Electronic negotiation is a rapidly developing domain where people communicate by email or other exchange of text which accompanies formal means of negotiation. The Management Science and Artificial Intelligence communities actively investigate the process and data of electronic negotiation, but nobody seems to have applied natural language processing (NLP) techniques to text data gathered through a negotiation process.

Natural language that people use in electronic communication is far from perfect. Text data obtained through electronic channels are characterized by a large number of spelling and grammatical errors, due to the lack of editing, and the uncontrolled use of informal and slang expressions. This also applies to the data of electronic negotiation.

We analyze characteristics of the language data obtained from electronic negotiation. We introduce a novel procedure for extracting and building a lexicon from raw noisy data. The data belong to a closed domain, which allows us to perform domain-dependent word-sense disambiguation. The procedure itself is domain-independent and should work with data from various similar text collections. We present the results of an application of our procedure to a text corpus collected by an electronic negotiation support system.

The procedure results allow us to use Machine Learning and statistical methods to predict success of negotiations from the data. We present a procedure to find language patterns corresponding to successful negotiations.