On Context-Free Grammar Induction by Incremental Compression
Thomas Peetz

Abstract:
State merging algorithms are predominant in Grammar Induction of
finite state machines. This thesis extends the state merging approach
to context-free grammars. It connects current standard implementations
of state merging to compression-based learning. The exact same design
principles are applied to the construction of a context-free grammar
induction algorithm. The resulting induction algorithm is analyzed in
terms of convergence, runtime, and data requirements.