ATRANS Automatic Processing of Money Transfer Messages

Unformatted natural-language money transfer messages play an important role in the international banking system. Manually reading such messages and encoding them in the format understandable by a bank's automatic payment system is relatively slow and expensive. Due to the very restricted nature of the domain, the problem lends itself naturally to a Conceptual Dependency (CD), script-style solution. This paper illustrates the solutions to a number of problems that arise when an academic theory is applied to a real-world problem. In particular, we concentrate on the problem of context localization in the absence of reliable syntactic clues, such as sentence boundaries.