Answer Set Programming with External Sources

Answer set programming (ASP) is a declarative programming approach which has gained increasing attention in the last years. It is useful for many tasks in artificial intelligence, and many language extensions have advanced the paradigm into a strong modeling language. While the ASP programming paradigm has proved to be fruitful for a range of applications, current trends in distributed systems and the World Wide Web, for instance, revealed the need for access to external sources in a program, ranging from light-weight data access (e.g., XML, RDF, or relational data bases) to knowledge-intensive formalisms (e.g., description logics). To this end, HEX-programs are an extension and generalization of answer set programs by external sources which can be tightly coupled to the reasoner. This is realized by external atoms, whose truth value is not determined within the logic program, but by a background theory, which is technically realized as a plugin to the reasoner. The traditional evaluation algorithm for HEX-programs uses a translation approach which rewrites them to ordinary ASP programs. The fundamental idea is to replace external atoms by ordinary ones whose truth values are guessed. The resulting program is then evaluated by a state-of-the-art ASP solver. The resulting model candidates are subsequently checked for compliance with the external sources, and are discarded if the guesses value differs from the real truth value. While this approach is intuitive and natural, it turned out to be a bottleneck in advanced applications. It does not scale well, as the number of candidate answer sets grows exponentially with the number of external atoms in the program. Moreover, the traditional algorithms also impose very strong syntactic safety conditions on the input program, which restricts the language. This motivates the development of novel evaluation algorithms for HEXprograms, which treat external atoms as first-class citizens and build models from first principles; it is expected that this increases scalability and expressiveness. The thesis consists of two major parts. In the first part, we present new algorithms for ground HEX-programs, i.e., programs without variables. Conflict-driven learning techniques will be an important basis for our algorithms,