I claim to be an intelligent entity (if not a real AI programmer), and one of my more valuable tools is a common dictionary, whether paper or electronic. There are many words I know, learned from the dictionary and/or from context, that I cannot pronounce since I did not learn them verbally. This occasionally comes to light when one of my English students here in Tianjin asks me for an English word and I have to reply, "I know the word, but I don't know how to pronounce it."
This intelligent entity (me) HAS an essential natural language DB in its system. I also WANT to have a better, more complete natural language DB. It would be great to have the contents of a good dictionary implanted in my brain if it were possible. It would seem to me that words are a good starting place for symbols that represent things. Why reinvent the wheel? Especially since we would like to communicate with the AI entities we create, if only to hear or see "king's pawn to king four." Certainly natural language is rather messy with its synonyms, homonyms, and multiple meanings per word. The WordNet mappings to synonym sets seems to be a good way to start approaching the problem, especially with its inclusion of small word groups such as "part of speech." I realize that there is a big conceptual difference between data and the ability to process it. But one cannot be useful without the other. Sometimes data and processing ability can be closely combined. Perhaps word definitions can be considered data. Then grammar and parts of speech that construct sentences may be considered a type of functionality. Several words in a definition map to a single word (or synonym set). Words are linked with grammar rules to form a sentence. One sentence, or perhaps a few in a paragraph map to a "thought." This is a system that works well for US, so has potential application to an AGI. Of course, "grounding" the words used to define other words is a big challenge. But this difficulty does not lead to the conclusion that natural language DB's should not be a very important part of an AGI. Perhaps the "grounding" and "real understanding" we instinctively strive for is a bit of an illusion. Maybe intertwined but ultimately circular and imperfect definitions are all we need. Sensory grounding would be nice, but perhaps not necessary. I can certainly understand things with very little sensory grounding, such as the general theory of relativity (well, a little understanding anyway). And does it really matter that I perceive red the same way as everyone else, so long as the perception is consistent? Things like mathematics and chemistry have their own "languages" that at least to some extent can be approached and explained using the "basic" natural languages. While this type of career plan is not to be recommended, I hope to draw on my varied background as practical engineer, research engineer, lawyer, patent lawyer, and import/export business (not to mention a variety of hobbies) to guide my thinking about thinking. We often intelligently use things we do not understand. Computers, automobiles, our brains, quarks, and so on. Why can't an AGI use words it does not actually understand, so long as it uses the word properly and accomplishes the desired result? I have seen expert systems and databases do truly amazing things in my various experiences. But nothing so amazing as seeing EllaZ deal blackjack, or place the entire contents of Kant's Critique of Pure Reason into a single scrolling browser textbox :-) Catch you all later . . . Kevin Copple P.S. I have one of the reduced Loebner contest versions of the EllaZ system (a/k/a Kip) running the Oddcast animated head now at www.EllaZ.com. The implementation is still a bit rough, but the AT&T Natural Voices TTS is quite good. [EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
