nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (04/15/88)
NL-KR Digest (4/14/88 20:28:08) Volume 4 Number 42 Today's Topics: Semantic Networks : HELP !!! Sapir-Whorf Request for software which perform morphological analysis ELIZA in Prolog ? What are grammars (for)? Re: Representing archiphonemes From CSLI Calendar, April 7, 3:23 BBN AI Seminar -- Jeff Van Baalen AI Seminar: Dave Schaffer From CSLI Calendar, April 14, 3:24 Lang. & Cognition Seminar Submissions: NL-KR@CS.ROCHESTER.EDU Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU ---------------------------------------------------------------------- From: rich <rich%EASBY.DURHAM.AC.UK@CUNYVM.CUNY.EDU> Subject: Semantic Networks : HELP !!! I am investigating search techniques over realistically sized semantic net structures as part of the work for my Ph.D. thesis. I would be very glad to hear from anyone who can provide a copy of a large semantic network on which to test my techniques, as the time overhead in building my own semantic net is prohibitive. The language of implementation of the network and its content is immaterial - any size of network would be most welcome. I would also appreciate hearing from anyone who is undertaking work on semantic network search strategies and the use of semantic network structures as the basis for commercial databases. Hopefully we may be able to help each other. Thanks very much......... Richard. E-MAIL: UUCP ...!mcvax!ukc!easby!rich ARPA rich%EASBY.DUR.AC.UK@CUNYVM.CUNY.EDU BITNET rich%DUR.EASBY@AC.UK FROM: Mr. R.M.Piercy, Computing Dept, S.E.A.S, Science Laboratories, South Road, DURHAM DH1 3LE, ENGLAND. ------------------------------ Date: Wed, 6 Apr 88 17:50 EDT From: Starbuck <pearl@topaz.rutgers.edu> Subject: Sapir-Whorf Some people have mentioned the Sapir-Whorf hyothesis. What is it? Can someone give me a quick explanation of it and where I might read more about it? Thanks, Steve NAME: Stephen Pearl (Starbuck) VOICE: (201)932-3465 UUCP: rutgers!topaz.rutgers.edu!pearl ARPA: pearl@topaz.rutgers.edu US MAIL: LPO 12749 CN 5064, New Brunswick, NJ 08903 QUOTE: "Works for me!" -Rick Hunter (The Cop, not the Robotech Defender) "What is Starbuck-ing?" -Adultress 19 ------------------------------ Date: Thu, 7 Apr 88 09:10 EDT From: fosli@ifi.uio.no Subject: Request for software which perform morphological analysis I'm working on a project to translate words and phrases from English to Norwegian. Anyone done something similar? In particular I would like: 1. A morphological analyzer to get the root and the features for a given word. 2. A dictionary to use with 1. ------------------------------ Date: Sun, 10 Apr 88 00:10 EDT From: Kanna Rajan <CS_5331_13@uta.edu> Subject: ELIZA in Prolog ? Does anyone know of ELIZA being written in PROLOG (any version either on mainframes or desk tops) ? Would appreciate if you could let me know at the earliest. Thanks in advance. ----Kanna Rajan cs_5331_13@uta.edu Computer Science Dept. Univ. of Texas @ Arlington ------------------------------ Date: Thu, 7 Apr 88 12:22 EDT From: HESTVIK%BRANDEIS.BITNET@MITVMA.MIT.EDU Subject: What are grammars (for)? Rick Wojcik writes: >AH> ... But the sentence 'How did you wonder whether Bill >AH> fixed the car', with the intended reading that 'how' is a question about >AH> the manner of fixing... RW>(Please continue to use the expression 'with the intended reading that'. RW>It serves to remind us all that grammaticality judgments don't exist RW>outside of pragmatic contexts.) Not quite true. You don't need any pragmatic context to decide that 'He likes John', with JOHN and HE coreferent is illformed. The same goes for the above sentence with 'how...' (etc. etc.) "Intended reading" is not "pragmatic con- text". >My understanding of generative >grammar is that no structural analysis is possible for the intended >reading. That is how it gets recognized as ungrammatical. That's exactly wrong. (It's true of formal grammars that you learn about in discrete mathematics, but not of the study of mentally represented grammars.) By having the grammar give structural analyses for illformed strings, we are able to pinpoint and discover in a systematic fashion WHY they are ungrammatical. If the grammar gave it no structural analysis, but simply made a yes/no distinction between well- formed and ill-formed strings, then we could never understand why it was ungrammatical; we would simply have to list it as an irreducible fact (and end up with context free PSGs). In fact, by giving a structural analysis, we find that the WH-word is related to a position in the embedded clause, and the relation crosses another WH-word, and such crossings are generally bad (which is how it gets recognized as ungramma- tical), etc. etc. I.e., the fact that it is ungrammatical TELLS US something very significant about the nature of human grammars, but we cannot discover what it is unless the grammar gives an analysis of the sentence. Rick Wojcik and Stan Friesen writes: >>>AH> least for part of the field). Rather, the main interest is to try to >>>AH> understand the very nature of grammars ... namely the >>>AH> psychologically represented mechanism that underlies e.g. language >>> acquisition >>>AH> and language processing. >>RW>Here we agree totally. This is why I believe that generative theory needs >>RW>a coherent position on the way in which grammars interact with linguistic >>RW>performance. It's not an a priori requirement on generative grammar that it has a coherent position on the way it interacts with performance. Rather, this interaction is an empirical question and people are working on figuring it out (see reference below). It is a fallacy to believe that we need to understand everything about B to work on A. People understood the movements of the planets before they understood what planets were! In linguistics it is more the opposite: We have a pretty good understanding of what sentences are, but we have (pretty much) no idea about how they are processed in the brain. >SF> Or why generative grammars must be thrown out, since they do not seem >SF> to correspond to any real psychological process! If you actually read any of the scientific literature, you would find that generative grammar is NOT meant to correspond to any real psychological PROCESS, but rather to real psychological KNOWLEDGE. The former view was a misunderstanding in the early years by psychologists, called the Derivational Theory of Complexity. They thought that a sentence with many transformations would be harder/take longer to process than a sentence with few derivations. This turned out to be wrong. There are some reasons to think that there is not a direct correspondence between the grammar and the processor. For starters, think of the fact a sentence like "This is the man that the cat that chased the rat that ate the cheese scratched" is really hard to process; in fact, this kind of sentences where the processor actually FAILS. Do we want the grammar to say that the sentence is illformed? Or consider the sentence "The child seems sleeping." This sentence is perfectly well understood by English speakers; they can assign it a specific meaning. Do we want linguistic theory to say that it is well- formed? Notice that the theory of grammar has as one of its tasks to aid a child an acquiring language, i.e., to instruct the child on what a grammar may possibly look like. I.e., it may say that "If AGR is rich, then you have pro-drop". This is a very different function from say, parsing a sentence. It's not very process-like. To read about the relation between grammar and processing, see ch. 2 (pp35-82) in Robert Berwick & Amy Weinberg (1984) "The Grammatical Basis of Linguistic Performance: Language use and acquisition". Cambridge, Mass.: MIT Press. On the "how-..." sentence and why it is ungrammatical, see e.g. N. Chomsky "Barriers", MIT Press 1986. Arild Hestvik Department of Psychology Brandeis University ------------------------------ Date: Sat, 9 Apr 88 10:56 EDT From: John Chambers <jc@minya.UUCP> Subject: Re: Representing archiphonemes > I don't see why your chart of Spanish graphemes contains archiphonemes. > Although 'm' in 'cambiar' cannot contrast with /n/, that does not make > it a different sound from 'm' in 'matar'. I would maintain that > positions of phonemic neutralization exist, but not archiphonemes. > Can you think of any real justification for underspecifying segments > that occur in neutralized positions? Bear in mind that no alphabetic > writing system has special symbols for them. They seem to have no > behavioral correlates. Of what use are they? Maybe I misunderstand, but it seems to me that there are a few examples of such representation of archiphonemes. For instance, in Russian, [v] and [f] are definitely in contrast, albeit weakly. The writing system has a letter for [f] (which looks like the Greek phi), and one for [v/f] (which looks like 'B'); there is no letter for [v] alone. When you see the [v/f] letter (which is usually transliterated as 'v', causing English-speaking people to mispronounce Russian names), you have to determine from context which of the two phonemes it stands for. When you see the 'f', you know it's [f]. There is confusion here with the representation of morpho-phonemes, such as the English /-s/ suffix. This is written "s", even when pronounced [z]. But this isn't an example of an archiphoneme; it is rather a morpho-phoneme conditioned by its environment. The Russian v/f case is more of a real archiphoneme. The two sounds aren't in contrast in "native" words; the contrast arises only from borrowings that violate the voicing rules (which is easy, as v/f is the only voiced/voiceless pair that isn't in contrast in the native vocabulary). An example is the common name "Fyodor", derived from the Greek "Theodore". The initial [f] violates the conditioning rules for the [v/f] archiphoneme, so it's spelled with an 'f'. There is even an example in English, though we use a digraph: The symbol 'th' stands for two different phonemes that *are* in weak contrast. A thousand years ago, they weren't in contrast, but other changes (such as loss of final weak vowels) has produced contrasts such as mouth/mouthe and breath/breathe. Anyhow, it seems straightforward to classify 'th' as an English archiphoneme, with a single symbol representing both phonemenes, and spelling conventions to distinguish them. There are other examples. Germans use 's' to represent both [s] and [z], with spelling conventions to disambiguate them. As in the above examples, s/z are barely in contrast, so you don't often need separate symbols. Mostly they just use 'ss' to represent [s] when the rules would give [z]. On the other hand, we haven't had any clear definitions of what constitutes an "archiphoneme" or a "morphophoneme", so I could be using different definitions that yours... -- John Chambers <{adelie,ima,maynard,mit-eddie}!minya!{jc,root}> (617/484-6393) ------------------------------ Date: Mon, 11 Apr 88 12:18 EDT From: Rick Wojcik <rwojcik@bcsaic.UUCP> Subject: Re: Representing archiphonemes In article <557@minya.UUCP> jc@minya.UUCP (John Chambers) writes: >... we haven't had any clear definitions of what >constitutes an "archiphoneme" or a "morphophoneme", so I could >be using different definitions that yours... >John Chambers <{adelie,ima,maynard,mit-eddie}!minya!{jc,root}> (617/484-6393) Archiphonemes often get confused with morphophonemes. The key to understanding them is the concept of automatic phonemic neutralization. For example, English does not allow syllable-internal obstruent clusters of mixed voicing. These are eliminated by a process of progressive voice assimilation. Thus, we can have initial /sp/ clusters, but any attempt to pronounce /sb-/ results in a devoicing of the /b/. Similarly, plural /-z/ is automatically devoiced after voiceless obstruents. N. Trubetzkoy proposed the use of capital letters to represent segments of indeterminate voicing for cases such as this. Thus, {spill} might be represented as /sPil/, and {cats} as /kaetZ/. Trubetzkoy took archiphonemics, together with phonemics, to constitute phonological theory. Morphophonemes stood for nonautomatic alternations between distinct phonemes (e.g. /f/~/v/ in 'leaf/leaves'). Although Trub. took morphonology to be a separate component of grammar from phonology, most other phonologists came to conflate morphonemes with archiphonemes. Hence, the confusion. JH> When you see the [v/f] letter (which is usually transliterated as JH> 'v', causing English-speaking people to mispronounce Russian names), JH> you have to determine from context which of the two phonemes it JH> stands for. When you see the 'f', you know it's [f]. Final /v/ gets pronounced [f] in Russian by automatic final devoicing, so one could use an archiphoneme /F/ to represent both phonemes in that position. As for /f/, it gets voiced before words beginning with voiced obstruents--e.g. 'graf byl' [grav byl]. So when you see {f}, you don't necessarily know that it's [f]. English-speaking people mispronounce Russian names because they lack the automatic final devoicing. There is good evidence that Russians think they are pronouncing final /v/'s in those words. JH> There is confusion here with the representation of morpho-phonemes, JH> such as the English /-s/ suffix. This is written "s", even when JH> pronounced [z]. But this isn't an example of an archiphoneme; it JH> is rather a morpho-phoneme conditioned by its environment. No. It's a candidate for archiphoneme, if you believe in such things. Some people only allow 'p' in 'spill' to be an archiphoneme, because it doesn't involve alternations. So, by some lights, your calling the {-s} suffix a morphophoneme is ok, because it involves alternations. JH> Anyhow, it JH> seems straightforward to classify 'th' as an English archiphoneme, JH> with a single symbol representing both phonemenes, and spelling JH> conventions to distinguish them. I know of no phonological theories that would take modern English 'th' to be an archiphoneme. I can think of no cases in English where the dental fricative occurs in a position of automatic neutralization. -- Rick Wojcik csnet: rwojcik@boeing.com uucp: {uw-june uw-beaver!ssc-vax}!bcsaic!rwojcik address: P.O. Box 24346, MS 7L-64, Seattle, WA 98124-0346 phone: 206-865-3844 ------------------------------ Date: Wed, 6 Apr 88 20:09 EDT From: Emma Pease <emma@russell.stanford.edu> Subject: From CSLI Calendar, April 7, 3:23 [Excerpted, etc.] THIS WEEK'S CSLI SEMINAR The Texture of Intelligence Alexis Manaster-Ramer (amr@csli.stanford.edu) April 7 A and B engage in conversation in French with a group of Frenchmen. However, while A speaks passable French, he does not understand spoken French well, and B understands colloquial French reasonably well, but does not speak it. So, A does the listening and B does the talking, communicating with each other in English when necessary. As far as the French interlocutors are concerned, A+B "knows" French. What I want to argue is that theories of intelligent human behavior should adopt the Frenchmen's point of view. Intelligence exists in culture. What seems to make human beings an intelligent species, biologically, is that we have evolved the ability--and the necessity--of living in a culture. In general, the subject of the study of human intelligence must then be interaction of groups of people. As a result, a proper explanatory theory of intelligent behavior must be HISTORICAL in nature (much as biology and physics are historical sciences). While we need to understand how an individual represents knowledge, reasons, speaks, etc., our theories must also capture the fact that no individual is capable of creating English or developing French cuisine, say, from scratch. Whether we want cognitive science or AI, we should think of simulating cultures evolving through time rather than individuals. Obviously, many of the processes we need to model do take place within individual human beings. These must be understood in terms of the interaction of the different mechanisms, which are postulated to account for specific patterns in the data rather than in terms of a priori mental faculties such as "grammar," "world knowledge," "commonsense reasoning," etc. In studying the individual, we must again develop theories that are historical (ontogenetic) in nature, since people's reasoning and language use, for example, both seem to depend to a large extent on how and when various skills and information happen to be learned. Moreover, the components of the theory of individuals cannot all be qualitatively alike. Some are physical, others cognitive, and others in between (as in my theory of TACTICS, the lowest level of language). The theories of the individual, as well as those of the cultural, phenomena can--and should--be formal (symbolic) without our having to assume that the object being studied is symbolic and represented in individual minds in symbols. The usual kinds of formalisms (say, automata) can be used to model interactions among individuals or cultural evolution or the states of a physical system such as the vocal tract, for example, just as easily as they can to represent the alleged cognitive faculties of individual people (like "grammar"). This approach seems to close the gap between the two main positions on the nature and definition--as well as the possibility of artificial simulation--of intelligence. We accept the "pessimistic" view on the scope of the subject to be modeled but adopt the "optimistic" view on the symbolic representation of the models (NOT of the objects of study). Results in different areas emerge immediately from this perspective, such as my work on tactics. -------------- NEXT WEEK'S CSLI TINLUNCH Reading: "Language and Interpretation: Philosophical Reflections and Empirical Inquiry" by Noam Chomsky Discussion led by Sylvain Bromberger (sylvain@csli.stanford.edu) April 14 Once upon a time there were serious people who tried to figure out what the heavenly spheres are made of. They never succeeded. There are no heavenly spheres. Quine, Davidson, Dummett, and Putnam hold views about language and its study that imply that much of what passes for serious linguistics---at least at MIT---should be dismissed like celestial sphereology, as based on delusion. These are prominent philosophers. They should be right. Chomsky does not think that they are. In this paper he tries to prove that they are mistaken. Are they? -------------- NEXT WEEK'S CSLI SEMINAR On Acting Together: Joint Intentions for Intelligent Agents Phil Cohen (pcohen@ai.sri.com) April 14 No one wants to build just one lonely autonomous agent. If we are successful, we will want our creations to be available to help each other, and us. In short, they should be able to act jointly with other agents. Obvious examples of joint action in human society include pushing a car, playing a duet, executing a pass play, engaging in a dialogue, and doing research (for example, this research was done jointly with Hector Levesque, Department of Computer Science, University of Toronto). Analogues of such "team play" can easily be created for any task requiring more than one agent for its accomplishment, and for those in which agents need to divide up the work. In a recent paper, we argued that intention is a derived concept, founded on the idea of an internal commitment, or "persistent goal," i.e., goals kept through time. In this talk, we develop an analogous concept, that of a "joint commitment," that can serve as the basis of a concept of joint intention. We show how joint commitments lead to synchronization, agreements to commence action and to terminate, individual actions by the collaborators, and communication. Finally, we show how the analysis compares with recent proposals by Searle and by Grosz and Sidner for describing joint intentionality. ------------------------------ Date: Fri, 8 Apr 88 14:25 EDT From: Marc Vilain <MVILAIN@G.BBN.COM> Subject: BBN AI Seminar -- Jeff Van Baalen BBN Science Development Program AI Seminar Series Lecture REPRESENTATION DESIGN FOR PROBLEM SOLVING Jeffrey Van Baalen MIT AI Laboratory (jvb@HT.AI.MIT.EDU) BBN Labs 10 Moulton Street 2nd floor large conference room 10:30 am, Thursday April 14 It has long been acknowledged that having a good representation is key in effective problem solving. But what is a ``good'' representation? In this talk, I overview a theory of representation design for problem solving that answers this question for a class of problems called analytical reasoning problems. These problems are typically very difficult for general problem solvers, like theorem provers, to solve. Yet people solve them comparatively easily by designing a specialized representation for each problem and using it to aid the solution process. The theory is motivated, in large part, by observations of the problem solving behavior of people. The implementation based on this theory takes as input a straightforward predicate calculus translation of the problem, gathers any necessary additional information, decides what to represent and how, designs the representations, creates a LISP program that uses those representations, and runs the program to produce a solution. The specialized representation created is a structure whose syntax captures the semantics of the problem domain and whose behavior enforces those semantics. ------- ------------------------------ Date: Tue, 12 Apr 88 08:33 EDT From: Dori Wells <DWELLS@G.BBN.COM> Subject: AI Seminar: Dave Schaffer BBN Science Development Program AI Seminar Series ADAPTIVE KNOWLEDGE REPRESENTATION: A CONTENT SENSITIVE RECOMBINATION MECHANISM FOR GENETIC ALGORITHMS J. David Schaffer Philips Laboratories North American Philips Corporation Briarcliff Manor, New York BBN Laboratories Inc. 10 Moulton Street Large Conference Room, 2nd Floor 10:30 a.m., Tuesday, April 19, 1988 Abstract: This paper describes ongoing research on content sensitive recombination operators for genetic algorithms. A motivation behind this line of inquiry stems from the observation that biological chromosomes appear to contain special nucleotide sequences whose job is to influence the recombination of the expressible genes. We think of these as punctuation marks telling the recombination operators how to do their job. Furthermore, we assume that the distribution of these marks (part of the representation) in a gene pool is determined by the same survival-of-the-fittest and genetic recombination mechanisms that account for the distribution of the expressible genes (the knowledge). A goal of this project is to devise such mechanisms for genetic algorithms and thereby to link the adaptation of a representation to the adaptation of its contents. We hope to do so in a way that capitalizes on the intrinsically parallel behavior of the traditional genetic algorithm. We anticipate benefits of this for machine learning. We describe one mechanism we have devised and present some empirical evidence that suggests it may be as good as or better than a traditional genetic algorithm across a range of search problems. We attempt to show that its action does successfully adapt the search mechanics to the problem space and provide the beginnings of a theory to explain its good performance. ------------------------------ Date: Wed, 13 Apr 88 20:05 EDT From: Emma Pease <emma@russell.stanford.edu> Subject: From CSLI Calendar, April 14, 3:24 [Excerpted from CSLI Calendar] Connections between Linguistics and Computer Science: Some Topics in the Mathematics of Language Bill Rounds (rounds@csli.stanford.edu) University of Michigan CSLI and Xerox PARC April 21 In this talk I will discuss some similarities and analogies between grammar formalisms, situation theory, database theory, and the modal logic of programs. The focus of the talk will be on a simple graphical representation of linguistic structures, essentially as state graphs of nondeterministic finite automata, and I will describe several kinds of logical statements useful for speaking about these structures. When a construct for specifying structures recursively is added to the basic logic, one obtains a fairly powerful declarative mechanism similar to Prolog. Unification of the extended structures can be thought of as forming a join operation in a suitable ordering of the structures. It turns out that in one such ordering, unification corresponds to taking the join of database relations. This ordering has also proved useful in the specification of concrete data types. Most of the talk will consist of examples and pictures, and only a nodding familiarity with any of the above topics will be presumed. ------------------------------ Date: Thu, 14 Apr 88 14:11 EDT From: Dori Wells <DWELLS@G.BBN.COM> Subject: Lang. & Cognition Seminar BBN Science Development Program Language & Cognition Seminar Series HOW LANGUAGE STRUCTURES ITS CONCEPTS: THE ROLE OF GRAMMAR Leonard Talmy Program in Cognitive Science University of California, Berkeley BBN Laboratories Inc. 10 Moulton Street Large Conference Room, 2nd Floor 10:30 a.m., Wednesday, April 20, 1988 Abstract: A fundamental design feature of language is that it has two subsystems, the open-class (lexical) and the closed-class (grammatical). These subsystems perform complementary functions. In a sentence, the open-class forms together contribute most of the *content* of the total meaning expressed, while the closed-class forms together determine the majority of its *structure*. Further, across the spectrum of languages, all closed-class forms are under great semantic constraint: they specify only certain concepts and categories of concepts, but not others. These grammatical specifications, taken together, appear to constitute the fundamental conceptual structuring system of language. I explore the particular concepts and categories of concepts that grammatical forms specify, the properties that these have in common and that distinguish them from lexical specifications, the functions served by this organization in language, and the relations of this organization to the structuring systems of other cognitive domains such as visual perception and reasoning. The greater issue, toward which this study ultimately aims, is the general character of conceptual structure in human cognition. ------------------------------ Date: Fri, 1 Apr 88 06:28 EST From: Francis LOWENTHAL <PLOWEN%BMSUEM11.BITNET@CUNYVM.CUNY.EDU> ANNOUNCING A CONFERENCE : LANGUAGE AND LANGUAGE ACQUISITION 4 ============================================================= CALL FOR PAPERS FIRST ANNOUNCEMENT Dear colleague, I have the pleasure to invite you to the fourth conference we organize on Language and Language Acquisition at the University of Mons, Belgium. The specific theme of this conference will be : "LANGUAGE DEVELOPMENT AND COGNITIVE DEVELOPMENT" Date : From August 22 to August 27, 1988 Place : Mons University. The aim of this meeting is to further an interdiscipli- nary and international collaboration among researchers connec- ted one way or the other with the field of communication and subjacent logic : this includes as well studies concerning normal children as handicapped subjects. Five topics have been chosen : Mathematics, Philosophy, Logic and Computer Sciences, Psycholinguistics, Psychology and Medical Sciences. During the conference, each morning will be devoted to two 45-minutes lectures on one of these domains, and to a wide discussion concerning all the papers already presen- ted. The aftrnoon will be devoted to short presentations by panelists and to further discussions concerning the panel and everything that preceded it. There will be no parallel sessions and, as the organi- zers want to favour as much as possible discussions between the participants, it has been decided to reduce the number of par- ticipants to 70. The selection procedure will be supervised by an international committee. Further informations and registration forms can be obtained by old fashioned mail or by E-mail from : F. LOWENTHAL Universite de l'Etat a Mons Laboratoire N.V.C.D. Place du Parc, 20 B-7000 MONS (Belgium) tel : (32)65.37.37.41 TELEX 57764 - UEMONS B E-MAIL : PLOWEN@BMSUEM11 Please, feel free to communicate this call for papers to other potential interested researchers. Thank you for your help and best wishes for 1988. F. LOWENTHAL JANUARY 7, 1988 ------------------------------ End of NL-KR Digest *******************