Return-Path: Date: Wed, 23 Oct 91 09:52:51 -0400 From: cbmvax!uunet!grebyn.com!lojbab (Logical Language Group) Message-Id: <9110231352.AA10485@grebyn.com> To: gilson%61510.decnet@ccf2.nrl.navy.mil Subject: Re: He posted it? Cc: cowan@snark.thyrsus.com Status: RO X-From-Space-Date: Thu Oct 24 11:59:20 1991 X-From-Space-Address: cbmvax!uunet!grebyn!lojbab My mistake. What he said was virtually identical to what Carter had posted citing Peter, so I had thought that Carter had posted the same message verbatim to the list. I discovered I was wrong when I reread off line, but still think he said the identically same thing as Carter. I replied to him privately, citing as an example the word "far". The factors in deciding whether a given distance is "far" are so numerous and context- dependent, as well as observer-based, etc., that the word truly has no fixed meaning in the sense taht the ideal AI processor would need. You live "far" from you work, but not as far as Pete does (he lives in Atlanta and is working in LA), but in some contexts, he would say he lives "near" because he is referring to his local crash space instead of his home. But that same d distance might suddenly become "far" if his car breaks down. In Lojban you can express all of these conditions by attaching BAI phrases, but to say that they are ALL always part of "far" is incorrect, because in most cases, they are irrelevant. The AI processor needs to know that Atlanta is "far" from LA regardless of whether Pete's car is working. So for Lojban we use the minimum necessary places to setermine "far" in an optimal context, and allow attachment of extra places as needed for special contexts. This minimizes memorixation. But it also parallels English and other NLs. In most cases two 'places' of the English verb are unmarked, the nominative and the accusative. But in some cases a third place can also be unmarked: "John gives Mark the book". So Lojban merely carries an NL feature to an extended conclusion - that for some words it is plausible to have and use as many as 5 or even more unmarked places. The only such word that has proven problematical in terms of large place structure in real usage has been "fanva" = translate; and even though we seldom get the places in the right order for this word, the context provides enough redundancy to allow commu7nication. Will computers be able to do this? Well they aren't artificially intelligent unless they can at least sense a potential nonsense and ask about it. And eventually we'll know Lojban well enough to not screw the places up for fanva, anyway. Aftyer all, I never screw up klama. It is merely a mattrer of using a word enbough to "know what it means", which in Lojban includes knowing the places. lojbab