Problems and Review of Statistical Parsing Language Model

The lexical language model is recently the hotspot in grammar research, which is promoted by incorporating the phrase head with statistics. This paper summarizes about four improving language models which belong to this kind of model: they have utilized heads that is extracted by CFG and calculated the probability between the heads or inside CFG. Different from N-gram and SCFG, the probability calculation change SCFG´s manner to the calculation among vocabulary. Corresponding experiments have shown improvement not only in precision but also in the language complexity. This paper has analyzed these models and also pointed out their problems. Especially, more detailed discussion was given about the immediate-head language model. But, it still has many defects. Finally, the potential idea for Chinese natural language model was promoted, the feasibility and the actuality of Chinese deep grammar analysis have discussed, too.