tailieunhanh - Báo cáo khoa học: "What's the Difference?"

Are these differences real or apparent? Riesbeck's parser i~ implemented as n production system, in which input text can either ssti~{y the condition side of any production rule within ~ packet of currently-active rules, or else interrupt processing by disabling the current packet of rules and enabling ('triggering') a new packet of rules. In operation, the main verb of each segment of text is located, and a pointer to its lexical decomposition (canonical form) is established in memory. | Schank Riesbeck vs. Norman Rumelhart What s the Difference Marc Eisenstadt The Open University Milton Keynes ENGLAND This paper explores the fundamental differences between two sentence-parsers developed in the early 1970 s Riesbeck s parser for Schank s conceptual dependency theory 5 and the LNR parser for Norman and Rumelhart s active semantic network theory 3 . The Riesbeck parser and the LNR parser share a common goal -that of transforming an input sentence into a canonical form for later use by memory inference paraphrase processes. Kor both parsers this transformation is the act of comprehension although they appear to go about it in very different ways. Are these differences real or apparent Riesbeck s parser is implemented as a production system in which input text can either satisfy the condition side of any production rule within a packet of currently-active rules or else interrupt processing by disabling the current packet of rules and enabling triggering a new packet of rules. In operation the main verb of each segment of text is located and a pointer to its lexical decomposition canonical form is established in memory. The surrounding text primarily noun phrases is then systematically mapped onto vacant case frame slots within the memory representation of the decomposed verb. Case information is signposted by a verb-triggered packet of production rules which expects certain classes of entity . animate recipient to be encountered in the text. Phrase boundaries are handled by keyword-triggered packets of rules which initiate and terminate the parsing of phrases. In contrast to this the LNR parser is implemented as an augmented transition network in which input text can either satisfy a current expectation or cause backtracking to a point at which an alternative expectation can be satisfied. In operation input text is mapped onto a surface case frame which is an n-ary predicate containing a pointer to the appropriate code responsible for decomposing .