Workshop on Human Reasoning and Computational Logic
The slides will be online. You can find them under Talks.
9th and 10th of February 2017, we organize a workshop on Human Reasoning and Computational Logic. The workshop will be held in in Dresden, Germany. The goal of this workshop is to provide a platform for the scientific exchange with respect to Human Reasoning between the areas of Cognitive Science and Computational Logic.
- 9th and 10th of February, 2017
- Dinner on 9th of February, location tba
The workshop is held at the Computer Science Faculty building of Technische Universität Dresden, Nöthnitzer Straße 46, Dresden-Räcknitz .
How to reach us
Commonsense Reasoning meets Theorem Proving by Claudia Schon
The area of commonsense reasoning aims at the creation of systems able to simulate the human way of rational thinking. We propose to combine automated reasoning with methods from the area of machine learning for tackling commonsense reasoning benchmarks. For this we use a benchmark suite introduced in the literature. Our goal is to use general purpose background knowledge without domain specific hand coding of axioms, such that the approach and the result can be used as well for other domains in mathematics and science. We furthermore report about preliminary experimental results.
Conditionals under Week Completion Semantics by Isabelly Rocha
Conditionals are sentences of the form if condition than consequence, which in classical logic is semantically equivalent to not condition or consequence. However, psychological experiments have repeatedly shown that this is not the way humans understand and use conditionals. An approach to evaluate this conditionals in human reasoning has been developed based on the Weak Completion Semantics. This approach can successfully model known examples in the literature, but still has many open questions which will be discussed in this talk.
Conditionals represent defeasible rules, encode recommendations for successful acting and provide basic structures for processing knowledge on machines. With their trivalent evaluations, they leave more semantical room for modeling acceptance than material implications, allowing to distinguish between cases where the conditional rule its refuted and cases where the conditional is not applicable. In many applications, conditionals are enriched with a numeric value out of an (un)countable set of weights, encoding how plausible, possible, or probable the conditional is believed to be by an agent. In this talk, we concentrate on the structural information of the conditionals in a knowledge base. We introduce mutually incomparable abstract impacts for the verification, falsification and non-applicability of a conditional under a possible world. We show that using these impacts, both high qualitative inference and revision can be realized with strictly qualitative information, not using any numeric measurement of plausibility, possibility, or probability. However, as a second step, we show that these abstract impacts can be linked to various semantics to provide even more sophisticated approaches to plausible inference and belief revision (so-called c-representations and c-revisions). We illustrate this in the framework of ordinal conditional functions, proving that not only the AGM postulates of belief revision, but also the advanced postulates of Darwiche and Pearl for iterated revision can be satisfied by our approach.
Syllogistic Reasoning under Weak Completion Semantics 2.0 by Ana Costa
In previous work we modeled human syllogistic reasoning under the Weak Completion Semantics. The principles outlined in that work turned out to successfully model this cognitive task. The accuracy of predictions derived from our modelling outperform the accuracy of the, by then, best theories. In this talk I will present an analysis of some of the problems of our first approach. From this analysis new ideas for the encoding of this cognitive task under weak completion semantics emerged. This talk will include a discussion of their implementation and their impact on the accuracy of our predictions.
The Quest for sensible Benchmark Problems in Human non-monotonic Reasoning by Marco Ragni
abstract coming soon...
Contextual Reasoning by Emmanuelle Dietz
We have shown in the past that abduction within the Weak Completion Semantics seems to adequately model the results of various human reasoning tasks, such as the suppression task, the selection task and the belief bias in syllogistic reasoning. So far, we have assumed that explanations are preferred to others, if they are minimal. However, it seems that humans favor certain explanations above others depending on the context. Furthermore, they seem to assume exceptions to occur only if there is some evidence for these exceptions. This more subtle notion of reasoning cannot be modeled within the Weak Completion Semantics straightaway. In the talk, I will introduce contextual programs. With help of these contextual programs, we can now define the notion of contextual abduction, which allow us to distinguish between exceptions and regular cases.
Weak Completion Semantics and Abduction by Steffen Hölldobler
I have extended abduction in allowing defeaters for assumptions to be among the abducibles. In the talk I will discuss various questions related to this extension: Is it needed? Do humans prefer minimal extensions? How shall we handle exceptions? Do humans prefer some abducibles over others? I will illustrate the questions and some hypothesis by means of examples from the suppression task and the ‘birds usually fly’ domain.
Evaluating Cognitive Theories using MPTs by Nicolas Riesterer
Note that on Friday, the talks already start at 9 o'clock!
|Thursday February 9||Friday February 10|
|9:30 - 10:15||
|9:00 - 09:45||
|10:45 - 11:30||
|10:15 - 11:00||
|11:30 - 13:30||Lunch||Lunch|
|13:30 - 14:15||
|14:45 - 15:30||
|16:00 - 16:45||