A Logic For Decidable Reasoning About Actions

We consider a modified version of the situation calculus built using a two-variable fragment of the first-order logic extended with counting quantifiers. We mention several additional groups of axioms that can be introduced to capture taxonomic reasoning. We show that the regression operator in this framework can be defined similarly to regression in Reiter’s version of the situation calculus. Using this new regression operator, we show that the projection and executability problems (the important reasoning tasks in the situation calculus) are decidable in the modified version even if an initial knowledge base is incomplete. We also discuss the complexity of solving the projection problem in this modified language in general. Furthermore, we define description logic based sub-languages of our modified situation calculus. They are based on the description logics ALCO(U) (or ALCQO(U), respectively). We show that in these sub-languages solving the projection problem has better computational complexity than in the general modified situation calculus. We mention possible applications to formalization of Semantic Web services and some connections with reasoning about actions based on description logics.

A Logic For Decidable Reasoning About Actions

We consider a modified version of the situation calculus built using a two-variable fragment of the first-order logic extended with counting quantifiers. We mention several additional groups of axioms that can be introduced to capture taxonomic reasoning. We show that the regression operator in this framework can be defined similarly to regression in Reiter’s version of the situation calculus. Using this new regression operator, we show that the projection and executability problems (the important reasoning tasks in the situation calculus) are decidable in the modified version even if an initial knowledge base is incomplete. We also discuss the complexity of solving the projection problem in this modified language in general. Furthermore, we define description logic based sub-languages of our modified situation calculus. They are based on the description logics ALCO(U) (or ALCQO(U), respectively). We show that in these sub-languages solving the projection problem has better computational complexity than in the general modified situation calculus. We mention possible applications to formalization of Semantic Web services and some connections with reasoning about actions based on description logics.

What Does My QA Model Know? Devising Controlled Probes Using Expert Knowledge

Open-domain question answering (QA) involves many knowledge and reasoning challenges, but are successful QA models actually learning such knowledge when trained on benchmark QA tasks? We investigate this via several new diagnostic tasks probing whether multiple-choice QA models know definitions and taxonomic reasoning—two skills widespread in existing benchmarks and fundamental to more complex reasoning. We introduce a methodology for automatically building probe datasets from expert knowledge sources, allowing for systematic control and a comprehensive evaluation. We include ways to carefully control for artifacts that may arise during this process. Our evaluation confirms that transformer-based multiple-choice QA models are already predisposed to recognize certain types of structural linguistic knowledge. However, it also reveals a more nuanced picture: their performance notably degrades even with a slight increase in the number of “hops” in the underlying taxonomic hierarchy, and with more challenging distractor candidates. Further, existing models are far from perfect when assessed at the level of clusters of semantically connected probes, such as all hypernym questions about a single concept.

Cross-cultural variation in the development of folk ecological reasoning

Two studies examined children’s reasoning about biological kinds in populations that vary in formal education and direct experience with the natural world, a Western (urban U.S.) and a Non-Western population (Tanna, Vanuatu). Study 1 examined children’s concepts of ecological relatedness between species (N=97, 5- 13-year-olds). U.S. children provided more taxonomic explanations than Ni-Vanuatu children, who provided more ecological, physiological, and utility explanations than U.S. children. Ecological explanations were most common overall, and more common among older than younger children across cultures. In Study 2, children (N=106, 6- 11-year-olds) sorted pictures of natural kinds into groups. U.S. children were more likely than Ni-Vanuatu children to categorize a human as an animal and the tendency to group a human with other animals increased with age in the U.S. Despite substantial differences in cultural, educational, and ecological input, children in both populations privileged ecological reasoning. In contrast, taxonomic reasoning was more variable between populations, which may reflect differences in experience with formal education.

A classification and constraint-based framework for configuration

One of the main difficulties with configuration problem solving lies in the representation of the domain knowledge because many different aspects, such as taxonomy, topology, constraints, resource balancing, component generation, etc., have to be captured in a single model. This model must be expressive, declarative, and structured enough to be easy to maintain and to be easily used by many different kind of reasoning algorithms. This paper presents a new framework where a configuration problem is considered both as a classification problem and as a constraint satisfaction problem (CSP). Our approach deeply blends concepts from the CSP and object-oriented paradigms to adopt the strengths of both. We expose how we have integrated taxonomic reasoning in the constraint programming schema. We also introduce new constrained variables with nonfinite domains to deal with the fact that the set of components is previously unknown and is constructed during the search for solution. Our work strongly focuses on the representation and the structuring of the domain knowledge, because the most common drawback of previous works is the difficulty to maintain the knowledge base that is due to a lack of structure and expressiveness of the knowledge representation model. The main contribution of our work is to provide an object-oriented model completely integrated in the CSP schema, with inheritance and classification mechanisms, and with specific arc consistency algorithms.