nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (10/05/88)
NL-KR Digest (10/04/88 20:44:15) Volume 5 Number 17 Today's Topics: Theft or Honest Toil, (was Re: Pinker & Prince Reply (long version)) BBN AI Seminar -- Mary Haper CfT: Machine Learning SUNY Buffalo Linguistics Colloquia CSLI Calendar, September 22, 4:1 Unisys AI Seminar: The SB-ONE Knowledge Representation Workbench From CSLI Calendar, September 29, 4:2 Submissions: NL-KR@CS.ROCHESTER.EDU Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU ---------------------------------------------------------------------- Date: Fri, 2 Sep 88 18:35 EDT From: Martin Taylor <mmt@client2.DRETOR.UUCP> Subject: Theft or Honest Toil, (was Re: Pinker & Prince Reply (long version)) Harnad characterizes learning rules from a rule-provider as "theft", whereas obtaining them by evaluation of the statistics of input data is "honest toil". But the analogy is perhaps better in a different domain: learning by evaluating the statistics of the environment is like building up amino acids and other nutritious things from inorganic molecules through photosynthesis, whereas obtaining rules from rule-providers is like eating already built nutritious things. One ofthe great advantages of language is that we CAN take advantage of the regularities discovered in the data by other people. The rules they tell us may be wrong, but to use them is easier than to discover our own rules. It is hardly to be taken as an analogy to "theft". If we look at early child learning, the "theft" question becomes: Has evolution provided us with a set of rules that we do not have to obtain from the data, so that we can later obtain more rules from people who did themselves learn from data? Obviously in some sense the answer is "yes" there are SOME innate rules regarding how we interpret sensory input, even if those rules are as low-level as to indicate how to put together a learning net. Obviously, also, there are MANY rules that we have to get from the data and/or from people who learned them from the data. The question then becomes whether the "rules" regarding past-tense formation are of the innate kind, of the data-induced kind, or of the passed-on kind. My understanding of the developmental literature is that children pass through three phases: (i) correct past-tense formation for those verbs for which the child uses the past tense frequently; (ii) false regularization, in which non-regular past tenses (went) are replaced by regularized ones (goed); (iii) more-or-less correct past tense formation, in which exceptions are properly used, AND novel or neologized verbs are given regular past tenses (in some sense of regular). This sequence suggests to me that the pattern does not have any innate rule component. Initially, all words are separate, in the sense that "went" is a different word from "go". Later, relations among words are made (I will not say "noticed"), and the notion of "go" becomes part of the notion of "went". Furthermore, the notion of a root meaning with tense modification becomes part of verbs in general. Again, I will not say that this is connected with any kind of symbolic rule. It may be the development of net nodes that are activated for root parts and for modifer parts of words. It would be overly rash to claim either that rules are involved or that they are not. In the final stage, the rule-like way of obtaining past tenses is well established enough that the exceptions can be clearly distinguished (whether statistically or otherwise is again disputable). One thing that seems perfectly clear is that humans are in general capable of inducing rules in the sense that some people can verbalize those rules. When such a person "teaches" a rule to a "student", the student must, initially at least, apply it AS a rule. But even in this case, it is not clear that skilled use of what has been learned involves continuing to use the rule AS a rule. It may have served to induce new node structures in a net. In "The Psychology of Reading" (Academic Press, 1983), my wife and I discussed such a sequence under the heading of "Three-phased Learning", which we took to be a fairly general pattern in the learning of skilled behaviour (such as reading). Phase 1 is the learning of large-scale unique patterns. Phase 2 is the discovery of consistent sub-patterns and consistent ways in which the sub-patterns relate to each other (induction or acquisition of rules). Phase 3 is the incorporation of these sub-elements and relational patterns into newly structured global patterns--the acquisition of true skill. "Theft," in Harnad's terms, can occur only as part of Phase 2. Both Phase 1 and Phase 3 involve "honest toil." My feeling is that current connectionist models are mainly appropriate to Phase 1, and that symbolic approaches are mainly appropriate to Phase 2, though there is necessarily overlap. There should not be a contention among models using one or other approach, if this is so. They are both correct, but under different circumstances. -- Martin Taylor DCIEM, Box 2000, Downsview, Ontario, Canada M3M 3B9 uunet!mnetor!dciem!client1!mmt or mmt@zorac.arpa (416) 635-2048 ------------------------------ Date: Tue, 13 Sep 88 15:58 EDT From: Marc Vilain <MVILAIN@G.BBN.COM> Subject: BBN AI Seminar -- Mary Haper BBN Science Development Program AI Seminar Series Lecture THE REPRESENTATION OF PRONOUNS AND DEFINITE NOUN PHRASES IN LOGICAL FORM Mary P. Harper Brown University Computer Science Dept. (MPH%cs.brown.edu@RELAY.CS.NET) BBN Labs 10 Moulton Street 2nd floor large conference room 10:30 am, Thursday September 15 Initially, I will discuss the representation of pronouns in logical form. Two factors influence the representation of pronouns. The first factor is computational. This factor imposes certain requirements on the logical form representation of a pronoun. For example, the initial representation of a pronoun in logical form should be derivable before its antecedent is known. The antecedent, when determined, should be specified in a way consistent with the initial representation of the pronoun. The second factor is linguistic. This factor requires that the representation for a pronoun should be capable of expressing the range of behaviors of a pronoun in English, especially in the domain of verb phrase ellipsis. I will review past models of verb phrase ellipsis. These models do not provide a representation of pronouns for computational purposes, and accordingly fail to meet our computational requirements. Additionally, I will show that these models fail to represent pronouns in a way which captures the full range of behaviors of pronouns. I will then propose a new representation for pronouns and show how this representation meets our computational requirements while providing a better model of pronouns in verb phrase ellipsis. The representation of definite noun phrases will also be discussed. As in the case of pronouns, there are two factors which influence this representation (i.e. modeling definite behavior and obeying our computational guidelines). I will discuss several examples which argue for representing definites as functions in logical form before pronoun resolution is carried out. I will discuss the actual representation I chose, and illustrate its use with an example. ------------------------------ Date: Wed, 14 Sep 88 10:51 EDT From: Alberto M. Segre <segre@gvax.cs.cornell.edu> Subject: CfT: Machine Learning Call for Topics: Sixth International Workshop on Machine Learning Cornell University Ithaca, New York; U.S.A. June 29 - July 1, 1989 The Sixth International Workshop on Machine Learning will be held at Cornell University, from June 29 through July 1, 1989. The workshop will be divided into four to six disjoint sessions, each focusing on a different theme. Each session will be chaired by a different member of the machine learning community, and will consist of 30 to 50 participants invited on the basis of abstracts submitted to the session chair. Plenary sessions will be held for invited talks. People interested in chairing one of the sessions should submit a one-page proposal, stating the topic of the session, sites at which research is currently done on this topic, estimated attendance, format of the session, and their own qualifications as session chair. Proposals should be submitted by November 1, 1988 to the program chair: Alberto Segre Department of Computer Science Cornell University, Upson Hall Ithaca, NY 14853-7501 USA Telephone: (607) 255-9196 Electronic mail should be addressed to "ml89@cs.cornell.edu" or "segre@gvax.cs.cornell.edu". The organizing committee will evaluate proposals on the basis of perceived demand and their potential impact on the field. Topics will be announced by early 1989, at which time a call for papers will be issued. Partial travel support may be available for some participants. ------------------------------ Date: Thu, 15 Sep 88 09:17 EDT From: William J. Rapaport <rapaport@cs.Buffalo.EDU> Subject: SUNY Buffalo Linguistics Colloquia SUNY BUFFALO LINGUISTICS COLLOQUIA Fall 1988 Co-sponsored by the Graduate Linguistics Club and the GSA Wine and cheese to follow, courtesy of the Graduate Linguistics Club ========================================================================= Thursday, Sept. 15 4:30 pm 684 Baldy John Whitman Cornell University "Korean-Japanese Historical Comparisons" ========================================================================= Friday, Sept. 16 3:00 pm 684 Baldy John Whitman Cornell University "Lexical Passives and Causatives in Korean and Japanese" ========================================================================= Friday, Sept. 30 3:00 pm 684 Baldy Geoffrey Huck University of Chicago Press "Discontinuous Constituency: Fact or Artifact?" ========================================================================= Friday, Oct. 14 3:00 pm 684 Baldy Lou Ann Gerken Dept. of Psychology SUNY Buffalo "What Children Don't Say: Competence or Performance?" ========================================================================= Friday, Oct. 28 3:00 pm 684 Baldy Donald G. Churma Dept. of Linguistics SUNY Buffalo "On `On Geminates'" ========================================================================= For further information, contact Donna Gerdts, Dept. of Linguistics, SUNY Buffalo, (716) 636-2177, linger@ubvmsc.cc.buffalo.edu ------------------------------ Date: Thu, 22 Sep 88 13:16 EDT From: Emma Pease <emma@csli.Stanford.EDU> Subject: CSLI Calendar, September 22, 4:1 [Note: anyone interested in subscribing should send mail to emma@csli.stanford.edu. Future forwardings will only include the seminar abstracts. - BWM] C S L I C A L E N D A R O F P U B L I C E V E N T S _____________________________________________________________________________ 22 September 1988 Stanford Vol. 4, No. 1 _____________________________________________________________________________ A weekly publication of The Center for the Study of Language and Information, Ventura Hall, Stanford University, Stanford, CA 94305 ____________ [...] -------------- ANNOUNCEMENT CSLI Thursdays 1988-89 in the Cordura Hall Conference Room This year the traditional TINLunch time (12:00-1:15) will be used for three different types of activities---TINLunches, TINLectures, and TINLabs---in approximately equal numbers and on an approximately rotating schedule. TINLunches will follow the familiar format; TINLectures will feature talks by outside speakers; and TINLabs will provide an introduction to research at CSLI. The TINLineup begins on 29 September with a TINLecture by Angelica Kratzer from the University of Mass; her talk will be on "Stage-level and Individual-level Predicates." On 6 October Martin Kay will lead a TINLab on "What is Unification Grammar?" Watch the CSLI Calendar for coming attractions. CSLI SEMINAR (2:15-3:45) will focus each quarter on a particular issue or problem. CSLI researchers and others will discuss the bearing their work has on the issue. The fall quarter seminar is being organized by Ivan Sag, Herb Clark, and Jerry Hobbs; the issue is: The Resolution Problem for Natural-Language Processing. How can communication proceed in light of the fact that interpretation is radically underdetermined by linguistic meaning? Languages exhibit massive ambiguity: I forgot how good beer tastes. [Structural ambiguity] The pen is empty. [Lexical ambiguity] Dukakis agrees to only three debates. [Scope ambiguity] I saw her duck. [Lexical and structural ambiguity] Kim likes Sandy more than Lou. [Ellipsis ambiguity] And massive parametricity (expressions whose interpretation relies in part on contextually fixed parameters): He is crazy. (Who's he?) John is in charge. (John who? In charge of what?) She ran home afterwards. (After what?) The nail is in the bowl. (Which nail? Nailed into the bowl, or just inside of it?) She cut the lawn/hair/cocaine/record. (What kind of cutting?) John's book. (The book he owns?/wrote?/edited?) The Resolution Problem for natural language is the question of how diverse kinds of knowledge (e.g., knowledge of local domains, context of utterance, the plans and goals of interlocutors, and knowledge of the world at large) interact with linguistic knowledge to make communication possible, even efficient. In this seminar, which will include presentations by the instructors, by Michael Tanenhaus (University of Rochester), and by David Rumelhart, we attempt to clarify the nature of the Resolution Problem and to consider a diversity of approaches toward a solution. This course is listed as Linguistics 232, Psychology 229, and Computer Science 379 and may be taken for 1-3 units by registered Stanford students. TEA will be served in the Ventura Lounge following these events at 3:45. -------------- STASS SEMINAR Counterfactual Reasoning Angelica Kratzer, UMass Linguistics Department The STASS seminar meets every other Thursday, 4:00-5:30, in the Cordura Conference Room. Everybody is welcome. ------------------------------ Date: Tue, 27 Sep 88 11:21 EDT From: finin@PRC.Unisys.COM Subject: Unisys AI Seminar: The SB-ONE Knowledge Representation Workbench AI SEMINAR UNISYS PAOLI RESEARCH CENTER The SB-ONE Knowledge Representation Workbench Alfred Kobsa International Computer Science Institute, Berkeley (on leave from the University of Saarbruecken, West Germany) The SB-ONE system is an integrated knowledge representation workbench for conceptual knowledge which was specifically designed to meet the requirements of the field of natural-language processing. The representational formalism underlying the system is comparable to KL-ONE, altough different in many respects. A Tarskian semantics is given for the non-default part of it. The user interface allows for a fully graphical definition of SB-ONE knowledge bases. A consistency maintenance system checks for the syntactical well-formedness of knowledge definitions. It rejects inconsistent entries, but tolerates and records incomplete definitions. A partition mechanism allows for the parallel processing of several knowledge bases, and for the inheritance of (incomplete) knowledge structures between parititons. The SB-ONE system is being employed in XTRA, a natural-language access system to expert systems. The use of SB-ONE for meaning representation, user modeling, and access to the expert system's frame knowledge base will be briefly described. 10:00am Friday, October 14 BIC Conference Room Unisys Paoli Research Center Route 252 and Central Ave. Paoli PA 19311 -- non-Unisys visitors who are interested in attending should -- -- send email to finin@prc.unisys.com or call 215-648-7446 -- * COMING ATTRACTION: On October 19, Marilyn Arnott (PhD from Texas in * * Chemistry) will speak on the topic of an expert system for predictive * * toxicology. The seminar will be held at 2:00 PM in the BIC Conference * * Room. An exact title and an abstract will be distributed when they * * become available. * ------------------------------ Date: Wed, 28 Sep 88 19:58 EDT From: Emma Pease <emma@csli.Stanford.EDU> Subject: From CSLI Calendar, September 29, 4:2 NEXT WEEK'S CSLI TINLAB What is Unification? Martin Kay (kay.pa@xerox.com) October 6 Unification is an operation on a pair of objects---usually expressions in a formal language---that yields a new object of the same kind. It comes up in logic, programming, and in several theories of linguistics. In particular, it comes up in the kinds of linguistic theories that are most often incorporated in computer programs. This is not because it makes for obviously "procedural" theories---quite the contrary. Why, then, does it appeal so strongly to computationalists? I will try to answer this question after first attempting to convey the basic intuition behind unification. ______________ NOTE Cordura Hall is our new building where we now hold our Thursday events. It is on the corner of Panama Street and Campus Drive, next to Ventura Hall, on the west side of the Campus. ------------------------------ End of NL-KR Digest *******************