|
Abstract : |
Abstract. In this paper a method based on abduction for explaining and repairing the errors detected in the analysis of natural language sentences is presented. This method builds the most plausible corrections for any unrecognized sentence. It is a declarative approach and,as such, departs from the traditional heuristic relaxation based approaches. In the proposed method, the ungrammaticality ofasentence is viewed as a diagnosis problem and abductive reasoning is used to obtain the explanations for the detected errors. The problem is represented as an abduction problem and is implemented using logic programming with explicit negation and with integrity constraints. The existence of an error in a sentence will cause a contradiction to become apparent dueto the violation of an integrity rule. Contradictions are removed according to a revision process developed for contradictory programs in the sense of the well founded semantics for extended logic programs. A contradiction can be eliminated by adding rules that force the change of the logic value of some literals. The plausibility criteria among the di erent correction hypothesis is related to the minimality of the revisions and to the number of fault modes in the revisions. In order to improve the e ciency of the rst reported experiment [5], a rst bottom-up parse, using a Datalog grammar [11], is now proposed. Then, building on the partial results obtained during this phase, the proposed methodology for detecting and repairing errors is launched. 1, |
