I recently advanced the idea that more complicated grammars could be taught to a program that learned incrementally, (through trial and error), by first teaching it simpler grammars. How might this work? This instruction would be associated with particular statements, so the grammars would be acquired as it learned about specific 'objects' of reference. But, a particular statement might refer to objects of generalization as well as specific objects. For example, 'my car' refers to a specific car, 'the car' can refer to some specific car which is not fully specified (by the phrase), so it is a little like a variable that refers to 'some car', and the term 'a car' usually refers to a car which is not going to be fully specified. These simple syntactic distinctions are not consistently upheld in natural language and that is part of the problem, but I am just using them as examples. Further examples of syntactic markers can be found in early AI. The phrase, 'is a' can often refer to a higher level of generalization which might be used as a category. 'A cat is an animal' is an example. The term, 'has a', also used in early AI, is often a way of denoting that some object of reference has some characteristic or property. There were many problems with the overly-simplistic use of syntactic markers. One is that they are not used consistently and the second is that the statements in which they appear are not usually universally true (which makes logical deduction problematic). 'A leopard has spots' can be true, but I have a specific memory of a black leopard that I saw (because it made me think of a much larger version of a black Burmese house cat that we had) that did not have spots. Since my AI / AGI program would be designed to look for common words that can be found within different kinds of sentences (and text) it will be able to detect potential candidates that might be used as generalizations in more complicated sentences.
It is my feeling that by using previously acquired simple grammatical forms I should be able to direct my program to be able to effectively use the relative generalization level of the sentences that I would use with it. And since I am designing the program to look for reason based reasoning, I will also be able to use simpler grammatical forms to emphasize relations that can be tied together by true reasoning. And I will also be able to use simpler grammatical forms to direct its attention to the connections of anaphoric-like relations in the text. I realize that I haven't convinced many of the people who will read this, but that is not my interest. I am trying to give the few people who might actually be interested some insight into what I am working on. I should have some more substantial examples, whether they work or not, sometime next year. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
