关键词: 神经网络, 递归神经网络算法（LMBP）, 模糊文法推导

Abstract: The second-order recurrent neural nerworks，based on the Real-Time Recurrent Learning（RTRL） algorithm and the Real-Coded Genetical Algorithm（RCGA），Fuzzy Grammar Inference（FGI） has the advantage of high accuracy，yet has the drawback of low convergence rate.However，the Levenberg-Marquard（LMBP） which is the fast algorithm of neural networks at present has not been paid more attention.The algorithm is described and the advantages and drawbacks are analysed by experiments.It is obvious that the convergence rate is high to perform the FGI.It also has the ability of processing super-long strings，too.

Key words: neural networks, Levenberg-Marquard（LMBP） algorithm, fuzzy grammartical inference