The Recognition And Analysis Of Spatial Expressions From Texts Based On SNoW

The recognition and analysis of spatial expression is an important part of the Nature Science Foundation"the 3D visualization of spatial relationships in text based on ontology", which includes three parts: the extraction of spatial relation, the quantification of spatial relation, and the 3D visualization of spatial relation. According to the linguists'analysis, we divide the spatial description into two parts, including the spatial expression and the corresponding trajectory. Trajectory is the focus of spatial relation, and the spatial expression,which is used to describe the spatial position of the trajectory, consists of localizer and landmark.Similar to other problems in NLP, the recognition and analysis of spatial expression also has two methods: rule-based method and statistic-based method. While rule-based approaches have achieved great success in precision, but the shortcoming of this method is that the rules cost too much manpower and material resource. On the contrary, the statistic-based method can be effective in reducing human and material cost. Meanwhile, the result achieved by this method is also very close to or even better than rule-based methods when the scale of the corpus is appropriate. What's more, it has particularly advantages when we do the transplant.With these experiences, this thesis presents a SNoW-based method to recognize and analyze the spatial expression, and implements a recognition and analysis system.In detail, this dissertation has conducted the following researches:1. Corpus construction for spatial relation. Although the linguists had more detailed introduction about the spatial description in the modern Chinese language, there is none corpus about spatial relation, we firstly annotate spatial relation corpus.2. The research on the recognition of the spatial expression and the landmark. In this thesis, we present a SNoW-based method, which also uses some rules, to recognize the spatial expression and landmark, then implement a system for the recognition.