DOCUMENT RESUME ED 039 510 U1TFOR Spinoza IT: Conceptual .
DOCUMENT RESUMEAL 002 310ED 039 510U1TFOR71TTLF"rNSTITU7IONSPONS AGENCY7.7.PORT NOPUB DATENOTEEDFSDESCRIPTORSIDENTIFIERSSchank, :Roger C.; And OthersSpinoza IT: Conceptual Case-Based Natural LanguageAnalysis.Stanford Univ., Calif. Artificial IntelligenceProject.Denartment of Defense, Washington, D.C. AdvancedResearch Projects Agency.; National Inst. of MentalHealth (DHEW) , Bethesda, Md.M-AIM-109Jan 70107p.FDRS Price MF- 0.50 HC Not Available from FDRS.*Case (Grammar), *Computational Linguistics, ConceptFormation, Conceptual Schemes, Deep Structure,*Linguistic Theory, Semantics, *Structural Analysis,*Thought Processes, Translation, Verbs*Conceptual Dependency Theory, Spinoza IIABSTRACTThis paper presents the theoretical changes thathave developed in Conceptual Dependency Theory and theirramifications in computer analysis of natural language. The majoritems of concern are: the elimination of reliance on "grammar rules'for parsing with the emphasis given to conceptual rule based parsing;the development of a conceptual case system to account for the powerof conceptualizations; the categorization of ACT's based onpermissible conceptual cases and other criteria. These items aredeveloped and discussed in the context of a more powerful conceptualparser and a theory of language understanding. (Author/AMM)
STANFORD ARTIFICIAL INTELLIGENCE PROJECTMEMO AIM 109JANUARY 1970SPINOZA II:WCONCEPTUAL CASE-BASED NATURAL LANGUAGE ANALYSIS,HEALTH, EDUCATION& WELFAREU.S, DEPARTMENT OfOFFICE Of EDUCATIONTHIS DOCUMENT HASBEEN REPRODUCEDPERSON OR ORGANIZATIONORIGINATING IT.STATED DO NOT NECESSARILY REPRESENTPOSITION OR POLICY.BYROGER C. SCHANKLARRY TESLERSYLVIA WEBERCOMPUTERSCIENCEDEPARTMENTSchool of Humanities and SciencesSTANFORD UNIVERSITYEXACTLY AS RECEIVED FROM THEPOINTS Of VIEW OR OPINIONSOFFICIAL OFFICE OF EDUCATION
STANFORD ARTIFICIAL INTELLIGENCE PROJECTMEMO AIM-109SPINOZA II:JANUARY 1970Conceptual Case-Based Natural Language AnalysisbyRoger C. SchankLawrence TeslerSylvia WeberABSTRACT:This paper presents the theoretical changes that havedeveloped in Conceptual Dependency Theory and theirramifications in computer analysis of natural language.The major items of concern are: the elimination ofreliance on 'grammar rules' for parsing with the emphasisgiven to conceptual rule based parsing; the developmentof a conceptual case system to account for the power ofconceptualizations; the categorization of ACT's based onpermissible conceptual cases and other criteria. Theseitems are developed and discussed in the context of amore powerful conceptual parser and a theory of languageunderstanding.This research is supported by Grant PHS MH 066-45-09 from the Nationalinstitute of Mental Health, and in (in part) by the Advanced ResearchProjects Agency of the Office of the Secretary of Defense (SD-183).Reproduced in the USA. Available from the Clearinghouse for FederalScientific and Technical Information, Springfield, Virginia 22151.Full size copy S3.00; microfiche copy .65.Price:
TABLE OF CONTENTSSection1.Pa eIntroduction11.1Goals11.2Conceptual Rule Parsing32.The Conceptual vs. the Linguistic3.The Semantics of English Verbs103.1Pseudo-state verbs103.2Linguistic Experience134.5.The Parsing Theory154.1Attributes of Spinoza II154.2An Example1725The Use of Case in Conceptual Dependency5.1Conceptual Prepositions255.2The Particular Cases315.2.1The Instrumental Case315.2.2The Recipient Case315.2.3The Objective Case355.2.4The Directive Case366.Conceptual ACT Categories by Sylvia Weber407.The Parser537.1Introduction537.2. The verb-ACT Dictionary7.38.55Operation of Spinoza 1.25962Etcetera8.1Attribute Statements and Tense Modifications.62
TABLE OF ave688.4Causation698.5Some Funny Words708.5.1ZFA's.,7o8.5.2Relative Adjectives72New Approaches to Conceptual Dependency Analysisby Larry Tesler739.1Types of Conceptualizations ransition859.10 Remarks10.Instruments; Stations73EndAppendix I - The verb-ACT DictionaryAppendix 11 - Output of Spinoza IBibliography868889103107
1. Introduction1.1 GoalsIn order to expect to do really relew.nt work in computationallinguistics it is necessary to design a computational linguistics model.This model would be a model of the human linguistic processing ability toas large an extent as is feasible, always utilizing the consideration thatthe model must be algorithmic in nature.It seems clear that there is an underlying conceptual basis to natural language, and that this conceptual basis is the same in all languages.We can say that a model of this conceptual basis of language would be infact a language -free, representation of any linguistic input or potentialoutput.A basis of this kind is necessary in order to account for theability of humans to translate and paraphrase.That is, in order for ahuman to recognize that two linguistic inputs are equivalent, whetherthey are in different languages or not, he must process the meaning ofthese inputs in such a way as to render their content the same.TheConceptual Dependency model (see [7]) is intended as a simulative modelfor computational linguistics that will perform this task.The modelcontains as an inherent part the ability to perform various tasks thatwe recognize to be within a human's linguistic ability.That is, themodel is not concerned with 'linguistic' problems such as acceptabilityor grammaticality but rather it is intended to model a human in a dialogue situation.Thus it considers all 'conceptually correct' input andis capable of interpreting a sentence if there is a missing word or theinput is in a 'queer' form (that is, it does not correspond to certaingrammatical rules).The model is thus concerned with 'understanding'-1-
rather than with 'accepting' a sentence.It is reasonable therefore, to inquire what we know about a 'humanparser'.First, he hears the sentence and may be said to understand itconceptually.That is, he has the ability to associate a linguistic in-put with some conceptual structure and to combine these conceptualstructures in accordance with the gramar rules of the language, thelanguage-free conceptualization rules, and his 'conceptual experience' or'knowledge of the world'.Thus his 'understanding' finds a meaning forthe sentence by discovering the propositions or beliefs expressed by thatpiece of discourse.This meaning expresse,) what has been said (as opposedto what to do with it) and has been checked against the human's knowledgeof previous propositions.The new information has been verified as to itsconceptual validity or if no valid alternative exists then the new information has been added to the experience.We also know that a human finds only one analysis of a sentence whenanother if prompit is expressed within e discourse, but that he can findted to do so.Thus, he would find only one analysis of 'time flies likean arrow' given the usual context.We know in addition that his analysiscan be based on the context of the previous discourse, the situation, andthe identity of the speaker.Primarily, the 'human parser' is concernedwith interpretation of an input rather than discovering hidden ambiguities.Another important ability that a human has is the recognition of sentences that are 'laughers'.That is, certain constructions in a languagelead to predictable blind alleys that nearly always provoke a laugh.Ina sentence such as 'I saw the Grand Canyon flying to New York', it is thislaughing likelihood that gives us an insight into the human processing of-2-
this sentence.We can see, for example that there is a predictable orderedprocessing here that causes one of the two possible grammar rules thatapply to be tried first, producing the laugh.But from a conceptual pointof view, we can predict this with a conceptually-based processor that isnot bound by grammar rules.The conceptual parser described here is intended to produce as output a language-free conceptual network representative of the meaning ofthe input.Such a network is potentially useful in translation, para-phrasing, and all computational work involving natural language.In orderto achieve this goal we intend to simulate what we know exists, namely thehuman ability to understand.Thus, our simulative theory employs a worldmodel, an interlingua, an ability to map into and out of that interlingua,and an ability to reject possible interpretations of an input on the basisof its linguistic and conceptual experience.Thus the model is stratified,with meaning at the highest level, employing syntax as a finder.1.2. Conceptual Rule ParsingThe first version of Spinoza I(see [9]) has made obvious some in-consistencies in the underlying theory as so far developed.Of primaryimportance in the consideration of revisions of Spinoza I is the desireto create any future version of this parser as one that more closely parallels a human parser.The major theoretical discrepancy between Spinoza I as it now standsand our perception of a human involved in the same task is the relianceon realization rules.Since the realization rules may be construed tobe the grammar rules of a language, it seems reasonable that a fluent* Spinoza stands for 'Semantic Parser Involving Neo-stratificationalOrganiZation And conceptual dependency heuristics.-3-
speaker of a language is in full possession of these rules.It doesnot necessarily follow that he employs these rules in parsing.there is evidence that he does not.In fact,For example, we are familiar withthe fact that it is much easier to understand a foreign language thanit is to speak it.Whereas, we need the 'grammar' rules of a languageto generate from our conceptual base, it seems plausible that the process of understanding can work sufficiently well with a knowledge of thewords of this foreign language and a very few of the major realizationpatterns.This is because the conceptual base into which we are mappingduring the process of understanding this foreign language is the same oneas we ordinarily utilize.parts (namely concepts).It has the same rules of organization of itsIf we are aware of the word-concept couplingsof this foreign language, we now only need to arrange these conceptsaccording to our usual (i.e. language-free) manner.Thus, it would seemthat humans can fare rather well without realization rules during parsing.If this is the case, we must require of any simulation that it do likewise.Although the use of realization rules in Spinoza I works well enoughthere are more intuitive reasons for the elimination of the reliance onrealization rules during the parse.(1)Consider for example, the sentence:I saw the Grand Canyon flying to New York.Spinoza I parses this sentence correctly by the use of two realization rulesand the elimination of the inapplicable one by a check with the semantics.PP(R1)PP1ACT,.PP2ACTing ACT1 4-02ACT(R2)PP,.ACT,.PP2 ACyng PP1 a ACT1T whilea ACTPP12PP22
The conceptual dependency PP2ACT2, (Grand Canyon flies) derived from Riis eliminated from consideration by examination of the possible actionsfor a 'location'.successful.Since 'fly' is not one of these, R2is tried and isThere is no reliable weighting system for realization rulesin Spinoza I, so it is perfectly possible that R2 would be selected firstand RIwould therefore never be tried.This element of randomness seemsquite unlike a human in the same situation.A second p -oblem in the effective simulation of a human by Spinoza Iis with regard to the conceptual semantics (see [6]).Again we are facedwith the difference between generation and parsing of coherent discourse.Although these processes are similar enough to enable our system to beeffective while making double use of certain features, it seems clear thatFor example, in parsing 'green horse'there are exploitable differences.it is unnecessary to know that 'horses' are not 'green' in order to dealwith this construct.That is, while information of this kind is a neces-sary part of the random generation process, it has little to do with theproblem of parsing except when there is a more attractive alternative parsefor the same set of concepts.'Similarly, 'the park with the girl' is an acceptable possibility asa construction, and we would only want it to be ruled out of considerationconceptually in the view of more favorable alternatives.Thus, the sen-tence:(2)'I went to the park with awill not utilize the dependent 'park' in its conceptual networks only be11 withgirlcause the available alternatives are more highly valued.-5-That is, it is
not the conceptual experience that should be ruling out this alternative.In any event, it would be rather difficult to have an experienc,, filespecify the things that a 'park' can be with, since this list is practicallyinfinite.But the mechanisms of the,generative conceptual semantics must beemployed in order to correctly differentiate the senses of 'with' in a'boy with a knife' and a 'boy with a girl'.The information utilized inthis differentiation process in Spinoza I must also decide between 'park''fl withgirlwithand 'gogirl'.However, while the modification 'I'is incorrect'fl withgirlbecause 'girl' is not a descriptor of 'I', the 'girl' in this sentencealso 'went to the park' so it is necessary to introduce a notion of 'accompaniment' as a sense of 'with' which would function similarly to alogical 'and'.toIA4. goparkgirl,Clearly, the 'go with girl' sense would only be acceptable only in theevent that 'girl' functioned as an instrumental.Thus, we will also needinformation about possible instruments frJr various ACT's.This is exempli-fied by the problem presented in [7]'He hit the boy with long hair'.(11)'He hit the boy with a wrench'.While these sentences are both ambiguous, it is unlikely that a human wouldnotice that upon encountering them.Furthermore, the first analysis ineach cas:, is predictable and corresponds to the second analysis it. the-6-
other sentence.We require that Spinoza be able to make the appropriatechoice in each instance.This would have to be done by the establishmentof an instrumental case dependent on the ACT.A further problem for Spinoza I is presented by a sentence of the type'he grew plants', where, in the most likely underlying conceptualization,it is the 'plants' that 'grew' and not 'he'.In order to recognize theproblem here it is necessary to reorient the parser to be more dependenton the ACT, and in particular for English to have the system's linguisticexperience file expressed as information in the form of expectations whencertain verbs are encountered.Thus, Spinoza II will be a system containing various levels of information.rules.The parsing operation will function mainly using conceptualThe object of the conceptual rules will be to point the way to theunderlying conceptual subject-verb-object (or actor-action-object) combinations present.The parser will look for these S -V -O constructionswhenever possible, and check to see if they are in accord with the system'sexperience.We will not need to check the semantics unless we have a choiceof rules, in other words, when a decision is to be made.This implies thatthe parser will never be able to make final decisions as to dependencysince it may (as in the 'Grand Canyon' case), be searching for a part ofthe S -V -O that is more acceptable and thus would rewrite a piece of theold S -V -O.The discovery of a conflict of rules would indicate a need forresolution by the semantics.The rest of this paper is concerned with changing the theory of SpinozaI such that Spinoza II will be a conceptual rule, verb-based parser thatis concerned with case restrictions and a more realistic conception of-7
semantics.2. The Conceptual vs. the LinguisticConsider the problem of 'Ken, saw Larry in the park'.This sentenceis unambiguously parsed by a human as opposed to the case of the sentence'Ken saw the boy in the park'.' The reason for this is 'boy' may have adescriptor whereas 'Larry' may not.completely.Here 'Larry' identifies the objectNow certainly 'Larry' could have been 'in the park', but theconceptual apparatus that humans employ makes a distinction between decriptive information and additional information.This is seen in thedifference between non-restrictive relative clauses and ordinary prenominal adjectives in English.Thus, the point of 4- PA's and 4 PP's (below the line) is to furtherdescribe a PP such as to explicate which of the set of PP's called by thatname is the referent.A a construction on the other hand is intended toprovide additional information about the PP.The theoretical point here is that there is a great deal of importantinformation inherent in the words themselves that can aid in our conceptualexpectations during the parse.Here, 'Larry' is the type of Noun (Proper)that tells us that conceptually we do not expect any descriptions.Then,if there are any descriptions present we can attach them elsewhere if therewould have been an otherwise equally likely alternative.This conceptual expectation abVity is important at a deeper level inthe parse.For example, (5)'I am in love' presents this type of problem.'Love' in the conceptual dependency framework is an ACT no matter whatsyntactic realizates are being used.(This is the case throughout.assumption is that all syntactic forms of any concept have the sameThe
conceptual realizate.)Furthermore, 'love' is a transitive ACT.ThisThat is, if someone isis important in an expectation-oriented system.the recipient of 'love' in all cases, then we can look for this recipient,or if none is to be found, at least know that some 'human' fits.Thus, theC-diagram of (5) is:I 4* lovePPhumanClearly, it is necessary that this be the parse.Since the informationthat 'PP human' satisfies the conditions of the object had to come fromlooking up 'love', we can allow ourselves the luxury of picking up addiIn this case, we may pick uptional information by consulting the verb.the advice that the PP on.the left is commonly of a different sex thanthat on the right.This would allow us to 'guess' that 'Joan' in thesentence 'Joan is a darling girl' following (5) can be placed as theobject in (5).An intelligent parser needs to know what to expect at any point inthe parge.If that information is there (that is, if humans would havesome guesses as to what follows at a given point in a sentence) then wecan provide some of that information to the parser.A similar case can be found in the construct'I cg, run'.The expecta-tions to be found here are made clear by the seeming unacceptability of'I run'.the simple realizate of this construct; (6)comfortable here if the sentence were: (7)'We would be morerun) orI am running' (Itnow(8)'I run to the store on Tuesdays' (I a runTuesdayeach-9-tostore)
The discomfort caused by (6) is representative of an important facet ofthe concept 'run'.This concept implies a destination or at least adirection ('around the block') was part of the conceptualization.Similar-ly, conceptualizations that are formed from present-tense linguistic real-izates that are not indicated as presently taking place (by '-ing' forexample) require a time.In other words, there are certain characteris-tics of a conceptualization that we can expect to be mentioned in a discourse in some way.Furthermore, the verb used indicates certain dependentconstzucts that are always present in the underlying conceptualizationeven if they are not present in the sentence itself.This was the casewith the expectation of the transitive ACT 'love' in that we required anobject conceptually even though the language did not.For 'run' we maysay for the moment that it is intransitive and takes dative case.Similar-ly, in the sentence 'I hit the boy', we can expect an instrument waspresent and we thus require that'hit' take instrumental case.We thusestablish a verb-dependent case system (with some similarity to Fillmore[4]).This will be delved into i
*Case (Grammar), *Computational Linguistics, Concept Formation, Conceptual Schemes, Deep Structure, *Linguistic Theory, Semantics, *Structural Analysis, *Thought Processes, Translation, Verbs *Conceptual Dependency Theory, Spinoza II. This paper presents the theoretical changes that have developed in Conceptual Dependency Theory and their
Std Pk Mstr Ctn Prod Code Price Each 435-002SRL 1/4 25 450 039 12.95 435-003SRL 3/8 25 300 039 12.95 435-005SRL 1/2 25 300 039 7.67 435-007SRL 3/4 25 300 039 9.60 435-010SRL 1 25 200 039 11.09 435-012SRL 1-1/4 10 80 039 17.04 435
Cashier’s Office SSC-115 (510) 215-4892 Childcare Services & Lab ELC (510) 215-4885 College Skills Center LLRC (510) 215-4898 Cooperative Education LLRC-212 (510) 215-4916 Counseling SSC-108 (510) 215-3935 Custodial Services C (510) 215-4809 Disabled Students Progra
County Contact List . Jurisdiction Disease/Role Name E-Mail Phone Number Fax Number . Alameda Accounts Emily Yette Emily.Yette@acgov.org 510-268-2092 510-2678085 Alameda CD Munira Shemsu Munira.Shemsu@acgov.org 510-208-2084 510-273-3744 Alameda CD Nicole Huerta Nicole.Huerta@acgov.org 510-267-3250 510-273-3744
Section 410.3 Sea Level Rise Section 410.3 Summary Chapter Five. Living Resources and Critical Habitats Section 510. Findings of Fact Section 510.1 Narrow River Ecosystem Section 510.2 Wetlands Habitats Section 510.3 Open Water and Aquatic Habitats Section 510.4 The Terrestrial Habitat Section 510.5 Summary
510-25451-91 Pilot Transfer Valve (Schrader Bellows) 510-05071-03 Pilot Transfer Valve (Alkon Corp. SP-5046) (replaces airman pilot valve) 510-05072-00 Pilot Transfer Valve (Norgren) 510-05588-20 Pilot Valve (Hendrickson) (one up / one down) 510-05314-20 Dial-A-Ride Suspension Control Panel (Hendrickson) PAGE 8
Our Lady of Guadalupe School 40374 Fremont Blvd. Fremont, CA 94538 (510) 657-1674 FAX (510) 657-3659 Email: LKeltie@csdo.org Web Site: www.guadalupe-school.com Extended Care Office (510) 657-1683 Extended Care Cell (510) 853-8294 Faith Formation Office (510) 651-4966 Our Lady of Guadalupe Parish Of
The cobas 4800 system software integrates the sample preparation with nucleic acid amplification and detection to generate test results. 510(k) Premarket Notification 510(k) Summary Report Page 4. Roche Molecular Systems, Inc. cobasQ) CT/NG Test Pleasanton, CA 94588-2722 Section 5: 510(k) Summary 510(k) Summary Report 2. .
DOCUMENT RESUME. CE 039 359. Rubber Tire Dozer Operator. Open Pit Mining Job. Training Series. British Columbia Dept. of Education, Victoria. ISBN-0-7718-8336-6 82 83p.; Joint project of the Ministry of Education and the member companies of the Mining Association of Bri
