From cbmvax!uunet!CUVMA.BITNET!LOJBAN Thu Feb 27 12:37:16 1992 Return-Path: Received: by snark.thyrsus.com (/\==/\ Smail3.1.21.1 #21.19) id ; Thu, 27 Feb 92 12:37 EST Received: by cbmvax.cbm.commodore.com (5.57/UUCP-Project/Commodore 2/8/91) id AA03770; Thu, 27 Feb 92 12:28:00 EST Received: from cunixf.cc.columbia.edu by relay1.UU.NET with SMTP (5.61/UUNET-internet-primary) id AA17904; Thu, 27 Feb 92 12:20:08 -0500 Received: from cuvmb.cc.columbia.edu by cunixf.cc.columbia.edu (5.59/FCB) id AA07162; Thu, 27 Feb 92 12:17:16 EST Message-Id: <9202271717.AA07162@cunixf.cc.columbia.edu> Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 5334; Thu, 27 Feb 92 12:15:50 EST Received: by CUVMB (Mailer R2.07) id 4248; Thu, 27 Feb 92 12:14:46 EST Date: Thu, 27 Feb 1992 09:14:47 -0800 Reply-To: cbmvax!uunet!MATH.UCLA.EDU!cuvmb.cc.columbia.edu!jimc Sender: Lojban list From: cbmvax!uunet!MATH.UCLA.EDU!cuvmb.cc.columbia.edu!jimc Subject: Thesaurus categories To: John Cowan , Eric Raymond , Eric Tiedemann Status: RO Naturally :-) I applaud John Cowan's effort to thesaurus-ize the gismu list. As he indicated, the Carter thesaurus categories below 100 are all for cmavo, and the categories will not be too useful for Lojban. People may wonder how I chose those particular divisions for the gismu categories. I tried to split up the Old Loglan gismu into easily learned lists of about ten gismu each, but with as much commonality of place structures as I could find, as well as semantic similarity. Thus, for example, "rutma-route" (now pluta) was in the motion word category because it had motion word places, which in Old Loglan were identical for almost all words -- even though "route" is semantically tenuously related to the other motion words. Of course, as new words were added some categories grew larger. Repartitioning may be in order now. One limitation was that the encoding only allows nine subcategories in any larger category, a problem in 420 "tools, machines and buildings". Words were assigned in multiple categories because semantically they belonged there. A few categories such as 150 pull together words whose "homes" are elsewhere, but which are useful for particular purposes such as modal operators or dikyjvo. I quickly discovered the impossibility of my goal, a "procrustean bed" approach to place structures in which all words within broad categories would have the same places. However, many categories did fit this model very well, such as the motion words, words for knowledge, parts of objects, and directional properties. And when I revised the vocabulary for -gua!spi, sticking close to the Loglan model, I found it useful and effective to modify a number of nonconforming words to match their list partners. My first contact with this generalization was when I had to translate "flight path". "fleti-fly" was a normal motion word except that JCB had not provided a "route" argument. After some struggle with ridiculous tanru I realized that if I assumed a "route" argument the converted result would be natural and obvious. I recommend to Lojbab a wholesale application of this insight. When place structures are uniform (particularly as to ordering) it is much easier to learn the words; and places which may not obviously be relevant have a habit of being needed eventually. -- jimc