So far, we have modified the solution synthesis algorithms to use maximally
interacting circles of variables in order to exploit their
disambiguating power as early as possible. Combined with CSP
techniques, these algorithms give fast and completely
reliable answers to simplified computational semantic problems.
The difficulty is that these problems are simplified. The
constraints used in the example above used yes-no constraints. In
computational semantics, however, semantic constraints must be used only as
tendencies, not sure-and-fast answers. Therefore, for computational semantics, bare CSP
techniques are not that helpful.
In this section, we will
demonstrate that branch-and-bound techniques can be combined with
solution synthesis to overcome this problem. We will see that the
circles created above to aid constraint satisfaction are also optimal
with regard to branch-and-bound techniques.