Conditional Argumentative Reasoning

能够针对不确定和矛盾的信息提供决策支持是现代和未来人工智能系统的核心功能之一。这一挑战不仅要求方法能够处理海量数据，而且要求方法能够象征性地推理从数据挖掘的可废止规则和从这些规则构造的参数。在人工智能领域，形式论证的研究领域最近得到了越来越多的关注。形式论证的计算模型能够构建、比较和分析论证，从而提供了一种根据相互矛盾的信息进行理性决策支持的方法。相比之下，处理类似问题的其他研究领域-如缺省推理、可废止推理，特别是条件推理-侧重于规则在进行推理时的作用，特别是规则适用性的不确定性。为了能够应对处理不确定和矛盾信息的挑战，必须考虑这两个方面。CAR项目旨在为形式论证和基于规则的推理的综合方法建立理论基础。在技术上，我们将考虑抽象辩证框架(ADF)和条件逻辑(CL)的方法，并重点研究以下两个问题。首先，在ADF中，对论点的接受是通过所谓的接受条件来定义的。人们可以将这些接受条件解释为规则，这就产生了CL中的知识库。现在，人们可以应用CL的推理机制--如系统Z--并将结果与原始的ADF推理机制进行比较，特别是以一般性的论证术语分析结果。其次，CL中的任何知识库都可以用同样的方式解释为ADF。现在可以应用ADF推理机制-例如稳定语义-从而为CL定义一种新的推理机制。翻译和研究问题都提供了比较不同方法的方法。对这些问题的研究将使人们深入了解这两种方法是如何联系在一起的，更重要的是，它们如何从彼此中受益。这两个研究领域都制定了不同的评估标准-例如玩具例子和合理性假设-对于具体的推理方法，通过我们的翻译，将分别为这两个领域提供新的标准。在这个项目中，我们将详细解决上面概述的两个研究问题。更具体地说，我们将基于CL推理机制开发新的ADF推理机制，反之亦然，并分别使用其他领域提供的评估标准来评估这些方法和现有的方法。

Being able to provide decision-support in the light of uncertain and contradictory information is one of the core functionalities of modern and future AI systems. This challenge calls for methods not only capable of handling huge amounts of data but, in addition, methods being able to reason symbolically with both defeasible rules mined from the data and arguments constructed from these rules. Within AI, the research area of formal argumentation has recently gained increasing attention. Computational models of formal argumentation are able to build, compare, and analyse arguments, thus providing an approach for rational decision-support in the light of contradictory information. In contrast, other research areas addressing similar problems---such as default reasoning, defeasible reasoning, and, in particular, conditional reasoning---focus on the role of rules when performing inference and particularly the uncertainty of the applicability of rules. In order to be able to address the challenge of handling both uncertain and contradictory information, both aspects have to be taken into account.The project CAR aims at establishing a theoretical basis for integrative approaches of formal argumentation and rule-based reasoning. Technically, we will consider the approaches of Abstract Dialectical Frameworks (ADFs) and Conditional Logic (CL) and focus on the following two research questions. First, in an ADF, acceptance of arguments is defined through so-called acceptance conditions. One can interpret these acceptance conditions as rules and this yields a knowledge base in CL. Now one can apply reasoning mechanisms from CL - such as System Z - and compare the results with the original ADF reasoning mechanisms and, in particular, analyse the results in general argumentative terms. Second, any knowledge base in CL can be interpreted as an ADF in the same way. Now one can apply ADF reasoning mechanisms---such as stable semantics---and thus define a new reasoning mechanism for CL. Both translations and research questions provide ways to compare the different approaches. Investigating these will bring insights on how these two approaches relate and, more importantly, how they can benefit from each other. Both research areas developed diverse evaluation criteria---such as toy examples and rationality postulates---for concrete reasoning approaches and through our translations, new criteria will be available for both areas, respectively.In this project, we will address both research questions outlined above in detail. More concretely, we will develop novel reasoning mechanisms for ADFs based on CL reasoning mechanisms and vice versa, and evaluate those and existing approaches with evaluation criteria made available by the other area, respectively.