I am not at all convinced that this is a solved problem. (Therefore, it must be an unsolved problem, IMHO.) I believe that there is a grey area between Natural Language Processing and so-called Natural Language Semantics. To some extent the Semantic Web initiatives have attempted to bridge this.
Once again, I will state that I believe that until now NLP has been more about understanding natural language than about generating natural language, in this case abstracting semantics. Somehow this kind of semantic abstraction is the threshold to artificial general intelligence.
Thequestion answering system, and its successor , have made headway into this “grey area”. There are other experimental systems in this class. has written her 2011 PhD thesis on , including Aqualog, PowerAqua, and others.
See also my quick and dirty webpages: