File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/79/j79-1017_metho.xml
Size: 110,813 bytes
Last Modified: 2025-10-06 14:11:09
<?xml version="1.0" standalone="yes"?> <Paper uid="J79-1017"> <Title>A REPORT ON THE TUTORIAL ON COMPUTATIONAL SEMANTICS Institute for Semantics and Cognitive Studies</Title> <Section position="1" start_page="0" end_page="0" type="metho"> <SectionTitle> NEWSLETTER OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS VOLUME 12 - NUMBER 2 JULY 1975 </SectionTitle> <Paragraph position="0"> Released for publication on June 20, 1975.</Paragraph> <Paragraph position="1"> This number of The Finite String contains news items, four short technical contributions, and a description of the constituent societies of the American Federation of Information Processing Societies.</Paragraph> <Paragraph position="2"> Volume 12, Number 3 of The Finite String is distributed in the same packet of A J C L . It contains a short survey paper and current bibliography.</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="metho"> <SectionTitle> AMERICAN JOURNAL OF COMPUTATIDNAL LINGUISTICS is published by </SectionTitle> <Paragraph position="0"/> </Section> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> FOEQfULAE IN'COHEKENT TEXT Felix Dreizin . . 70 </SectionTitle> <Paragraph position="0"> A F I P S Constituent S~~cieties . Purposes. Membership Reauirements. Activities. Publications. Dues. Addresses 86 American Journal of Computations Linguistics Microfiche 17 .- 3</Paragraph> </Section> <Section position="4" start_page="0" end_page="0" type="metho"> <SectionTitle> PERSONAL NOTES </SectionTitle> <Paragraph position="0"> ~pecialfies and changed or improved addresses received since publication of the 2974 Membership Directory are givt?n in asterisked entries.</Paragraph> <Paragraph position="1"> BUSA, REV. ROBERTO, SI * Department of Philosophy, ALOISIANLFf (College) , 21013 Gallarate, Italy. Index Thornisticus Inventory and census of large quantity of natural text, lemmatized and codified as to text typolagy.</Paragraph> <Paragraph position="2"> DQUVILLE, MRS. JUDITH A. *Metals Information Center, Olin Corporation, 91 Shelton Avenue, New Haven, Connecticut 06504. Indexin.g, abstracting, chemical literature searching; organization and maintenance of scientific information centers.</Paragraph> <Paragraph position="3"> ENGELS, LEOPOLD-KAREL * Applied Linguistics; Katholieke Cnivessiteit Leuven, 61/3-Tiense Vest, 3200 Kessel-Lo, Belgium. Automatic syntactic analysis of English; discourse analysis.</Paragraph> <Paragraph position="4"> JOSH1 , ARAVIND K. &quot;Department of Computer and Information Science, University of Fennsylvania, R. 268 Moore School, Philadelphia 19174, syntactic and semahtic representations; mathematical linguistics and logic; artificial intelligence.</Paragraph> <Paragraph position="5"> KAPLAN, RONALD PhD 1975, Psychology, Harvard University.</Paragraph> <Paragraph position="6"> LONGYEAR, CHRI STOPHER R. *Formal pragmatic representations; natural language models; da-ta base structur~s.</Paragraph> <Paragraph position="7"> MATHIAS, GERALD B. To Department of East Asian Languages, Moore Hall 370, University of Hawaii at Manoa, 1890 East West Road, Honolulu 96822, from Indiana University.</Paragraph> <Paragraph position="8"> NEEDHAM, DR. KA~REN SPARCK JONES * Semantics, information retrieval, ~Tassifkation.</Paragraph> </Section> <Section position="5" start_page="0" end_page="0" type="metho"> <SectionTitle> PERSONAL NOTES 4 NEEDHAM, ROGER *Computing, operatingsystems. I </SectionTitle> <Paragraph position="0"> PHILLIPS, BRIAN PhD 1975, State University of New York, Buffalo, Linguistics (Topic analysis) .</Paragraph> <Paragraph position="1"> SAGVALL. FIL. DR. ANNA-LENA *Department of Slavonic Languages and Data Center, Uppsala University. Automatic text analysis, applied mainly to Russian text. Automatic text understanding, applied to Swedish medical text.</Paragraph> <Paragraph position="2"> SALKOFF, MORRIS *Laboratoire d14utomatidue Documentaire et Linguisrique, Universite de Paris 7. 2, Place Jussieu, Paris 5, France. Automatic syntactic analysis of French; compilation of a dictionary of F~ench verbal constructions. SCHUEGRAF DR. ERNST *Ihformation retrieval; statistical linguistics. SHAPIRD, STUART C. &quot;Semantic networks, representing and carrying out inferences computer assisted instruction.</Paragraph> <Paragraph position="3"> SILVA. GEORGETTE *System Development Corporation, 2500 Colorado Avenue, Santa Monica, California 90406. Natu~al language processing; linguistics.</Paragraph> <Paragraph position="4"> SP ITZBARDT, PROF. DR. HARRY 'Automatic morpheme analysis; English, Indonesian.</Paragraph> <Paragraph position="5"> SUPPLE. JAMES P. *Product Support, Compute1 Systems Ltd Scientific computer languages (Fortran, Algol, APL, Snobol); AI (visual); Biopnysics (radiaL,distribution function) .</Paragraph> <Paragraph position="6"> WEBB, FREDERICK N. *Computer Systems Division, Bolt Beranek and Newman Inc. Progrgrnming languages, syntax descriptior</Paragraph> </Section> <Section position="6" start_page="0" end_page="0" type="metho"> <SectionTitle> UNIVERSITY OF OTTAWA: DEPARTMENT OF LINGUISTICS. IN COLLABORATION WITH PROBLEMS AND METHODS SEMANTICS: LOGIC AND A I SEMANTICS: LINGUISTIC PARS I NG AND SYNTHES I S LEXICOGRAPHY AND STYLI ST1 CS SPEECH RECOGNITION AND SYNTHESIS PAPERS ON MT AND MAT ARE WELCOME IN ALL THEME AREAS </SectionTitle> <Paragraph position="0"/> </Section> <Section position="7" start_page="0" end_page="0" type="metho"> <SectionTitle> COLING 76 ADDRESSES REGISTRATION COLTNG 7 6 </SectionTitle> <Paragraph position="0"/> </Section> <Section position="8" start_page="0" end_page="0" type="metho"> <SectionTitle> LANGUAGES FEES </SectionTitle> <Paragraph position="0"> FRENCH, ENGLISH The organizers will attempt to provide simultaneous interpretation in Russian</Paragraph> </Section> <Section position="9" start_page="0" end_page="0" type="metho"> <SectionTitle> ARRIVAL DEMONSTRATION </SectionTitle> <Paragraph position="0"> Student residences on campus Lord Elgin Hotel, 5 minute$ from campus Cafeterias on campus Restaurants in Ottawa and Hull Banking and exchange facilities on campus Mirabel Airport, serving Montreal and Ottawa, is one hour from t.he campus by a road to be opened in 1976- null May I cslarify a little the two sentences of mine about the LOGOS Machine Translation system that you were kind enough to publish and which provoked Mr. Scott s mere extended reply? r feel sure that the differences betyen him and me are only matters of definition of what is unrestricted natural language and it may be worth making that clear. Let me also add that nothing I said was meant to deny that the cammercial MT companies like his own have done-excellent work, and that I wish them well in the future. But Whether they have solved the MT problem for natural language in the sense in which that problem was understood the last time round this cycle in the Fifties and Sixties, is another matter, and I remain to be convinced.</Paragraph> <Paragraph position="1"> For thoge who have just joined in, let me remind them that the intractable problems its first phase were word sense anibiguity, case ambiguity (of prepositions, and referential ambiguity (roughly, pronouns). Anyone who. claims to have solved thoseproblems without making any general theoretical claims about natural language in the process is either dealing with restricted languaqe, or is in much the same position as one who arrives to demonstrate a perpetual motion machine. In the latter case, he is entitled to a respectful hearing, but there is nonetheless a certain scepticism in the audience. No amount of talk about millions of dollars spent, or important contracts obtained makes that hard fact any softer, Mr. Scott says. that UN treatises shaulsd be a test case of what is natural, rather than restricted, language. I quite agree, and if his system can translate an unseen UN treatise chosen by a neutral party to the satisfaction of a neutral audience then I will back down. He is careful not to say he has done it, and I personally believe that</Paragraph> </Section> <Section position="10" start_page="0" end_page="0" type="metho"> <SectionTitle> LETTERS </SectionTitle> <Paragraph position="0"> he cannot do it, armed with a phrase structure grammar.</Paragraph> <Paragraph position="1"> a semantic categorisation system and nothing more. The reasons why are set out in any standard paper on Artificial Intelligence and Natural Language. They involve the essential role of semantic structures, inference and knowledge of the world in understanding and so in translation. I will be happy to send him a bibliography.</Paragraph> </Section> <Section position="11" start_page="0" end_page="0" type="metho"> <SectionTitle> HARRY GOODE MEMORIAL AWA'RD KENNETH El IVERSON </SectionTitle> <Paragraph position="0"> AFL won its inventor the eleventh award presented by AFIPS for outstanding contributions to computing,.</Paragraph> <Paragraph position="1"> Dr. Iverson, IBM Fellow and Manager of the APL Design Group at IBM's System Development Division in Philadelphia, was formerly on the faculty of applied mathematics at Harvard.</Paragraph> </Section> <Section position="12" start_page="0" end_page="0" type="metho"> <SectionTitle> AFIPS DISTINGUISHED SERVICE AWARD YIORTON No ASTRAHAN </SectionTitle> <Paragraph position="0"> His key role in the formation of AFIPS and his influence on the growth, programs; and service of the organization earned Dr.</Paragraph> <Paragraph position="1"> Astrahan the third AFIPS award for service to the computing field through accomplishments on behalf of the Federation.</Paragraph> <Paragraph position="2"> Dr. Astrahan organized and was first chairman (1952-53) of the Institute of Radio Engineers Professional Group on Electronic Computers, predecessor to the.IEEE Computer Society. He has been with IBM for more than 25 years: the 701, SAGE, associative memory using program interrupt for I/O control, and two years in France.</Paragraph> <Paragraph position="3"> American Journal of Computational Linguistics Microfiche 17 : 11</Paragraph> </Section> <Section position="13" start_page="0" end_page="1976" type="metho"> <SectionTitle> AUL CHAP </SectionTitle> <Paragraph position="0"> Dr. Chapin goes to the National Science F~undation in August from the University of California, San Diego, where he has been since 1967 except fok a year at the University of Hawaii (1971-72)-His doctorate is from MIT (1967). While a qraduate student, he worked in Donald Walker's sroup at the MITRE Corporation. At UCSD he has taught and conducted research in descriptive and theoretical syntax. computational linguistics, psycnolirrguistics, and comparative Polynesian linguistics. Since 1973, he has been an Assistant Provost ~f John.Muir College, one of UCSD's four undergraduate cluster coxleges.</Paragraph> <Paragraph position="1"> In Hawaii Dr. Chapin studied Polynesian history and culture.</Paragraph> <Paragraph position="2"> with the support of an ACLS Study Fellowship.</Paragraph> <Paragraph position="3"> Dr, Chapih succeeds Alan Bell as NSF Program Director for Linguistics; br. Be: -1s returns to the Department of Linguistics at the UniveYsity of lorado ado, Boulder.</Paragraph> </Section> <Section position="14" start_page="1976" end_page="1976" type="metho"> <SectionTitle> NATIONAL COMPUTER CONFERENCE NEW YORK CLTY JUNE 7-10 EXHIBTTSt CQLISEW HOTELS &quot; HI LTDN AMERJ CANA CONFERENCE GHA~ RMAN CARL HAMMER DIRECTQR OF COMPUTER SCIENCES SPERRY UNIVAC, WASHINGTON PROGRAM 4 CHA I RMAN STANLEY W INKLDR MANAGER OF APPLIED TE C'HNOLOGY IBM SYSTEMS DEVELOPMENT nsvfsro~ GAITHERSBURG, MARYLAND </SectionTitle> <Paragraph position="0"> DR. HAMMER is a.member of the AFIPS Board of Directors and Adjunct Professor at American University and the industrial College of the Armed Forces.</Paragraph> <Paragraph position="1"> DR. WINKLER is Adjunot Professor of Computer Systems at</Paragraph> </Section> <Section position="15" start_page="1976" end_page="1976" type="metho"> <SectionTitle> AAAS SECTION T INFORMATI ON r COM-PUTI'NG AND/ POMMUN I CAT1 ONS </SectionTitle> <Paragraph position="0"> The section has adopted a new name and is looking for ways to give its subject matter greater visibility in Science: reviews of the state of component arts; editorials; program and committee participation ; etc.</Paragraph> <Paragraph position="1"> The section, with 1234 members has 106 Fellows and a ouata of 8 nominations for election to fellowship this year The secretary of the section can supply information about the Congressional Science Fellowship Program which provides a stipend of about $15,000 to scientists and engineers who spend one year on the staff of a congressman,. a congress'ional committee, or the</Paragraph> </Section> <Section position="16" start_page="1976" end_page="1976" type="metho"> <SectionTitle> SEMWTICS OF HUMAN !..ANGUAGE </SectionTitle> <Paragraph position="0"/> </Section> <Section position="17" start_page="1976" end_page="1976" type="metho"> <SectionTitle> LJNGUISTIC BIBLIOGRAB~JES MACHINE TRANSLATION IN CANADA </SectionTitle> <Paragraph position="0"/> </Section> <Section position="18" start_page="1976" end_page="1976" type="metho"> <SectionTitle> PUBLISHERS AND SUPPLIERS #OF MATERIALS FOR LINGUISTICS DIRECT CURRENT SERVICES </SectionTitle> <Paragraph position="0"> Computer stored, frequently updated. Single copies on computer paper by surface mail free of charge. On good quality paper, air mail outside Canada, $2 prepaid.</Paragraph> </Section> <Section position="19" start_page="1976" end_page="1976" type="metho"> <SectionTitle> DIRECTORY OF LINGUISTIC ORGANIZATrONS CALENDAR DF LINGUISTIC EVENTS 1975-1977 CONFERENCE INTERPRSTERS~ GLOSSARY, ENGLISH - FRENCH ACCESS-ION LIST OF THE CENT~E SINCE 1974 DIRECTORY OF CANAD-IAN E~UCATION IN SPEECH PATHOLOGY ~~BLIOGRAPHY OF LINGUISTICS AND DOCUMENTATION LOCAL OR TELEPHONE CONSULTATION </SectionTitle> <Paragraph position="0"> Holdings include 200 bibliographies, 20 dictionaries of linguistic terminology, CAN/SDI-ERIC current-awareness cards, files of meeting programs, courses, job offers, serial publications, privately circulated papers, offprints, etc.</Paragraph> </Section> <Section position="20" start_page="1976" end_page="1976" type="metho"> <SectionTitle> WORLD INVENTORY OF ABSTRACTING AND IIjDEXING SERVICES </SectionTitle> <Paragraph position="0"> A machine-readable inventory was expected to be complete by July 1, 1975; publication is planned by the end of the year.</Paragraph> <Paragraph position="1"> Gaye Hoffman is Project Coordinator at the National Federation of Abstracting and Indexing Services and Toni Carbo Bearman is Principle Investigator of the National Science Foundation grant recently supplemented with $26,650. The Federation,Internationale de Documentation and UNESCO UNISIST are supporting.the development of the inventory.</Paragraph> </Section> <Section position="21" start_page="1976" end_page="1976" type="metho"> <SectionTitle> SUMMARY </SectionTitle> <Paragraph position="0"> ATEF converts an input string into a labeled tree; the label evolves under the control of a grammar. A set of labels is associated with each segment of the string, and several functions permit control of the number of alternative labels.</Paragraph> <Paragraph position="1"> CETA simulates a transformatianal grammar. It uses a set of grammars with conditional linkages. The applicabili&y of a transformation can be determined in part by conditions on the resulting tree.</Paragraph> <Paragraph position="2"> Computer processing 3f natural languages requires more or less polished algorithmic models. The two systems presented here represent a choice of a large class among the algorithms proposed in recent years to solve these problems. The principal choice determined by these systems lies in the formal use of labeled trees ( arborescences ) . Freedom of choice of these la-bels and possible structures gives these systems broad fields of applications in several domains and notably in that of the automatic processing of natural languages. The ATEF system has the purpose of transformFng a string of words into a tree which is manipulable by the CETA system. The definition of labeled trees determines what objects CETA can manipulate and the objecti'ves of ATEF. This note therefore begins with the definition of labeled trees. To obtain a tree of this type beginning with an input string, ArEF uses a dictionary and a finite-state grammar. The result of this system can be manipulated by CETA in order to obtain the desired type of structure. The example of analysis given here shows the possibilities of the CETA system w?th two different manipulative strategies: search for constituent or dependency structure .</Paragraph> </Section> <Section position="22" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 1, LABELED TREES </SectionTitle> <Paragraph position="0"> A rree is a set of points with which is associated a structure, that is to say a relation having the properties: The relation between two points is directed (one point depends on the other) A point Cannot depend on a point belonging to its own descent set (the descent set of a point is the set of points that depend on it, the points that depend on them, etc. ) A unique point descends from no other.</Paragraph> <Paragraph position="1"> It is possible to draw a tree placing below a point all of its descendants, linked by lines. (See the example on the next frame. ) A labeled tree is a tree such that with each of its points is associated a lab-el This label IS formed of a set of data. The figure below represents a labeled tree.</Paragraph> <Paragraph position="2"> fernthe,-nom commurr feminin, sin&ul.ier une, article irresisxible jeur~e rousse indef ini, f eminin adjectif, fcr-ir,iri adjectif, feminin adjectif,couleur singulier singu*ier singulier f5minin,singulier.</Paragraph> <Paragraph position="3"> derivation negative du verbe resister vers I adjemlf</Paragraph> </Section> <Section position="23" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 7: T-E ATEF SYSTEM </SectionTitle> <Paragraph position="0"> The purpose of the ATEF system is to transform an input sming of words into a labeled tree, each word in the string possibly,leading to one or several points in the final tree (ambiguity). The determination of the label originating in an input word results from its analysis. This analysis proceeds by segme~tation of the input word according to elements from different dictional: ies . A correct segmentation therefore gives. a label for a point of the final tree .</Paragraph> <Paragraph position="1"> Tn advance of any.</Paragraph> <Paragraph position="2"> analysis, the definition of the elements employed in the composition of different labels is required and is supplied by two Eiles called variable declaration files.</Paragraph> <Paragraph position="3"> A label will consist of a set bf variables. Each variable must be defined with its set of possible values. Thus if one defines the variable &quot;category&quot; the set of &quot;categories&quot; that can be used must be specified. The set is written category = (NOUN, ARTICLE, PRONOUN, ADJECTIVE, VERB, ...) (A constraint requires that the name of a variable must not be longer than 7 characters. Thus the preceding var2able could be written, for example, CAT = (NN, ART, Pm, ADJ, VRB, ...) ) The definitdon of a particular label consists in an enumeration of the variables relevant to the label. A set of labels can be predefined and is collected in a so-called format file. The ATEF system analyzes the words and thus employs dictionaries A dictionary is a set of segments (character strings), ~th each of which is associated a label, a processing pointer, and a lexical unit pointer. The processing pointer specifies the particular process which must be associated with the segment.</Paragraph> <Paragraph position="4"> The analysis of the input word by the ATEF system resides at first in a label processing, that is to say in an evolution of the empty label toward a final label characteristic of the analyzed word This evolution is controlled by the grammar, which at each moment has access to two labels. the label being developed (noted by the symbol C) and the label associated with the segment which was read in the dictionary (noted by A) The analysis of a word aims to produce a segmentation of the woru simultaneously compatible with the segments of the different dictionaries (the word must be an assembly of dictionary segments) and compatible with a correct evolution of the grammar Thus the segmentation of the input ~ord is tighcly bound to the evolution of the grammar which controls the coherence of the segmentation In the course of a segmentation operation the state of the system takes into account for the analyzed word the label resuLting from the analysis of the segments al~eady obtarned for this word the label associated with the segment found in a dictlonary the remaining characters of the input word the complete form of the input word Thus for example in the course of the analysis of the word irresistible and after analysis of the segment &quot;ible&quot; and in the course of reading the segment 'resist1' the following erements are obtained C the label resulting from the analysis of &quot;ible&quot; This label contains for example the variable derivation with value verb-adj, the variable gender with value masculine and feminine, the variable number with value singular.</Paragraph> <Paragraph position="5"> A the label associated with the segment''resist This label coatains notably the lexical unit &quot;resister&quot;, the varlable category with value verb The character$ IR The complete form IRRESISTIBLE The purpose of the grammar is to permit or prevent the evolution of label C starting wlth label A Here, the label will evolve and obtaln the variable category wlth value adjective. A rule associated with the segment &quot;resist&quot; by means of its pointer will therefore describe this evolution of the label C. When no evolution of the label C is possible, the corresponding segmentation is blocked and considered nonsignificant. The set of labels plays a fundamental role in this system and forms the set of sts~tes of the finite state transducer corresponding to the logical model of the system Each coherent segmentation of a word (a word can have several coherent segmentations leading to ambiguities) provides a labeled point in the final tree Three elements are fundamental to the system the choice and evolution of the segmentation the calculation of the set of labels associated with a word the positioning of the labeled points created by the analysis of a word in the final tree The choice and evolution of the segmentation has to do with the sequence of input characters. The segmentation forces, above all, a prior linguistic choice. Thus with the segment &quot;UN&quot; two possibilities can be conceived either accept &quot;UN E&quot; as a coherent segmentation or have the segment &quot;Ul?E&quot; in the dictionary and refuse the segmentation &quot;UN El' For each initial form several segmentations are possible to arrive at the same results arid only a linguie tic study of the phenomena permits a decision on the strategy to be adopted. In any event, this stLategy is left to the user of the system In the course of a segmentation the system can operate d~rectly on the nonsegmented chatacters in order to force them into a &quot;canonical'! form. Thus in the case of the word reel several possib~lities arise to accept a word like realite put the segment &quot;real&quot; in a dictionary as well as the segmenf &quot;reel&quot;, the former will generate words like realite, irrealite, etc.</Paragraph> <Paragraph position="6"> put the single segment &quot;reel&quot; in the dictionary and the analysis of the word realite Qill follow the schema realit& => 1st segment found &quot;lte&quot;, remainder &quot;real&quot; mddificat&on real ->r&el => 2nd segment found &quot;reel&quot; segmentation .&quot;riSel it&&quot; N B In thls analysis, it is to be noted that the search for successive segqents is performed from left to right PSox the input word. This depends on the strategy adopted and, for a given use, the direction of the segmentatson of a word can be either left to right or right to left.</Paragraph> <Paragraph position="7"> To avoid a proliferation of possible segmentations and therefore of possible solutions, several functions provide for intervention in the segmentation A first possibility is offered b:~ the management of the dictionaries In fact, the system includes several dictionaries and after isolation of each segment the I I system can open&quot; or &quot;close&quot; a dictionary This method makes it easy to avoid, for example, looking EUor two consecutive prefixes. Another mode of intervention uhich is more direct, is peovided by the presence of functions acting on the enumeration procedures by which the system counts ~ff solutions. For example, the system analyzes all possible segmentations starting with a given segment beginning with the segmentations containing most characters. An intervention at this level makes it possible hot to analyze but to reject subsegmentations of P segment. The analysis of the segment &quot;UNE&quot; can, for example, reject the analysis of the subsegment &quot;UN E&quot; (Observe that the segmentation of the word &quot;chacune&quot; will then be obtained as GAC UNE because th8 segmentation CHACUN E will. be rejected as a subsegmentation of &quot;UNE&quot; This problem can easily be resolved because these functions appear in the rules of the grammar and are consequently conditional. One can at the same time f&quot;orbid the subsegmentation &quot;UN E-&quot; in the word &quot;UNE&quot; and aur;horize this. segmentatian in the word &quot;CHACUNE&quot;) The calculation of the set of labels associated ~ith a word is produced and controled by the grammar. This calculation corresponds above all with a conditianal modification of the label C or current state starting from the label A or argument state. The condition for the evolution of this label is such that if no evolution is possible then the corresponding segmentation Ls rejected. This condition can refer to the labels of the preceding analyzed words and can condition its result on the analysis of the following form. Thus for example in the course of the analysis of the word &quot;LA&quot; in the sequence &quot;il la voit&quot;, the segmentation taking &quot;la&quot; as article can be rejected. The transfer of information to different labels can be realized through assignmenttothe following label S. When this label has been assigned in the course of the analysis of a word the analysis of the follow'ing word will begin with the assigned label instead ef the null label.</Paragraph> <Paragraph position="8"> The final result of the system is a labeled tree. With no supplementary specification in the course of analysis, this tree appears in the following form: w ---- 43 w n Au .p - -A fi, A >* solution 1 I solution & phrase I ph~ase I phrase p phrase p The solution for a sentence (~hrase) consists of a string of labels (one for each word of the sentence), each of which represents an interpretation of a ward of this sentence. In this case, the sentence is not structured; simply the ambiguities are sepa~ated. Th the course of the analysis of the words, a first sketch of a construct5an can be made and give as result a more developed tree. These functions specify the position that the point to which the calculated mask applies must take in the final tree. This position is determined in all cases below a point w, and is relative to the root (first point on the left below wi) and to the rightmost point of the tree already constructed. Thus this point can become itself the root, the rightmost leaf, etc.</Paragraph> <Paragraph position="9"> With, for example, the analysis of the string &quot;une belle maison&quot;, we can. have during the analysis of une&quot;, no tree during the analysis of &quot;belle&quot;, the tree eontains the single point &quot;une&quot;. A function can render the point &quot;belle&quot; as root and give belle une during the analysis of &quot;maison&quot;, if the constructed tree is belle,. a function can provide for swapping we the root with the occurrence in work and give the tree une belle.</Paragraph> <Paragraph position="10"> In this case, the result for the system will be une belle</Paragraph> </Section> <Section position="24" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 5, THE CETA SYSTEM </SectionTitle> <Paragraph position="0"> The CETA system provides for writing and s5mulating a transformational grammar. This system manipulates labeled trees of the type described above (labeled trees produced by the ATEF or other system). To construct a transformational grammar with this system two complementary elements are necessary: the set of rules used defines the set of prfmitives of the system for a given application The set of grammars and the definition of their linkage defines the mode of use of the primitives The definition of a transformation rule defines a mode of potential transformation of the tree considered. A rule is defined by a le'fthand part representing the subtree to be modified and a righthand part defining the resulting subtree. For example, let the following be two rules:</Paragraph> </Section> <Section position="25" start_page="1976" end_page="1976" type="metho"> <SectionTitle> ART NMC ART [UM~ ART NMC NMC ART </SectionTitle> <Paragraph position="0"> On the tree resulting from the analysis by the ATEF system of the senterice &quot;une irresistible jeune femme rousse&quot;, we will have the following applications: une irrgslstible jeurre femme rousse irresistible .ibunz GT*' rousse art. edj. adj. NVC adj. adj.</Paragraph> <Paragraph position="2"> uns irresistible jeune femme 'ousse irr6sistible Jeune femme rousse art. adj. ad j ..</Paragraph> <Paragraph position="4"> In fact, the~defin'itton of a transformation can call on a hierarchical set of subtrees. In the example taken here, the input tree is not very &quot;deep&quot; and most often only one-level trees are applicable. However, in the course of development of a cornplete structure, tshe considered tree is arbitrary and the definition of a complex transformation constructed beginning with several subtrees is very refined. The subtrees defined in a rule can likewise be considered. ordered or unordered Let rule R3 below be considered as unardered:</Paragraph> <Paragraph position="6"> uns ~1-~-~~1s~ible jeune f~mme rousse art. adj.</Paragraph> <Paragraph position="7"> aQJ.</Paragraph> <Paragraph position="8"> NMC adj.</Paragraph> <Paragraph position="9"> I f 1 I uric' irrisistible jeune femme* art. adj.</Paragraph> <Paragraph position="11"> adj.</Paragraph> <Paragraph position="12"> The linkage of the different rules previously described is defined by the set of elementary grammars. An elementary grammar c~~n~ists of ordered rules. A rule Ri will be applied prior to an R. if the order of R. is less ahan the order of R J 1 j An elementary grammar has furthermore a mode of execution. Atl elementary grammmar unitarily executable is such that its result will be obtained after an application of a part of the set of rules mentioned. (An application of the rules mentioned can cause to appear new possible applications which will not be performed in this case.) Anoth-er mode of application of an elementary grammar is exhaustive. In this mode, the set of rules of the gzammar will be applied up ro the maximum but the application of a given rule has the effect of eliminating it from this elementary grammar. (That is, for a given point.) With this second mode of application, the number of possible steps for a given tree is always finite. Within an elementary grammar which is unitarily or exhaustively exechtable, the presence of recursive rules makes. it possible to obtain complex constructions by simulating repetitive procedures. A recursive rule is charaoterized by a call to a new grammar (which can obviously be the same as thae in which the recursive rule is found). The result of'the application of a recursive rule consists of the tree obtained after transformation by the called grammar of rhe tree transformed by the rule in question.</Paragraph> <Paragraph position="13"> For example, let R3 and R2 be the-rules previously described. The elementary grammar G consistiag of these two rules will furnish as result, In unitary mode: application of R3. (Priority is given by the order of enumeration of the rules.) une irrgsistible jeune femme rousse une irrgsistible femme rousse art. adj. adj. NMC adj'. art. adj. NMC In exhaustive mode: application of R3 then R2.</Paragraph> <Paragraph position="14"> (figure at top of next frame) une irresistible jeune fe&ne rousse adJ. rdMC adj.</Paragraph> <Paragraph position="15"> \ i t une irrgsistible femme rousse art. aqj.</Paragraph> <Paragraph position="16"> a~t. NMC - III adj . adj. adj.</Paragraph> <Paragraph position="18"> a, 1.</Paragraph> <Paragraph position="19"> Application of rule R3* corresponds to the1 recursive call of this rule, terminating when the rule is nb longer applicable.</Paragraph> <Paragraph position="20"> With rules R4, R5, and R6, the construction is The definition of a CETA grammar consT-s-t-s of a set of elementary grammars and a conditional linking procedure over them. The linking must he such that the Corresponding graph is loop-free. An elementary gxamar from which no linking is possible yields as result the input tree in place of the transformed tree. This procedure permits one to obtain a methad of analysis involving several criteria of acceptance, each consisting in the presence of a tree schema in the terminal tree.</Paragraph> <Paragraph position="21"> Translation of a tcx~ prepared for the First National Gonference Congress in Munich, the first informal meeting was held of what became Research Cornittee Number One of the Association, the Committee on Conceptual and Tem~nological Analysis (COCTA) This committee (which includes political scient ists, sociologists, anthropologists, linguists, and philosophers) has been moving toward several objectives of concept clarification in political and social analysis. COCTA has organized panels at many polirical science and socialogy associations, including its formal, association with the C mparative Interdisciplinary Studies Section of the International Studies Association as the Internet on Conceptual and Terminological Analysis. Over the half-decade of its existence, COCTA has developed sever31 separate stages of concepmal analysis including special foci on metalinguistics, concept construction and reconstruction, and clarification of the theoretical usages of concepts. Underlying these and other interebts is a prerequisite need for an inventory of concepts-in-use. The rationale for attempting to develop the inventory, and discussion of its potential usages, are fully stated in my Commencement of a Systematic Concept</Paragraph> </Section> <Section position="26" start_page="1976" end_page="1976" type="metho"> <SectionTitle> COCTA </SectionTitle> <Paragraph position="0"> Cozlection This statement sets forth the description of the resulting official COCTA Concept inventory2.</Paragraph> <Paragraph position="1"> The inventory is a rather ambitious project that will depend upon the contributions of interested scholars. It will begin with special focal points within political science and sociology as a pilot project. The logic of this pilot collection, however, is to provide a framework within which che collection can be expanded into other social sciences and related fields in the humanities. The immediate task is to commence the collection of social science concepts-in-use and to demonstrate the inventory's utility. Since the inventory can be commenced only by volunteers, the aid of scholars from several disciplines is essential to its success. Any concepts can be listed by interested scholars.</Paragraph> <Paragraph position="2"> The present proceduxes for entering concepts into the inventory are simple. Scholars in the fields record concepts and related information according to the inventory's format and mail them .tro me. These materials will be edited and sent to Carl Beck at the University of Pittsburgh where the concepts and information will be recorded and stored (the Pittsburgh '~ittsburgh: Univer.sity Center for International Studies, No. 9, 1974. See also the other COCTA papers listed therein.</Paragraph> <Paragraph position="3"> 2~he fhal design of the collection has seriously benefitted from comments from Fred W Riggs, from those who attended a special workshop on the inventory at the 1975 International.</Paragraph> <Paragraph position="4"> Studies Association Meeting in Washington (including Carl Beck, James Bj orkman, Judy Bertelsen, Ray Corsado , David Hays, Ray Johnston, R. J. KirkbrPde, David Nasatir, Stephenie Neuman, Jona than Pool, Char les Powell, Fred Riggs , Henry Teurie, Theodore Bukahara, and Alan Zuckerman), and special responses from David Hays and Glenda Patrick.</Paragraph> <Paragraph position="5"> COCTA 43 system also houses, among other important resources, the United</Paragraph> <Section position="1" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> States Political Science Information S~S~~~--UPSIS)~ Except </SectionTitle> <Paragraph position="0"> for the labor and postage costs, to be absorbed by scholars one way or another, Beck's technical and storage assistance permits commencing the inventory without funds. Once the inventory is seriously commenced, funds should quickly follow.</Paragraph> </Section> </Section> <Section position="27" start_page="1976" end_page="1976" type="metho"> <SectionTitle> INVENTORY FORMAT </SectionTitle> <Paragraph position="0"> For each definition of a concept from the literature, the following information should be recorded by typing the information on- 8% x 11 inch paper. The identification of field and its contents should follow as below, with the information replacing the field descriptions. The information for some fields may either not be available or not be relevant, but NO RECORD WILL BE STORED THAT DOES NOT COMPLETE THE INFORMATION F8R THE FIRST. SEC~D, AND THIRD FIELDS. Each definition of a concept will be assigned an entry number when placed in the inventory because of multiple definitions for a specific term, but this will not affect the records sent from the field.</Paragraph> </Section> <Section position="28" start_page="1976" end_page="1976" type="metho"> <SectionTitle> FIELD DESCRIPTION OF CONTENTS </SectionTitle> <Paragraph position="0"> THE TERM USED BY THE AUTHQR TO REFERENCE A CONCEPT, e.g consensus' (IE the term is not English, it shauld be followed by a coma and the closest English translation</Paragraph> </Section> <Section position="29" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 2 ORIGINAL LANGUAGE DEFINITION DIRECTLY FROM THE TEXT. </SectionTitle> <Paragraph position="0"> If the term and definition are in a language other than English, the definition should be followed by an EXACT 3~he UPSIS is a special abstracting and retrieval system of political science articles, books, papers, etc., published in the United States which are indexed and retrieved by using the American Political Science Association's Political science Thesaurus, eds. Carl Beck, Eleanor D. Dym, and 3. Thomas McKechnie (Washington, D.C.: American Political Science Association, 1975).</Paragraph> <Paragraph position="1"> COCTA 44 English translation. (Because exact translation may require familiarity with the article, these entries require exceptional care.)</Paragraph> </Section> <Section position="30" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 3 The source of the definition should be fully cited by AUTHOR, TITLE OF PUBLICATION (article and journal title </SectionTitle> <Paragraph position="0"> if appropriate), PUBLICATION INFORMATION (full standard references for book, journal, or other paper or publication), and PAGE(S) from which the DEFINITION is drawn If guidance for full citations is needed, the most complete reference is A Manual of Style, (12th ed. ; Chicago and London: University of Chicago Press, 1969)</Paragraph> </Section> <Section position="31" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 4 RELATED CONCEPTS should be noted by identifying terms </SectionTitle> <Paragraph position="0"> associated with the meaning identified by the definition.</Paragraph> <Paragraph position="1"> The use of a term is, of course, arbitrary since rneaningful associations must be with other concepts, but the associations of terms will provide guidelines specified by the individual recording the entry. Each related concept (identified by terms) listed should be preceded by BC, NC, RC, or OC as follows: BC BROADER CONCEPT of which the recorded concept is a less extensive definition NC NARROWER COWCEPT of which the recorded concept is a more extensive definition RC RELATED CONCEPT of which the recorded concept is on the same LEVEL of extension, though different in extension OC OVERUAPPING CONCEPT of which the recorded concept shares extension 5 Category of concepts as either THEORETICAL or OPERATIONAL INDICATOR should be noted simply by entering either '!theoreticalw or &quot;operational&quot; in this field. This will facilitate searches for sets of measures far concepts in devising research designs.</Paragraph> </Section> <Section position="32" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 6 ENGLISH LANGUAGE DESCRIPTIONS OF THE USE O'F THE CQNCEPT </SectionTitle> <Paragraph position="0"> should be recorded. For example., &quot;Revolution is defined only for use when analyzing third world nations from the COCTA 45 perspective of demographic measures. &quot; These descriptions should attempt LO characterize the type and level of theory employed as completely as is possible. Several sentences can be used. Retrieved definitions then CAN be limited to only those concepts which ALSO have description terms of interest in this file For example: REVOLUTION/THIRD WORL,D/DEMOGRAPHIC. (Since the collection will be stored in the same retrieval network as USPSIS, the APS Thesaurus terms provide useful guides for types of descriptors that can be used in both</Paragraph> </Section> <Section position="33" start_page="1976" end_page="1976" type="metho"> <SectionTitle> systems .) 7 IF A TERM FOR THE CONCEPT IS INCLUDED IN ESTABLISHED RETRIEVAL THESAURI, THESE SHOULD BE LISTED. The term </SectionTitle> <Paragraph position="0"> associated with the definition may or may not be listed in the Political Science Thesaurus of the American Political Science Association, or some other thesauri IF THE TERM IS LISTED in any available thesaurus, the name(s) of the thesaurus should be listed, If in more than one, a coma should separate each listing. IF THE TERM IS</Paragraph> </Section> <Section position="34" start_page="1976" end_page="1976" type="metho"> <SectionTitle> KNOWN NOT TO BE LISTED IN A THESAURUS, the recorder is. </SectionTitle> <Paragraph position="0"> asked to select the term(s) closest to the assigned term and list it, followed by the thesaurus s name (e .g. , &quot;APPEASEMENT, political Science ~hesaurus&quot;) . The internal structure of the thesaurus will provide, without recording for the storage system, broader, narrower, and related TERMS, in contrast with the recorder-listed set of related CONCEPTS recorded under 4.</Paragraph> </Section> <Section position="35" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 8 THE NAME AND LOCATION OF THE INDIVIDUAL 'RECORDING THE CONCEP~'S DEFINITION. </SectionTitle> <Paragraph position="0"> The clarification of concepts will inevitably lead to restatements of definitions from the literature, to metalinguistic information worth storing, etc. Any restatements not contained in papers, articles, books, etc., can be sent in the same for-</Paragraph> </Section> <Section position="36" start_page="1976" end_page="1976" type="metho"> <SectionTitle> COCTA </SectionTitle> <Paragraph position="0"> mat as the above with the third category filled in as a COCTA PARTICIPANT RESTATEMENT If the restatement is in a form subject to citation, it is simply entered as any other conceptin-use .</Paragraph> <Paragraph position="1"> Because the COCTA Concept Inventory is designed to facilitate research and concept clarification in the social and related sciences, the COCTA ~oard~ and the Director of the COCTA Concept Inventory hope to draw upon and share the mutual rewards and costs with active scholars. The enterprise depends upon scholars taking the time to record the concepts they are using and promises, in return, to facilitate the efforts of scholars by providing an expanding list of concept meaningsin-use. null &quot;~eneral information about activities can be received.</Paragraph> <Paragraph position="2"> The Institute, a branch of the Fondaeione dalle Molle, is carrying on research on arttfiaial intelligence (AI); about ten scholars devote themselves to the study of comhunication between man and machine, under the direction of Manfred Wettler.</Paragraph> <Paragraph position="3"> The tutorial was a week of lectures, seminars, and discussiws conducted by the staff of the Institute, supplemented by evening discussions and presentations of their own results by participants. About 100 persons from Germany, Great Britain, Italy, Holland, Denmark, France, Belgium, Switzerland, Norway, Israel, Canada, and Japan attended. They were teachers, students, or researchers with various fields of interest and</Paragraph> </Section> <Section position="37" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 1.UTORIAL ON COMPUTATIONAL SEMANTICS </SectionTitle> <Paragraph position="0"> background: linguist$cs, psychology, philosophy, automatic translation, computer science, social sciences, engineering., etc. The courses offered embraced a wide range of topics related to semantics. Some of them were inrroductory courses, others were survey courses including the lecturers' own scientXic results and discussions of these in relation to recent research. This variety of fields taught at different levels was well suited to the audience.</Paragraph> <Paragraph position="1"> Below we will account for the lectures chronologically, describing at greater length those which were most relevant to us</Paragraph> </Section> <Section position="38" start_page="1976" end_page="1976" type="metho"> <SectionTitle> PARSING ENGLI S-H - Yorick Wilks </SectionTitle> <Paragraph position="0"> A survey and comparison of some of the better known A1 systems, this course began with certain fundamental concepts and general characteristics of relevance for all the-systems in question. A principal issue is parsing. Wilks defined it as &quot;prooedural ascription of structures to sentences, where the structures are pot smtactic at all, but semantic.</Paragraph> </Section> <Section position="39" start_page="1976" end_page="1976" type="metho"> <SectionTitle> 1 I </SectionTitle> <Paragraph position="0"> Parsing may be done in two different ways: TOP-DOWN or BOTTOM-UP. Bottom up is the more straightforward way. The words of the sentence are listed and each word is replaced by its category Then pairs of category symbols (for instance Verb + NP) are rewritten by reversing the grammar's rewrite rules (Verb + NP --> VP) until the final sentence symbol S is reached. The lines of the derivation can then be considered as the parsing.</Paragraph> </Section> <Section position="40" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS. </SectionTitle> <Paragraph position="0"> Top-down parsing is the reverse procedure starting with the generations and continuing from left to right until the last word is reached. Another important pair of teahnical terms is BREADTH-FIRST and DEPTH-FIRST. Breadth-first is the parallel treatment of all possible alternative structures at a given time, none of which is given precedence. In depth-first pafses,, the akternative structures are treated sequentially. So far the description may apply to any kind of parsing, but it was Wilks's aim to demonstrate parsing procedures where the structures are not syntactic but semantic. He described his own view of semantics as a version of the &quot;meaning is procedures&quot; attitude, i e. the procedures of its application give a pgrsed structure its s-ignificance.</Paragraph> <Paragraph position="1"> After mentioning what he called the &quot;problem of natural lang~sge&quot;, by which he meant the problem of systsmatic ambiguity, Wilks gave a brief historical sketch of the first approaches to machine translation, the failure JPS which he put down to the ambiguity problem.</Paragraph> <Paragraph position="2"> Terry Winograd has proposed a distinction between &quot;first&quot; and &quot;second&quot; generation CI language systems. This distinction that seems no* to be wfdely acceptad also lies behind the survey below, where the systems of Winograd and Woods are considered first-generation and those of Simmons, Schank, Charniak, and Wilks belong to the second ge~eration. Winograd's well-known dialogue system SHRDLU operates in a closed world of colored blocks and</Paragraph> </Section> <Section position="41" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIOhAL SEMANTICS 50 </SectionTitle> <Paragraph position="0"> pyrdmids. The gratnmatj of SHRDLU is not the conventional list of rules but s~ll subprograms that actually represent procedures for imposing the desired grammatical structure. In terms of: the notions set out earlier, Winograd's parsing is top-down and depth-first. After the syntactic parsing a number of &quot;semantic specialistst1 attach semantic structvres to specific syntactic structures. These semantic structures can then be used by the deductive component of the system. Woods's system, tob, is considered first-generation, but both Woods and Winograd have argued that their systems are essentially equivalent, which is the reason why Wilks described only one of them in detail What the second-generation systems have in common is the assumption that understanding. systems-must be able to manipulate very complex linguistic abjects, or semantic structures, and that no simplistic approach to understanding language with computers rill work. A common Peature in connection with second-generation systems is what Rinsky (1974) calls a FRAME. It is described as a data-structure representing a stereotyped. situation and attempting to specify in advance what is going to be said, and h~w.the world encountered is going to be structured.</Paragraph> <Paragraph position="1"> Colby s system, too, is a dialogue system, by which an interview between a doctor and a paranoid patient called PARRY is carried out. The input text is segmented by a heuristic that breaks it at any occurrence of key words. Patterns are then matched with each word string segment. Stored in the same format as the patterns are rules expressing the conseque-xes</Paragraph> </Section> <Section position="42" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS. </SectionTitle> <Paragraph position="0"> fo~ the 2atient of detecting aggression and overfriendliness in the interviewer's questions and remarks. The matched patterns are then tied directly, or via these inference rules, to che response patterns which are generated.</Paragraph> <Paragraph position="1"> A very interesting aspect of the PARRY system is the fact that the answers of the system cannot be distinguished from those of a human patient This fact suggests that many people on many occasions seem to understand the information they receive in the same way that PARRY does.</Paragraph> <Paragraph position="2"> Schank's is a rich system of semanti.: representation. It consists of the following three components: 1. an ANALYZER of English, due to Riesbeck 2. a SEMANTIC MEMORY Component, due to Rieger 3. a GENERATOR OF ENGLISH, due to Goldman The aim of Schank's system is to provide a representation of meaning in terms of which different kinds of analysis and machine translation can be carried out; a representation, moreover, that is independent 0.f any particular language ,. and of syntax, and, indeed, of all traces of surface structure After a detailed description of Schank's so-called CONCEP-TUALIZATIONS, built up by conceptual categories, primitive acts, cases, etc., Wi-lks gave his own comments an Schank's system. Like that of Schank, Wilks's system has a uniform representation, in the shape of structures and primitives, for the content of natural language. It is uniform in that the</Paragraph> </Section> <Section position="43" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 52 </SectionTitle> <Paragraph position="0"> information that might conventionally be considered syntactic, semantic or factual is all represented within a single structure of complex entities (called FORMULAS and PARAPLATES), all of which are in turn constructed from 80 primitive semantic entities. The formulas are tree structures of semantic primitives, stored in the dictionary of the system. The main element in any formula is its &quot;head&quot;, i. e. the fundamental category to which the formula belongs. Sentences and their parts are represented by the socalled TEMPLATE STRUCTURES, built up as networks of formulas. Templates always consist of an agent node, an action node, and an object node, and other nodes that may be governed by these. A formula or, say, the noun.&quot;drinkl' can be thought of as an entity at a template action node, selecting a liquid object, that is to say a formula with FLOW STUFF as its head, to be put at the object node of the template (sentence structure). This seeking is preferential in that formulas not satisfying a given requirement will be accepted, but only if nothing satsifying it can be found. The template ultimately established for a fragment of text is the one in which the most formulas have their preferences satisfied. This preference principle is of essential importance in connection with solving the many ambiguity problems in natural language texts. When the local. inferences have been done that set up the agentr action-object templates for fragments of input text, the system attempts to tie these templates together so as to provide an overall initial structure for the inaut called a CASE TIE.</Paragraph> </Section> <Section position="44" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON CUMPUTATIONAL SEMANTICS 53 </SectionTitle> <Paragraph position="0"> Case ties are made with the aid of angther class of ordered structures called PARAPLATES, each of which is a string of functions that seek inside templates for information. The last step in the parsing is the inference procedure in which commonsense inference rules attempt by a simple strategy to construct the shortest possible chain of rule-linked template forms, on the principle of preference.</Paragraph> <Paragraph position="1"> The other main section of this course was a comparison of the parsing systeps described, including Charniak's system.</Paragraph> <Paragraph position="2"> This,comparison was based on the following principal aspects: LEVEL OF REPRESENTATION. At this point there qre two Opposite views: that language can be realized or represented at different levels depending on the subject matter, or that the appropriate level of computation for inferences about natural language has to be to some degree reduced. The different level attitude is supported mainly by Colby and Charniak, while Schank and Wilks hold that a certain primitivization is necessary CENTRANTY OF INFORMATION. This aspect concerns the degree of specificity of the information required. Some systems, Iike Charniak's, are based on infomation highly specific to particular situations, while the sorts of information central to Sohank's and Wilks's systems are of a much niore general nature, consisting mainly of partial assertions about hman wants, expectations, and so on. This problep of centrality is of great theoretecal importance, which Wflks illustrated by an example: A person might know nothing of a particular type of</Paragraph> </Section> <Section position="45" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 54 </SectionTitle> <Paragraph position="0"> situation, for instance a birthday party, but could not for this reason be accused of not understanding the language.</Paragraph> <Paragraph position="1"> Yet, if he did not have available some very general inference such as for instance people gettihg sleepy at night, then it is possible that his failure to understand quite srmple sentences would cause observers to think that he did not know the language. Wilks went on: An interesting and difficult question that then arises is whether those who concentrate on central and less central areas of discourse could, in principle, weld their bodies of inference together in such a way as to create a wider system; whether, to put the matter another way, natural language is a whole that can be built up fr~m parts.</Paragraph> <Paragraph position="2"> PHENOMENOLOGICAL LEVEL. This is a question of degree of explicitness. Here Schank s system is distinctive. Wilks's opinion is that the amount of detailed inference that a system may perform must be llmited not to go beyond 'commpn sense'.. As an example he mentioned Schank's analysis of the action of eating (performed by moving the hands to the mcuth) and de- null scribed it as Ugoing too far from the meaning' of eating, whatever that may be, towards generally true information about the act which, if always inferred about all acts aPS eating, will carry the system unmanageably far. . . . There clearly is a danger of taking inferences to a phenomenological level beyond that of common sense,&quot; he concluded.</Paragraph> </Section> <Section position="46" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 55 </SectionTitle> <Paragraph position="0"> DEGOUPLING.</Paragraph> <Paragraph position="1"> The issue is whether the actual parsing of text fnto an understanding system is essential.. Charniak and Minskr believe that this initial parsing can be decoupled. In Wilks's opinion this is not so, because he belkeves semantic analysis to be fundamental and because many of the later inferences would actually have to be done already, in order to have achieved the initial parsing. Also the problem of systematic ambiguity may be met much more efficiently with a system that does not decouple the parsing from rhe inference procedure.</Paragraph> </Section> <Section position="47" start_page="1976" end_page="1976" type="metho"> <SectionTitle> AVAILABILITY OF SURFACE STRUCTURE. In first and second </SectionTitle> <Paragraph position="0"> generation systems it is generally accepted that word-sense is a.</Paragraph> <Paragraph position="1"> closely associated with the surface structure of the sentence, but Schank has made a point of the-nonavailability of the surface structure, on the grounds that an ideal representation should be totally independent of the input surface structnre and words. In connection with this claim of Schank's, Wilks pointed out two things: in many cases the order of the sentences in.a text is part of its surface structure, and this information should be available in some way. The other point conceined the form of =epresentation employed Wilks was not sure that a structure of primitlves. is sufficient for specifying and distinguishing word senses adequateLy without transferring information specifically associated with the input word.</Paragraph> <Paragraph position="2"> APPLICATION.</Paragraph> <Paragraph position="3"> This concerned the way in which different systems display, in the structures they manipulate, the actual procedures of application-of those structures to input text or</Paragraph> </Section> <Section position="48" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS </SectionTitle> <Paragraph position="0"> dialogue Here the most distinctive. system is that of Wfnograd where the procedural notation, by its nature, tends to make clear the way in which the structures are applied. In hi6 view, as stated in some of his more recent writings, the control structure of an understanding program is itself of theoretical significance, for only with a control structure, he believes, can natural language programs of great size and complexity remain perspicuous.</Paragraph> <Paragraph position="1"> FORWARD INFERENCE. IS it neaessary to make massive forward inferences as one goes through a text., as Charniak and Schank do, or can one adopt some laziness hypothesis' about understanding and generate deeper inferences only qhen the system is unable to solve, say, a referential problem by more superficial methods? Charniak's argument is that, unless forward inferences are made during the analysis, the system will not in general be able to solve ambiguity or reference problems that arise later. Wilks had some theoretical difficulties. tn arguing against this view, and he admitted the difficulty of defining a degree of forward inference that aids the solution of later semantic problems without going int~ unnecessary depth THE JUSTIFICATION OF SYSTEMS. Finally Wilks tried to contrast. the different modes of justification implicitly appealed to in terms of the power of the inferential system employed, of the provision and&quot;formalizatfon, of a system's actual performance, and of the linguistic or psychological.</Paragraph> <Paragraph position="2"> plausibility of the proffered system of representation.</Paragraph> </Section> <Section position="49" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 57 </SectionTitle> <Paragraph position="0"> In his conclusion Wilks concentrated on those areas where the greatest problems within the field of A1 are found.</Paragraph> <Paragraph position="1"> The following needs seem to be the most pressing ones the need for a good memory model (stressed by Schank), the need for an extended procedural theory of texts and for a more sophisticated theory of reasons, causes, arid motives for use in a theory of understanding. Wilks ended his survey by stressing the fact that there is an AI paradigm of language understanding which embraces first and second generation approaches and goes back to a considerable amount of earlier work in computational linguistics</Paragraph> </Section> <Section position="50" start_page="1976" end_page="1976" type="metho"> <SectionTitle> INFERENCE AND KNOWLEDGE - Eugene Charniak </SectionTitle> <Paragraph position="0"> Why do we make inferences? We do when we use language and when we decode the information conveyed by language, i . e. in the case of structural disambiguation as well as in word-sense disambiguation, reference determination, question answering, translation, smarizfng, etc., everywhere a thing not stated explicitly has to be assumed. In so doing we are looking Eor a piece of information, for knowledge beyond the given text or situation Charniak poses five questions about how knowledge is used to make inferences: 1 What concepts, and in what combinations, do we need to record our impressions of the wgrld? (semantic representation) 2. Under what circurnstanees and why do we make inferences? (inference tr ipgering) TUTORIAL ON COMPUTATIONAL SEMANTICS 58 3. How do we 1ocate.the needed information? (organization) 1, 4. Once located, how do we know how to vSe the information? (inference mechahism) 5. @hat is the knowledge that we have of the world that enables us to understand language? (content) After this program had been put forth, Charniak presented ~ww partial answers to the questions the first order predicate calculus (FOPC) and the programing language PLANNER. FOPC consists of a+ language for expressing facts and rules for deriving new facts from old. The language consists of constants, variables, predicates, functions, logical connectives, and quantifiers. There are rules for inference. Charniak then outlined RESOLUTION THEOREM PROVING. It is a system for setting up proofs for deciding which rule of inference to use. Charniak proceeded to look at the five questions he had set forth and examined what answers. FOPC provides to them. He concluded that FOPC is primarily a theory ot inference mechanism,, but that it says very little about semantic representation. As FOPC dogs not tell how one is to locate the facts which are to be used to prove the derived result, theoretically we come up against a huge amount af possibilities when we combine the number of possible clauses.with the number of possible resolutions. This is called the &quot;combinatorial explosion'&quot; and is a serious problem in most inference systems, not only far FOPC.</Paragraph> <Paragraph position="1"> Charniak then examined the ~roblern of when we make inferences. There are two obvious occasions when we may make one:</Paragraph> </Section> <Section position="51" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS </SectionTitle> <Paragraph position="0"> 1. When a question is asked which requires an inference to be made (question time) 2. When the system has been given edough input i-nformation to make the inference (read time) Although the inference making restricted to question time would seem to be more ,economical since inference.is done only when we</Paragraph> <Paragraph position="2"> must, in oraer to answer the system user's question', there is some ovide~ca that inference is done at reading time (e.g. psych~logical experiments on recall of texts). Furthermore, it is not possible to do word sense or steuctural disambiguation.</Paragraph> <Paragraph position="3"> withovt making inferenc-es. Wilks makes a distincti~n between 'broblem occasioned&quot; and &quot;nonproblem occasioned&quot; inference. A typical example of the latter is given in -&quot;Janet shook ner piggy-bank. There was no sound.. &quot; We assurnel that there is nothing In the piggy-bank although the problem has not yet arisen in the story. Charniak believes that to do question answering on complex stories the system must perform nonproblem occasioned inference. ne glves examples rrom children's stories where persons lie about things and where the system has to.guess why the person is lying An alternative to FOPC is to use the natural properties of some programming language to make inferences.</Paragraph> <Paragraph position="4"> Bertram Raphael (1968) did this .ih the system SIR when he used LISP to construce a data base. Another way is making the programing languagesmore suited to the needs of inference making. Such a system has t&rl designed but not implemented: PLANNER</Paragraph> </Section> <Section position="52" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 60 </SectionTitle> <Paragraph position="0"> (Hewi'tt 1969) In this system we are able to pick up an assertion by knowing parts of it If no appropriate assertion can be made, we can try to have theorems (i.e. programs) investigated An antecedent theorem is one where we are given the antecedent and'we assert the consequent, while with a consequent theorem we are asked to p-row the coasequen-t and we try to find the antecedent. PLANNER has the ability to choose which theorems to use on the basis of their patterns. This is called PATTERN DIRECTED INVOCATION. Furthermore, the system can back up to see if any earlier choices night be changed.</Paragraph> <Paragraph position="1"> This feature is somewhat controversial, since it might encourage the construcLi0n of programs which depend on blind search.</Paragraph> <Paragraph position="2"> PLANNER'S advantage over FOPC is that it offers several built-in organizational features, tne primary one being pattern directed invocation. A disadvantage about it as theory of knowledge and inference is thar: it is too vague Charniak (197.2-) illustrates the pros and cons of PLANNER using children's stories. Given a piece of simple narration, the system should be able to answer reasonable quegtions about it. Charniak Stresses the need for looking ahead in thg story to make inferences For this he uses. an anteceden~ theorem or a &quot;demon&quot;. The routines which are available to set up demons he calls ensE ROUTINES.</Paragraph> <Paragraph position="3"> In addition- he makes use BOOKKEEPING for updating the assertions and of Consequent theorems pealled FACTFI-NDERS: the basic idea behind faetfinders is that they are used to establish facts whieh-are'not too important so that we do not want to</Paragraph> </Section> <Section position="53" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL .SEMANTICS </SectionTitle> <Paragraph position="0"> nssert thew and store them in the data base.. The main advantage of this system is that it provides a good theory of srganization. It states in particular that &quot;given a particular assertion. the way we find those facts which we should use to nake inferences from the assertion is to 'look in two places-.</Paragraph> <Paragraph position="1"> first the base routine for assertions of that form, & second for any demons.which happen to have been activated which are looking for assertions of that ford&quot;' Charniak concluded his lectures by examining the recent works of three scholars: 1. McDermott ' s system TOPLE (1.974) is rnalnly concerned with the problem of 'beliefs, describing a simple world consisting of a monkey and an experimenter in a single room. The program listens to a present-tense account of what is happening in the zoom; it tries to understand why things happen and what can be sxpected to happen as the story poas on. It tells us at t.he end of every sentence what new assertions it has assumed as a rbesult of hearing.</Paragraph> <Paragraph position="2"> TOPLE's restrictions are the following: it does not answer questibns, it does not handle actual natural language but rather a formal-looking input language. On the otrrer hand, it tries to visualize concretely a situation. It is based on a &quot;multiple wo.rld structure&quot; 2. Rieger (1974) is the first to have attempted to use Schank's conceptual dependency theory within a theory af inference and knowledge. Rieger!~ program has as its main purpose tc make reasonable inferences from the input it is given. The</Paragraph> </Section> <Section position="54" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 62 </SectionTitle> <Paragraph position="0"> input is expressed in a suitable formalism, i.e. conceptual dependency representation. It is also designed to understand stories, engage in dialogues, figure out references and word-sense ambiguity, answer questions about the way the world normally is 3. Minsky ' s (1974) frames are reinterpreted by Charniak as &quot;a collect-ion or questions to be asked about a hypothetical situation. Frames specify issues to be raised and methods to be used in dealing with them.</Paragraph> <Paragraph position="1"> It Charniak also gave a double lecturn on SYNTAX IN LINGUISTICS. This was an introduction to generative grammar for those who had not had a. formal course in linguistics.</Paragraph> </Section> <Section position="55" start_page="1976" end_page="1976" type="metho"> <SectionTitle> MEMORY MODELS - Greg W- Scraqg. </SectionTitle> <Paragraph position="0"> After introducing SEMANTIC NETS, Scragg discussed their most important properties and compared several systems indluding some with partial semantic nets, some with partially quantitied semantic nets, some with fully quantified semanttc nets, and some with executable semantic nets.</Paragraph> <Paragraph position="1"> He compared semantic net representations and predicate calculus tepresentations.</Paragraph> <Paragraph position="2"> Attempts to construct proofs in the predicate calculus will show the difficulty CXE sklecting the relevant infqrrnation for making a particular deduction from a specif-ic fact. The techniques currently employed in theborem proving programs are even less efficient,a.t selecting the most relevant material.</Paragraph> </Section> <Section position="56" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS </SectionTitle> <Paragraph position="0"> In comparison 3f predicate calculu-s and semantic nets, most problems center around the question of quantification.</Paragraph> <Paragraph position="1"> How does one quantify relations in a semantic net? Scragg mentions three different approaches.</Paragraph> <Paragraph position="2"> 1. There are six possible quan~~ficatio~s fur a two-place predicate Pxy vxv yPxy , ZxVyPxy , 3yVxPxy, Vz3yPxy, *VyIIxPxy, 3?$3yPxy In Scragg (1973) the claim is made that #the first three forms are so rare in everyaay (nonscientif?~) s-ituations that they may be ignored. The rematning ones may be distinguished with a type-token flag.</Paragraph> <Paragraph position="3"> 2. palme (1973) tries to represent quantification by introducing a third quantifier, ITS (meanins spme-thing like the possessive pronoun &quot;its&quot;). With three quantifiers, he now cam define six separate relations for each pwvious relation: Quantifying with FOR-ALL or EXISTS on the Left and FOR-ALL., EXISTS, or ITS on the r-ight of the old relation. One disadvantage of this is that he potentially has six times as many relations to work with and has to keep erack of the relationships between each of the six versions of the same-relation.</Paragraph> <Paragraph position="4"> 3. Schubert- (1975) treats quantifiers in a different way. He first puts the predicate calculus representation of the statement into SKOLEM FORM (a form which has no existential quantifiers and with all- universal quantlfiers outside 05 the body of the express-ion), Any node that is existentially</Paragraph> </Section> <Section position="57" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORItAL ON COMPUTATIONAL SEMANTICS 64 </SectionTitle> <Paragraph position="0"> quantified but dependent an a universally quantified node is connected to that governing node. An event is asserred if and only if there is no arrow pointing to that node in the diagram.</Paragraph> <Paragraph position="1"> The semantic net structures here tend to become very complex.</Paragraph> <Paragraph position="2"> ~---~~&quot;--&quot;&quot;--------&quot;----&quot;&quot;&quot;&quot;&quot;i.iiiiii&quot;iiiiiiiii&quot; 1 .., a&quot;&quot;. - . .. .&quot; .... .- . -1 - I - - . . - ---- .. .L . . . . . --dl&quot;. .-----It is not clear that any of the three approaches give really practical (or int~~tively satisfying) results.</Paragraph> <Paragraph position="3"> What we need at present is a theory of more conplex actions.</Paragraph> <Paragraph position="4"> For example, how do we link the descriptions of the various substeps of the pro.cess of cake making into a single desciiption of the overall action of making a cake? There arb those who claim that all knowledge is stored in the form of procedures and there are those who clraim that it is stored as a collection of facts.</Paragraph> <Paragraph position="5"> Scragg (1974; see also Nonuan 1973 and Norman et a1 1975) takes an intermediate approach by making use of ambiguous (data or procedure) representat-ions to store information about actions.</Paragraph> <Paragraph position="6"> The system knows how to simulate various human actions-such as toasting bread, making spaghetti sr cleaning up the kitchen.</Paragraph> <Paragraph position="7"> The information ab~ut how to perform these siinulatibns is: stored as procedures. However, these procedures can be used as data by other parts of the system to answer such questicms.</Paragraph> <Paragraph position="8"> as &quot;'HOW do you make a ham and cheese sandwtich?&quot;., &quot;How many utensils do you use if you make a mushroom omelette?&quot; or &quot;Why did Don use a knife?&quot;</Paragraph> </Section> <Section position="58" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS SEMANT I CS I'N LI NGU I ST I CS SEMANTIC MARKERS AND SELECTIONAL RESTRICTIONS. Phillip </SectionTitle> <Paragraph position="0"> Hayes discussed in detail the influential paper by Katz and Fodor (1963). He concluded that their semantic theory is n0.t qufte adeouate even for the purely linguistic system they try to outline. everth he less, it can be a useful component of an A1 theory of natural language comprehension.</Paragraph> <Paragraph position="1"> GENERAT~IVE SEMANTICS. Margaret King outlined the defining characteristics.of this theory and then concentrated on its relationship with AI. As a conclusion, she stated that the definition of grammar logically should be extended to embrace not only wellformedness and semantic acceptability but also all possible aspects of the context of use of a sentence. This is contradictory to the traditional view of grammar-understood aF the sole means of determin'ing which sentences are grammatical for the majority of speakers of the standard form oE the language.</Paragraph> <Paragraph position="2"> CASE GRAMMAR. Wolfgang Samlowski snrv-eyed- Fillmore's theory with special reference to Its -intLuence on American linguistic theories of semantics and on leading researchers within AI.</Paragraph> <Paragraph position="3"> The survey consisted of a presentation of case grammar, an examinatlon of some explicit and implicit traces lett in A1 by the Case grammar theory, and a demonstration of some of the complications that the acceptqnnre of the case gradrmar theory by language-understanding researchers would. cause</Paragraph> </Section> <Section position="59" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTURIAL ON COMPUTATIONAL SEMANTICS DIVERSE </SectionTitle> <Paragraph position="0"> PHILOSOPHY of= LANGUAGE. Yorick Wilks, in a double lecture, compared and contrasted modern philosophy with relation to linguistics, in particular systems of formal logic, represented by the works of Ludwig Wittgenstein and Richard Montague. The survey had special reference to the application of such systems of formal logic to the preparation of language understanalng system-s.</Paragraph> </Section> <Section position="60" start_page="1976" end_page="1976" type="metho"> <SectionTitle> PSYCHOLOGY OF LANGUAGE AND MEMORY.: Walter Bischof gave a </SectionTitle> <Paragraph position="0"> selective historical survey of the prevailing concepts in the field: association, organization aPS data, Gestalt, meaningfulness of data, temporal structure of memory, reaction-time paradigm to investigate semantic memory and the network models of representation as proposed by C.ol'15ns and Quillian (1969) Recent work based on the same assumption has shown that the structure of semantic memory is not quite the logical, bier.archical and economical structure proposed by Collins and Quillian. Bischof gave a list of possible relationships between artificial intelligence and cognitive psychology and concluded that these two disciplines have Little to say to each other because of their different aims and because the available experimental tools proposed by psychology are too poor.</Paragraph> <Paragraph position="1"> LISP. Margaret King taught an &quot;0-level&quot; course and Philip Hayes a more advanced introductory course, to this programming language, which is being used widely by AT researchers, in its original form or in some of its extensions (CONNIVER, PLANNER).</Paragraph> </Section> <Section position="61" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS 67 </SectionTitle> <Paragraph position="0"> TUTORIAL GROUPS. Work consisted of discussions between participants in smaJler groups and one or two of the lecturers Some evenifig lectures were given by the participants.</Paragraph> <Paragraph position="1"> These included H. Harrell, R. Giintermann, and G. Zifonun, who presented ISLIB ('information System on a Linguistic Base), a system for answerifig questions to an input in restricted German, carried out at-.the Institut-fUf aeutsche Sprache at Mannheim. A. McKinnon 3f McGill University, Montreal. discussed his work on the Kierkegaard indices. Some lectures caused vivid discussibn. For example that of V. V. Raskin, Hebrew University, Jerusalem-, advocated corpus dependent semantic.models and recommended his own &quot;f,estric~ed sublanguages&quot;</Paragraph> </Section> <Section position="62" start_page="1976" end_page="1976" type="metho"> <SectionTitle> APPRECIATION </SectionTitle> <Paragraph position="0"> Altogether, tne tutorial in Lugana was very inspiring and profitable for the participants. It was well organized and gave good opportunity for discussions. The teachers in the tutorial being familiar with each other's work succeeded in giving a comprehensive view on the topic of computational semantics. some or us, .however, felt a need for more precise definttions of standard notions, this being a very acute problem in view-of the heterogeneity of the participants' backgrounds. We are, however, aware that this is an inherent and recurring problem at such gatherings, where people with different qualifications meet to dis-cuss comon problems We would like to express the wish tha-t the Fondazione dalle Molle will be able to arrange more tutorials of a similar kind in the future.</Paragraph> </Section> <Section position="63" start_page="1976" end_page="1976" type="metho"> <SectionTitle> TUTORIAL ON COMPUTATIONAL SEMANTICS &quot;FORMULAE&quot; IN COHERENT TEXT : LINGUISTIC RELEVANCE OF SYMBOLIC INSERTIONS </SectionTitle> <Paragraph position="0"> Some difficulties in automatic analysis ~nd translation bound to symbolic insertions in mathematical texts are discussed. Rules dealins with these difficulties are proposed, These rules are based on the use of the whole text of the a~Cicle incorporating a formula.</Paragraph> <Paragraph position="1"> For satisfactory automatic analysis of texts, it is necessary to provide in the dictionary exhaustive serriantical Information ascribed to its entries. But this information can appear to be insufficient in cases where the meaning of! linauistic elements is ascribed to their occurrences by the very text in whlch they. are encountered cf . I or example, pronouns.</Paragraph> <Paragraph position="2"> The other example is provided by symbolic insertions in mathematical texts, which we shall call 'If ormulaeN .</Paragraph> <Paragraph position="3"> So not only 'a= b , 'X 2 Y' etc., but also Ox* @ @ and so on are nforrnulaew.</Paragraph> <Paragraph position="4"> Mathematical formula resembles pronouns in one respect: it is semantically *voidqt being out of context.</Paragraph> <Paragraph position="5"> For example, @G9 may be *setH, &quot;~ubset*, N~rouplt , %peratorn, Mfunctim*, &quot;stringw, *elementu, 9qrule of grammarH, eto.</Paragraph> <Paragraph position="6"> The meaning is ascribed ta a formula by the context.</Paragraph> <Paragraph position="7"> There are a few types of fo~mulae with fixed meanings.</Paragraph> <Paragraph position="8"> F'or example, *dx/dyf Is @derlVatlve8. But this sltbtion 1s not typical.</Paragraph> <Paragraph position="9"> One of the basic usages of formulae conSist's of naming by formula A some individual object a belonging to some class b of objects such that there exists some noun block Cl(A) that names b.</Paragraph> <Paragraph position="10"> For example, i'n the expression 'set R' the formula *R' names some individual set belonging to the class of %etsN-. Noun block @seta (consisting in this case- of a single -noun) names this class.</Paragraph> <Paragraph position="11"> So here c~(R) = *sete.</Paragraph> <Paragraph position="12"> Consider some difficultAes arising in translation because of the absence in a source sentence of the Cl(f) for a formula E.</Paragraph> <Paragraph position="13"> Let US try to translate from, Engllcsh-to 'Russian the sentence n\yn --Wcirst find an xment r of Rv. (1) (Previous surface syntactical analysis is, assumed, its results being represented in dependency-tr-ee form). Syntactically, this sentence (is very simple, but even an experienced &quot;humanq* interpreter would not be able to properly understadd this expression and translate it. h Russian. the element corresponding to the English preposition ?ofv is, generally, the grammatical meaning wgenltivell. We can ascribe this meaning to the formula *Rn [.governed by the prep~sitiorr 'of'): --*We first find element r R + genitive'.</Paragraph> <Paragraph position="14"> (Syntactical links are also shown).</Paragraph> <Paragraph position="15"> At ^this point, the process of translation is suspendea because of the fact that in Russian two non-coordinate formulae cannot depend on the same now.</Paragraph> <Paragraph position="16"> Similar examples are provicted by other languages:. German : In jeder Umgebung V von o X* .</Paragraph> <Paragraph position="17"> A human interpreter does not usually hesitate to properly translate such expressions only because he understmnds their meaning from a general background or vast context. We can point out some characteristic construe-tions in mathematical texts that are sufficient as contexts in such cases. Consider, for example, such a context.(i.e. an expression From the same text): 'Let R be a ring with a unity I*. (2 Ear expression (I), and let us formulate.;~very simple rules t *Let f bea N- Cl(f) = *Ne; (R Here f is some flformulan, N - a noun block, means syntactical irnlr, - reads: 'if . . .</Paragraph> <Paragraph position="18"> then*, and icllb means substitutability.</Paragraph> <Paragraph position="19"> With the aid of the raes R and R we can &quot;f ormulaeN depend on different nouns ) .</Paragraph> <Paragraph position="20"> Of course, 'ri'ng R* Is not substitutable for R in the expression 'ring R' The expression *Components x are nonnegative *</Paragraph> <Paragraph position="22"> w.ith the aid of the re R for the French and German examples abqve, Let us now try to trans1at.e to Russian, the following expressions: 'H is cyclic' (3); 'A smallest k* (4). A predicative adjective in Russian must be put In grammatical agreement With the subject of the seritence; an attributive adjective - with the qoverning noun. That Is, the Russian adjectives for cyclicp In (3) and for *smallestm in (4) mus* asree in gender with 'H' and *kt correspondingly. It is clear that the inf brmation about the gender of a Itf omulafl can be proyided by, Cl(f). Having defined. Cl(H) = 'matrixw for wh1chb..t;he translation 'KATRXTSAw is. remlnine, we receive for (3) the translation There exist numerous other ex,pre.ssions for which the finding of Cl(f) is very desirable, for example: We deflne 3 and k by j = m + n; k = m - no. (5) The &quot;direct* translation of (5) to Russian: 'Qpredeltm j i k putjorn j = m + n; k = m - nb is not smooth enough; the translation: 'Opredelim J i k s pomosh ju so~tn~shenij</Paragraph> <Paragraph position="24"> ~l(f] can be sometimes defined from the very formula f. For example, 'a = b' is nequalityl,, ,, ,*a ) b&quot; fs &quot;inequalityqt, and so on. Somet%mns t-he .he&quot;meming18 of a formula f ban be derived from words syntactically linked to this formula or from a more complex fo2niula F incorprating f.</Paragraph> <Paragraph position="25"> For example, from the e2presSion we can derive that T is a wtransformatianw and that A and @BW are wsetsM. From the expression we can derive that 'B' is a &quot;setu and that 'ap is an 'Subset of A' A@ is a *.setN. In 'Differentiation (.or : integration.') with respect to xv, 'x' IS a l%arlabldN, and so on.</Paragraph> <Paragraph position="26"> Cl(f) for a formula f can be sometimes a more or less bulky expression consisting of a noun with words depending on the noun directly or Indirectly.</Paragraph> <Paragraph position="27"> 'Tous les ensembles Li d indices lnferieurs a un nombre donne IS@ (c~(L~) is underlined).</Paragraph> <Paragraph position="28"> In this case we can reduce C1(L ) to only one word i *ensembles*. But in rare cases such reduction will produce absolutely inadequate translations,</Paragraph> <Section position="1" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Example t </SectionTitle> <Paragraph position="0"> In the expression 'Pour les fonctlons x(t) de Lg with a context ,La partie cornwe L de tous les ensembles Li ' (Cl(L) is underlined), we cannot reduce C1(,L) to only one word upartlee, which Is its syntactical governor, 'St is very difficult to formulate a general Pule ta discriminate between cases of typee (6) and (7). The expression. (7) om be translated using a synonym for C~(L), for example. 'ensemble', having in mind that the intersection of several sets is also a set. The computability of such synonyms can, of course, be questioned, Now we shall consider some proDlems arising in translation of Russian mathematical texts into European and other languages.</Paragraph> <Paragraph position="1"> The construction in Russian has two syntactical meanings, (a) appositive:, (b) genitive:' The cause of thls difficulty is the omission of Cl(f) in r;ne surfaoe syntactical structure of some Russian sentences: lpodmnozhestvo mnozhestva Bv- 'podrnozhestvo Bp &quot;SubseT (of 1 set B&quot; &quot;Subset (of) Bn. In such cases me qenitive link is rare (5% of n all occurrences of constructions of type N f, i.em several-dozen occurrences in a mathematical article). The task of automatic cho1c.e here is very difficult. It was solved only partially. We can ch~ose from the text of an article about 70% of all occurrences of the appositive links and also sowe occurrences of geni't-ive links, The rest of occurrences remain ambiguous. The proposed procedures were checked in exhausting manual experiments, hue their aaaptation for computer is quite feasible.</Paragraph> <Paragraph position="2"> Choice of appositive links Let us consider the following empirically stated ax h oms val id : (A ) In the same Russian text every two different occurrences of t.he same expsre,ssion of type .</Paragraph> <Paragraph position="3"> N f are or both appositive or both So, if we have s-ucceeded in clarifying the meaning of a link in one occurreme of a construction, we can ascribe this meaning to every occurrence of the same construction. (A~) In a.construction of the type where fl and f2 are two syntactt;cally co_T null ~rdinate formulae, the two links are both appositive or both qenitive.</Paragraph> <Paragraph position="4"> For example, havirlg a construction (&quot;sets A and BH or '@sets of and 13&quot;) and knowing that in 'mnozhesf va Am the link is appositiye, we can consider the link in *mnozhestva Be to also be appositive (1.e. &quot;sets A and B&quot;). (A ) Let us call constructions of the type f jest' N* and *oboznachim N cherez f* (I*~et us desiqnate N by f&quot;) introductory construotions. Every introductory construction ascr-ibes the meanizg to the formula which It introduces. null n In every construc,ti.on N f, for which m. lntroductory construction exists in the same text. the (A ) Sometimes the meaning is ascribed to a formu- null la without any introductory con.struction. The link in an occurrence r of a construction n oft eN fLf the expression f has not occurred in the text berore. r,. In this case the formu3.a f must also not occur before r as any coherent part (subformula) of some other formula F, Because t-Re meaning can be ascribed to a formula f by its place In F tsee above). Sut to use the distinction between a coherent and a non-coherent part of a formula (Cf. 'a + b' in '(a + b)/d8 and in 'ca +bd9), we need a calculus of' all mathematical symbolic notations, of which only small portions exist (Cf. arithmetic expresssionS 0f prograhming languages ) . Becaus-e of this Ah was formu1.ated in the above form).</Paragraph> <Paragraph position="5"> (A ) Sometimes there occur in mathematical texts expressions where verbal and symbolic parts are interwoven so that irl syntactic analysis a s-ymbolic insertion appears not as a single unit but as a complex construction having its own struct'ure. Some parts of tr formula can have links of their om wish the external verbal parts of the sentence. null</Paragraph> </Section> <Section position="2" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Examples : </SectionTitle> <Paragraph position="0"> (tmFunotion L H(I1)&quot; ).</Paragraph> <Paragraph position="1"> He re * is the predicate of .the sentence, 'Funstidn9 is iks subject and CH(R)* - an indi~ect object, The sentence Can be read as' 'F.unction L belon~s to H(EI)'* ( &quot;For every 1 6 B&quot;.. . ) Here is anmattribute *of 1 and can be read as 'belonging to'.</Paragraph> <Paragraph position="2"> 3. 'funktsija LE H(R) opredeljajetsjam..,. &quot;Function LO H(R) is defined by*... ) Here @ H(R) @ is an attribute of *Function9, and @t9 is an apposition modif-ying the same word. But the whole string L E H(R) can also be corlsidered an apposition modifying the word 'functioh' . So, we can formulate a rule: m If the link in some construction of the type N -A f Is apposiCt;'ve, then the link of the same N with the formula -f R-- f * . where R is one of the ~ymbols~,*,<,&~~,,3.~ a.c,C.f.% or 3. and f! is a (coherent) part of the formula f R f' Ls also appositive.</Paragraph> <Paragraph position="3"> The inverse also holds true.</Paragraph> <Paragraph position="4"> Using the axioms A1 to A5 cyclicly , we receive the 70% mentioned above.</Paragraph> </Section> <Section position="3" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Example : </SectionTitle> <Paragraph position="0"> Let us assume that the following Russi.an expresslions belong to the same mathematical text (and the preliminary syntactical analysis has already been done): (I) @Oboenachim etu tsepochku cherea A' (&quot;Let us designate thsiS string by A&quot; ) ; (2) 'tsepochki A i B = D@ (&quot;strings. A and B = p&quot;? Strlngs of A and B = D&quot;?); (3) tsepochki B 1 F' (&quot;strings B and Fw? &quot;Strings of 3 and F&quot;?); (4) tsepochka Fg (&quot;string Fg'? &quot;Stride; of Fw?)</Paragraph> </Section> <Section position="4" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Using axioms </SectionTitle> <Paragraph position="0"> A 39 A=, A29 kg* A19 Az and A1 we can ascribe the meaning &quot;appositive&quot; to the link in (4). Assuming that In a text expressions (2), 3 and (4) are present, and thar; the occurrence of @ in (4) 1s the f1rs.t; occurrence of this formula, we Can ascri3e to the A, A5 and A2 .</Paragraph> <Paragraph position="1"> So, we receive for expressio2s (3) to (4) translations: &quot;strings -A and B = DM; *stxings B and FH ; *string F* It is worthwhile to mention that the same formula may occur in a text being linked appositively to several d%ffe,rent noms, for example, and 'mno~oobrazi je it (l'manlfold a R* ) .</Paragraph> <Paragraph position="2"> -f Different N in expressions of the type N (with the same f) can r'ef'er to each other as genus and species or can name obdects for which the fact of their identlty has been proven in the text. Using tho axiom A4 we can (very rarely) make an error. An error can occur in a case where the f~rmula has the meaning specified once and for all independently from the text. So, wit,haut any previous definition of the meaning in an introductory construction or in a construction wftn tne apposftive link, a formula can at once be linked ~enit1vel.y to a noun. This situation is not typical in mathematical texts, Ih thls case we have a hieroqlyphio word (cf. '&', @$' in common Enqlish) and not a freely chosen notation. Such a word must be storea in the dictionary (wlth the epecific meaning ascribed to it). For example, 'dx/dyW is 'derivativeq.</Paragraph> <Paragraph position="3"> Using Cl (f ) in every case of occurrence of e-yery formula, authors of mathematical cexts would nake the above procedures unnecessary. The problem qf standardizing the lanquage of scientifFc publications is not new, and in many cases some format of texts is prescribed.</Paragraph> <Paragraph position="4"> The problem of choos-ing occurrences of qenitive links in constructlons of the type N- f from the set of all occurrences af such constructions in mathematical texts and, also, of choosinq the only semantically relevant governor for a formmla which has several formally equivalent ones Is considered in (1). The qeneral procedure for resolving ambiguities in surface syntactical arialysis using broad context Is proposed in (2 ).</Paragraph> </Section> </Section> <Section position="64" start_page="1976" end_page="1976" type="metho"> <SectionTitle> AFIPS CONSTITUENT SOCIETIES </SectionTitle> <Paragraph position="0"> Contents.</Paragraph> <Paragraph position="1"> The Arnedcan Federation of lnformafion Processing Societies acts on behalf of 15 national organizati~ns engaged in the &sigh and/or application of computers and information processing systems. These societies range in areas of interest from the highest degree of technology in softwate and hardware to accounting and education, and they represent a tdtal memberains of U&aaO,--- .-Inherent in the relationship between AFfPS and its Constituent Societies are the aammon goals of promoting understanding between societies and the genaral public.</Paragraph> <Paragraph position="2"> This brochure provides a short overview of each societ& its goals, membership requirements, aatlvitles and publications. Every individual society has more extensive informatlon available for you, should you be rnterested in learning further about its specific areas of Intarest and expertise.</Paragraph> <Paragraph position="3"> Page: 1. Introduction 5. American lnstitute of Aeronautics and Astronautics 6. American lnstitute of Certified Public Accountants, 7. American Society for Information Science 8. American Statisticai Assobiation 9. Association far Computational Linguisttcs lo. Association for Computing Machinew mon intefest in space, the atmosphere and the sea. They see in the exploitation of these elements an opportunity to expand and enrich human life in countless ways and have set themselves to the study of the physical characteristits of these elements and to the develo~ment of machinery that will bring them more fully to humanvy's service, AlAA s objective is to the advancement of the' profession and the individual ih these pursuits.</Paragraph> <Section position="1" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Membership Requirements </SectionTitle> <Paragraph position="0"> All persons engaged in the profesbional practice of {hearts, sciences or technology of aeronautics, astronautics or hydronautics are eligible for,memberskip in AIAA: Others whuae work cantributes to the advancement of these fields are also eligible.</Paragraph> </Section> <Section position="2" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Activities </SectionTitle> <Paragraph position="0"> Each year AlAA sponsors or co-sponsors from 25 to 30 national meetings in different parts of the country at which AlAA members have an opportunity to hear, present and d~scuss pape'rs of importance to the advancement of aerospace science and engineering Many of the meetings include aerospace exhibits and field trips to nearby aerospace plants and laboratories.</Paragraph> <Paragraph position="1"> The American Institute of Certified Public Accountants is the national professional association pf CPA s. The many activrties of the Institute are Uesigned to help member9 improve the quality of their professional services. the effectiveness with which they manage their practices. and their status as C;PA1s In the communities they serve. The Institute serves lo. unite the profession and to maintajn the staddards OF the CPA qualification aid the practice of accounting in the</Paragraph> </Section> <Section position="3" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> United States. Membership Requirements </SectionTitle> <Paragraph position="0"> An applicant must be a-certified Public Accountant and be enga8ed in work teiated to accounting. Inte~ national Aseociate memberships are also available.</Paragraph> <Paragraph position="1"> ActiGi ties The Institute r annual meetings and conferences are a~med at keeping professional Issues and problems before its members. The Institute prepares and grades ihe CPA examination used thrbughout the United States and supports the Financial Accounting Standards Board which sets accounting standards. Other activities are promulgatron of auditlng standards, creation and administration of continujhg education courses and programs, establishment ef rules of professional conduct and conducting rnves ligations in connection with alleged violabions, maintenance of 3 Washington: D.C. office for haison with the Internal Revenue Service. Securities and Exchange Cornmlssion, and other federal agencies, monitoring of stgte legislation pertaining to CPA's, operatlon of an on-I~ne, real-time computer-based ~nforrnation retrieval system,. and research into accounting, auditing and computer subject areas. &quot;</Paragraph> </Section> <Section position="4" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Publications </SectionTitle> <Paragraph position="0"> The Journal of Accountancy - Monthly The Tax Adviser - Monthly The CPA Letter - Semimonthly Special techntcql publ~cations,.books, and pam~hlets.</Paragraph> </Section> <Section position="5" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Dues </SectionTitle> <Paragraph position="0"> Established by the Councll, dues aje levied on a graduatedscale, according to longevity as a CPA, pos~tion in practice. and occupationaJ status.</Paragraph> <Paragraph position="1"> The AMerican Society for Information Science is a nonprofit national and professional association organized for scientific, literary. and educational purposes and dUicate4.to.the creation, organization. d~sseminalicn. and appncalion :of knowledge concerning information ana its Iransrer.:ASIS is dedicated to the imp~ovtment of the information-transfer process through research, development,. application gnd educatjon. The Society acts as a bridge between research and developq-ient and the requirements of d~vetse types of information systems. ASlS provldes a forum for the discussipn, publication and critical analysis of work deal~ng wUh the theory and practice of all elements involved in the cornmunicat~on of information.</Paragraph> </Section> <Section position="6" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Membership Requirements </SectionTitle> <Paragraph position="0"> Regular membership in ASIS-is open to any in-terested person who applies for membership and pays the prescribed dlles, No formal educational quallflcations for membership exist, Student memberships are available tor a period of'not more than three years to sersdns regylarly enrolled at a college or un~versity In one or mom-courses ~f training or study for whiCh degree credits are given in the fblds of documentation. library science or in'formation scienee. Institutufional memberships are.also available.</Paragraph> </Section> </Section> <Section position="65" start_page="1976" end_page="1976" type="metho"> <SectionTitle> Activities </SectionTitle> <Paragraph position="0"> The Annual meeting of the Soc~ety. usually held In October. provides a .focal point for the discussion oi formal papers and an Opportunity for informal talks with people of diverse ihterests. The Society alqo conducts a mid-year meeting. usually in May; part~cipates in programs of other professional societies: and, is an active partidpant in the National Computer Conference. AStS operates a member. placement service 'The Soc~ety has 24 major regianal. local chapters throughout the U.S., Canada and Europe.</Paragraph> <Paragraph position="1"> lives of the American Statistical Association, a nonsrofrt organi'ratlon, .shall be to fester. In the broadest nanner, stat~stics and its applications, to promote ~nity and eifectiveness o,f Bffort among all concerned with statistical problems, end to increase the contribution of stAtistics to human welfare.&quot; The Association is composed of persons interested. in statMics, applied or theoretical. Through the Association, members mutually help each other w~th the exchange of professional Knowleam and tho reporting of new developments,,insuring twit stat~stlcal techniques 'discovered in one field are. made kdown to workers in'others,</Paragraph> </Section> <Section position="66" start_page="1976" end_page="1976" type="metho"> <SectionTitle> Membership Requirement5 </SectionTitle> <Paragraph position="0"> There are several tyges of memberships available in ASA, including Regular and Institutignal. Contact ASA*Headsuart&s for further information on qualifications tor membels hip.</Paragraph> </Section> <Section position="67" start_page="1976" end_page="1976" type="metho"> <SectionTitle> Activities </SectionTitle> <Paragraph position="0"> The annual meeting of the Association 'IS h,eld in August, typically in conjunction with annual meetings of other scientific societies-qoncerned dtth statistical practice. In addition. Ibwl rnaetkgs ar0 held by the ASA chapters and regional meetings are arranged when desired within the geographical districts into which the Association 'is divided. Since these meetings are smaller'th'an the annual meetings, they further opportunrties for still more intimate discussion on statistical matters of local or regional interest.</Paragraph> <Paragraph position="1"> founded in 962 by am rod^ bf researchefs wno shared a corhmon rnterest inmroad class of problems ihvolving both languages ana computation. Their purposes were:l(l) to promote research and development activities in tfie field pf computational fin'guist~cs, (2) to plomote cooperation and information exchange among related profess~onal and technital societies; (3) to represent computational llnguistlca to foundations and government agenues and lo represent tho United States to similar organizations in other nations and in international organlzatlohs which rncludc Cbmputa,tional lingulstlcs as a proper concern.</Paragraph> <Section position="1" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Membership Reqyirements </SectionTitle> <Paragraph position="0"> Any person fino IS intetested in computational llr~ guistics-is invited to ioin the Assoclatlon.</Paragraph> <Paragraph position="1"> Antlvi ties The Associatlon m#e?s.snnually. usually with the Nertional Cornputq~ Qonf.etsnce and Exposition or with, the Lingu~qic Soqiety of lcrmerica at 'their summer meeting.</Paragraph> <Paragraph position="2"> The American Journal of Computational Linguist~cs, the primary journal of the Association, appears quarterly. The AJCL is piublished sn 4&quot; x 6': units, each an ind-ek card or a miqrofiche. For each article, blbl~ography, or survey. two unlts are supplied - an index card bearlng a summary and a microfiche'conta~ning the full text. Announcements and advertisements appear on index cards. A yearly index is prov~ded on tabbed Index cards.</Paragraph> <Paragraph position="3"> TO advance the sciences and arts of information proaessing includtng the study, designi development. construction, ana application of modern machinew. cbmputrng techn~ques and appropriate languages ror general information processing. storage, retrieval,and processing of data of all,kinds, and for the automatic. control and simulation or process.</Paragraph> <Paragraph position="4"> TO promote rrie rree interchange of informatlqn ahout the sciences and-arts of information prme~sing 1-0 develop and rnaiataln the lntegrlty ana competence of individuals engaged in the practlce 04 the sciences and arts: of information processing.</Paragraph> </Section> <Section position="2" start_page="1976" end_page="1976" type="sub_section"> <SectionTitle> Mernberlhtp Requirements </SectionTitle> <Paragraph position="0"> Meniher: Persons qualified to be members: a) supscri~v the purposes of ACM; b) have attained prafessi~nal statur,e as,denlonstrated by intellectual competence and ethjcal conduct In tlt'e hrts and sciences of ~nformation processing; c) have earnedla.8achelors Degr,ec or academlc equivalent, or have four years' fulr-time experience i~ rformatioq processing; d) are endarsed by two members of ACM and who attest to the above. Associate Member: Persons qualified to b.e associate memb-ers s'ubscrlbe to the purposes of the Associatlon. Student Members: \ndividuals registered in an accredited educational instltutwn full-t~me are qualified for student membershim</Paragraph> </Section> </Section> class="xml-element"></Paper>