|
Current computational model for Nature Language Processing
tends to combine statistics-based model and rule-based model.
It is easy to keep the benefits of both rule-based and
statistics-based language models in such a kind of models. Word
segmentation systems also tend to be implemented in this way.
Although the word segmentation systems are more mature than
before, there also exist an serious problem. Proper noun
identification is a bottle neck and should be resolbed. In
this thesis, we propose models to identify some proper nouns.
The key point is feature assignment. Not only are those proper
nouns identified, but also some special features are assigned.
Chinese surname-names are more than the other proper nouns in
general Chinese texts. We propose a new model to resolve this
problem. By the way, we also assign those proper nouns
features. Transliterated person-names are parts of person-
names in general texts. In the view point of Chinese surname-
name, the structures of transliterated person-names are more
complex. We propose a complete new thinking for resolving this
problem. Based on the structures of proper nouns, we know
organization names are much more complex than the former two.
Sometimes, an organization name may be composed of other proper
nouns, e.g., Chinese surname-names, transliterated person-
names, place names, and so on. In this thesis, we propose a
new model to identify these proper nouns. Finally, we will
demonstrate some examples to explain our approaches and discuss
the experimental results.
|