Within artificial intelligence, a research paradigm sometimes known as experimental programming is common. In this, a computer program is used to explore ideas, rather than to provide a final and conclusive test of a single well-articulated theory or to operate as a polished piece of software. Although the methodology owes a lot to the notion of an ``experiment'' in traditional science, it does not rely on critical experiments to falsify abstract theories (although it would be satisfying if this were the case). Rather, the researcher attempts to clarify his/her ideas by posing the question: ``what would it take to have a computer program perform this task?''. Concrete fleshing out of embryonic ideas then consists of trying to construct a computer program. This not only forces a degree of detail and precision (cf. comments above on linguistic methodology), it also provides a readily testable version of the draft theory. Running the program and observing its behaviour (not just its final results but also how it achieves them) can provide useful insights into weaknesses of the ideas, and may even inspire possible amendments or extensions to the proto-theory. (See [Rit94][NS76][Buc88] for further discussion of this approach).
The work described here can be seen as exemplifying this approach. The central task of the project was to design, implement, and test a computer program, but the important product of the work was not the program itself; rather, it was the set of ideas that we developed in the course of the work.