4.2.1 Models of Dominance Constraints

A model of a dominance constraint \phi is a pair (T,I) of a tree T and an interpretation of I mapping variables of \phi to nodes in T, and such that \phi is satisfied. We write (T,I)\models\phi to say that \phi is satisfied by (T,I) and define it in the usual Tarskian way as follows:

\begin{array}{ll}
(T,I)\models \phi\wedge\phi' &\text{ if }
(T,I)\models\phi\text{ and }(T,I)\models\phi'\\
(T,I)\models x\mathbin{R} y&\text{ if }
I(x) \text{ is in relation } r \text{ to } I(y) \text{ in } T,
\text{ for some } r\in R
\end{array}

Solving dominance constraints is NP-hard. This was shown e.g. in [KNT98] by encoding boolean satisfiability as a dominance problem. However, the constraint-based technique first outlined in [DG99] has proven quite successful and solves practical problems of arising e.g. in semantic underspecification very efficiently. The technique is formally studied and proven sound and complete in [DN00]. An extension of this technique to handle general descriptions with arbitrary Boolean connectives was presented in [Duc00].


Denys Duchier
Version 1.2.0 (20010221)