Language is a part of a communication system [Rus98]. Besides it, there is a world which all the language users share, which is encoded as common sense among the community. To really understand the language component of this system, you cannot separate it from the world it mentions. So the translation and the NL understanding systems which ignore the world or a model of it, cannot be good.
A sucessful example of my statement is the SHRDLU project (Procedures as a representation for data in a program computer for understanding natural language [Winograd70]). In SHRDLU project we have a model of a mini world of blocks. (Need to understand the model and the syntax they used in the mathematical view).
Another good project to look at is the BORIS program in the In-depth undestanding book [Dyer83].
To model the mini world, there several efforts, such as: Verb Frame (for verbs), WordNet (for nouns, verbs and adjective), another attemp is in the KBMT project is FRAMEKIT [Goodman91], a language for meaning representation. In the FRAMEKIT language, we also use an ontology and a set of frame filters to describe the world. Because of the importance of meaning representation, a recent attemp is MultiNet [Helbig06] in the book “Knowledge Representation and the Semantics of Natural Language”. MultiNet is claimed to be one of the most comprehensive and thoroughly described knowledge representation systems[Wikipedia].
Books to read:
The model we look at here is model which is mentioned in the book “From Discourse to Logic” [Kamp93].
The period of investigation is about 10 years. Compare them together pros and cons, what is the state of the art. How they offer treatments to pros and cons of the previous methods. Why do you choose a specific model? What is your proposal?
About neutralizing writing style (or characterizing writing style), you should look at the keywords: stylistics and linguistics