Natural language processing is, no doubts, one of the most frequent
application field of machine learning, and especially of inductive
logic programming(ILP). ILP is very convenient for an automatic
synthesis of rule-based NLP systems, the most promising ones being
part-of-speech taggers. Problems already addressed by ILP techniques
include context-sensitive spelling checkers, part-of-speech tagging as
well as grammar learning. We give a summary of applications showing both
their promises and their limits. Then an applicability of those methods
in inflective languages is discussed. Several projects working with
Czech corpora are being solved. We conclude with describing the used ILP
methods, the main difficulties and the results reached.