Traditional syntactic parsing such as using Probabilistic context free grammar (PCFG) parser, Stanford parser, etc. can achieve good performance on parsing natural language sentences. However, they usually suffer ambiguous problems in dealing with situations such as PP attachment that need semantic information to resolve. We propose a novel data-oriented fragment-based adaptive parsing method that combines both syntactic and semantic information with the help of parsing fragments and a recursive auto-associative memory (RAAM) that can disambiguate by selecting the most semantic plausible parse tree from ambiguous candidates.