Starting from first principles, we have designed a rule-based first-order logic (FOL) knowledge-based system (KBS), called EXPARS, and coupled it with an original inductive learning module for the purpose of grammatical inference. While we stress the flexibility of the KBS approach asapplied to parsing, we also raise the issues imposed by such a coupling. The interaction between EXPARS and the learning module contributes to the learning of one grammar from positive examples. The proposed approach represents a contribution towards the design of future intelligent parsers.