From hombre!uunet!cbmvax!snark!lojbab Tue Nov 6 04:13:00 1990 Return-Path: Received: by marob.masa.com (/\=-/\ Smail3.1.18.1 #18.1) id ; Tue, 6 Nov 90 04:12 EST Received: by hombre.MASA.COM (smail2.5) id AA06602; 6 Nov 90 03:51:38 EST (Tue) Received: from cbmvax.UUCP by uunet.uu.net (5.61/1.14) with UUCP id AA24721; Tue, 6 Nov 90 00:03:07 -0500 Received: by cbmvax.cbm.commodore.com (5.57/UUCP-Project/Commodore Jan 13 1990) id AA06988; Tue, 6 Nov 90 00:00:38 EST Received: by snark.thyrsus.com (smail2.3) id AA08681; 6 Nov 90 00:02:21 EST (Tue) To: math.ucla.edu!jimc, major@pta.oz.au Subject: Re: response to J. Prothero book review and comments of 12 Oct 90 Cc: lojban-list@snark.thyrsus.com Date: 6 Nov 90 00:02:21 EST (Tue) From: uunet!cbmvax!snark.thyrsus.com!lojbab Message-Id: <9011060002.AA08681@snark.thyrsus.com> Status: RO >I'm not sure that an algorith exists for reliably dealing with antecedents in something like this "Joe likes Jerry. He says that he tells cool jokes." Two answers to this. The first is that the example is hopelessly ambiguous. Without any context, >I< can't tell what the second sentence pronouns refer to. Why expect a computer to do so. The computer, absent any heuristics, would simply ASK "Who says this?". With heuristics, the computer would learn to make the guess (which is all that it is, since a human can;t be sure either) that the first "he" is Joe, because there has been no change in focus, or indication of change. Guesasing is, of course, the second answer. However, I think it depends on the apllication whether guessing in such an instance is acceptable. I'm not looking for a Turing type of HAL-9000, but an intelligent program that can recognize when it doesn't know something and ask, possibly indicating its best guess for confoirmation. This should be much easier. I've heard that there is someone programming a computer with 'common sense' (a Discover from a couple issues ago, I think) understanding of English. It is a slow process to write all the rules, as jimc seems to be trying to do. A heuristic program would >learn< by experience. -lojbab