小暮潔
戦略的遅延逐次複製グラフ単一化手法
Abstract:Two feature structure unification methods called the lazy incremental
copy graph unification method and the strategic incremental copy graph
unification method have been developed. The former method achives structure
sharing with constant order data access time which reduces required memory.
The latter method uses an early failure finding strategy which tries to unify
substructures tending to first fail in unification. This method is based on
stochastic data and reduces unnecessary computation. The two methods can be
combined into a method called the strategic lazy incremental copy graph
unification method. The combined method not only makes each feature structure
unification efficient, but also reduces garbage collection and page swapping
occurrences, thus increasing total efficiencies of TFS unification-based natural
language processing systems such as a spoken Japanese sentence analysis system
based on HPSG ..