nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (01/13/88)
NL-KR Digest (1/12/88 20:19:31) Volume 4 Number 3 Today's Topics: Dependency Grammar / Variable Word Order empirical science of language Natural vs. Programming Languages Re: Language Learning Re: online dictionaries Seeking machine readable dictionaries in French Submissions: NL-KR@CS.ROCHESTER.EDU Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU ---------------------------------------------------------------------- Date: Fri, 1 Jan 88 00:29 EST From: Michael Covington <MCOVINGT%UGA.BITNET@forsythe.stanford.edu> Subject: Dependency Grammar / Variable Word Order [Excerpted from PROLOG Digest] I would like to hear of any work, published or unpublished, on the following topics: (1) Parsing using a dependency rather than a constituency grammar, i.e., by establishing grammatical relations between individual words rather than gathering words into groups. (2) Parsing strategies that have been used successfully with languages that have variable word order, such as Russian or Latin. I am familiar with GPSG, HPSG, and ID/LP grammars, and with various proofs that (some) dependency grammars are equivalent to (some) con- stituency or phrase-structure grammars. However, there must have been some good ideas in circulation much earlier, perhaps in connection with early machine translation work. ------------------------------ Date: Tue, 5 Jan 88 11:50 EST From: Bruce E. Nevin <bnevin@cch.bbn.com> Subject: empirical science of language To Walter Rolandi <rolandi@gollum.columbia.nrc.com> The status of linguistics as a science has been a vexed question for a very long time. There are a number of good reasons. Probably the central one is this: in all other sciences and in mathematics, you can rely on the shared understanding of natural language to provide a metalanguage for your specialized notations and argumentation. In linguistics you cannot without begging fundamental questions that define the field. There is an exactly parallel difficulty in psychology: a psychological model must account for the investigator on the same terms as it accounts for the object of investigation. The carefully crafted suspension of subjectivity that is so crucial to experimental method becomes unattainable when subjectivity itself is the subject. (See Winograd's recent work, e.g. _Understanding Computers and Cognition_ for reasons why computer modelling of natural language is not possible, on the usual construal of what computer modelling is. I have references to work that gets around this "Framer Problem" if you are interested.) For a more explicit and comprehensive critique of generative linguistics, I recommend Maurice Gross's paper "On the Failure of Generative Grammar" in _Language_ 55.4:859-85 (1979). Despite its prominent publication in the most prestigious journal in the field by an acknowledged expert, there has been NO REJOINDER to this paper. It is hard to construe this rather astonishing silence other than as a tacit confession on the part of generativists. Gross also offers very specific and detailed alternatives to the Generativist programme, so we are not just talking about carping negativity here. You wanted an empirically based natural science of language. For more on the empirically based paradigm of linguistics as a natural science that Gross draws upon, you should become acquainted with the recent work of Zellig Harris: _Mathematical Structures of Language_ (1968); _A Grammar of English on Mathematical Principles_ (1982--I reviewed this in _Computational Linguistics_ in 1984); _The Form of Information in Science_ (with Gottfried, Ryckman, and others, 1987); a book on the theory of language forthcoming from Oxford whose title I forget. Harris is at Columbia where you are, in the Center for the [Study of the?] Social Sciences, so it shouldn't be too hard to find out what he's about. He gave the Bampton Lectures a year ago October at Columbia, and his series on language and information will be published by Columbia. Chomsky was Harris's grad student in the late '40s and early '50s when Harris began to develop the notion of linguistic transformations as a way of regularizing texts for discourse analysis. Chomsky has attributed his breach with Harris to a conflict between his own Rationalist philosophy and (what he considers) Harris's Empiricist position. Chomsky borrowed notation from mathematics (Post's "production systems") to represent a form of syntactic analysis that had been developed in the 1920s and '30s and formalized in 1946 in a couple of papers by Harris and Rulon Wells. (Harris noted the major deficiency--inability to account for the heads of endocentric constructions in a natural way--in his 1946 paper, and his workaround was borrowed as the X-bar notation when the generativists realized something was wrong almost thirty years later. Harris developed string grammar in the late 1950s as an alternative. Joshi proved Harris's suggestion that you could refine the word subclasses of a string grammar to get a transformational grammar, and Sager & Grishman exploited this in their very successful LSP system. Joshi's tree-adjoining grammars are a more or less direct development from string grammar, combining very restricted rewrite rules for exocentric constructions with adjunction rules for endocentric constructions.) Because of his presupposition of innate hard-wired neural correlates of linguistic entities, Chomksy shifted attention from language to the now familiar rewrite rules and phrase structure trees of his notation. (Backus and Naur developed BNF notation 5 or 6 years later as a metalanguage to account in a unified way for the proliferating variety of computer languages. The affinity with trends in nascent computer science appears to have lent at least credibility to the rapid expansion of Generativist missionaries. Certainly, Harris supported his (in the event ungrateful) student, for example recommending Chomsky as a substitute for himself to give the keynote address at the 1962 (?) International Congress of Linguists.) A historically arbitrary choice of notation has become reified as the form of data in linguistics. Many problems stem from the characteristics of PSG: that there is no principled way to control proliferation of nonterminal symbols; that nonterminal symbols have no semantic value (names of nonterminals are arbitrary, relation to words of "surface strings" is indirect; that the rules "overgenerate" and extraneous structures must be filtered, and so on. I can't go into a detailed comparison of the two paradigms here. For Chomsky, transformations were derivation operations on the abstract objects generated by rewrite rules. For Harris, transformations were mappings in the set of sentences from which, over the next thirty years, he was able to develop a mathematical theory of language and of linguistic information with rules of composition and derivation. For Chomsky, having something in mind to say and putting it into words is a matter for a separate theory of performance, which his theory of competence merely constrains. For Harris, this obfuscatory dualism is unnecessary, a fact that alone recommends him to computational linguists. For Harris, it is essential that the grammatical description add no unnecessary structure to the redundancy that language uses for informational purposes (extrinsic redundancy), and the product of a maximally efficient grammar is a representation of the objective information in texts. For Chomsky, the great complexity of his grammatical machinery is yet another evidence that it corresponds to something innate, else how could a child possibly learn a language? (It is a fascinating exercise to compare what Harris demonstrates is sufficient for language to what Generativists claim is necessary for language.) According to Kuhn, a scientific revolution takes at least two generations. For old die-hards the new paradigm just doesn't make sense, they can't make the "paradigm shift" and remain unconverted; for subsequent generations of entrants to the field, the new paradigm seems obviously better, and they take over as the die-hards die off. But the Generativist revolution in linguistics has failed to become the obvious way to talk about language, so we have to say that it was more a political revolution than an intellectual one. It has, however, done enormous damage to the field. Linguistics now has lost credibility. Perhaps it _is best that computer scientists re-invent the field under a new name. It is unfortunate in the process to reinvent a number of cubical and hyperbolic wheels, but there may be benefit even there--sort of like the value of recessive genes in the gene pool that are dysfunctional at time x and adaptive at time y. Certainly, we can now more easily find out what American structuralists were really about, for example, without the rhetorical distortions of the Generativist gospel. PS: I just noted an article by Victor Yngve in the last issue of Theoretical Linguistics, "Linguistics among the sciences." I have not yet read it. bn@cch.bbn.com <usual_disclaimer> ------------------------------ Date: Tue, 5 Jan 88 13:06 EST From: Rick Wojcik <rwojcik@bcsaic.UUCP> Subject: Natural vs. Programming Languages >From: John Chambers <jc@minya.UUCP> [remarks that linguistics had little to contribute to computer modelling of natural language] >This has had a large shock effect on linguistics, but linguistics has >had little influence on computing. For people that claim to be scientists, >this is pretty damning. I'm sure that you don't mean to ignore the important contributions made by Noam Chomsky on the formal description of artificial languages. I don't think that you can study compiler design without learning something about the value of formal linguistic theory in computing. The real problem in the computer community is that natural language models have ignored the important differences between natural and artificial languages. I take it as a criticism of Chomsky that he has sometimes been guilty of the same sin. >In the last few years, I've tried occasionally to interject into some >of the discussions in this newsgroup some examples tieing computer >languages into the discussions on human languages. The responses have >been quite revealing. Time and again, people have sent me some really >good flames about how stupid I was to think of computer languages and >human languages together. (After all, can you write poetry in a computer >language? :-) I don't think that it is stupid to think of computer languages and human languages together. How else can you learn what the differences are? I have seen some computer programs that I consider poetry. >Not all of these flames were from novices; some were from professional >linguists. Nearly all make the claim that a "natural" human language >is somehow different from "artificial" computer languages. They sure are. After all these years, can't you think of any differences? >My main reaction to all these flames is: Who do you think created the >computer languages? Computers? In a manner of speaking, yes. The structure of computer language follows very much from the nature of computers, just as the structure of natural language follows from the nature of humans. The evolution of computer languages has always been in the direction of trying to make them more like natural language. But the similarities are quite superficial. Computer languages are not designed to be unambiguous, but they are designed to be disambiguated from "local context"--i.e. by reference to computer code alone. Natural languages are far more ambiguous than nonlinguists tend to imagine. The reason for this is that speakers perceive natural language as relatively unambiguous. They do so because human minds resolve ambiguity in pragmatic and semantic contexts. Cognitive scientists have really only begun to understand the mechanisms that are needed to disambiguate human language. Recent work at Berkeley (cf. Lakoff's Women, Fire, and Dangerous Things) has added an additional wrinkle: the role of metaphor in natural language understanding and cognition. Nothing like metaphorical reasoning is to be found in computer languages. >It's also fun to taunt the linguists with a challenge: If you really >understand how languages work, then obviously you should be able to >express your theory as a "model" in the form of a computer program. >Most other scientific fields consider computer modeling to be routine >nowadays. So let's see those working computer models of English, >Japanese and Quechua! For a linguist to claim to "understand how languages work" would be to beg the question. Several natural language parsers are loosely based on linguistic theories--GPSG and GB, for example. Bresnan's LFG theory grows directly out of a computational model. The point about computer modelling is good, but it ignores the limitations of computer modelling. There need to be many more advances in computer hardware and software before our models become reasonable approximations of what the human mind can do. =========== Rick Wojcik rwojcik@boeing.com ------------------------------ Date: Tue, 5 Jan 88 16:32 EST From: Rick Wojcik <rwojcik@bcsaic.UUCP> Subject: Re: Language Learning (a Turing test) I have come across a couple of items that should be of interest to this news topic. One is Rene Coppieters' Sept. '87 Language article "Competence differences between native and non-native speakers". Coppieters studied the intuitions of native French speakers and "Near Native" French speakers who learned French after puberty. She found that the NS group had substantially different intuitions of grammaticality than those in the NNS group. This supports the Critical Period Hypothesis (CPH). The second item relates to a paper presented at the recent LSA meeting in San Francisco: Mark Patkowski's "Age and accent in a second language: A critical review." (Mark is an assistant professor of linguistics at Brooklyn College.) His review concludes that "the literature provides support (albeit indirect) for the notion of an 'optimal' period for the acquisition of phonology in a second language in particular, and for the 'critical period hypothesis' in general." After the paper, a commentator remarked that research by Newport and others shows a critical period effect in the acquisition of American Sign Language (ASL). This work is about to be published in Science and in Cognitive Psychology. If true, such work could be even more damaging for those who minimize differences between adult and child acquisition. The ASL study shows a deficit in the acquisition of syntax, not just phonology. -- =========== Rick Wojcik rwojcik@boeing.com ------------------------------ Date: Tue, 5 Jan 88 20:38 EST From: Rob McConeghy <malibo@arizona.edu> Subject: Re: Language Learning (a Turing test) In article <120@psc90.UUCP>, tos@psc90.UUCP (Dr. Thomas Schlesinger) writes: > > My own children were 3 and 4 respectively when I spent three years in > Frankfurt, Germany. It fascinated me to watch how they had no inkling > of their total bilinguality (yes, I know a horrible "word"). But > when a German spoke to them they automatically answered in German, and > when an American spoke to them they answered in English. If I spoke > to them in the "wrong" language, i.e. German, they'd get angry at me. > But they didn't know why... they didn't really know what "languages" > were and that they were "bilingual." Pardon our curiosity, but did your children remain bilingual after you left Germany ? I had cousins who at slightly older ages spent a year in Lausanne Switzerland where they attended the local schools. When they returned they had a fairly low level command of French - i.e. their accents were good and they could understand simple everyday statements and questions and respond in a semi- coherent manner. They had had no previous exposure to French and did not live in a bilingual household, only English was spoken at home. After a year or two back in the states they had completely forgotten all the French they had picked up. This possibly illustrates two aspects of childhood language acquisition. First of all it is not instantaneous. It does take quite a long time, even in a high immersion environment, a year is not sufficient. Secondly, motivation is a factor. My cousins were not really highly motivated to learn French. They knew they would be going home at the end of a year, and they could still speak English at home to their parents and each other. They also had no reason to retain or expand on their French after they returned to the US. Of the three cousins, only the youngest, (about 5 or 6 as I recall), really learned quite a lot of French, the 9 or 10 year old also learned a lot, the 14 or 15 year old not much at all (or at least we couldn't get any out of her, pretty typical non-communicative teenager). This is mainly as was to be expected except for the difference between the 5 year old and the 10 year old. If we accept the theory that there are automatic language acquisition processes that are at work in children's heads, we should consider whether various parts of them shut down at various times rather than all at once at age 12 or so. As any elementary school teacher or parent knows, children do not master all aspects of their native language at once, nor do they master all aspects of thinking with equal speed. Language acquisition in children probably needs to be studied in close relation to the child's acquisition of other mental abilities. Does anyone know of studies where this has been done, especially in relation to the acquisition of multiple languages by children? ------------------------------ Date: Thu, 7 Jan 88 19:53 EST From: Doug Rudoff <doug@wiley.UUCP> Subject: Re: Language Learning This is slightly off the subject but ... When I was a junior in college, I spent a year studying electrical engineering in Lausanne, Switzerland. Over the course of the year I learned to speak French somewhat fluently (i.e. I had no problem thinking in French without resorting to thinking of the English translation). It has been about four years since then, and due to lack of speaking French, I really have to struggle and think if I try to speak it. Also, even when I was able to speak French fluently, I had an atrocious accent. On occasion I have dreams in which people speak French to me. When they speak French, it is perfect and unaccented. Yet, when I speak French in my dreams, I speak as well (or really as poorly) as I do during my waking hours. From this observation I get the impression that speaking a language and understanding a language by listening are totally separate brain processes. Have studies been made to this effect ? Another observation I have is related to my ability to learn languages by speech or writing (I've studied Spanish, German and French). I have always found that I have an easy time learning a language in writing. I have a very hard time picking things up by ear. It seems that it is easier for me to convert visual information (writing) to language concepts than it is to convert aural information (speech) to language concepts. Other people I know are the are the exact reverse. =-=-=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Doug RUDOFF TRW Inc, Redondo Beach, CA {cit-vax,trwrb,uunet}!wiley!doug H: (213) 318-9218 W: (213) 812-2768 wiley!doug@csvax.caltech.edu =-=-=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- ------------------------------ Date: Wed, 6 Jan 88 14:29 EST From: Mary Patricia Lowe <mary@csd4.milw.wisc.edu> Subject: Re: online dictionaries In article <29@gollum.Columbia.NCR.COM> rolandi@gollum.UUCP () writes: > > the Microsoft CD ROM version of the American Heritage Dictionary > the OED from Oxford University Press > >If anyone can locate these sources, I would appreciate what they find out. In the January 1988 issue of IEEE Spectrum, the section on Tools and Toys (p. 73) contains a short blurb on the Microsoft Bookshelf. The CD-ROM includes the following reference works: The World Almanac and Book of Facts, The American Heritage Dictionary, The U.S. ZIP Code Directory, The Chicago Manual of Style, Bartlett's Familiar Quotations, Roget's II: Electronic Thesaurus, Houghton Mifflin Spelling Verifier and Corrector, Houghton Mifflin Usage Alert, Business Information Sources. For more information, contact: Microsoft Corp., Box 97017, Redmond, WA. 98073, (206)-882-8088. -Mary Mary Patricia Lowe mary@csd4.milw.wisc.edu ...ihnp4!uwmcsd1!mary ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ------------------------------ Date: Fri, 8 Jan 88 09:51 EST From: Craig Stanfill <craig@think.COM> Subject: Re: online dictionaries In article <1092@pbhyd.UUCP> rjw@pbhyd.UUCP (Rod Williams) writes: >My understanding is that the online Oxford English Dictionary (OED) >is still a work-in-progress and is not yet commercially available. There is a new edition of the OED, which is currently in preparation, and will eventually be available in electronic form. There is also the old (1932?) edition plus numerous supplements, which is available in electronic form through Oxford University Press. ------------------------------ Date: Fri, 8 Jan 88 09:24 EST From: Gilloux <rutgers!mimsy.umd.edu!uunet!mcvax!cnetlu!gilloux> Subject: Seeking machine readable dictionaries in French I am in search of machine readable dictionaries in French. The intended use is to extract automatically semantic information needed in a NL parser. Any help would be appreciated. Michel Gilloux Centre National d'Etudes des Telecommunications LAA/SLC/AIA Route de Tregastel, BP 40 22301 Lannion Cedex FRANCE UUCP: mcvax!inria!cnetlu!gilloux ------------------------------ End of NL-KR Digest *******************