Learning context-free grammars from structural data in polynomial time
Yasubumi Sakakibara · Conference on Learning Theory · 1988
Abstract We consider the problem of learning a context-free grammar from its structural descriptions. Structural descriptions of a context-free grammar are unlabelled derivation trees of the grammar. We present an efficient algorithm for learning context-free grammars using two types of queries: structural equivalence queries and structural membership queries. The learning protocol is based on what is called “minimally adequate teacher”, and it is shown that a grammar learned by the algorithm is not only a correct grammar, i.e. equivalent to the unknown grammar but also structurally equivalent to it. Furthermore, the algorithm runs in time polynomial in the number of states of the minimum frontier-to-root tree automaton for the set of structural descriptions of the unknown grammar and the maximum size of any counter-example returned by a structural equivalence query.