Analysis Algorithms, Based On Grammatical Functions To Match The Syntax

Chinese parsing is an important foundational topic in Chinese Information Process, simultaneously is also a recognition difficult problem, in the machine translation, the text abstract, the information filtration, the automatic question and answer has the widespread application.Based on the primary method of Grammar Function Match (GFM) as the syntactic analysis, this paper realized two algorithms, which viewed the TCT 973 as principal resource to survey the grammatical function.1. Parsing algorithm based on GFMThe algorithm's basic principle is: From the neighboring word or phrase which has analyzed, discovers dynamicly all possible new matches and joins them to the sharing forest, circulates unceasingly until no longer has the new node production, obtains certain different trees which can cover the entire sentence. The algorithm carried on superiorly in 4 aspects reduced the match number of times 1, the match threshold value; 2, chunk blocking; 3, ahead of time pruning; 4, control inflation. The disambiguation mainly depends on the tree's production probability to choose a biggest probability tree. Proposed two kind of computation model based on the static probability and the dynamic probability, and had extracted the corresponding probability dictionary and compared their disambiguation effect.2. Error recovery algorithm based on GFMThe algorithm's basic principle is: To parsing defeat's sentence, first finds certain active subtrees in the sharing forest and activates. The words they can't cover are considered as the key point that creates the sentence parsing defeat. Synthesis following three resources: 1, own grammatical function; 2, grammatical function similarity; 3, the word grammatical function cluster, makes the adjustment to the key point words' probability distribution, and takes them as the new nodes insertion sharing forest, then restarts the parsing. To which is still unable to complete the parsing, chooses certain widest subtrees compulsion matched and then outputs.At last we carried on the plan comparison and the parameter establishment through the experiment. The result indicates that the parser not only efficiently reduced the fake ambiguities, but also had a favorably analyzed efficiency, which analyzed results included abundant and accurate grammatical information. Using the PARSEVAL appraisal system, the rate of phrase precision and recall reaches 88.95%, 89.13% and 76.75%, 77.07% respectively in close test and open test.