Theoretical Issues In Natural Language Processing最新文献

{"title":"Connectionist Models: Not Just a Notational Variant Not a Panacea","authors":"D. Waltz","doi":"10.3115/980304.980317","DOIUrl":"https://doi.org/10.3115/980304.980317","url":null,"abstract":"Connectionist models inherently include features and exhibit behaviors which are difficult to achieve with traditional logic-based models. Among the more important of such characteristics are 1) the ability to compute nearest match rather than requiring unification or exact match; 2) learning; 3) fault tolerance through the integration of overlapping modules, each of which may be incomplete or fallible, and 4) the possibility of scaling up such systems by many orders of magnitude, to operate more rapidly or to handle much larger problems, or both. However, it is unlikely that connectionist models will be able to learn all of language from experience, because it is unlikely that a full cognitive system could be built via learning from an initially random network; any successful large-scale connectionist learning system will have to be to some degree \"genetically\" prewired.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"519 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123439198","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The Definitional Power of Words","authors":"B. Boguraev","doi":"10.3115/980304.980306","DOIUrl":"https://doi.org/10.3115/980304.980306","url":null,"abstract":"I am deliberate in introducing ambiguity in the title. Par t of my thesis in this brief note is going to be that there is a wealth of information relevant to a range of natural language processing functions available and extractable from the definitions of words found in obvious places like dictionaries. By dictionaries here I mean monolingual dictionaries of the style exemplified by, e.g., The Longman Dictionary of Contemporary English, The Collins English Dictionary, or Webster's Seventh New Collegiate Dictionary. This is hardly surprising, given that what is to be found in a dictionary is essentially the result of a substantial amount of work on analysing and collating data about real language and elicitating collocational and distributional properties of words and applying certain common principles of defining their meaning. Furthermore, [ am going to argue that carefully exploited interplay between notions of \"words\" and \"primitives\", can add substantial leverage to the functionality and coverage of a natural language processing system.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121480835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Connectionism and Explanation","authors":"Eugene Charniak","doi":"10.3115/980304.980319","DOIUrl":"https://doi.org/10.3115/980304.980319","url":null,"abstract":"More generally, the problem of representing new proposit.ions in connectionist networks is a real mess. The connectionists know about this however, and t, here has been some work on the topic. The basic idea is tha t one uses the s ta te of the entire network to represent a proposition ra ther than concentra t ing it a t a node. So, one might have one set o1\" nodes which represent the first a rgument of a proposition, one for the second, and one for the predicate, and each set of nodes could indicate different individuals, depending on the pat tern . The networks could be tra.ined so that , say, if \" fa ther -of\" was the propositi.ou, it, would tend to like \" j ack\" as argument one, and \" a n n \" as a rgument two, assuming one wanted to store the fact that, Jack is the father of Ann. Using the connectionist abili ty to complete pat terns, one can a.lso see how such a network might fill in \" j a c k \" if both \" fa ther -of\" and \" a n n \" where put into the appropr ia te places. David McClelland does something like this in his work on case assignment using connectionist networks. There are other ways to repreresent propositions as well (perhaps best being the work of David Touretzky) , but they all loose what is so nice about the unsophisticated version of the networks. Before, the network as a whole represented the situation as a whole, and filling in","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"95 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128590386","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Determiners, Entities, and Contexts","authors":"D. Dahl","doi":"10.3115/980304.980340","DOIUrl":"https://doi.org/10.3115/980304.980340","url":null,"abstract":"Abstract : I am concerned with the relationship between the forms of linguistic expressions noun phrases in particular, and the discourse entities to which they refer. That is, when does a noun phrase introduce a new referent into the discourse? My concern in particular is to specify the role that the discourse context plays in answering this question. A simple first approach to the relationship between noun phrases and discourse entities might suggest that definite noun phrases refer to entities which are assumed to be mutually known to the speaker and hearer, and indefinite noun phrases refer to entities which are not mutually known, and thus, that discourse context plays no role at all. This discussion will point out problems with this approach for both definite and indefinite noun phrases. I will describe examples where definite noun phrases are used to introduce new referents, and, conversely, where indefinite noun phrases do not introduce new referents. In the first case, the local focus structure provides a guide to recognizing that a new entity is involved, and in the second case, the recognition that no new entity is introduced is based on the given/new status of propositions in the discourse. I will begin by describing certain definite descriptions that introduce new entities. I will then describe some examples where indefinite descriptions do not introduce new entities. In each case, I will discuss some related processing issues. I will restrict the current discussion to deal with cases where the mutual knowledge is based on the discourse context, rather than on knowledge that the speaker and hearer bring to an interaction. In the cases of indefinites, I will also restrict my discussion to sentential contexts where an indefinite could introduce a new entity; in other words, to specific contexts, as distinguished from non-specific contexts as discussed in [Prince1981].","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125908680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"“No Better, but no Worse, than People”","authors":"David D. McDonald","doi":"10.3115/980304.980354","DOIUrl":"https://doi.org/10.3115/980304.980354","url":null,"abstract":"It might also be in the modularity of the underlying system's information: Is it propositional and easily mapped onto kernel clauses and noun phrases, or does it have some drastically different organization? Generation research today has the lion's share of the important computational linguistics problems. As more and more people work in it, it will quickly become the cutting edge, forcing extensions on understanding and knowledge representation if they are to match it as a source of insight into the nature of language and thought in the human mind. There is no appropriate goal for generation research short of matching human performance, part of which entails coming to understand the limits on that performance. We don't really know how good people are at using language; our experiments with mechanical speakers may someday tell us.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125157041","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Let’s Put the AI Back in NLP","authors":"L. Birnbaum","doi":"10.3115/980304.980331","DOIUrl":"https://doi.org/10.3115/980304.980331","url":null,"abstract":"Artificial intelligence is, or should be, at the heart of natural language processing research. After all, it is AI more than any other cognitive science that has made processing a central issue in the study of the mind. Yet, it seems to me that there has been a tendency recently on the part of many natural language researchers -even, rather inexplicably, on the part of some within the AI community itself -to view AI as playing a secondary role in the study of language, at best a useful engineering adjunct to the more important \"theoretical\" studies carried out elsewhere. One need not reach as far back as the blunderbuss attack of Dresher and Hornstein -an attack whose ferocity in fact reflected a certain amount of healthy respect for AI, or at least anxiety about its success .to find signs of this tendency. Consider, for example, the title of a recently published book, Natural Language Parsing: Computational, Psychological, and Theoretical Perspectives. It seems reasonably clear what the \"computational\" and \"psychological\" perspectives mentioned in the title are intended to refer to, but what does \"theoretical\" mean in this context? That becomes clearer, perhaps, when we observe that the book was edited by several linguists. The use of the term \"theoretical\" in place of \"linguistic\" reflects, I suspect, not only a desire to rhyme but also an unconscious assumption on the part of the editors that they amount to the same thing when it comes to the study of parsing.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133335892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Information, Unification and Locality","authors":"Fernando C Pereira","doi":"10.3115/980304.980311","DOIUrl":"https://doi.org/10.3115/980304.980311","url":null,"abstract":"In a unification framework we deal with a domain of descriptions P tha t are used to classify objects from some class under analysis 0 , u t terances and their f ragments in the case tha t concerns us here. Classification is given by a description relation ~ between objects in 0 and elements of P. If d is a [partial] description of e, we write e ~ d, e satisfies (or is described by) d. The set of all objects tha t satisfy a description d is wri t ten [dl. Descriptions are in general partial, that is, a description d may in general be extended to a more specific (more informative) description d I. With suitable technical assumptions, this gives a partial order d ~ d' on descriptions. In terms of the description relation ~ , d E d' iff for every object e, e ~ d whenever e ~ d ~. Two descriptions d and d' are compatible if there is a description d\" such tha t d _C d\" and d ~ _C d\", tha t is if d and d' can both be extended to a single description more informative than both. If two descriptions d and d ~ are compatible, it is reasonable to assume tha t there is a least specific description d U d' more specific tha t both d and d'. In other words, d U d' contains all the information in d and d', bu t no more. For historical reasons, d t.J d' is called the unification of d and d'. In more s tandard mathemat ica l terminology, d u d' is the join of d and d'. In terms of the description relation, if e ~ d it d', e ~ d and e ~ d'. Fur thermore , we want unification to behave like logical conjunction: if e ~ d and e ~ d', e ~ d U d'. Thus lid U d'~ = [[d] n [d'~ holds for any compatible descriptions d and d'. The domains of objects, descriptions and the description relation are usually infinite, even though there may be some way of finitely characterizing the description relation. Such a characterization is a grammar . To write grammars, we need to be able to constrain entities to satisfy arbitrari ly complex descriptions without giving the descriptions in full. Our main ins t ruments for this are parameterized descriptions and rules. A parameter ized description d(pl,. . . ,pk) is not a description itself, but ra ther an encoding of a function from k-tuples of descriptions to descriptions. An object e satisfies such a parameter ized description iff there are descriptions f l , . . . , fk such tha t e satisfies d( f l , . . . , f~). Given a family of parameter ized descriptions (di)iel with parameters (Pi)i~s and a set C of constraints involving the parameters , a family of objects (ei)ie I satisfies the parameter ized descriptions relative to the constraints iff there are descriptions (fi)ieJ tha t can be uniformly replaced for the parameters in such a","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"CATV-2 4","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1987-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114123887","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Taxonomic Lattice Structures for Situation Recognition","authors":"W. Woods","doi":"10.3115/980262.980268","DOIUrl":"https://doi.org/10.3115/980262.980268","url":null,"abstract":"The kinds of intelligent computer assistants that we would like to be able to construct are very much like intelligent organisms in their own right. Imagine for a moment an intelligent organism trying to get along in the world (find enough food, stay out of trouble, satisfy basic needs, etc.). The most valuable service played by an internal knowledge base for such an organism is to repeatedly answer questions like \"what's going on out there?\", \"can it harm me?\", \"how can I avoid/placate it?\", \"Is it good to eat?\", \"Is there any special thing I should do about it?\", etc. To support this kind of activity, a substantial part of the knowledge base must be organized as a recognition device for classifying and identifying situations in the world. The major purpose of this situation recognition is to locate internal procedures which are applicable (appropriate, permitted, mandatory, etc.) to the current situation.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"128 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1978-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124223343","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"What has language to do with perception? Some speculations on the Lingua Mentis","authors":"Z. Pylyshyn","doi":"10.3115/980262.980290","DOIUrl":"https://doi.org/10.3115/980262.980290","url":null,"abstract":"The topic under consideration in this conference session (viz. Language and Perception) is not the one to which the greatest amount of attention has been devoted in philosophy of mind and philosophy of language. There a major concern has been the relation between language and thought. As everyone knows there has been a long standing dispute regarding whether or not it makes sense to view thought as being carried out in the medium of natural language or whether some other form of representation is involved. There has not been a comparable dispute over the relationship between perception and language. For one thing no one to my knowledge has proposed that perception occurs through the medium of natural language (though some early behaviorist writings come close, especially in respect to memory for perceptual events). What I propose to consider in this brief note are some respects in which the language-thought relationship is similar to the language-perception relationship.","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1978-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131249411","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Description Formation and Discourse Model Synthesis","authors":"B. Webber","doi":"10.3115/980262.980270","DOIUrl":"https://doi.org/10.3115/980262.980270","url":null,"abstract":"Many researchers in linguistics, psychology, philosophy and artificial intelligence have recently begun to abandon a purely linguistic approach to definite anaphora in favor of a notion of reference into some kind of model of the discourse, cf. Karttunen [1976], Levin & Goldman [1978], Lyons [1978]. Stenning [1975]. My own research on definite anaphora (cf. Webber [1978a&b]) follows this approach, in particular making the following five assumptions:","PeriodicalId":439225,"journal":{"name":"Theoretical Issues In Natural Language Processing","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1978-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131475875","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}