What you are trying to do is very hard. Statistical NLP is the way to go for unbounded domains. You might be able to do better semantic analysis for smaller clo

What you are trying to do is very hard. Statistical NLP is the way to go for unbounded domains. You might be able to do better semantic analysis for smaller closed domains. There is a relatively new great book on NLP out now that I suggest you take a look at. Particularly the chapters semantics are very useful, but they should give you an idea of how incredibly difficult what you're trying to do is. Book: http://www.amazon.com/Language-Processing-Prentice-Artificial-Intelligence/dp/0131873210/ref=sr_1_1?ie=UTF8&s=books&qid=1221233467&sr=1-1 http://www.amazon.com/Language-Processing-Prentice-Artificia...