rwojcik@bcsaic.UUCP (Rick Wojcik) (09/24/89)
In article <567@ariel.unm.edu> bill@wayback.unm.edu (william horne) writes: >This example is relavant to AI, because it questions the validity of the >Turing Test as a test of "understanding", as well as questioning the >legitimacy of rule based systems as models of intelligence. One serious flaw in the Chinese Room Problem is that it relies on the so-called 'conduit metaphor' (originally described by Michael Reddy in A. Ortony's _Metaphor_and_Thought_ Cambridge U. Press 1979). That metaphor assumes that meaning is essentially contained in the linguistic expression. A logical consequence of this belief is that one can devise a set of principles for translating from one language into another without losing any of the semantic 'stuff' that a linguistic expression conveys. The conduit metaphor is very powerful and useful as a means of illuminating the behavior of language, but, like all analogies, it breaks down. Those who deal with real language to language translation know that there is no one-to-one match between expressions in one language and those in another. An alternative view of linguistic communication is to assume that linguistic expressions merely help to shape the flow of mental pictures (alas, another metaphor :-) that constitute the end product of communication. Therefore, there is no necessary one-to-one correspondence between linguistic expressions in one language and those in another. The trick to translation is to construct expressions in the target language that evoke the same thoughts as those in the source language. And this may even be impossible without modification of the target language (i.e. the creation of new words to fit new experiences). So I claim that the Chinese room problem rests on incorrect assumptions about the nature of language and understanding. -- Rick Wojcik csnet: rwojcik@atc.boeing.com uucp: uw-beaver!bcsaic!rwojcik
sp299-ad@violet.berkeley.edu (Celso Alvarez) (09/26/89)
In article <15157@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: >. . . The trick to translation is to construct expressions in the >target language that evoke the same thoughts as those in the source language. Much more than thoughts are evoked by language. How do you translate the signalling of identity, roles, and social relationships? >And this may even be impossible without modification of the target language >(i.e. the creation of new words to fit new experiences). So I claim that the >Chinese room problem rests on incorrect assumptions about the nature of >language and understanding. I'm not familiar with the Chinese room problem, but where do you/Searle leave the question of interpretation? There is more to language than understanding. Celso Alvarez sp299-ad@violet.berkeley.edu
rwojcik@bcsaic.UUCP (Rick Wojcik) (09/29/89)
Celso Alvarez (CA) writes: me>. . . The trick to translation is to construct expressions in the me>target language that evoke the same thoughts as those in the source me>language. CA> Much more than thoughts are evoked by language. How do you translate CA> the signalling of identity, roles, and social relationships? I think that such concepts have to be represented as thought structures, since they have an impact on language structure. But your question may be filed under my general question: Just what do 'Chinese Room' debaters think a translation is? What criteria do you use to judge that a translation from one language to another is successful? My position is that there is no such thing as translation in an absolute sense. A seemingly trivial example is the translation of expressions that refer to language-specific grammatical structure. Thus, there is no way to translate French 'tutoyer' directly into English. You must rely on circumlocution. It means roughly 'use the intimate 2nd person singular form of the verb'. But practical translators might take an equivalent French expression to 'Don't tutoyer me' into English as 'Don't use that tone of voice with me', or some such thing. But it is difficult to say what makes one such translation better than another. People can get into heated arguments over such questions. N. Boubaki (NB) writes: >...Those who deal with real >language to language translation know that there is no one-to-one match >between expressions in one language and those in another. NB> But this difficulty would affect the native Chinese speaker and the NB> Chinese Room Demon equally. That is one premise of Searle's NB> argument - the "mechanical" system is presumed to be just as competent NB> (not necesarily perfect) at translation as the "understanding" system. I know, but I think that Searle, like most of us, has implicitly adopted the conduit metaphor in his conceptualization of the problem. He really believes that there is some absolute sense whereby an expression in one language corresponds to one in another. This seems clear from his insistence that the translation itself be 'mechanical'--in other words, symbol manipulation. Those involved in translation understand that the translation process requires editing and revision. Who determines that the "mechanical" system is "just as competent" if there is no mechanical basis for judging competence? But that is just what you need to do in order to bring about translation. You need mechanize the ability to judge and revise. That would be tantamount to mechanizing the understanding process, since it is only by understanding expressions in two different languages that you can judge their equivalence. I want to be careful to distinguish modern Machine Translation efforts, which do not attempt to automate the revision process (rather they attempt to facilitate it), from an ideal MT system, which would require mechanized understanding to do its job properly. So I agree with you that Searle ultimately begs the question. The question is whether or not 'understanding' is a mechanizable process. He either assumes that it is not, or he doesn't have a proper conception of the nature of translation. Ray Allis (RA) writes: RA>It seems to me your position is in fact very close to Searle's. The problem RA>I have with his little parable is that he pretends that the output from RA>the Chinese room is satisfactory (or rather lets us assume so). I believe RA>that if the room does not "understand" Chinese, and he argues that it does RA>not, the output will not be satisfactory... From my above remarks, you should see that I am closer to your viewpoint than Searle's. In fact, I find myself largely in agreement with most of what you said. I would only quibble on the issue of whether or not modern NLP efforts, including MT, are futile. The pragmatic purpose of such work is to increase human efficiency in language-intensive work on computers. There are many good things you can do without addressing the need for full language understanding. MT (really Machine-Assisted Translation) can improve the output of a human translator, even though the MAT system may produce some pretty bad translations. Our grammar-checking system is proving useful in the writing of aircraft maintenance manuals. But this takes us away from the philosophical question of whether or not you can mechanize language understanding. -- Rick Wojcik csnet: rwojcik@atc.boeing.com uucp: uw-beaver!bcsaic!rwojcik
rwojcik@bcsaic.UUCP (Rick Wojcik) (09/29/89)
In article <822kimj@yvax.byu.edu> kimj@yvax.byu.edu writes: >Could you elaborate what you mean by "the semantic stuff"? Say I translate >"kick the bucket" into "die" in Chinese. Does the translation lose what >you call "the semantic stuff"? I want to recommend to you Ronald Langacker's tour de force Foundations_of_ Cognitive_Grammar. v.1. Stanford U. Press, 1987. I particularly call your attention to the discussion on and around p. 93, where he lays out a clear distinction between literal and figurative senses. He argues quite convincingly that you can take neither a purely compositional, nor a purely conventional, approach to meaning. I do not know how his work, and that of other 'cognitive grammarians' will end up affecting the world of computational linguistics, but it does help to point up many areas for future research. I do not think that there is any precise way to translate 'kick the bucket' into Chinese, and I don't think that the opening scene of the movie 'It's a Mad, Mad World' can be properly understood by Chinese speakers, even in its dubbed version. (That scene has a great sight gag involving the 'kick the bucket' idiom.) Semantic stuff is very often lost when idioms get translated. But it is the compositionally-derived stuff that gets lost, not the conventional stuff. -- Rick Wojcik csnet: rwojcik@atc.boeing.com uucp: uw-beaver!bcsaic!rwojcik
dmocsny@uceng.UC.EDU (daniel mocsny) (10/01/89)
In article <15336@bcsaic.UUCP>, rwojcik@bcsaic.UUCP (Rick Wojcik) writes: > What criteria do you use to judge that a translation from one > language to another is successful? How about: the same criteria (note: plural) we use to judge whether native language speakers can communicate with each other successfully. > My position is that there is no such thing > as translation in an absolute sense. Two languages aren't even necessary. Two people who speak the "same" language can misunderstand each other. Two translation steps are already going on there---from the speaker's thoughts into a serial symbol string, and then from the string to the hearer's thoughts. If the hearer's thoughts differ substantially from the speaker's, then the translation has failed. However, I think absolute translation *must* be possible in principle, unless we believe that the human mind has an infinite information content. That is, if we view communication as a thought-transfer between two thinkers, then some finite serial data stream must represent the thoughts of the speaker in sufficient detail to allow the hearer to reconstruct them with arbitrary accuracy. We may not know how to move thoughts from one person to another as one would copy files between computers, but the materialist assumption says it must be possible. (If the brain turns out to be not a very convenient medium to "write" on, then one might have to resort to physically reconstructing features of the sender's brain in the recipient. "Let me give you a piece of my mind..." This won't be an easy trick, but it can't be impossible.) > A seemingly trivial example is the > translation of expressions that refer to language-specific grammatical > structure. Thus, there is no way to translate French 'tutoyer' directly into > English. You must rely on circumlocution. It means roughly 'use the intimate > 2nd person singular form of the verb'. But practical translators might take > an equivalent French expression to 'Don't tutoyer me' into English as 'Don't > use that tone of voice with me', or some such thing. But it is difficult to > say what makes one such translation better than another. People can get into > heated arguments over such questions. I'm not sure what you mean by "directly." Perhaps you should use "concisely." After all, we are English/American speakers here, and Lo! we can certainly grasp some idea of the action "tutoyer" refers to from your brief description. Since we start off without the necessary concepts, you simply have to hand us the underlying knowledge heirachy for us to understand. That doesn't make your translation "bad." It simply means you can't lop off the top floor of a skyscraper, ship it across town, and expect it to float the same distance above bare ground. You can speak concisely when you share a large base of common knowledge/experience with your hearer. If you don't, then you must recursively expand your high-level expressions until you reach the level of your listener's available knowledge. Consider how differently you might describe what you did at work today to your boss, to a coworker, to a casual friend, and to your mother. Thus translating a static word-string from one language into another should be, in general, about as hard as inferring a person's hair color by observing the mud they have tracked onto a carpet. A person's language is essentially a set of high-level pointers into the large knowledge network they cart around in their head. Without an accurate model of that network, translating those pointers into another language (with all its different cultural baggage) will be tough. In any case, the "goodness" of the translation must always depend on the recipient. Dan Mocsny dmocsny@uceng.uc.edu
sp299-ad@violet.berkeley.edu (Celso Alvarez) (10/05/89)
In article <15336@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes:
CA> Much more than thoughts are evoked by language. How do you translate
CA> the signalling of identity, roles, and social relationships?
RW>I think that such concepts have to be represented as thought structures,
RW>since they have an impact on language structure.
If you're talking about (mental) typifications of social relationships,
that's one thing. Typifications generate expectations which underly the
production and interpretation of talk. But the signalling of identity
etc. is a situated process, and by this I imply that there may be no
matching between typifications and actual behavior. Additionally,
certain linguistic markers of social dimensions are inherently ambiguous
(this ambiguity may be of a different kind than lexical ambiguity).
I don't know how you can translate social markers unless you establish
certain universals (or, at least, certain generalizable transcultural
principles) of socio-interactional meanings. I don't think you can do
this without incorporating context as a variable in those universals.
And I'm not sure that even then you can give account of the dynamic
reconstitution of those general principles (thought structures?)
in and through context, during the course of an interaction.
Celso Alvarez
sp299-ad@violet.berkeley.edu
rwojcik@bcsaic.UUCP (Rick Wojcik) (10/07/89)
In article <1989Oct5.080214.7683@agate.berkeley.edu> sp299-ad@violet.berkeley.edu (Celso Alvarez) writes: >CA> Much more than thoughts are evoked by language. How do you translate >CA> the signalling of identity, roles, and social relationships? >RW>I think that such concepts have to be represented as thought structures, >RW>since they have an impact on language structure. >If you're talking about (mental) typifications of social relationships, >that's one thing. Typifications generate expectations which underly the All I meant was that anything which influences linguistic structure ipso facto has to be represented as some kind of thought structure. However you want to represent those thought structures is open to debate. I agreed with your implicit point that they are not represented well in modern linguistic theory. But I don't think that my statement should have generated any controversy. -- Rick Wojcik csnet: rwojcik@atc.boeing.com uucp: uw-beaver!bcsaic!rwojcik
sp299-ad@violet.berkeley.edu (Celso Alvarez) (10/09/89)
In article <15578@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: >>RW>I think that such concepts [as social identity, etc.] have to be >>RW>represented as thought structures, >>RW>since they have an impact on language structure. CA>If you're talking about (mental) typifications of social relationships, CA>that's one thing. Typifications generate expectations which underly the RW>All I meant was that anything which influences linguistic structure ipso RW>facto has to be represented as some kind of thought structure. However RW>you want to represent those thought structures is open to debate. That's why I opened the debate. Am I sure that I do want to represent those categories/typifications as thought structures? Is it analytically or heuristically productive to work with a notion such as `thought structure' to help explain linguistic behavior? (because we're talking about this, aren't we?). Between linguistic action and cognition there's still a missing link, both in Searle and beyond Searle, in discourse analysis or ethnomethodology. RW>I agreed with your implicit point that they are not represented RW>well in modern linguistic theory. But I don't think that my statement RW>should have generated any controversy. And I agree with your view on translation, however different our approaches may be. But I'm not that interested in contributing to fill the holes in modern linguistic theory. In other, socio-interactionally oriented linguistic disciplines, yes. It's not my intention to create unnecessary controversy. I'm just trying to translate your language (`thought structures', social relationships as `notions' and not actions) into mine. Celso Alvarez sp299-ad@violet.berkeley.edu