FOA Home | UP: Text-based intelligence


Grounding symbols in texts

According to Harnad's grounding hypothesis, if computers are ever to understand natural language as fully as humans, they must have an equally vast corpus of experience from which to draw [REF425] . We propose that the huge volumes of natural language text managed by hypertext systems provide exactly the corpus of ``experience'' needed for such understanding. Each word in every document in a hypertext system constitutes a separate experiential ``data point'' about what that word \means. The exciting prospect of using search engines as a basis for natural language understanding systems is that their understanding of words, and then concepts built from these words, will reflect the richness of this huge base of textual ``experience.'' Their are of course differences between the text-base ``experience'' and first-person, human experience, and these imply fundamental limits on language understanding derived from this source.

In this view, the computer's experience of the world is second-hand, via documents written by people about the world and subsequently through users' queries of the system. The ``trick'' used is to learn what words mean by interacting with users who already know what the words mean, with the documents of the textual corpus forming the common referential base of experience.

The hypertext itself is in fact only the first source of information, viz., how authors use and juxtapose words. The second, ongoing source of experience is the subsequent interactions with users, a new popualtion of people who use these same words and then react positively or negatively to the system's interpretation of those words. Both the original authors and the browsing users function as the text-based intelligent system's ``eyes'' into the real world and how it looks to humans. That insight is something no video camera will ever give any robot.


Top of Page | UP: Text-based intelligence | ,FOA Home


FOA © R. K. Belew - 00-09-21