Philosophie du Web et Ingénierie des Connaissances

Abstract

Imagine a library of captioned images and a user who comes along and types in a word, phrase, or sentence asking for images. Today's software would have to do Boolean searches based on keywords in the query and the captions, perhaps broadening the search a bit by looking up synonyms in a thesaurus or definitions in a dictionary. Or consider the World Wide Web, whose keywordbased indexing is the only way to search through that immense information space. That's fine if you want to match 'A bird in water' against 'A duck in a pond', but it takes something like CYC to match 'A happy person' against 'A man watching his daughter take her first step'. CYC uses common sense to do matches of that sort. Similarly, CYC matched the query 'a strong and adventurous person' to a caption of 'a man climbing a rock face'. To do that, it used a few rules of the sort: 'If people do something for recreation that puts them at risk of bodily harm, then they are adventurous.' […] The key point here is that if you have the necessary common-sense knowledge-such as 'deadly pastimes suggest adventurousness'-then you can make the inference quickly and easily; if you lack it, you can't solve the problem at all. Ever » (Lenat, 1997) « L&F's [Lenat & Feigenbaum] system is liable to resemble nothing so much as an electric encyclopedia. No wonder its semantics will be derivative. Now it's possible, of course, that we might actually want an electric encyclopedia. In fact it might be a project worth pursuing-though it would require a major and revealing revision of both goals and procedure. Note that L&F, on the current design, retain only the formal data structures they generate, discarding the natural language articles, digests, etc., used in its preparation. Suppose, instead, they were to retain all those English entries, thick with connotation and ineffable significance, and use their data structures and inference engines as an active indexing scheme. Forget intelligence completely, in other words; take the project as one of constructing the world's largest hypertext system, with CYC functioning as a radically improved (and active) counterpart for the Dewey decimal system. Such a system might facilitate what numerous projects are struggling to implement: reliable, content-based searching and indexing schemes for massive textual databases, CYC's inference schemes would facilitate the retrieval of articles on related topics, or on the target subject matter using different vocabulary. And note, too, that it would exploit many current AI techniques, especially those of the