From lojbab@lojban.org Fri Nov 30 20:37:45 2001 Return-Path: X-Sender: lojbab@lojban.org X-Apparently-To: lojban@yahoogroups.com Received: (EGP: mail-8_0_0_1); 1 Dec 2001 04:37:43 -0000 Received: (qmail 92916 invoked from network); 1 Dec 2001 04:37:42 -0000 Received: from unknown (216.115.97.171) by m10.grp.snv.yahoo.com with QMQP; 1 Dec 2001 04:37:42 -0000 Received: from unknown (HELO stmpy-4.cais.net) (205.252.14.74) by mta3.grp.snv.yahoo.com with SMTP; 1 Dec 2001 04:37:44 -0000 Received: from bob.lojban.org (43.dynamic.cais.com [207.226.56.43]) by stmpy-4.cais.net (8.11.1/8.11.1) with ESMTP id fB14baH11997 for ; Fri, 30 Nov 2001 23:37:36 -0500 (EST) Message-Id: <4.3.2.7.2.20011130223529.00d82e30@pop.cais.com> X-Sender: vir1036@pop.cais.com X-Mailer: QUALCOMM Windows Eudora Version 4.3.2 Date: Fri, 30 Nov 2001 23:37:22 -0500 To: lojban@yahoogroups.com Subject: Re: [lojban] morphology In-Reply-To: <9u8vf1+b093@eGroups.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; format=flowed From: "Bob LeChevalier (lojbab)" At 09:58 PM 11/30/01 +0000, thinkit8@lycos.com wrote: >is it true that the lojban morphology is exaustively proven? that >is, can it be shown that when following the morphology rules, audio >visual isomorphism is assured? It depends on what you mean by proven. We have a morphology algorithm that so far as we know follows the rules of the language that we have stated, and it will make a decision about whether words are or are not of various classes. But that algorithm does not distinguish between the various types of brivla - gismu, lujvo and fu'ivla are all members of "selma'o" BRIVLA (ignoring the fact that BRIVLA is not a category of cmavo). Thus "slinku'i" is a valid brivla form, but is not any of the valid types of brivla. The algorithm has not been mathematically verified - we need someone with the right sort of mathematical proof-of-correctness skills to do that sort of thing, as the combinatorics form a VERY bushy tree (I think I estimated several million possible combinations to test in a string of only a dozen phonemes). However, we have dummy-checked all the common cases and any errors, if they exist, will involve weirdnesses in combinations of experimental cmavo and type IV fu'ivla and the like. Nora has built a program that implements the algorithm, but since it does not distinguish between fu'ivla and lujvo and gismu, and you have to type stress explicitly, it is not a particularly useful thing. We've never found a bug when we've tried though. >i was under the impression that it >wasn't, and this was what was holding up cultural fu'ivla. No. What holds up cultural fu'ivla is the very looseness of the algorithm for distinguishing classes of brivla. The three forms of brivla have a hierarchy. There are gismu first and foremost, then there are lujvo, which fit certain forms that do not break down in combinations with other words into a cmavo and a different lujvo and/or gismu, and the space of anything that is left, which will not break down alone or in combination into a gismu and/or lujvo and something else is a fu'ivla. Given the hierarchy, there is little doubt that the algorithm will succeed in resolving all conflicts. However, since fu'ivla wordforms are defined in the algorithm only by what they are NOT, we do nopt have a definitive algorithm for testing a wordform to see if it is a valid fu'ivla. We know that "slinku'i" words fail the "slinku'i" test. We don't know what other wordforms within brivla space are not lujvo or gismu and which do not fail the test except by trial and error. The human process of testing words for possible breakup is sufficiently unintuitive and subject to error, that we simply avoid the difficulty for now, especially since type 3 fu'ivla are good enough for everyone except Pierre %^). The classic example of this problem is the brivla "iglu". JCB and Nora went back and forth for years over whether it was a valid fu'ivla in TLI Loglan and never reached an agreement - but then JCB in his later years wasn't as committed to audiovisual isomorphism as he was in 1960. (I'll let someone else look in the archives to see where iglu ended up in Lojban - I never can remember and Nora has gone to sleep). We adopted the type 3 fu'ivla form because we WERE able to satisfy ourselves that it always gave a valid fu'ivla, and we are almost as confident about the special experimental fu'ivla "gismu" mentioned in the book, even though we spent only a short time checking it. But no one has solved the general case for the validity of a type 4 fu'ivla form and it has not been a priority, since such words are by definition barely part of the language in the first place. >furthurmore, can it be proven within reasonable limits, such as not >allowing fu'ivla? I'm not sure what this means. There is no doubt that the algorithm could be mathematically proven by someone with the skills to do so - the proof complexity might be akin to the proof of the 4-color map theorem of topology though (I think that ran several hundred pages) - it is a simple problem with a very bushy tree. It is probably easier to come up with a definitive test for whether a word is a valid fu'ivla, given some maximum length to test, but no one has done so because the test isn't worth much. The critical point of fu'ivla will be nonce usage when you want to communicate a new concept that has not been analyzed, and we don't believe that it is possible to have human algorithm for making fu'ivla of arbitrary form. If you always have to test fu'ivla against a computer program, then adding fu'ivla will tend to be a "Lojban academy" function, and right now there is little sentiment to create such an academy, even without all the mutterings about "Cabals" and "Lojban Central". lojbab -- lojbab lojbab@lojban.org Bob LeChevalier, President, The Logical Language Group, Inc. 2904 Beau Lane, Fairfax VA 22031-1303 USA 703-385-0273 Artificial language Loglan/Lojban: http://www.lojban.org