evidence, we impose constraints on the inferences that may be performed with IGs. Our IG-formalism is intended to facilitate the construction of formal representations within AI systems by serving as an intermediary formalism between analyses performed using informal reasoning tools and formalisms that allow for formal evaluation. In this paper, we investigate the use of the IG-formalism as an intermediary formalism in facilitating Bayesian network (BN) graph construction. We propose a structured approach for automatically constructing from an IG a directed BN graph, together with qualitative constraints on the probability distribution represented by the BN. Moreover, we prove a number of formal properties of our approach and identify assumptions under which the construction of an initial BN graph can be fully automated.

We study the dynamic argumentation task of detecting stability: given a specific structured argumentation setting, can adding information change the acceptability status of some propositional formula? Detecting stability is not tractable for every input, but efficient computation is essential in practical applications. We present a sound approximation algorithm that recognises stability for many inputs in polynomial time and we discuss several of its properties. In particular, we show under which constraints on the input our algorithm is complete. The proposed algorithm is currently applied for fraud inquiry at the Dutch National Police - we provide an English demo version that also visualises the output of the algorithm.

In this paper, we propose the information graph (IG) formalism, which provides a precise account of the interplay between deductive and abductive inference and causal and evidential information. IGs formalise analyses performed by domain experts in the informal reasoning tools they are familiar with, such as mind maps used in crime analysis. Based on principles for reasoning with causal and evidential information given the evidence, we impose constraints on the inferences that may be performed with IGs. Moreover, we propose an argumentation formalism based on IGs that allows arguments to be formally evaluated.

In this paper, we present a learning-based approach to determining acceptance of arguments under several abstract argumentation semantics. More specifically, we propose an argumentation graph neural network (AGNN) that learns a message-passing algorithm to predict the likelihood of an argument being accepted. The experimental results demonstrate that the AGNN can almost perfectly predict the acceptability under different semantics and scales well for larger argumentation frameworks. Furthermore, analysing the behaviour of the message-passing algorithm shows that the AGNN learns to adhere to basic principles of argument semantics as identified in the literature, and can thus be trained to predict extensions under the different semantics – we show how the latter can be done for multi-extension semantics by using AGNNs to guide a basic search. We publish our code at https://github.com/DennisCraandijk/DL-Abstract-Argumentation.

Crime investigation and criminal intelligence analysis often rely on the notion of modus operandi. We propose modelling such MOs as story schemes and show how real-life terrorist incidents can be assigned to such schemes. We discuss several requirements of MO schemes and present an implementation in the form of an OWL ontology. This ontology is intended as a conceptual model to be used to support a sensemaking process, whereby crime intelligence analysts populate the ontology with instance data.

We propose an agent architecture for transparent human-in-the-loop classification. By combining dynamic argumentation with legal case-based reasoning, we create an agent that is able to explain its decisions at various levels of detail and adapts to new situations. It keeps the human analyst in the loop by presenting suggestions for corrections that may change the factors on which the current decision is based and by enabling the analyst to add new factors. We are currently implementing the agent for classification of fraudulent web shops at the Dutch Police.

Bayesian networks (BNs) are powerful tools that are increasingly being used by forensic and legal experts to reason about the uncertain conclusions that can be inferred from the evidence in a case. Although in BN construction it is good practice to document the model itself, the importance of documenting design decisions has received little attention. Such decisions, including the (possibly conflicting) reasons behind them, are important for legal experts to understand and accept probabilistic models of cases. Moreover, when disagreements arise between domain experts involved in the construction of BNs, there are no systematic means to resolve such disagreements. Therefore, we propose an approach that allows domain experts to explicitly express and capture their reasons pro and con modelling decisions using argumentation, and that resolves their disagreements as much as possible. Our approach is based on a case study, in which the argumentation structure of an actual disagreement between two forensic BN experts is analysed.

This paper briefly describes AIF-EL, an OWL2-EL compliant ontology for the Argument Interchange Format.

In this paper, we propose a structured approach for transforming legal arguments to a Bayesian network (BN) graph. Our approach automatically constructs a fully specified BN graph by exploiting causality information present in legal arguments. Moreover, we demonstrate that causality information in addition provides for constraining some of the probabilities involved. We show that for undercutting attacks it is necessary to distinguish between causal and evidential attacked inferences, which extends on a previously proposed solution to modelling undercutting attacks in BNs. We illustrate our approach by applying it to part of an actual legal case, namely the Sacco and Vanzetti legal case.

In this paper, we propose a way to derive constraints for a Bayesian Network from structured arguments. Argumentation and Bayesian networks can both be considered decision support techniques, but are typically used by experts with different backgrounds. Bayesian network experts have the mathematical skills to understand and construct such networks, but lack expertise in the application domain; domain experts may feel more comfortable with argumentation approaches. Our proposed method allows us to check Bayesian networks given arguments constructed for the same problem, and also allows for transforming arguments into a Bayesian network structure, thereby facilitating Bayesian network construction.

Contains fulltext : 161870.pdf (Publisher’s version ) (Open Access)

In the process of proof alternative stories that explain 'what happened' in a case are tested using arguments based on evidence. Building on the author's earlier hybrid theory, this paper presents a formal theory that combines causal stories and evidential arguments, further integrating the different types of reasoning in a framework for structured argumentation. This then allows for correct reasoning with causal and evidential rules, and further integrates arguments and stories by grounding them both in well-known dialectical argumentation semantics.

Stories and legal cases have much in common, but there are also differences. Both can be seen as a sequence of events, but in a legal case the facts and events are legally qualified. Moreover, the point of a story is usually implicit, whereas the outcome of a legal case is explicitly explained. Stories have been mainly used in AI and Law to explore the evidence presented in legal cases, but here we will explore the relationship on the assumption the facts of the case have already been established, and so include legal qualification and the decision. We illustrate our approach the well known wild animals and Popov v Hayashi cases.

We apply an existing formal framework for practical reasoning with arguments and evidence to the Goal-oriented Requirements Language (GRL), which is part of the User Requirements Notation (URN). This formal framework serves as a rationalization for elements in a GRL model: using attack relations between arguments we can automatically compute the acceptability status of elements in a GRL model, based on the acceptability status of their underlying arguments and the evidence. We integrate the formal framework into the GRL metamodel and we set out a research to further develop this framework.

The process of proof is one of inference to the best explanation, in which alternative scenarios are supported and attacked by arguments. This combination of scenarios and arguments was previously presented as a formal hybrid theory. In this paper, the aim is to further integrate scenarios and arguments by defining a notion of attack between alternative explanations. Thus, scenarios and arguments can be incorporated in the same dialectical framework.

In this paper, we show two important connections between computational models of narrative and computational models of argumentation. First, we show how argumentation techniques can be applied to enrich story understanding, especially where an understanding the story requires understanding of the motives of its characters. This also helps to explain how stories can themselves be seen as as arguments for a particular ordering on values within a value based argumentation framework. We illustrate our discussion using biblical parables, taking as our main example the parable of the Good Samaritan.

In this paper, we present the Dialogue Game Execution Platform (DGEP), which is able to process and execute any dialogue game specified in a general description language and build arguments and dialogue histories in the language of the AIF ontology. Thus, DGEP allows us to generalise techniques for generation and investigation of dialogues and, through a set of web services, connect a wide variety of multi-agent systems and human-computer interfaces for dialogue.

In this paper, we present a software tool for ‘ArguBlogging’, which allows users to construct debate and discussions across blogs, linking existing and new online resources to form distributed, structured conversations. Arguments and counterarguments can be posed by giving opinions on one’s own blog and replying to other bloggers’ posts. The resulting argument structure is connected to the Argument Web, in which argumentative structures are made semantically explicit and machine-processable. We discuss the ArguBlogging tool and the underlying infrastructure and ontology of the Argument Web.

The Argument Interchange Format (AIF) has been devised in order to support the interchange of ideas and data between different projects and applications in the area of computational argumentation. In order to support such interchange, an abstract ontology for argumentation is presented, which serves as an interlingua between various more concrete argumentation languages. In this article, we aim to give what is essentially a logical specification of the AIF ontology by mapping the ontology onto the logical ASPIC+ framework for argumentation. We thus lay foundations for interrelating formal logic-based approaches to argumentation captured by the ASPIC+ framework and the wider class of argumentation languages, including those that are more informal and user-orientated.

We have implemented a translator that translates Carneades Argument Graphs as specified in the LKIF files of the Carneades editor to a database specification of the Argument Interchange Format and vice versa. In this paper the algorithms for this translation are presented.

Dialogue systems attempt to capture structured communication with the aim of understanding, improving, and automatically recreating su communication. This paper discusses dialogue templates: blueprints that can be instantiated and combined to form argumentative dialogues. These templates provide a generic way of representing individual dialogue systems and allow us to generalise techniques for investigation, generation and recognition of dialogues.

In this paper, we present Arvina, an online discussion tool supporting mixed initiative argumentation. Arvina allows stored arguments in the Argument Web to be introduced by software agents which human participants can then interact with.

This paper introduces AIFdb, a database solution for the Argument Web. AIFdb offers an array of web service interfaces allowing a wide range of software to interact with the same argument data.

In this paper, we present ArguBlogging, a simple tool that allows blog users to directly respond to text on a web page, publishing the response to their blog while simultaneously capturing the argumentative structure in the Argument Web via the Argument Interchange Format.

In this paper, we continue our research on a hybrid narrative-argumentative approach to evidential reasoning in the law by showing the interaction between factual reasoning and legal reasoning. We therefore emphasize the role of legal story schemes (as opposed to factual story schemes that formed the heart of our previous proposal). Legal story schemes steer what needs to be proven, but are also selected on the basis of what can be proven. They provide a coherent, holistic legal perspective on a criminal case that steers investigation and decision making. We present an extension of our previously proposed hybrid theory of reasoning with evidence, by making the connection with reasoning towards legal consequences. We discuss the phenomenon of legal shifts that shows that the step from evidence to (proven) facts cannot be isolated from the step from proven facts to legal consequences. We show how legal shifts can be modelled in terms of legal story schemes. Our model is illustrated by a discussion of the Dutch Wamel murder case.

When reasoning about the facts of a case, we typically use stories to link the known events into coherent wholes. One way to establish coherence is to appeal to past examples, real or fictitious. These examples can be chosen and critiqued using the case-based reasoning (CBR) techniques from the AI and Law literature. In this paper, we apply these techniques to factual stories, assessing a story about the facts using precedents. We thus show how factual and legal reasoning can be combined in a CBR model.

In order to support the interchange of ideas and data between different projects and applications in the area of computational argumentation, a common ontology for computational argument, the Argument Interchange Format (AIF), has been devised. One of the criticisms levelled at the AIF has been that it does not take into account formal argumentation systems and their associated argumentation-theoretic semantics, which are part of the main focus of the field of computational argumentation. This paper aims to meet those criticisms by analysing the core AIF ontology in terms of the recently developed ASPIC argumentation framework.

In this paper, we provide a formal logical account of the burden of proof and proof standards in legal reasoning. As opposed to the usual argument-based model we use a hybrid model for Inference to the Best Explanation, which uses stories or explanations as well as arguments. We use examples of real cases to show that our hybrid reasoning model allows for a natural modeling of burdens and standards of proof.

Motives play an important role at every stage of a crimi-nal investigation. In this research abstract we provide an overview of an account of motivations based on a general approach to practical reasoning. 1.

In this paper we present an approach to abductive reasoning in law by examining it in the context of an argumentation scheme for practical reasoning. We present a particular scheme, based on an established scheme for practical reasoning, that can be used to reason abductively about how an agent might have acted to reach a particular scenario, and the motivations for doing so. Plausibility here depends on a satisfactory explanation of why this particular agent followed these motivations in the particular situation. The scheme is given a formal grounding in terms of Action-based Alternating Transition Systems and we illustrate the approach with a running legal example.

In the present paper, we provide a formalised version of a merged argumentative and story-based approach towards the analysis of evidence. As an application, we are able to show how our approach sheds new light on inference to the best explanation with case evidence. More specifically, it will be clarified how the events in a case story that are considered to be proven abductively explain the otherwise unproven events of the case story. We compare our approach with existing AI work on modelling legal reasoning with evidence.