Millstream Systems and Graph Transformation for Complex Linguistic Models

Abstract

Allan Turing [5] suggested to regard the ability to communicate in human language as an indication of true intelligence. If a computer would be able to engage in such a communication with human beings without them being able to identify the computer, the computer should be considered to be intelligent. Although it is debatable whether this conclusion could really be drawn from the Turing test (see also [6]), it shows how complex human language is and how many facets it has. Some of the most important dimensions of language are phonology, morphology, syntax, semantics, and pragmatics. Pragmatics includes the whole field of contextual and ontological knowledge. These dimensions are not orthogonal, but are intertwined in many ways. Even if we restrict ourselves to text input and output, thus disregarding phonology, this creates an amazingly complex structure. While computational linguists try to make progress understanding the relation between the various dimensions, we usually restrict ourselves to syntax in natural language processing, sometimes extended by limited attempts to make a semantic interpretation or to make use of ontological knowledge. The reason for this is, of course, the descriptional and computational complexity of the models required.