Generating explanations for answer set programming applications

Abstract

The paper presents an enhancement of xASP, a system that generates explanation graphs for Answer Set Programming (ASP). Different from xASP, the new system, xASP2 , supports different clingo constructs like the choice rules, the constraints, and the aggregates such as #sum, #min. This work formalizes and presents an explainable artificial intelligence system for a broad fragment of ASP, capable of shrinking as much as possible the set of assumptions and presenting explanations in terms of directed acyclic graphs.

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

“…Lines 13-14 contain weight body. Given an ID i that does not occur in any output statement [3,9], we use l(i) to denote the corresponding literal. Constraint r 4 is shown in Line 8.…”

“…During the preprocessing, the set of all negation atoms in P, NANT (P) = {a | a ∈ r − ∧ r ∈ P} [6,9], is computed. For P 1 and the answer set A = {n(1), n(2), c, m(1)}, we have NANT…”

“…• A tentative assumption set TA [6,9] is computed, which is the superset of minimal assumption sets. • E r(P) in Sec.…”

“…Given E, the algorithm from [9] will find the explanation graph of literal in P, taking into consideration the additional types of nodes and links. Example 3 For program P 1 , the explanation graph of m(1) is shown in Fig.…”

“…In this paper, we present an improvement over our previous system, exp(ASP) [9], called exp(ASP c ). Given an ASP program P, an answer set A, and an atom a, exp(ASP c ) is aimed at answering the question "why is a true/false in A?"…”