Reconstructing Repairs of Census Data Forms as Database Repairs Leopoldo Bertossi 1 , Enrico Franconi 2 , and Andrei Lopatenko 2, 3 1 Carleton University, School of Computer Science, Ottawa, Canada, bertossi@scs.carleton.ca 2 Free University of Bozen–Bolzano, Faculty of Computer Science, Italy, franconi@inf.unibz.it, alopatenko@unibz.it 3 University of Manchester, Department of Computer Science, UK Abstract. In the context of consistent query answering (CQA) from inconsistent databases, the notion of repair is fundamental. A repair is a new database instance that minimally differs from the original, incon- sistent database, but does satisfy the integrity constraints. Minimality usually refers to a minimal set of tuples on which the two instances differ. In this paper we reexamine the process of correcting census data forms, where the notion of minimal number of changes is natural. Underlying assumptions are made explicit, and on the basis of this, corrections of census questionnaires are characterised as database repairs as introduced in the context of CQA. Minimal number of changes on census question- naires are represented as database repairs. Other interesting issues ad- dressed here, that are also relevant in the context of database repairs, are the notions of hard and soft constraints for the repair or correction pro- cess, and new specific, but natural ways of minimising changes. Finally, on the basis of the database representation, an answer set programming approach to census questionnaires corrections is presented. 1 Introduction The main goal of this paper is to characterise the problem of amending incorrect census questionnaires—a problem first formalised in [10]—as a special database repair problem, so that a comparison with existing database repair semantics could be carried out. Given a relational database schema, S , that includes a fixed infinite database domain D, we can consider the first-order language, L(S ), constructed using the predicates in S . Integrity constraints can be expressed as sentences in L(S ). A database instance r can be seen as a set of ground atoms of L(S ), or, alter- natively, as a first-order Herbrand structure compatible S . Given a fixed finite set, IC, of integrity constraints, a relational database instance is consistent if it satisfies IC, i.e. r  IC. Otherwise, we say that r is inconsistent. We assume that the set IC is logically consistent in the sense that there is a database instance that satisfies IC. There may be built-in predicates in addition to those in S , but they have the same, fixed extension in every instance.