Learning of Structurally Unambiguous Probabilistic Grammars

The problem of identifying a probabilistic context free grammar has two aspects: the first is determining grammar's topology (the rules grammar) and second estimating weights for each rule. Given hardness results learning context-free grammars in general, particular, most literature concentrated on problem. In this work we address We restrict attention to structurally unambiguous weighted (SUWCFG) provide query algorithm strucuturally (SUPCFG). show that SUWCFG can be represented using co-linear multiplicity tree automata (CMTA), polynomial learns CMTAs. learned CMTA converted into grammar, thus providing complete strucutrally (both weights) structured membership queries equivalence queries. demonstrate usefulness our PCFGs over genomic data.