# Text-based intelligence

Knowledge representation has always been a central issue for AI, and as a sub-discipline within Computer Science it's primary contribution is probably the beginnings of a computational theory of knowledge. While it is still too early to speak of such a theory, some key aspects of good knowledge representation are becoming clear [REF438] .

The text captured in document corpora was not entered with the { intention} of being part of a knowledge base. These are documents written by someone as part of a natural communication process, and any search engine technology simply gives this document added life. Alternatively, we can say that the document {\em was} intended to become part of a knowledge base,'' but one that pre-dates (at least the AI) use of that term: people publish their documents with the explicit hope that their ideas can become part of our collective wisdom and used by others.

Note the ease with which author-as-knowledge engineer can express their knowledge. Hypertext knowledge bases are accessible to every writer. In this view, hypertext solves the key AI problem of the KNOWLEDGE ACQUISITION BOTTLENECK , providing a knowledge representation language with the ease, flexibility and expressiveness of natural language --- by actually using natural language! The cost paid is the weakness of the inferences that can be made from a textual foundation: contrast the strong theorem-proving notions of inference of Section §6.5.1 with the many confounded associations which arise in Swanson's analysis of latent knowledge in Section §6.5.3 .

## Subsections

FOA © R. K. Belew - 00-09-21