nl-kr-request@CS.ROCHESTER.EDU (NL-KR Moderator Brad Miller) (01/05/88)
NL-KR Digest (1/05/88 00:08:23) Volume 4 Number 2 Today's Topics: Language engineering Computer Languages as Social/Philosophical Models Early Linguistic Emphasis Re: Linguistics & artificial language design ("linguistic science") Re: Language Learning (a Turing test) Seminar - Intelligent Agents as NL Interfaces Seminar - The lexicon in human speech understanding Seminar - Lang. & Cognition Thiersch Seminar Submissions: NL-KR@CS.ROCHESTER.EDU Requests, policy: NL-KR-REQUEST@CS.ROCHESTER.EDU ---------------------------------------------------------------------- Date: Wed, 30 Dec 87 10:34 EST From: WATKINS@rvax.ccit.arizona.edu Subject: Language engineering I believe there is some (very small-scale) language engineering going on out there, on English idioms anyway--done by advertisers, mostly. I am interpreting engineering as deliberate creation of new mechanisms intended to serve a purpose specified in advance (though of course serendipity applies). If one is simply tinkering with language to figure out how it ticks and/or how it can tick, that's closer to experimentation...though where language is concerned it's hard to construct experiments with adequate controls, repeatability, and other characteristics required by the physical sciences. Thus one measure of an engineered product lies in its ultimate use in order to do something else (fly, heal the injured, teach children, communicate accurately); and the test of the product's success lies in the success of that use. On these grounds I find such products as Esperanto and Loglan marginally successful at best; most computer languages eminently successful; and the products of a few (lucky? ingenious?) advertisers quite good too. But this attitude toward engineering makes us all engineers of a sort (more or less deliberate, more or less successful). I engineer our family dialect quite deliberately sometimes, as my husband engineers our door latches for our small children. Maybe it would be useful to define more closely what we mean when we talk of someone engineering a language (bearing in mind also that many engineering projects are group efforts). ------------------------------ Date: Sun, 3 Jan 88 14:41 EST From: Ken Laws <LAWS@IU.AI.SRI.COM> Subject: Computer Languages as Social/Philosophical Models As I recall earlier discussions of computer languages (or attempts to avoid such discussions), the claim was made that since such languages are artificial there is nothing unknown and in need of study. Since modern CLs were designed with BNF grammars in mind, there doesn't seem to be much point in studying their syntax. Their semantics as formal systems also seems well understood, although subroutine calls and other interfaces create interesting complexities. The linguistically interesting thing about these languages is, or should be, their adequacy. CLs and Computer Science are largely concerned with precise characterization of process and information flow. LISP, in particular, has been touted as an appropriate representation language for both structured knowledge and operations on that knowledge -- a strong claim, similar to those made for various logics. The "connectionists" have recently begun making similar claims for some of thier networks. They do not offer well developed languages or logics yet, but are moving in that direction. Efforts are being made to embed LISP in connectionist representations and vice versa. Rather than wish the connectionists luck, I hope that they fail in the unification but discover spatio-temporal representations of process and information beyond anything yet available -- or at least comparable to the power of natural language. Another modern trend in computer science is the study of distributed systems, either as interacting autonomous agents or as cooperating processes. Contract nets and other models of human social organizations are being tested and characterized, and someday we may develop effective languages and logics for this work. This experimental science should be of interest to many linguists. Current computer languages are very limited compared to natural ones, but in their own domains they model the structure and possibilities of the real world better than do natural languages. Would that we could develop such rigor in human thought and communication, where algorithmic notions are poorly expressed. The study of natural languages tends toward ever-increasing discussion (knowledge?) of many inconsistent views of reality, both because these views (i.e., languages) are inherently interesting and because we hope that their intersection will lead to a consistent view of reality. The design of computer languages tends toward ever-decreasing complexity as we learn to combine the functions of older languages in more powerful ones, thus homing in on the consistent view that linguists and philosophers seek. -- Ken Laws ------------------------------ Date: Sun, 3 Jan 88 15:27 EST From: Ken Laws <LAWS@IU.AI.SRI.COM> Subject: Early Linguistic Emphasis A hypothesis in cognitive psychology: We all know that first/only children go to MIT and get listed in Who's Who. This may be due to many environmental factors, particularly those related to interaction with siblings. A linguistic influence (for general IQ) has also been posited: that first children are exposed to adult conversation whereas later children are exposed to childish speech and thought patterns. I would like to suggest another linguistic influence, one that might be tested by experiment. I notice that my third child is a happy baby, exchanging eye contact and grins with everyone around him. I suspect that he is going to be a "people person" rather than a "nerd". Is this genetic or environmental? What makes him different from my first child, also a boy, who strikes me as more interested in the physical world? A linguistic difference in their upbringing is that the first words we are teaching the third child, after "mama" and "dada", are the names of his siblings. We have also been pointing out Big Bird and other characters on TV, but have made no special effort to name the dozens (hundreds?) of toys lying around the house. My first born, on the other hand, was taught the names of physical objects long before other people or TV characters became important in his life. Each new toy was presented as something special, to be learned and studied. He was drilled on the alphabet rather than learning it from simple exposure as my second child did. I will never know whether differences between my children are due to such differences in parental emphasis, but I hypothesize that such an effect could be measured by a controlled experiment. The linguistic experiment, then, involves the raising of two groups of children who are matched for number of siblings and other environmental factors. One group of parents and siblings would be asked to stress object names, the other to stress people names. Only the earliest linguistic training would be modified (e.g., the first nine months), so parents needn't feel that their assigned role might ruin the kid. Any significant difference in the kids' later school performance or personality profiles could be attributed to the early linguistic training. Such experiments aren't my field, so I offer the idea gratis to anyone who wants it. Perhaps such studies have already been done, in which case I'd be interested in reading a summary. -- Ken Laws ------------------------------ Date: Sun, 3 Jan 88 19:28 EST From: Mark William Hopkins <markh@csd4.milw.wisc.edu> Subject: Re: Linguistics & artificial language design ("linguistic science") In article <37343@sun.uucp> landauer@sun.UUCP (Doug Landauer) writes: >I'm not a linguist; sci.lang and comp.lang.* are just about the only >reading I do in anything resembling linguistics. I'm ignoring the >questions Walter raised about the study of language as a behavioral >science (not because it's not a good question, but because it's not >what I'm most interested in). I'm also almost ignoring John Chambers' >questions and comments about computer languages, because I believe that >computer languages are (currently) too small and too precisely >specified to be particularly relevant to the study of the languages >that people use to talk with one another. In fact, I don't even care >whether linguistics is a science. > >Walter Rolandi thought ... >> that the purpose of linguistics was to determine a causal >> explanation of linguistic phenomena by means of the scientific method. > >My first reaction is that because languages are man-made, there *could English is not man-made in the sense that Esperanto is. None of its features was ever designed. Everything arose as a product of historical evolution, by accident, as it were. Languages tie in closely to cultures (but to a lesser degree for world languages, such as English, French or Spanish.) For example, the classification of words in many languages has ties to the underlying mythology and world-view of the culture in which the language is spoken. Even the noun classification of German may have arisen for a mythology and religious tradition that regarded things as masculine or feminine. >be* much more of an engineering aspect to linguistics, or to some >sub-branch of linguistics. Maybe "linguistics" is a specialized branch >of the more general study of languages and of language features. There >might be sort of a spectrum, from descriptive linguistics (what most >linguists do today), through prescriptive (grammarians), through >reformers (not too many of these around any more), to language >engineers (language designers -- e.g., Zamenhof, Jesperson, James Cooke >Brown, Dennis Ritchie, Niklaus Wirth, John Backus, et al.). > > >The question I have is -- why do linguists show such little interest in >language design? Why do linguists totally ignore Esperanto, Loglan, >and all computer languages? Perhaps I can offer an answer here. To design a language a viable semantic theory is necessary. I come from the "school of thought" that says that a language is an instrument of communication before all else. Therefore, in my frame of mind, the design of a language has to be based on a semanatic model before all else. The problem, in both Programming Languages and in Linguistics is that people have been concentrating too much on Syntax at the expense of Semantics. It is a semantics that makes a language, the syntax is just there as "sugar-coating". Also, there are not very many people in Linguistics with an expertise in Computer Science or one of the Engineering disciplines or vice versa (like myself), except in Computational Linguistics. There is also the problem of where to start, as a natural language is a very big system. Finally, there is the question of how to deal with the "corrupting" influences of inevitable historical change and cultural and religious factors. If this is not taken into consideration, then you'll end up designing another Latin. I offer an answer to the question of where to start, and that is to start with the adverbials that denote space, time, causality and logic. As soon as one can reconstruct the underlying semantics to this group then one has the skeleton of a natural language. Language design has been one of my pet projects for a very long time. In a language like Hungarian, the space and time denoting words already show a regularity that makes it possible to organise them into paradigms. This needs to be done with other languages to come up with either language Universals or a classification of the different types of semantics that exist in today's languages. A good place to start is with English with the following set of words: TIME & SERIES: SPACE: until,during,since,while, where,here,there, before,early,first, whence,hence,thence, after,late,last, whither,hither,thither, when,now,then, in,into,out,out of, always,sometimes,never,ever, on,onto,off,off of, again,once,twice,thrice,often,rarely at,to,from,away from, near,far,away, LOGIC & CAUSALITY: which way,toward,onward,upward, unless,if,then, downward,forward,backward, why,because,how,so,thus, everywhere,somewhere,nowhere,anywhere wherefore,therefore anyhow,somehow,no how,not I am not a linguist (meaning I have no PhD in Linguistics) but I have a very broad background in the field accquired on and off for the last 10 to 15 years. So, to answer your questiuon, there IS somebody currently working on language design. Another answer to your question is that if you want to see something done, you might have to do it yourself (designing a language, that is). ------------------------------ Date: Thu, 31 Dec 87 12:36 EST From: Robert K. Coe <bobcoe@cca.CCA.COM> Subject: Re: Language Learning (a Turing test) In article <3111@bcsaic.UUCP> rwojcik@bcsaic.UUCP (Rick Wojcik) writes: >So the issue is not just degree of fluency. Why is there such a >general discrepancy between adult and child language learning? What >is it that so severely damages our ability to acquire a foreign language? I don't understand why the subscribers of this newsgroup find the answer to this question so elusive. The young child readily learns the patterns of his native language because he lacks preconceived notions of how a language ought to sound. Once a set of speech patterns becomes ingrained over time, a person tends to react to what he thinks he heard, based on his previous experience, rather than on what he really heard. Phonetic distinctions that are not phonemic in a speaker's native language are often overlooked. Examples in American English include nasalation of vowels, aspiration of stops, and devoicing of "l" and "r". As far as the degree to which an adult can achieve command over a foreign language is concerned, one should be careful not to draw unwarranted conclu- sions from apparent "evidence". Note that when you hear someone whom you don't know speak, you really have no idea what his native language is. Full native command over more than one language may be more common than we think. -- * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * | Robert K. Coe | bobcoe@cca.cca.com | | Computer Corporation of America | | | 4 Cambridge Center, Cambridge, Mass. 02142 | 617-492-8860, ext. 428 | ------------------------------ Date: Sat, 2 Jan 88 20:40 EST From: Dr. Thomas Schlesinger <tos@psc90.UUCP> Subject: Re: Language Learning (a Turing test) Ref. posting ventured that there may be more second (or multi-?) native language learning than is commonly thought. AMEN! Think about Africa, e.g. Nigeria with 200 languages plus English; South Africa, where every Black person grows up with their own Black language plus either Afrikaans or English, sometimes both. India with app. 200-500 languages -- 14 official languages, NOT counting English... where vast numbers of people grow up bi-lingual or multi-lingual from the start. One could go on and on... hundreds of thousands in Islamic countries learn Arabic as a second language in Koranic school at age 3 or 4 in addition to their mother tongue. 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." But I believe that literally hundreds of millions of children in the world grow up that way. ------------------------------ Date: Tue, 22 Dec 87 19:19 EST From: David N. Chin <chin@renoir.Berkeley.EDU> Subject: Seminar - Intelligent Agents as NL Interfaces Doctoral thesis Seminar 1-3pm, Wednesday, December 23, 1987 206 Evans Hall Intelligent Agents as a Basis for Natural Language Interfaces David Ngi Chin Computer Science Division University of California, Berkeley Berkeley, CA 94720 ABSTRACT Typical natural language interfaces respond passively to the user's commands and queries. They cannot volunteer information, correct user misconceptions, or reject unethical requests. In order to do these things, a system must be an intelligent agent. UC (UNIX Consultant), a natural language system that helps the user solve prob- lems in using the UNIX operating system, is such an intelligent agent. The agent component of UC is UCEgo. UCEgo provides UC with its own goals and plans. By adopting different goals in different situa- tions, UCEgo creates and executes different plans, enabling it to interact appropriately with the user. UCEgo adopts goals from its themes, adopts sub-goals during planning, and adopts meta-goals for dealing with goal interactions. It also adopts goals when it notices that the user either lacks necessary knowledge, or has incorrect beliefs. In these cases, UCEgo plans to volunteer information or correct the user's misconception as appropriate. These plans are pre- stored skeletal plans that are indexed under the types of situations in which they are typically useful. Plan suggestion situations include the goal which the plan is used to achieve, the preconditions of the plan, and appropriateness conditions for the plan. Indexing plans by situations improves efficiency and allows UC to respond appropriately to the user in real time. Detecting situations in which a plan should be suggested or a goal adopted is implemented using if- detected daemons. The user's knowledge and beliefs are modeled by the KNOME (KNOwledge Model of Expertise) component of UC. KNOME is a double- stereotype system which categorizes users by expertise and categorizes UNIX facts by difficulty. KNOME deduces the user's level of expertise during the dialog with the user. After UCEgo has selected a plan, it is refined through the pro- cess of answer expression by the UCExpress component. UCExpress first prunes the answer to avoid telling the user something that the user already knows, and to mark where to use anaphora or ellipsis in generation. UCExpress also uses specialized expository formats to express different types of information in a clear, concise manner. The result is ready for generation into English. ------------------------------ Date: Fri, 1 Jan 88 16:36 EST From: Yigal Arens <arens@vaxa.isi.edu> Subject: Seminar - The lexicon in human speech understanding Speaker: William Marslen-Wilson, Cambridge Title: Human speech understanding: The role of the lexicon Place: ISI, 11th floor conference room Time: Wednesday, January 6, 1988, 3-5pm Biographical information: William Marslen-Wilson received his Ph.D in experimental psycholinguistics from MIT in 1973. His principal interest is in the comprehension of spoken language, and he has worked on several different aspects of this problem, ranging from the nature of the acoustic-phonetic interface with the mental lexicon, to the manner in which listeners construct a mental model of the current discourse. He taught at the University of Chicago from 1973 to 1977, before moving to the Max-Planck Institute in Nijmegen from 1977 to 1982. Following a spell teaching at the University of Cambridge, he returned to Nijmegen as director from 1984 to 1987. He now works in the Medical Research Council Applied Psychology Unit, in Cambridge, England. Abstract The process of spoken word-recognition breaks down into three basic functions -- of access, selection, and integration. Access concerns the mapping of the speech input onto the representations of lexical form, selection concerns the discrimination of the best-fitting match to this input, and integration covers the mapping of syntactic and semantic information at the lexical level onto higher levels of processing. The lecture will present a "cohort" based approach to the lexical processing problem, showing how it embodies the the concepts of multiple access and multiple assessment, allowing a maximally efficient recognition process, based on the principle of the contingency of perceptual choice. If any one wants some background reading, I recommend the paper by myself "Functional parallelism in spoken word-recognition", in Cognition, 25, 71-102, 1987. ------------------------------ Date: Mon, 4 Jan 88 10:02 EST From: Dori Wells <DWELLS@G.BBN.COM> Subject: Reminder Lang. & Cognition Seminar BBN Science Development Program Language and Cognition Seminar THE EMERGENCE OF UTTERANCE MEANING THROUGH SOCIAL INTERACTION Charles and Marjorie Goodwin Department of Anthropology University of South Carolina Columbia, South Carolina BBN Laboraatories 10 Moulton Street Large Conference Room, 2nd Floor 10:30 a.m., Thursday, January 7, 1988 Abstract: Using micro-analysis of video-taped materials, we will show how utterances (and the sentences being made visible through them) are shaped by ongoing processes of interaction between speaker and recipient(s) that is occurring while the utterance is being spoken. The emerging utterance is modified as various contingencies emerge within the interaction. For example as speaker moves his or her gaze from one possible recipient to another, the emerging sentence is changed so that it remains appropriate to its recipient of the moment. As the interaction unfolds new segments are added to the emerging utterance, other projected segments are deleted and the emerging meaning of the utterance is reconstructed. The utterance thus emerges not from the actions of speaker alone, but rather as the result of an collaborative process of interaction that includes the active participation of recipient(s) as well. For information about this Seminar Series contact Livia Polanyi at 873-3455 [lpolanyi@g.bbn.com] ------------------------------ Date: Mon, 4 Jan 88 10:08 EST From: Dori Wells <DWELLS@G.BBN.COM> Subject: Thiersch Seminar BBN Science Development Program Language & Cognition Seminar PARSING WITH PRINCIPLES & PARAMETERS: Prolegomena to a Universal Parser Craig Thiersch Kath. Universiteit Brabant Tilburg, Netherlands BBN Laboratories Inc. 10 Moulton Street Large Conference Room, 2nd Floor 10:30 a.m., Wednesday, January 13, 1988 Abstract: We have constructed a pilot parser for natural languages based on recent advances in linguistic theory, and using a different structural concept from that of previous natural language parsers. Many linguistic theories [e.g. Generalized Phrase Structure Grammar (GPSG)] can be regarded as rule-oriented, in that they propose to characterize the speaker's linguistic competence through a set of language specific rules. In contrast, some generative grammarians [now represented chiefly by the Government/Binding (G/B) school] now hypothesize that a speaker's competence consists of a set of (universal) principles governing the potential phrase-structures available to human language and conditions restricting their deformations and their semantic properties rather than collections of language- (and construction-) specific rules. The apparent wide variety of constructions occurring in natural languages is presumed to be the result of the interaction of these general mechanisms with (a) idiosyncracies of lexical items and (b) language specific parameters (or defaults), which are set early in the child's language acquisition process. More specifically, we assume [a] strict modularity of the components of the grammar; [b] projection from the lexicon; [c] use of general principles, rather than specific rules; [d] parametrizability across languages. We have built a first experimental pilot parser which adheres as closely as possible to these basic premises as a foundation for future research. We allow enough flexibility so that it is not bound, for example, to a particular instantiation of G/B theory, but can be used as a testing ground for various theoretical hypotheses in this general framework, across a wide variety of languages. This is joint work with Hans-Peter Kolb. ------------------------------ End of NL-KR Digest *******************