Some aspects on natural language context dependencies handling using adaptive technology.

by Moraes, Miryam de

Since low-complexity language formalisms are too weak to handle NL, stronger formalisms are required, most of them resource demanding, hard to use or unpractical. Structured pushdown automata are excellent to represent regular and context-free aspects on NLs by allowing them to be split into regular layer (implemented as finite-state machines) and a context-free one (represented by a pushdown store). Such devices accepts deterministic context-free languages in linear time, and is suitable as un underlying mechanism for adaptive automata, allowing handling - without loss of simplicity and efficiency - languages more complex than context-free ones. In this thesis context dependency is handled with adaptive technology. This work shows as a Natural Language rule described with a metalanguage can be converted into adaptive structured pushdown automata. It was possible to verify that complex problems in Natural Language parsing e.g., nondeterminisms and ambiguities present in agreement, subcategorization, coordination can be solved with efficiency. In fact, all adaptive mechanisms attached to these problems have O(n) performance. An adaptive architecture for NL Language processing is presented.