From lojbab Sat Jun 15 02:48:25 1991 Return-Path: Message-Id: Date: Sat, 15 Jun 91 02:47 EDT From: lojbab (Bob LeChevalier) To: lojban-list Subject: 'case' and Lojban Status: RO jimc has several times made a big deal of my avoidance of the term 'case' for sumti tcita - the tags that specify the roles of non-place-structure sumti. Some comments of the last couple of days have clarified in my mind what the reasons are. Let us see if I can explain. Linguistic case theorists postulate as few as 3 or 4 cases are necessary and sufficient to analyze language usefully. We have gismu with 5; this would seem to contradict the linguists if we say that each of the 5 is a separate 'case'. The reason, of course, is that these linguists use a broad and abstract view of case, and see absolutely nothing wrong with having more than one argument of a predicate having the same case. I am most wary about defining something contradictory to linguistic theory when it is not necessary, especially since the contradiction is due to different uses of the term 'case'. Natural language processing theorists use a slightly different definition of case, one closer to what jimc is thinking of, close to what Jim Brown came up with for Loglan, and close also to what we tried to allow for with BAI. These researchers define a number of cases as basic, usually between 5 and 20. Some of these cases are defined rather broadly - the case 'patient' or 'passive' in particular is typically a hodge-podge of semantic roles. For example, (and I think Jim Brown used this one but am note checking), if a bridge is across the river, the river is a 'patient'. Jim Brown analyzed all of his gismu in the mid-80s, and with another never-named person came up some large number like 31 or 47 cases. By sufficiently broadening the metaphorical definitions, he was able to reduce that number to 13, which is the number of words in his DIO case tag selma'o. The Procrustean bed of these compressions is truly torturous - the case tags prove to be fully as broad in meaning as English prepositions, and indeed his language with case tags in use looks like a prepositional language. Jim recognized that there were other roles/cases that were not part of any gismu place structure. This is known becuase he retained his equivalent of BAI, his 'modals'. Some of these overlap his case tags and some cover roles that never occur in a place structure. He then grammatically defined these as something separate from the DIO tags. DIO tags can only go on place structure sumti, 'modals' can go only on the added non-place structure sumti. We did not accept this scheme for Lojban for a couple of reasons: 1. Why should the set of possible case tags and the set of possible added sumti tags be different? This in effect suggests two different case systems are at work in one language. 2. Why should the set of cases be artificially constrained to a small number of cases, so broadly based that they are no longer intuitive? Thus we researched the current state of case theory (actually pc did). His conclusion was that there was no single case theory that had achieved predominance, and that researchers could not agree on any set of cases that was necessary and sufficient for all aspects of language analysis. Can Lojban, which stresses a minimum of metaphysical assumptions, impose a big one by specifying a limited number of cases that semantically covers all language needs? No! Especially when the sassumption would be one that neither linguists nor AI researchers can agree on. Not only is this philosophical against the underlying spirit of the language, but a simple demonstration shows it to be unworkable. I believe that jimc will find it equally unworkable for his diklujvo. Given that each gismu has a set of places, each of certain cases, what happens when you make a lujvo of any two. You get a place structure that in some way combines the two place strcutures of the separate gismu. Some places will be eliminated because the lujvo-making in effect allowed some variety of one of the terms to fill a place in the other term's place structure. In general, however, the place structure is likely to end up longer than either original. Now make a 3-term lujvo by combining this two-termer with a gismu. The result: still more places. Since there is no theoretical limit on the number of terms in a lujvo, then for any finite number of cases, you will eventually come up with a lujvo that have more places than there are cases. THus for Lojban, either there is no sufficient set of cases or you have to use one case for more than one place of some lujvo. I contend that this makes any case theory unworkable when applied to Lojban as a human language, because the primary use of cases would be to help keep the semantics of the place structures straight. (AI processing can probably allow for multiple places of the same case, and indeed Lojban can grammatically do so using subscripting. But this becomes a mechanical method of labelling and tracking places in manipulation more than it does a direct access to the semantics.) I'm willing to be told or proven wrong on this, but I still think the avaoidance of metaphysical assumptions is important. When we are so unsure of the place structures that alone of all features of the language we do not intend to baseline them - if for no other reason simply because of the impossibility of a comprehensive and consistent place structure analysis to be completed before we put out a dictionary. But a better reason is that we don;t yet really know what the Lojban words mean. If Lojban is not to be an encoded English, then the last stage of determining the meanings of each word will have to be done IN LOJBAN, without translation to English. Until then, we won;t know what the exact range of meanings covered by each word should be, and hence will not know the exact range of meaning of each place structure place. How can we before then, perform any systematic analysis of the places that will be binding after that in language analysis. And of course meanings will drift, and so probably will place structures. If so, then a pre-analysis of the places will inevitably be wrong, possibly before the 5 year 'freeze' is over. Indeed, I can say that I HOPE this is the case. lojbab