| { |
| "paper_id": "P82-1034", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T09:16:16.378000Z" |
| }, |
| "title": "DESIG~ DIMENSIONS FOR NON-NORMATIVE ONDERSTARDING SYSTEMS", |
| "authors": [ |
| { |
| "first": "Robert", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bobrow", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Bolt Beranek and Newman Inc", |
| "location": { |
| "addrLine": "10 Moulton Street Cambridge", |
| "postCode": "02238", |
| "region": "Massachusetts" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Madelelne", |
| "middle": [], |
| "last": "Bates", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Bolt Beranek and Newman Inc", |
| "location": { |
| "addrLine": "10 Moulton Street Cambridge", |
| "postCode": "02238", |
| "region": "Massachusetts" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "P82-1034", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "This position paper is not based upon direct experience with the design and implementation of a \"non-normative\" natural language system, but rather draws upon our work on cascade [11] architectures for understanding systems in which syntactic, semantic and discourse processes cooperate to determine the \"best\" interpretation of an utterance in a given discourse context. The RUS and PSI-KLONE systems [I, 2, 3, 4, 5] , which embody some of these principles, provide a strong framework for the development of non-normatlve systems as illustrated by the work of Sondhelmer and Welschedel [8, 9, 10] and others.", |
| "cite_spans": [ |
| { |
| "start": 179, |
| "end": 183, |
| "text": "[11]", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 402, |
| "end": 417, |
| "text": "[I, 2, 3, 4, 5]", |
| "ref_id": null |
| }, |
| { |
| "start": 587, |
| "end": 590, |
| "text": "[8,", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 591, |
| "end": 593, |
| "text": "9,", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 594, |
| "end": 597, |
| "text": "10]", |
| "ref_id": "BIBREF10" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "Here we pose a number of questions in order to clarify the theoretical and practical issues involved in building \"non-normatlve\" natural language systems.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "We give brief indications of the range of plausible answers, in order to characterize the space of decisions that must be made in deslgnlng such a system.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "The first questions cover what is intended by the illdefined term \"non-normatlve system\", beyond the important but vague desire for a \"friendly and flexible\" computer system. The remaining questions cover several of the architectural issues involved in building such a system, including the categories of knowledge to be represented in the system, the static modularization of these knowledge sources, and the dynamic information and control flow among these modules.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "The way the system is to deal with ill-formed input depends in a strong way on how much the system is expected to do with well-formed input. Ad hoc data base retrieval systems (a currently hot topic) pose different constraints than systems that are expected to enter into a substantlal dialogue with the user.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "When the behavior of the system is severely limited even given perfect input, the space of plausible inputs is also limited and the search for a reasonable interpretation for ill-formed input can be made substantially easier by asking the user a few well-chosen questions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "In the dbms retrieval domain, even partially processed input can be used to suggest what information the user is interested in, and provide the basis for a useful clarification dialogue.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I. Introduction", |
| "sec_num": null |
| }, |
| { |
| "text": "The system may be expected to understand the input but not provide direct feedback on errors (e.g. by independently decldlng on the (most plausible) interpretation of the input, or by questioning the user about possible alternative interpretations). Alternatively, the system might provide feedback about the probable source of its difficulty, e.g. by pointing out the portion of the input which it could not handle (if it can be localized), or by characterizing the type of error that occurred and describing general ways of avoiding such errors in the future.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "What is the system expected to do with illformed input?", |
| "sec_num": null |
| }, |
| { |
| "text": "What are the overall performance objectives of the system? Marcus has argued [7] that the \"wellformedness\" constraints on natural langua6e make it possible to parse utterances with minimal (or no) search. 2 The work we have done on the RU3 system has convinced us that this is true and that cascading semantic interpretatlon with syntactic analysis can further improve the efficiency of the overall system.", |
| "cite_spans": [ |
| { |
| "start": 77, |
| "end": 80, |
| "text": "[7]", |
| "ref_id": "BIBREF6" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "System performance goals", |
| "sec_num": "2." |
| }, |
| { |
| "text": "The question naturally arises as to whether the performance characteristics of this model must be abandoned when the input does not satisfy the well-formedness constraints imposed by a competence model of language.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "System performance goals", |
| "sec_num": "2." |
| }, |
| { |
| "text": "We believe that it is possible to design natural language systems that can handle well-formed input efficiently and ill-formed input effectively.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "System performance goals", |
| "sec_num": "2." |
| }, |
| { |
| "text": "In order to design a fault-tolerant language processing system, it is important to have a model for the component processes of the system, how they interact in handling well-formed input, and how each process is affected by the different types of constraint vlolatlons occurring In 111formed input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "What categories of knowledge are needed to understand well-formed input, and how are they used?", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Typically, a natural language understandlng system makes use of lexical and morphological knowledge (to categorize the syntactic and semantic properties of input items), syntactic knowledge, semantic knowledge, and knowledge of discourse phenomena (here we include issues of ellipsis, anaphora and focus, as well as plan recognition (\"why did he say this to me now?\") and rhetorical structure).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Of course, saying that these categories of knowledge are represented does not imply anything about the static (representational) or dynamic (process interaction) modularizatlon of the resulting system.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "We will assume that the overall system consists of a set of component modules.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "One common decomposition has each category of knowledge embodied in a separate component of the NLU system, although it is possible to fuse the knowledge of several categories into a single prooeas.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Given this assumption, we must then ask what control and information flow can be imposed on the interaction of the modules to achieve the overall performance goals imposed on the system. The other results of the parsing process are less essential.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Thus, for example, the case structure of clauses is often highly constrained by the semantic features of the verb and the constituent noun phrases, and it is possible to reconstruct it even with minimal syntactic guidance (e.g. \"throw\" \"the bali\" \"the boy\").", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "How can each component fill its role in the overall system when the constraints and assumptions that underlie its design are violated by ill-formed input?", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "The distinction between the information used within a component from the information which that component is required to provide to other components is critical in designing processing strategies for each component that allow it to fulfill its primary output goals when its input violates one or more well-formedness constraints. Often more than one source of information or constraint may be available to determine the output of a component, and it is possible to produce well-formed output based on the partial or conflicting internal information provided by Illformed input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "For example, in systems with feedback between components, it is possible for that feedback to make up for lack of information or violation of constraints in the input, as when semantic coherence between subject and verb is sufficient to override the violation of the syntactic number agreement constraint.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "When the integrity of the output of a component can be maintained in the face of ill-formed input, other components can be totally shielded from the effects of that input. In general, the problem to be solved by a non-normative language understander can be viewed as one of finding a \"sufficiently good explanation\" for an utterance in the given context. ~ A number of approaches to this problem can be distinguished.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "One approach attempts ~o characterize the class of error producing mechanisms (such as word transposition, mistyping of letters, morphological errors, resumptive pronouns, etc.).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Given such a characterization, recognition criteria for different classes of errors, and procedures to invert the error process, an \"explanation\" for an ill-formed utterance could be generated in the form of an intended well-formed utterance and a sequence of error transformations.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "The system would then try to understand the hypothesized well-formed utterance.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "While some \"spelling corrector\" algorithms use this approach, we know of no attempt to apply it to the full range of syntactic, semantic and pragmatic errors.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "We believe that some strategies of this sort might prove useful as components in a larger errorcorrecting system.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "A more thoroughly explored set of strategies for non-normative processing is based on the concept of \"constraint relaxation\". If a component can find no characterization of the utterance because it violates one or more constraints, then it is necessary to relax such constraints.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "A number of strategies have been proposed for relaxing well-formedness constraints on input to permit components to derive wellstructured output for both well-formed and illformed input:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "1. extend the notion of well-formed input to include (the Common cases of) illformed input (e.g. make the gr,m,~r handle ill-formed input explicitly);", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "2. allow certain specific constraints to be overridden when no legal operation succeeds;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "3. provide a process that can diagnose failures and flexibly override constraints.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Somehow the \"goodness\" of an explanation must be related to the number and type of constraints which must be relaxed to allow that explanation. How good an explanation must be before it is accepted is a matter of design choice.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Must it simply be \"good enough\" (above some threshold), or must it be guaranteed to be \"the best posslble\" explanation?", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "If it must be \"the best possible\", then one can either generate all possible explanations and compare them, or use some strategy like the shortfall algorithm [12] that guarantees the first explanation produced will be optimal.", |
| "cite_spans": [ |
| { |
| "start": 158, |
| "end": 162, |
| "text": "[12]", |
| "ref_id": "BIBREF12" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "While space prohibits discussion of the advantages and disadvantages of each of these strategies, we would llke to present a number of design dimensions along which they might be usefully compared.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "We believe that choices on these dimensions (made implicitly or explicitly) have a substantial effect on both the practical performance and theoretical interest of the resulting strategies. These dimensions are exemplified by the following questions: It may be that more robust parslr~ strategies can be found.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Without stating how the semantic component might relax its constraints, we might still point out the parallel between constraint violation in syntax and such semantic phenomena as metaphor, personification and metonymy. We believe that, as in the syntactic case, it will be useful to distinguish between the internal operation of the semantic interpreter and the interface between it and discourse level processes.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "It should also be possible to make use of feedback from the discourse component to overcome violations of semantic constraints.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "In the context of a waiter talking to a cook about a customer complaint, the sentence \"The hamburger is getting awfully impatient.\" should be understood.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Architectural issues", |
| "sec_num": "3." |
| }, |
| { |
| "text": "We believe that it will be possible to design robust systems without giving up many valuable features of those systems which already work on well-formed input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "In particular, we believe it will be possible to build such systems on the basis of competence models for various linguistic components, which degrade gracefully and without the use of ad hoc techniques such as pattern matching.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "One critical resource that is needed is a widely available, reasonably large corpus of \"illformed input\", exhibiting the variety of problems which must be faced by practical systems. This corpus should be sub-divlded by modallty, since it is known that spoken and typewritten interactions have different characteristics.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "The collections that we know of are either limited in modality (e.g. the work on speech errors by Fromkin [6] ) or are not widely available (e.g. unpublished material collected by Tony Kroch).", |
| "cite_spans": [ |
| { |
| "start": 106, |
| "end": 109, |
| "text": "[6]", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "It would also be valuable if this material were analyzed in terms of possible generative mechanisms, to provide needed evidence for error recovery strategies based on inversion of error generation processes.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "Finally, we believe that many error recovery problems can be solved by using constraints from one knowledge category to reduce the overall sensitivity of the system to errors in another category.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "To this end, work is clearly needed in the area of control structures and cooperative process architectures that allow both pipelinlng and feedback among components with vastly different internal knowledge bases.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "1The preparation of this paper was supported by the Advanced Research Projects Agency of the Department of Defense, and monitored by the Office of Naval Research under contract NO001q-77-C-0378.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "2The parser designed by G. Ginsparg also has similar search characteristics, given grammatical input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| }, |
| { |
| "text": "3What constitutes \"sufficiently good\" depends, of course, on the overall goals of the system. ~In almost any case, we believe, the information available at the Interface between components should be expressed primarily in terms of some competence model.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "q. Conclusions", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Tools for Syntactic and Semantic Interpretation", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Bates", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bobrow", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [ |
| "L" |
| ], |
| "last": "Webber", |
| "suffix": "" |
| } |
| ], |
| "year": 1981, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bates, M., Bobrow, R. J. and Webber, B.L. Tools for Syntactic and Semantic Interpretation. BBN Report ~785, Bolt Beranek and Newman Inc., 1981.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "The RUS Parsing System", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Bates", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bobrow", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bates, M. and Bobrow, R. J. The RUS Parsing System. Bolt Beranek and Newman Inc., forthcoming.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Parsing and Semantic Interpretation in the BBN Natural Language Understanding System", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bobrow", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "The", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "System", |
| "suffix": "" |
| } |
| ], |
| "year": 1978, |
| "venue": "", |
| "volume": "3878", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bobrow, R. J. The RUS System. BBN Report 3878, Bolt Beranek and Newman Inc., 1978. q. Bobrow, R. J. & Webber, B. L. PSI-KLONE - Parsing and Semantic Interpretation in the BBN Natural Language Understanding System.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "CSC3I/CSEIO Annual Conference, CSCSI/CSEIO", |
| "authors": [], |
| "year": 1980, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "CSC3I/CSEIO Annual Conference, CSCSI/CSEIO, 1980.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Knowledge Representation for Syntactic/Semantic Processing", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [ |
| "J" |
| ], |
| "last": "Bobrow", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [ |
| "L" |
| ], |
| "last": "Webber", |
| "suffix": "" |
| } |
| ], |
| "year": 1980, |
| "venue": "Proceedings of The First Annual National Conference on Artificial Intelligence", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bobrow, R. J. & Webber, B. L. Knowledge Representation for Syntactic/Semantic Processing. Proceedings of The First Annual National Conference on Artificial Intelligence, American Association for Artificial Intelligence, 1980.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Soeeoh Errors ssl~ Evidence", |
| "authors": [ |
| { |
| "first": "Victoria", |
| "middle": [ |
| "A" |
| ], |
| "last": "Fromkln", |
| "suffix": "" |
| } |
| ], |
| "year": 1973, |
| "venue": "Janua~, Series malor", |
| "volume": "77", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Fromkln, Victoria A.. Janua~, Series malor. Volume 77: Soeeoh Errors ssl~ Evidence. Mouton, The Hague, 1973.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "A Theory of Svntactic~ for Nstural LB\u00a3E~Ig~", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Marcus", |
| "suffix": "" |
| } |
| ], |
| "year": 1980, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcus, M.. A Theory of Svntactic~ for Nstural LB\u00a3E~Ig~. MIT Press, 1980.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "A Rule-Based Approach to Ill-Formed Input", |
| "authors": [ |
| { |
| "first": "N", |
| "middle": [ |
| "K" |
| ], |
| "last": "Sondhelmer", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [ |
| "M" |
| ], |
| "last": "Weisohedel", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "Proo", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sondhelmer, N.K. and Weisohedel, R.M. A Rule- Based Approach to Ill-Formed Input. Proo. 8th", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Conf. on Computational Linguistics", |
| "authors": [], |
| "year": 1980, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "46--54", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Int'l Conf. on Computational Linguistics, Tokyo, Japan, October, 1980, pp. 46-54.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "If The Parser Fails", |
| "authors": [ |
| { |
| "first": "Ralph", |
| "middle": [ |
| "M" |
| ], |
| "last": "Welsohedel", |
| "suffix": "" |
| }, |
| { |
| "first": "John", |
| "middle": [ |
| "E" |
| ], |
| "last": "Black", |
| "suffix": "" |
| } |
| ], |
| "year": 1980, |
| "venue": "Proceedln~s of the 18th Annual Meetlnsof the ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Welsohedel, Ralph M. and Black, John E. \"If The Parser Fails.\" Proceedln~s of the 18th Annual Meetlnsof the ACL (June 1980).", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "A Framework for Processing Ill-Formed Input. Department of Computer and Information Sciences", |
| "authors": [ |
| { |
| "first": "Ralph", |
| "middle": [], |
| "last": "Welschedel", |
| "suffix": "" |
| }, |
| { |
| "first": "Norman", |
| "middle": [], |
| "last": "Sondhelmer", |
| "suffix": "" |
| } |
| ], |
| "year": 1981, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Welschedel, Ralph and Sondhelmer, Norman. A Framework for Processing Ill-Formed Input. Department of Computer and Information Sciences, University of Delaware, October, 1981.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Cascaded ATN Grammars", |
| "authors": [ |
| { |
| "first": "W", |
| "middle": [ |
| "A" |
| ], |
| "last": "Woods", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Woods, W. A. \"Cascaded ATN Grammars.\" ~. E~LI~~, I (Jan.-Mar.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Optimal Search Strategies for Speech Understanding Control", |
| "authors": [ |
| { |
| "first": "W", |
| "middle": [ |
| "A" |
| ], |
| "last": "Woods", |
| "suffix": "" |
| } |
| ], |
| "year": 1982, |
| "venue": "Intelli=ence", |
| "volume": "18", |
| "issue": "3", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Woods, W. A. \"Optimal Search Strategies for Speech Understanding Control.\" Intelli=ence 18, 3 (June 1982).", |
| "links": null |
| } |
| }, |
| "ref_entries": {} |
| } |
| } |
