On Learning Attacks in Probabilistic Abstract Argumentation

European Conference on Artificial Intelligence, 2020

Argumentation is an important topic in the field of AI. There is a substantial amount of work about different aspects of Dung's abstract Argumentation Framework (AF). Two relevant aspects considered separately so far are extending the framework to account for recursive attacks and supports, and considering dynamics, i.e., AFs evolving over time. In this paper, we jointly deal with these two aspects. We focus on Attack-Support Argumentation Frameworks (ASAFs) which allow for attack and support relations not only between arguments but also targeting attacks and supports at any level, and propose an approach for the incremental computation of extensions (sets of accepted arguments, attacks and supports) of updated ASAFs. Our approach assumes that an initial ASAF extension is given and uses it for first checking whether updates are irrelevant; for relevant updates, an extension of an updated ASAF is computed by translating the problem to the AF domain and leveraging on AF solvers. We experimentally show our incremental approach outperforms the direct computation of extensions for updated ASAFs.

2012

In this paper, we extend Dung's seminal argument framework to form a probabilistic argument framework by associating probabilities with arguments and defeats. We then compute the likelihood of some set of arguments appearing within an arbitrary argument framework induced from this probabilistic framework.

2018

The combination of argumentation and probability paves the way to new accounts of qualitative and quantitative uncertainty, thereby offering new theoretical and applicative opportunities. Due to a variety of interests, probabilistic argumentation is approached in the literature with different frameworks , pertaining to structured and abstract argumentation, and with respect to diverse types of uncertainty , in particular the uncertainty on the credibility of the premises, the uncertainty about which arguments to consider, and the uncertainty on the acceptance status of arguments or statements. Towards a general framework for probabilistic argumentation, we investigate a labelling-oriented framework encompassing a basic setting for rule-based argumentation and its (semi-) abstract account , along with diverse types of uncertainty. Our framework provides a systematic treatment of various kinds of uncertainty and of their relationships and allows us to back or question assertions from the literature.

HAL (Le Centre pour la Communication Scientifique Directe), 2020

Recursive argumentation frameworks (RAF) take into account the notion of higher-order attacks. Their semantics are defined in terms of structures. A labelling version of these semantics is defined in this report, and their correspondence with structures is shown. The case of RAF with no recursive attacks is considered, confirming that RAF are a generalization of Dung's argumentation frameworks. A semantics which had not been considered so far for RAF, the eager semantics, is in addition introduced.

Lecture Notes in Computer Science, 2004

The problem of interpreting Natural Language (NL) discourse is generally of exponential complexity. However, since interactions with users must be conducted in real time, an exhaustive search is not a practical option. In this paper, we present an anytime algorithm that generates "good enough" interpretations of probabilistic NL arguments in the context of a Bayesian network (BN). These interpretations consist of: BN nodes that match the sentences in a given argument, assumptions that justify the beliefs in the argument, and a reasoning structure that adds detail to the argument. We evaluated our algorithm using automatically generated arguments and hand-generated arguments. In both cases, our algorithm generated good interpretations (and often the best interpretation) in real time.

2013

The computation of preferred labellings of an abstract argumentation framework (or briefly, AF) is generally intractable. The existing decomposition-based approach by exploiting strongly connected components (SCCs) of a general AF is promising to cope with this problem. However, the efficiency of this approach is highly limited by the maximal SCC of an AF. This paper presents a further solution by exploiting the most sceptically rejected arguments of an AF. Given an AF, its grounded labelling is first generated. Then, the attacks between the undecided arguments and the rejected arguments are removed. It turns out that the modified AF has the same preferred labellings as the original AF, but the maximal SCC in it could be much smaller than that of the original AF. Empirical results show that this new method dramatically reduce the computation time for some sparse AFs (when the ratio of the number of edges to the number of nodes of an AF is between 1:1 and 1.8:1).

International Journal of Approximate Reasoning, 2020

What does it mean that something is probably obligatory? And how does it relate to the probability that it is permitted or prohibited? In this paper, we provide a possible answer by merging deontic argumentation and probabilistic argumentation into a probabilistic deontic argumentation framework. This framework allows us to specify a semantics for the probability of deontic statuses. The deontic argumentation part builds on standard concepts from the study of computational models of argument: rule-based arguments, argumentation graphs, argument labelling semantics and statement labelling semantics. We then encapsulate this deontic composition with the approach of probabilistic labellings to probabilistic argumentation, in order to associate deontic statements with probability values. The framework is illustrated with a scenario featuring a violation and a contrary-to-duty obligation.

2010

For argumentation useful, the means must be found overcome the knowledge acquisition bottleneck between the source material and the translation into and out of an argumentation formalism. As source material is often expressed and understood in natural language, we should consider how to translate natural language arguments. The current state of the art in argumentation does not have any formal translation method. In this paper, we discuss this lacuna. First, there are robust, wide-coverage computational linguistic systems which automatically translate multi-sentential structures into first order logic. In restricted domain of discourse, this can be used to formalise the knowledge base. We then consider how such an approach could be extended to argumentation. We propose two different models of translation based on currently available argumentation formalisms-logicbased and logic graphs. The advantage of the logic-based approach is that the arguments input into the argumentation system are sets of first order logic expressions and what is inferred from them. The consistency of each argument is assured; the relationships between the arguments in an argumentation framework graph are determined by the logic-based approach. One disadvantage is that each argument cannot have any missing expressions (enthymemes) before it can be used in the argumentation formalism. A second disadvantage is that the theory requires a complex system to determine relationships between the arguments, which is particularly relevant as arguments are added over the course of a dialogue. The logic graphs approach allows reasoning to proceed even where there are enthymemes, and it is straightforward to determine the relationships between arguments in the network and * 2009 c Adam Wyner. The author thanks Tony Hunter for comments. This paper was created 14.06.09 and revised over the summer. The current paper reflect some minor revisions. Errors rest with the author. new arguments. On the other hand, it requires that sets of logically consistent statements be recalculated every time a statement is added to the argument network. To overcome this, we suggest that research be directed to support calculations over subgraphs. Additional problems for both approaches are outlined, providing an agenda for future research.

Proceedings of the Seventeenth International Conference on Principles of Knowledge Representation and Reasoning, 2020

Neural-symbolic systems combine the strengths of neural networks and symbolic formalisms. In this paper, we introduce a neural-symbolic system which combines restricted Boltzmann machines and probabilistic semi-abstract argumentation. We propose to train networks on argument labellings explaining the data, so that any sampled data outcome is associated with an argument labelling. Argument labellings are integrated as constraints within restricted Boltzmann machines, so that the neural networks are used to learn probabilistic dependencies amongst argument labels. Given a dataset and an argumentation graph as prior knowledge, for every example/case K in the dataset, we use a so-called K-maxconsistent labelling of the graph, and an explanation of case K refers to a K-maxconsistent labelling of the given argumentation graph. The abilities of the proposed system to predict correct labellings were evaluated and compared with standard machine learning techniques. Experiments revealed that ...