What EDI needs is a good semantics, that is, a workable formal theory of what EDI messages mean. As is widely recognized, the point applies to electronic commerce and to communications by artificial agents in general. Some progress has been made in this direction, but very much work remains to be done. In this paper we introduce and discuss the spanning hypothesis for agent (EDI, electronic commerce, . . . ) communication languages. The spanning hypothesis is a claim about the semantics for a given communication language, and we think it represents a necessary condition for truly successful artificial communication in electronic commerce. After discussing and clarifying the hypothesis, and how it might be confirmed, we present evidence in its favor from an analysis of several EDIFACT transaction sets.
Steven O. Kimbrough, Scott A. Moore