From a.rosta@xxxxx.xxxx Mon Jan 3 08:45:54 2000 X-Digest-Num: 328 Message-ID: <44114.328.1779.959273825@eGroups.com> Date: Mon, 3 Jan 2000 16:45:54 -0000 From: "And Rosta" >But if I wanted a logical lang that was only going to be used > >in-house, then numerous modifications to Lojban could be made to > >make the language easier to learn than Lojban, and yet as > >effective. Basically, the main modifications I'd make would > >involve discarding tons of stuff and changing the gismu forms to > >lightly modified English. > > "Discarding tons of stuff" means using a language subset. I wouldn't > expect usage of Lojban for patents to do anything other than use language > subsets. [...] > >I disagree on this. It depends on the nature of the more focused problem, > >but for the patent/legal problem, I claim that the lg could be done > >quickly, possibly be using parts of Lojban. > > And I have no qualms against using a language subset. It is changes and > not subsetting that would hurt the project. Use of a language > subset in a top down application will cause people to look at the rest of > the language for other, less constrained applications, including those where > a full range of communication is more desireable. Ah, well then: we agree that a Lojban-subset would be a fairly good solution. For me, a Lojban-subset is not Lojban, so my point that Lojban is not the ideal choice holds good. For you, what counts is that this would still be a major use and application of Lojban, and would contribute to Lojban- proper reaching critical mass. > As for the gismu forms, as I said: you are reproposing "Anglan" > and the advantages people see in the idea just don't pan out in actual > trial. I really find it hard to believe that Anglan wouldn't save loads of effort in learning vocab. > If you change the morphology, of course, you lose the > self-segregating aspects and thus spoken/written interchangeability. True. For an ideal loglang, I prefer self-segregating morphology. For any kind of IAL that actually has to get widely learnt, I think a more transparently a posteriori lexicon would be preferable. > > > By this argument, the efforts to develop an interlanguage for machine > > > translation would need only use predicate calculus. But instead the > > > closest that anyone has come to successfully using an > > > interlanguage was the DLT project that used Esperanto. > > > >I don't know why this was. Maybe they wanted to translate aspects of > >form as well as meaning. And very probably it was easier to work out > >the mapping from natlangs to Esperanto than from natlangs to something > >as stark as predicate logic form. > > I would contend that form sometimes is part of meaning. We know for > example that sumti/object ordering can put convey emphasis. > > But it is precisely the problem of working out mappings from natlangs to > predicate logic vs. Lojban that makes the difference. To map to > predicate logic, you have to explicitly think of things that are subliminal > in Lojban, like prenex/quantification issues. I'm skeptical about whether the difference really is "thought about subliminally" versus "thought about explicitly" as opposed to "thought about" versus "not thought about". > There is also brevity: I believe pc once said that quantifiers other than > the standard logical quantifiers in pure predicate calculus form get > extremely verbose. True, but not always a bad thing, in that explicitness can be preferable to brevity in some circumstances. > That is why learning something as a natlang or quasi-natlang is > critical to these sorts of applications. It is a matter of speed and > confidence that you've done it correctly. Programming languages come close > but fall a little short of the mark of knowing you've done it right just by > writing it out, as well as of fluency in writing. > > And the fact that not all programmers are good lawyers and vice versa > suggests that there is a distinction in skills between the kinds of > accuracy in logic that are needed. My experience of hearing lawyers talk about language is very limited, but depressing and deeply troubling. I think a language that forced its users to think in appropriate ways would be better than a more accommodating language that allowed users to continue to think in appropriate ways. > > > > In a Polish/Reverse Polish predicate > > > >logic notation ... > > > >Likewise, the entire syntax could be formulated in a single > > > >sentence. > > > > > > But could human beings use it to describe a patent? > > > >I don't see why not. > > I said human beings. And maybe I should have included "read" a patent > too. Human beings wrote Principia, but very few human beings can read it > with any understanding. I confess I have no intention of ever trying to read the Principia, so I don't know how readable it is, or even in what language or notations it is written in, but I am doubtful that the difficulty lay with the notation rather than the content. > > > And what happens when you need to translate an indirect > question? After > > > all, haven't you just found that it is a fairly intractible > problem for > > > predicate logic? > > > >Hopefully it is tractable. But I'd use a special WH quantifier if the > >language had to be done today. > > From what pc has said, different schools believe different things, and > that in itself means that the problem is intractable at a logical > level. At a language level it is tractable because language is more > tolerant of ad hoc conventions than predicate calculus. I was thinking that predicate logic notation makes a good foundation for a language because of its simplicity rather than because of its consensuality. > > > > If so, I guess that they problem with predicate calculus is > > > >that there's no fudgeability with it, which nonfudgeability > > > >is exactly why one wants a logical language. > > > > > > But fudgeability is fine for patent translation (maybe even > > > desirable to the lawyers), so long as the fudguing does not create an > > > ambiguity comparable to those of natlangs. > > > >By fudgeability, I mean the possibility of being ambiguous when you > >can't be bothered to disambiguate, or when disambiguation is more > >trouble than it's worth. > > Well then fudgeability is IMPORTANT for legal translation, because there > are places where lawyers WANT to leave meaning a little ambiguous. > Ambiguity can allow two sides in disagreement to come to terms, > leaving the ambiguous areas to be worked out later if necessary. > You need to be arbitrarily precise in some areas, but Lojban's allowing of > ambiguity when precision is too much trouble (or agreement is impossible) > is an advantage. Correct me if I'm wrong, but I think that with the exception of a few special cases, every Lojban sentence can unambiguously be translated into a pure predicate logic counterpart. I don't think Lojban has any mechanisms for fudging logical meaning. > > > You learn the forms of predicate logic, more or less (but how many > > > people actually use the full set of Lojban logical connectives, for > > > example?). But you do not learn to reason according to the rules of > > > inference along the lines of the predicate calculus. > > > >True, but that's also true even if you learn pred calc notation. > > I think though that you have to learn to reason with predicate > calculus in order to accurately write in it. You also need to be able to do > so with Lojban. In looking at the indirect question discussion that you > recently conducted, I can imagine identifying and discussing the logical > problems with everyone's formulations entirely in Lojban. I cannot > imagine doing so entirely confining ourselves to predicate notation. Maybe > some logicians can, but not me. I suppose that if you insist so strongly I should believe you. Still, I could write in something that would be both a version of predicate notation and proper Lojban. Would you predict that you would be unable to understand that? Maybe yes, but in that case it would be the case that only certain subsets of Lojban are comprehensible to you. > > > > And the original idea that a loglan-speaking community would test > > > > sapirwhorf, I've always regarded as a bit of blarney baloney by JC > > > > Brown who really wanted to invent a language but was trying to > > > > (a) gain respectability for an ill-respected activity, (b) > > > > differentiate the product from others, (c) attract adherents. > > > > > > I know the history enough to be sure that it was not blarney at the > > > time. Remember that in the mid-50s, testing SWH was a big deal. > > > Remember also that JCB came from the Campbell school of science fiction > > > which I think had a certain amount of SWH built into it. He does seem > > > to have conceived of Loglan before the SWH became big, but I think that > > > he seriously wanted to make the language a research tool. > > > >You know the history better than me. The main reasons why I came to the > >above conclusions are firstly that it seems improbable that anyone > >would seriously propose Loglan as a psycholinguistic experiment, since the > >experiment is so uncontrolled and it takes so much effort to set up, > > I think that JCB himself has a fundamental lack of understanding of the > problem of controls, as well as the nature of scientific testing. If he > had understood, he might never have started %^). Look at Chapter 7 of the > now on-line L1, and you can see what he envisions as a SWH test. But the > problems are tractable, IMO - they just need a LOT more thinking out. > > There were similar criticisms of JCBs scientific procedures during the > redesign of 1979-84. JCB conducted "taste tests" with Loglan users to > choose among various options for affix formation, but his experiments were > uncontrolled (not to mention only performed by English native speakers), > and his statistical conclusions were apparently, umm, skewed to get the > answers that he wanted. McIvor wrote a devastating criticism of the > methodology which was a key trigger starting of the political split; McIvor > later reconciled with JCB of course. I won't say any more, since he can > speak for himself if he wishes (I think he is subscribed at the > moment - RAM?) > > > and secondly that > >thereafter JCB showed himself much more interested in all the designing > >and tinkering and so forth that conlangers are so familiar with, and also > >introduced additional design goals. If he'd been serious, he could have > >genuinely invented a speakable predicate logic in a short span of time, > >and then, say, raised funds to pay people to learn and use it or to > >participate in experiments that oblige or incentivate them to learn and > >use it. > > Well here is where the history comes in. JCB did try to do these things. > His original Loglan, the 1956 version, "rattled around in people's > heads" (Cowan quoted this more accurately than I did), (I've never seen a > copy of the 1956 'book', of which only a couple hundred mimeo copies were > made), and people couldn't learn or use it effectively. So he started > adding flesh to the language, developed his peculiar learnability schemes > (the word making algorithm is something he actually claims to have tested > on groups of students, unlike most other aspects of the language, though > again we have no idea what typo of controls he used). I've done JCB a partial injustice, then. All the same, this first phase strikes me as the one that, had it been done thoroughly, would have been the most instructive. For example, was the problem really one of excessive simplicity, or was it the cumbersomeness of too few abbreviatory devices, or was it the fudgeabilitylessness making too many demands on the speaker's cerebrative powers? I doubt that it was the simplicity per se that was the problem. The cumbersomeness probably would be a problem. But the extent to which the fudgeabilitylessness was a problem is most interesting, because essentially logicality and fudgeabilitylessness are the same thing. Hence introducing fudgeability is a way of relaxing the requirements of logicality, and the need for fudgeability would show that speakers couldn't cope with a thoroughgoingly logical language. I suspect that even though Lojban is defined so that every sentence has an unambiguous logical meaning, its accommodations of more familiarly natlang-like forms will allow speakers to communicate in Lojban effectively without actually mastering its logicality. That is, its logicality is something that will be circumvented. Hence I predict that speakers' ability to communicate with each other in Lojban will prove nothing interesting. > With his first wife, JCB apparently did a fair amount of serious language > work more aimed at teaching rather than fiddling in the early > 60s. In many ways I think that we he went through then was something akin > to what we went through in the few years of writing the Book - questions > came up which he made decisions on, but he did not have the advantage we had > of an Internet community to poke holes in his occasional 'brain farts'. He > published the first L1 in microfilm in the mid 60s, which was the first > chance more than a couple people had to actually look at what he was > working on, but few changes were made in the language until after 1974. > > The book publishing in 1974-5 shows that he probably thought that the > language as "done" then. He DID stop fiddling, and started research grant > proposal writing to get funding for SWH experiments and such. Put > politely, his 1976 proposal was mabla. Meanwhile the computer people > showed him that the language wasn't as unambiguous as he had thought, the > attempts of the earliest to try to write in the language showed copious > flaws, and then the fact emerged that he had screwed up the > "pretty little girls school" analysis and omitted needed cases. Meanwhile NSF > told him the proposal was inadequate, and JCB reacted by producing two more > proposals that were even worse (not having worked in academia for 15 years, > he apparently did not know the standards for proposal writing), and then > declared war on the NSF evaluators, using its formal protest procedure to > the utmost while having no valid issues. That was the end of the > possibility for JCB/TLI ever conducting funded research. > > I think that JCB turned to language fiddling only because the research > aspect was closed to him and the fiddling was where the action was in the > community that was left. I think he has always wanted to see people using > his language, as the necessary step to wherever, and the fiddling was in > his mind probably kept to the minimum necessary to gain usage (as well as > to keep control once we started outdoing him). TLI's attitude towards > change was MUCH more conservative than LLG's until we baselined. I think > JCB honestly thought the language was done in 1974 and then again > in 1984, but in both cases his community showed him he was wrong; now he > cannot commit to saying it is done, because at that point he will lose > control (actually, he HAS lost control, because we exist and thrive > relatively speaking, but TLI has nothing left once JCB ceases to be the hub > of the wheel of change). Very interesting indeed. This should all be part of a proper documentation of the Loglan project. > >I'm probably displaying my ignorance of patent write-ups then. I was > >assuming that all that counts is their pure content, and that issues > >of style and the structures used in the source language are irrelevant. > > They might come to be. But at first the patent application will be as an > interlanguage for translating patents that were originally written in other > languages. This seems relevant only if a translation into Lojban has to preserve ambiguities in the original. That could be difficult. > > > Finally, looking at predicate calculus (or *any* calculus, for that > > > matter) is misleading because it ignores the question of *vocabulary* > > > (a point lojbab makes below). Try programming in Prolog with no > > > libraries. Sure, you can write quick-sort in a couple of lines, but what > > > are you going to *do* with it? And how do you map it to the real world? > > > > > > Lojban has that in the form of gismu - predicate calculus does not. > > > >It's precisely because this is an area of noncomparability that this issue > >is a red herring. Pred logic notation is proposed as an alternative to > >Lojban syntax and cmavo. As I originally said, you need predicates. I > >don't think Lojban gismu are adequate, because they're not yet defined, > >but if you think they're adequate, then the Lojban gismu could be used > >as the predicates in the language I'm mooting. > > If Lojban predicates are not sufficiently defined, I fail to see how any > alternative approach would be better. That wasn't the claim. I was saying that an alternative approach would in this respect not be worse. This was in response to you and Brook saying Lojban would be better. You yourself recognize that a legal language would have to define its own predicates: > > > > Of course the predicate words' senses have to be defined. But in > > > > Lojban the predicate words' senses are not defined -- this task has > > > > been left to 'usage' to achieve. > > > > > > And patent translation would be a large amount of usage. > > > >Exactly. I don't know anything about patents, but a great deal of law > >involves decreeing definitions of terms. > > Yes. And you can define Lojban predicate words, so that is not a > problem. Law would not rely on "usage" in Lojban any more than > it does in English. Important terms will be carefully defined. Quite so. > You mentioned using English-like gismu earlier, but that just reduces to > the problem of translating into legal English. The advantage of English-like gismu, in cases where the learners are anglophones, is that the gismu forms are easier to learn/remember. > English has ambiguous grossly polysemous words. Legal English has > conventions to deal with English grammatical ambiguities. Legal Japanese > has different and incompatible conventions in both words and grammar. To > write something intelligible to both requires choosing something that > tackles both sorts of ambiguity. In addition, for the European world, > English-like means un-sellable to the French. My point is nothing more than the one I made earlier -- that for most IAL purposes, more transparently a posteriori vocab would be better than what Lojban offers, all things considered. > > > > In other > > > > words, setting aside how variables are handled, you could have > > > > a language with only 3 cmavo! I'll admit that that number might > > > > be expanded a bit, e.g. to include numbers, but even an expanded > > > > cmavo inventory would be only a tiny proportion of Lojban's. > > > > > > A perfect example of going too far towards minimalism. Sure, > > > AND, OR, and NOT are all you need to make any other truth function > > > but you better believe people that build ICs for a living don't > > > recreate a one-bit adder from AND, OR, and NOT every time they need > > > one - they don't even use one-bit adders - they pull more useful > > > things from a library - 32-bit add, multiply, etc. and a whole lot > > > of other stuff. > > > >I can't think of which Lojban cmavo represent huge savings in > >convenience over a combination of predicates and basic connectives > >and quantifiers. > > The quantifier 5/mu is much briefer than the predicate logic method of > saying "5". I wouldn't treat numbers as quantifiers, but I would admit them as a distinct selma'o in a logical language. > Lojban's set selection operators and massifiers are much more flexible than > logical objects. I don't mean to challenge your assertion, but can you spell out more fully what you mean? > (pc established long ago that logical connectives are ineffective for > truth-tables of 4 or more elements and aren't that great even for 3). Does Lojban have logical connectives for truth-tables of more than 2 elements? At any rate, your point reminds me that I would subsume logical connectives into the class of predicates, and so since the class of predicates is open, so is the range of logical connectives that can be defined. > And then there is tense. I've looked at pc's book on tense > logic. I will never understand it. I don't see why tense can't be handled straightforwardly by the existing apparatus of predicates and quantifiers. > > > What? Okay, lojbab is hard at work on a dictionary (in his, I'm > > > sure, copious spare time) but gismu are as well defined as any word > > > you care to name in any language - better, in fact, because they only > > > ever have *one* definition (try looking up "run" or "fork"). > > > >Either you have in mind some notion of "defined" that I cannot apprehend, > >or what you say is wrong. Lojban Central correctly declares that gismu > >aren't defined and that definition will be left to usage. In other > >words, in its gismu definitions Lojban will operate like a natural > >language rather than an invented language. > > The place structures are baselined. But this I think is incidental since > the gismu alone are not adequate to a legal language. So there will be > need to coin new words and define new meanings in any event, and that > process will mimic the natlang way. > > > In natural languages, usage has already defined the meanings of words, > > For any given language. But the translation problem makes these > polysemous definitions less than ideal. I also think that natlang words are > often less well-defined than some Lojban gismu (in that you often don't > even know the implied place structure without contextual interpretation). That's a problem of ambiguity rather than undefinedness. > > and on them there is intersubjective agreement, such that they are a fit > > subject for rational debate and quasiempirical research (in the subfield > > of linguistics usually called 'lexical semantics'). > > Which field is understood and followed by lawyers, right? I wonder if any > percentage of lawyers have ever heard of Montague. No reason you should know this, but Montague was a formal semanticist rather than a lexical semanticist. Formal semanticists are concerned with 'logical meaning', while lexical semanticists are concerned with 'lexical meaning' (the senses of words). The little evidence available to me would indicate that lawyers are in fact shockingly ignorant of lexical semantics even though they practise a variety of it themselves. A good legal language should have an apparatus for defining words and for indicating how well-defined words are. Anyway, my point was that natlang words *are* fully defined, but not that these definitions are adequate for legal language; they're not, because they're not explicit enough and not precise enough. > > > Now, unless you believe that "subject", "verb", and "object" are > > > hardwired into brains, I'd submit that a young child exposed to a > > > fluent lojban speaker could pick it up easily enough (I'm not fluent, > > > but my three-year-old daughter seems to get the hang of lojban easily > > > enough). > > > >I don't understand. Since we agree that a subset of Lojban is a close > >approximation of predicate logic notation, surely if a child picks > >up Lojban, they have picked up predicate logic notation. > > No. They have picked up a linguistic manipulation of predicate logic > notation. I doubt that his daughter could read the notation (if she can > read at all). We may be getting into an issue over the primacy of speech > vs. writing here. There is a subset of Lojban that is an approximation of a speakable form of a version of predicate logic notation. There are official rules of Lojban that can translate any Lojban sentence into such a subset. If you master Lojban, then you master those rules, and hence you master the subset, and hence you master a form of predicate logic notation. > > > Um, but you may also recognize that "Better is the enemy of good > > > enough," and see the reduced costs of using something in existence, > > > even if it isn't quite what you were looking for. > > > >I do. > > I realize and appreciate that. At one time, you were our pet > iconoclast. Now you write Lojban. Not for an awfully long time, I'm afraid. Not enough time (and possibly not enough enthusiasm). I've started to forget stuff, in fact. > > > Flexibility isn't necessary for describing *inventions*? > > > >I wouldn't have thought so. By flexibility, I was thinking of "having > >many ways to say the same thing". Not "being able to say lots of > >things". > > Flexibility in being able to say things at an arbitrary (but variable) > level of specificity depending on the relative importance of > precision vs. ambiguity. For example, a patent on exercise equipment (say > a treadmill) may need to distinguish precisely distinguish between "run" > and "walk", but a patent on a procedure that involves a person travelling > on foot does not need to. For this kind of flexibility you just need appropriately defined predicates. But this is not what I meant. Again, your example is one of being able to say different things. An example of what I mean is the freedom of word order in Lojban, and its wellknown one-bird-with-many-stonesiness. > > > And cultural neutrality seems like a very desirable trait for patent > > > description in the *European Union*! Even more so for global > patent uses. > > > >For global, yes. But the EU is very eurocentric and there is no tradition > >whatever of cultural neutrality; rather, europeanicity is celebrated at > >every opportunity. > > But the EU has to deal with patents written in non-European languages, > which is where this thing started. Hartmut Pilch mentioned the > difficulty of translating Japanese patents into European languages, none of > which was particularly compatible with some of the problems in Japanese > language use. Lojban is MUCH better than any European language in > representing ARBITRARY language features unambiguously. IIRC, this cultural neutrality point arose with regard to the etymologies of the vocabulary. I don't think the europeanicity of IAL etymologies has any bearing on how easy it is to translate out of Japanese. > > > If you pared Lojban down to the smallest adequate portion you'd still > > > > be left with unnecessary stuff (e.g. zo'u, terminators) and > > > what remained > > > > > > Your particular examples of zo'u and terminators seem again, > perhaps not > > > strictly *necessary* but so useful to merit inclusion. > > > >Again, I don't see why, but I guess that would be more appropriate for > >discussion on some generic loglang list. > > There isn't any such thing, and I think expanding this could be > informative (outside the context of multihundred line posts on an unrelated > thread %^) Well, we've re-had the discussion many times about whether to split the list. All I can do is point out to you how many more people stay subscribed to this list now that I no longer post very much. > > > > Second, if it is possible to do better than Lojban, with the same > > > > set of goals, this is largely because it is possible to learn > > > > from Lojban's 'mistakes', i.e. it is by standing on Lojban's > > > > shoulders that Lojban can be bettered. > > > > > > Please, do so! :-) > > > >I'm not clear why you keep responding in this way. If you had a genuine > >idea to explore the way in which we think Lojban might be bettered, > > I think such a topic is germane to Lojban List (so long as we don't > actually try to change it). Knowing limitations and flaws is a useful > thing, especially since Lojban will see no end to criticism whether it > catches on or not. It's not so much 'limitations and flaws' -- it's more a question of, say, how we would do things differently if we could start over, knowing what we now know. Take Lojban phonology, for example: I don't think it has 'limitations', and there's nothing that a majority of us might consider a flaw, but nonetheless if we were starting from scratch we'd probably do it differently. > > > > Also, in a certain sense, it has been proved that it is easy to do > > > > better than Lojban, because over the years people have often proposed > > > > valid improvements that were not adopted (on the grounds that > > > > completion was a more important goal than improvement). > > > > > > A cost/benefit tradeoff - did the change improve things enough to > > > destabilize the design? I wasn't there, but I'll bet it didn't. > > > >Whether it did or didn't, that wasn't an issue. However much it did improve > >things, it would only be permitted to destabilize the design if the > >design were shown to be failing to meet the language's explicit goals > >(e.g. if a hitherto unnoticed syntactic ambiguity were discovered). The > >rationale for this was as I stated before. > > In practice, if you look at the set of changes that were incorporated > during the grammar baseline, you will see that we did add things > that were not "broken". My memory's leaky, but I don't recall much if anything being added since 1992 or so that was not to fix something quasibroken. {ce'u}, for example, which might have been one of the last cmavo to be added, is not so much a useful augment as a plug to fill a gap that ought to have been filled already. --And.