A System for Explainable Answer Set Programming

Abstract

We present xclingo, a tool for generating explanations from ASP programs annotated with text and labels. These annotations allow tracing the application of rules or the atoms derived by them. The input of xclingo is a markup language written as ASP comment lines, so the programs annotated in this way can still be accepted by a standard ASP solver. xclingo translates the annotations into additional predicates and rules and uses the ASP solver clingo to obtain the extension of those auxiliary predicates. This information is used afterwards to construct derivation trees containing textual explanations. The language allows selecting which atoms to explain and, in its turn, which atoms or rules to include in those explanations. We illustrate the basic features through a diagnosis problem from the literature.

exp(ASPc) : Explaining ASP Programs with Choice Atoms and Constraint Rules

“…E c(P 1 ) shows that atom b in r 4 makes the constraint satisfied while m(1) does not support the con-straint. Thus, {∼b} is the support set of triggered constraint(m(1)), and {triggered constraint(m(1))} is the support set of m (1). For the additional constraint of P 1 , l( 9) is true (encoded in (m(1), n(1))) w.r.t.…”

“…This type of reasoning falls within the scope of explainable Artificial Intelligence and is useful to enhance the understanding of the resulting solutions as well as for debugging programs. So far, only a limited number of approaches have been proposed [1,6,7]. To the best of our knowledge, no system deals directly with ASP programs with choice atoms.…”

Explainable Machine Larning for liver transplantation

“…we obtain an answer set including the two facts in the first line plus the atom bad(55) derived from the rule in the second line. xclingo [3] is an ASP tool built on top of clingo that explains why a given atom was derived by tracing the relevant fired rules. To this aim, we may use textual descriptions to annotate specific rules (with the %!trace rule directive) or any derivation of a given atom (with the %!trace directive).…”

“…The dataset was collected at the Digestive Service of the Coruña University Hospital Center (CHUAC), Spain. The method consists in representing the DT as a logic program and using the tool xclingo [3] to annotate the program with natural language tags and construct the compound explanations. We purpose two different translations into logic programming, each with its benefits and disadvantages, which will be discussed.…”

Justifications for Goal-Directed Constraint Answer Set Programming

“…Even with the more restrictive query ?-opera(tuesday), clingo still generates 32 models. As a consequence, an explanation framework for clingo such as xclingo [3], based on Causal Graph Justifications [2], takes 10 times longer than s(CASP) to justify the results of opera.pl because it needs to process every single model.…”