In natural languages, a semantically completely independent entity, settled
word combination, often arises from combination of several words (e.g. a
multi-word name of a municipal community or chemical compounds). Such entity is
than called collocation.

In this work, we concentrate on description of a system for saving individual
collocations that would be able to detect even the changes of inflection by
means of rules. For our system we further use the term database of
collocations.

We shall analyse the properties of the kinds of collocations described by us
for Czech and suggest a formalism for the database of collocations. We shall
describe algorithm by means of which it is possible to search the collocations
with low time demands. By means of implementation the database of collocations
and the described algorithm, a module shall arise that enables to improve the
quality of written text, for example to identify the collocations in
corpora. We have devised and implemented the database of collocations by a
tool for interactive edition.