From cbmvax!uunet!PICA.ARMY.MIL!protin Wed Aug 7 14:28:39 1991 Return-Path: Resent-From: cbmvax!uunet!PICA.ARMY.MIL!protin Resent-Message-Id: <9108071636.AA07914@relay1.UU.NET> 6 Oct 90 0:21 EDT To: lojban-list Subject: response to jimc on indicators #2 Date: 5 Oct 90 22:43:17 EDT (Fri) From: cbmvax!snark.thyrsus.com!lojbab Message-Id: <9010052243.AA17262@snark.thyrsus.com> Resent-Date: Wed, 7 Aug 91 12:32:15 EDT Resent-To: John Cowan Status: RO A response to Jim Carter in our continuing exchange on indicators: JC>For example, in typical (Indo-european) natural languages you have special grammar for timelike tenses, for gender, for number, for possessive, and so on ad ridiculosum. In Loglan most of this deadwood was pruned out. But meaning remained in a few areas, specifically, in attitudinal indicators. I'm not sure I agree on this. Loglan did not eliminate these in the grammar - it made the structurse optional. Lojban tries to continue the same practice. You CAN express everything in simple predicates, but you NEED NOT. I think I can summarize a lot of our agreement as being based on a different understanding of language. jimc divides language into grammar and semantics, and tries to put all 'meaning' into the semantics. My understanding of language includes recognition of pragmatics and supersegmentals, even though I don't know nearly enough about either to speak authoritatively. The essence of what I feel Lojban needs that jimc does not allow for derives from these areas. I DO NOT accept that the word 'Ouch' is identical semantically to 'I feel pain', or indeed to any other predicatible expression. On the other hand, you can say of x, who says "Ouch", that 'x indicates the emotion of pain', or, if you have no reason to suspect emotive acting: 'x feels pain'. But if x IS acting, the first is true and the second false. If indicators meant the same as the corresponding predicates, acting would consist of making false statements. I think our hangup here is simply that I will not accept that an indicator is an 'abbreviation', a shortened but otherwise identical form. A bridi that talks about one's emotions is not the same as the direct expression either logically or semantically. There are many places in the language where short forms are not the equivalent to longer forms. The one that comes up most is "lo .urnicorni cu barda" (Some unicorn is big) which is not identical to the more logically explicit "da poi .urnicorni cu barda" (something which is a unicorn, is big.) The latter logically implies that there exists a unicorn. The former is more like a conditional (If things exist that are unicorns, some are big.) But the former is NOT an >abbreviation< for a conditional, it is logically equivalent to one (I think). JC>While -gua!spi has literal quotes, the normal form for dialog is unlike English or any other natural language I am familiar with: the spoken material is put at the top level as it would be in the script of a play, and the speaker and listener are specified with -- what else? -- subordinate clauses, which can be positioned creatively. I should note that we started with jimc's suggestion along this line when we first started on Lojban, but it is flawed. You need some type of metalinguistic indicator to distinguish citations in quotes. Else "The lion (John said) ate the meat." works fine, but at the next level of quote, you have problems. If George wants to say: Mary said, "The lion (John said) ate the meat." he would need two inserted clauses to show the hierarchical citation, and I suspect that it is difficult to unambiguously isolate whether he was claiming that Mary says that John says, from John says that Mary says. Related to this, if subordinate clauses are embedded in a quote, you cannot tell whether they are part of the quoted text, or the statement of citation. You also run into the problem of distinguishing direct quotation from indirect quotation. I note by the way, that I don't consider it a virtue that a language to be used by humans does things 'unlike any natural language'. The only 'unnatural thing' in Lojban is the predicate grammar itself - we want the rest to be as natural as possible. JC>The same formalism is feasible in Loglan/Lojban; instead of clauses, "tagged sumti" would be used. It gives a very lively style to the dialog, I find, and is much easier to produce and to analyse (particularly algorithmically) than the usual English scheme with imbedded literal quotes. Yes, and we have that structure. But for indirect quotes, not for direct quotes. To indicate that something is a direct quote with an embedded tagged sumti, you need to insert a metalinguistic 'unquote', and then mark that this metalinguistic insertion is indeed metalinguistic, and not part of the quote. This is done with the editorial insert marker sa'a. There are examples of this in Athelstan's Saki translation. Even with sa'a there is a paradox possible if someone wants to quote someone else using sa'a in a sentence, and we had to come up with a convention for this - namely that you have to double the sa'a if you want to actually quote it similarly to how some editors require doubling a delimiter if it is embedded in a text command. jc>Since I pay a lot of attention to semantics, I have trouble with this particular example; how does the listener know that "x likes (z more than y)" means the same as "(x likes z) is more intense than (x likes y)"? In other words, how do you create unambiguous rules that guide the listener to backmap the sumti tcita "more than" to the related selbri, and then (this is the hard part) jack that selbri up to discourse level, replicate the original discourse bridi twice with z and y in X2, and drop the results in X1 and X2 of the comparative selbri? The easy answer is that you don't. We don't claim that Lojban has unambiguous semantics, or that any two different grammatical constructs can be equated semantically. Indeed, I suspect that in natural language, if two constructs exist, they are never quite identical semantically if only because there is a different emphasis. On the other hand, if you attach a sumti tcita to a sumti as a relative phrase, as in "x likes (z more than y)" [ko'a nelci ko'e ne semau ko'i] the phrase can be expanded to a parallel clause which is logically equivalent (if not semantically): [ko'a nelci ko'e noi ko'e zmadu ko'i zo'e] "x likes y (and incidentally y is more than z in some property - which context and convention will say is 'the amount of being liked by x'. Each such sumti tcita will have its own semantic conventiosn of this sort, which will probably be fairly systematic in most usages. We thus DO consider each sumti tcita to be equivalent to a corresponding bridi, but with only 1 place (the tagged sumti) explicitly specified, and the others elliptically omitted. JC>A logical language ought to have its semantics specified just as well as the grammar is. Perhaps, but I don't think that we understand the semantics of human languages well enough to do this, since grammar too often gets in the way. Studying users of Lojban, we may fianlly have a clear window on what is actually going on semantically. Then, someone could possibly define some rigid rules.