American Journal of Computational Linguistics 
THE FINITE STRING > 
NEWSLETTER OF THE ASSOCIATION FOR COMPUTATIONAL LI NGU I ST1 CS 
VOLUME 12 - NUMBER 5 
With this mailing, AJCL completes the substantive 
part of its second year of publleatien. 
Index 
guides will be mafled to all 1575 subscribers 
early in 1976; the topical index will again be 
printed on the tabbed cards. 
This mailing con- 
tains a questionnaire about the format and scope 
of AJCL; responses will give the Association and 
Editorial Board guidance in poIicy development. 
AMERICAN JOURNAL OF COMPUTATIONAL LINGUISTICS is published by the 
Center for Applied Linguistics for the Association for Computa- 
tional Linguistics 
EDITOR: David G. Hays, Professor of Lingu~stlcs and of Computer a,Sc~cnce, 
State University of New York, Buffalo 
EDITORIAL ASSISTANT: willlam Bcnzon 
EDITORIAL ADDRESS * 
Twin ~illbws, Wanakah, New York 14075 
MANAGING EDITOR : A. ~ood Roberts, ~cputy Dl rector, Ccnter far 'dpplied, 
Linguistics 
ASSISTANT MANAGING EDITOR : 
Penny Plcket t 
?RI)I)UCTION AND SUBSCRIPTION ADDRESS 
1611 ~orth ~cnt Street, 
nrlington, ~lrginia 22209 
Copyright @ 1975 by the Associa tion Por Computa tlonal LJ ngurstlcs 
TABLE OF CONTENTS 
ASSOCIATION FOR COMPUTATIONAL LINGUISTICS 
Officers 1976 .............. 3 
Executive C~mmittee Meeting ................ 4 
Financial Report ................. 6 
PERSONALNOTES ....................... 7 
NSF: Reorganization ........... ...... 8 
MT: LatsecShowsiqZurich .............. 9 
MEE-TINGS 
X'IIth International Congress of Linguists ....... LO 
ACH Conference - Employment Register ......... 12 
Compcon 76 ...................... 13 
Cybernetics and . Systems Research ........... 14 
Literary and Linguistic Computing ........... 16 
Charles S Peirce Society ............... .17 
Computer Applications to Learning .......... 18 
Historical Linguistics .............. 19 
Short announcements and reminders ............ 20 
GRANTS 
ACLS: Soviet Studies, East European Languages ... 22 
Special Libraries Association Scholarships ...... 23 
......... .... ACKNOWLEDGMENT ....... , 24 
COMPUTER GENERATION OF SENTENCES BY SYSTEMIC GRAMMAR 
John Self .................... 25 
TNTERPRETATPON & INTEGRATION 'OF . SENTENCES INTO A Ce-NET 
Th R.Ho£mann .................... 46 
Americau Journal of Computational Linguist icr 
Microfiche 29 : 
OFFICERS 
President 
Vice-President 
Secretary-Treasurer 
Executive Committee 
Nominating Committee 
STANLEY R, PETRICK 
IBM Research 
JOSEPH E, GRIMES 
Cornell University 
A, HOOD ROBERTS 
Center for Applied Linguistics 
TIMOTHY C, DILLER 
Sperry-Univac 
ARAVIND K, JOSH] 
University of Pennsylvania 
Continuing members of the Executive Comittee are Charles J 
Rieger 111 (through 1976) and Bonnie Nash-Webber (through 1977 
Continuing members of the Nominating Comittee are Robert 
Barnes (1976) and William A Woods, Jr. (1977) The Editor is 
a member of the Executive Committee ex. offici-o. 
American Journal of Computational Linguistics Microfiche 29 : 4 
Present at the meeting in Boston, October 30, 1975, were 
President Joshi, Pice-President Petrick, Executive Committee 
Member Nash-Webber , Editor Hays, Secretary-Treasurer Roberts, 
Pagt Presidents Barnes and Woods, and Martin Kay, guest. 
Current membership is 
Exchange, gift, life 
Individual, paid through 1974 
Individual, paid through 1975 
Institution, paid through 1974 
Institution, paid through 1975 
Nash-Webber reported that about 600 copies of the Proceedings 
of the June Interdisciplinary Workshop have been sold; the text 
is used in courses at Berkeley (O'Malley) and Wisconsin (Klein) 
Edited discussions are in preparation. 
The Committee decided to hold the 1976 meeting of ACL at San 
Francisco in October, jointly with the ASIS meeting. Discussion. 
with Kay did not le'ad to generally satisfactory proposals for a 
combination of the ACL meeting with the International Conference 
at Ottawa. 
MORE 
Executive Commitbee Meeting 
The Cmittee decided to establish permanently the reduction of 
meeting registration fees for students, but not to introduce a 
reduced membership fee 
The format of the Journal was discussed; a questionnaire is to 
be distributed to members, calling for their opinions about 
several combinations of printed and microfiched publication. 
The Committee decided to offer a limited subsidy to the news- 
letter proposed by Hans KarPgren of Stockholm; the proposal is 
for 10 issues per year, free distribution, and a publication 
limited to announcements of significant events and other matters 
of general interest, with occasional short technical notes. 
American Journal of Computat ianal Linguir tics Microfiche 29 : 6 
FINANCIAL REPORT 
Balance as of July 26, 1974 
RECEIPTS 
Membership dues '74, '75 
Redistribution of AFIPS shrplus 
Sales of back issues, TFS 
ACL '75 meeting receipts to date 
DISBURSEMENTS 
Admjnistratiite costs, office 
supplies, PO mailing, AJCL costs 
not covered by Acc ' t 11317 
TFS (micro£ iched , old is sues bound) 
Membership, ACAL 
AFTPS dues 1974 
AFIPS dbes 1975 
Annual meeting costs, 1974 and 1975 
to date 
Paid out of ACL membership receipts 
into CAL accounr: 11317 for AJCL: from 
1974 receipts 7,968.88 
1975 receipts 5,000.00 
$17,581.77 
Balance as of October 30, 1975 
SAVINGS 
Certificates and accounts now worth 
A. Hood Roberts, Secretary-Treasurer 
American Journal of Computational Linguistics 
Microfiche 29 : 7 
PERSONAL MOTES 
DILLER, TIMOTHY C. TO Speech Comunications Group, Sperry- 
Univac, Univac Park, P. 0. Box 3525, St. Paul. Minnesota 
55101. From SDC. 
ROOCH. MELVIN R., MAJOR. USDAO, Box 2, American Embassy, APO 
San Francisco 96262. Chinese and Japanese characters 
Artificial intelligence. 
SAGER, NAUM~. NYU Linguistic String Proj ect , Warren Weaver 
Half, 251 Mercer Street, New York 10012. (The Project has 
moved to quarters in the building cne the Courant Institute 
of Mathematical Sciences. ) 
SALTON. GERARD. ASIS Award for best information science sbdok of 
1975: Dynamic Information and Librafy Processing. 
SOMERVILLE. CRAIG A. TO 4 Sherry Road, Troy, New York f2180. 
From Birmingham, MI. 
WILKS, YORICK. TO Department of Artificial Intelligence, 
University of Edinburgh, Scotland. From Fondazione Dalle 
Molle, Castagnola, Swi~zerland. 
American Journal of Computational Linguistics Microfiche 29 : 6 
NATIOHAL SCIENCE FOUf4DATION 
REORGAN kZATION 
RSF is reorganizing itself into six directorates; the new 
arrangements are intended to be complete by October 31, 1975: 
Division of Mathematical, Physical, and Engineering Scienceg 
Dr. Edward Creutz, Assistant Director 
Division of Astronomical, Earth, and Ocean Sciences 
Dr. Robert E. Hughes, Assistant Di~mector 
Division of Biological and Social Sciences 
Dr. Richard C. Atkinson, Acting Assistant Director 
Division of Science Education 
Division of Research Applications 
Dr. Alfred J. Eggers, Jr., Assistant Director 
Division of Scientific, Technological and International Affairs 
Dr. Robert E. Hughes, Assistant Director (acting) 
Division of Administration 
Mr. Eldon T. Taylor, Assistant Director 
Americaa Journal of Comp~tational Linguistics 
Mi crof i che ,29 : 9 
LATSEC SHOWS MT Ill ZURICH 
A forthcoming report in the ALLC Bulletin written by Herbert 
Bruderer describes a demonstration canducted by Peter P. Toma 
of LATSEC, Inc., in the Enstitut fur Infomatik of the Univer- 
sity of Zurich. The report describes the systb as having 
been proved practical by the American Air Force since 1970, 
and says that the Systran system was used for the Apollb-Soyuz 
flight. 
Mr. Bruderer writes that the text was furnished by the Qrenoble 
MT group; he describes the program as requiring a large computer. 
According to his report, a Russian text of 30,060 words was 
translated into English in 6 minutes on an IBM 370/155. 
American Journal of Computations Linguistics 
Microfiche 29 : 10 
XI ITH INTERNATIOt4AL 
CONGRESS OF LINGUISTS 
VIENNA 
AUGUST 29 - SEPTEMBER 2, 2977 
President of the Organizing Cornittee is Professor Wolfgang U. 
Dressler; Secretary General, Dr. Oskar E. Pfeiffer; Secretary, 
Eva M. Schaup. 
Plenary sessions: Basic problems of semantics 
Language and satiety 
Logically-based syntax versus autonomous 
generative syntax 
Word formation 
Diachrony: reconstruction 
History of linguistics: aims and methods 
Round-tables : Linguistics as an empirical science 
Language and music 
Sections : All plenary topics; and also 
Indo-European studies: wave theory 
Textlinguistics: dialogue 
Linguistics and the computer 
Contributions of dialectology to linguistic 
theory 
MORE 
XIIth International Congress of Linguists 
working groups : 
Language and liternur e 
Contrastive Linguistics 
Grarmnar and pragmatics of deixis 
Semiotics of human and animal language 
The interplay between diachronic and 
synchronic phonology 
Speech acts 
Speech errors slips of the tongue') 
Lexicography 
Other topics to be arranged 
Registration: Send name, address, inst-itution, name of, any 
accompanying person, etc with the fee of AS 750 to the 
Secretariat, Congress of Linguists, Postfach 35, A-1095 Wien, 
Austria. (The congress account is No. 65-28715 Creditanstalt- 
Bankverein Vienna, branch Schubertring.) After May 1976, the 
fee is AS 900; at the Congress, AS 1000. Students half price. 
Contributions: Send title and short summary of a 15-minute 
lecture (for a section) or short statement (for a working group 
The second circular will be sent in May 1976 to registered 
participants. 
heriim Journal of Computational Linguistics ~icrofiche 29 : 12 
EMPLOYMEijT REGISTER 
Applicants and positions available 411 be listed in books open 
for consultation by any conference participant. Staff will 
operate a message desk and maintain employer sign-up sheets at 
the conference to f acilicate making contacts. 
Applicants and employers must register in advance. The charge 
is $5 for applicants (free to students; $5 additional for an 
anonymous listing) and $20 for employers. Deadline for sub- 
mission is January 20, 1976. 
Forms can be obtained from 
Orrkn E. Taulbee 
Computer Science Employment Register 
Department of Computer Science 
University of Pittsburgh 
Pittsburgh, Pennsylvania 15260 
American Journal of Computational Linguictics 
Microfiche 29 : 13 
TWELFTH I EEE COMPUTER SOC I ETY INTERNATIONAL CONFERENCE 
MICROPROCESSOR TUTOR1,ALr FEBRUARY 23 
KEYNOTE SPEAKER: 
HISTORICAL SPEAKER: 
SESSION TITLES 
The personal computer: dream or reality? - What will be the 
social impact of computers? - Is 'distributed computer systems 
just a buzzword? 
- When will.design automation come of age? - 
What will microprocessor hardtrare evolve into? - What is the futm 
of language directed machines? - Where is the money coming from? 
- What will happen with computer networks? - What can we expect 
in data communicatians techniques? - What wil-1 win the solid 
state memory race? - What's going on in the rest of the world? 
Computer science: is it related to computing? - What can we 
expect in micr~processor software? - Will software engineering 
get us good software? - What can we expect in hardware design 
techniques? 
SHORT CONTRIBUTIONSI DEADLINE J,ANUARY 1, 1976 
5-minute presentations of recent work or experience can still be 
offered. A 200 to 300 word summary, with name and affiliation, 
goes to Martin Graham, Department of Electrical Engineering and 
Computer Science, University of California, Berkeley 94720. 
FOR A PROGRAM, WRITE Tor SIDNEY FERNBACH, COMPUTER DEPARTMENT 
L-61, LAWRENCE LIVERMORE LABORATORY, PO BOX 808, LIVERMORE 94550. 
American Journal of Computational. Linguistics 
Microfiche 29 : 14 
THIRD EUROPEAN MEETING ON 
CYBERNETICS A-ND SYSTEMS RESEARCH 
LIST OF SYMPOSIA ON NEXT FRAME 
CONTRIBUTIONS 
Deadline: January 1, 1976 
Form: Abstract - one A4 page, including full title of 
the paper, author's name and aff lliaticm 
Full paper - 3 to 4,000 words on A4 paper 
Three copies of abstract and paper required 
FORMAT NOTES 
Full text double-spaced; abstract may be single-spaced. Typled 
symbols only; if handwritten symbols must be used, author is to 
specify precise shape and meaning on a separate sheet 
Photo- 
graphs to be on hard glossy paper; at least one copy of any 
drawing, map, or diagram must be in India ink on hard paper. 
(The announcement says that abstracts of papers should be re- 
ceived by January 1, but speaks of using full text for referee- 
ing; AJCL cannot resolve the conflict .) 
Advanco, notice of 
films or slides, noting size and quantity, is needed. 
ADDRESS 
Osterreichische Studiengesellschaft fur Kybernetik 
Schottengasse 3 
A-1010 Wien 1. Austria 
3rd European Meeting on Cybernetics 
SYMPOSIA 
General systems methodology 
G. Klir, USA 
Biocybernetics and theoretical neurobiology 
L. Ricciardl, Italy 
Cybernetics of cognition and learning 
G. Pask, UK 
Sttucture and dynamics of socio-economic systems 
K. A. Hammeed, UK 
Health-care systems 
J. Milsum, Canada 
Cybernetics in organization and management 
F. deP. Hanika, UK 
Engineering systems methodology 
F. Pichler, Austria 
Computer simulation methods and languages 
G. Chroust, Austria, and J. P. C. K1ei jnen, Netherlands 
Computer linguistics 
W. Dressier, Austria 
Cbmputer performance control and evaluation 
N. Rozsenich and L. Heinrich, Austria 
Fuzzy mathematics and fuzzy systams 
H. - 2. Zlmmermann , BRD 
FEE 
AS 1500, participants; AS 750, contributors 
American Journal of Computational Linguistics Microfiche 29 : 16 
ASSOCIATION FOR LITERARY AND LI NGUI STI c COMPUTING 
INTERNATIONAL MEETING : ANNUAL GENERAL MEETING 
DECEMBER 13, 1975 - INIVERSITY OF AMSTERDAM 
PROGRAM 
COMPUTER CONTROLLED SAMPLING FOR BILANGUAGE DICTIONARY COMPILATION 
R. D. Bathurst (United Ki~gdom) 
LES-ACTIVTVES DU LEXIQUE INTELLECTUEL EUROPEEN--TRAITEMENT 
ELECTROMI-QlJE DES TEXTES ID AUTEURS 
T Gregory (Italy) 
ON LEXICOGRAPHICAL COMPUTING--SOME ASPECTS OF THE WORK FOR A 
MEXICAN SPANISH DICTIONAXY 
M. Alinei (Netherlands) 
THE RECOGNITION OF FINITE VERBS IN FRENCH TEXTS 
J. S. PetBfi (West Germany) 
ADDRESS 
A. van Wijngaarden (Netherlands) 
Inquiries: Mrs. J. M. Smith, 6 Sevenoaks Avenue, Heaton Moor, 
Stockport, Cheshire SK4 4AW, England. 
American Journal of Compatational Linguir tics 
Microfiche 29 17 
INTERNATIONAL CONGRESS OF THE 
CHARLES S, PEIRCE SOCIETY 
PRAGMATISM AND SEMIOTIC 
About eight invited papers in plenary sessions and 
volunteered papers in concurrent sessions. 
Submissions and requests for information can be addressed to 
Professor Carolyn Eisele 
215 East 68th Strebt, Apt. 27E 
New York, New York 10021 
American Journal of Computational Linguistics 
Microfiche 29 : 18 
NATIONAL ASSOCIATION OE USERS OF 
COMPUTER APPLICATIONS TO LEARNING 
1975 CONFERENCE 
HOLIDAY INN, OTTAWA CENTRE 
OTTAWA, ONTARIO, CANADA 
NOVEMBER 20-22, 1975 
PROGRAM TOPICS 
CURRENT EVENTS IN CAI 
TE~ACHING A CAI AUTHPR LANGUAGE 
CAL DELIVERY SYSTEM 
DISTRIBUTED APPROACH TO CAL 
GUIDANCE INFORMATION RETRIEVAL 
TUTORIAL TECHNIQUES 
USING PLATO 
IMPACT OF EOUCATIONAL COMPUTING 
SYSTEMS APPROACH TO INSTRUCTION 
COMPUTER TECHNOLOGY FOR ELEMENTARY SCHOOLS 
COMPUTER BASED TESTING 
ADDRESS 
NAUCAL Registration Chairman 
Algonquin College 
1385 Woodroffe Avenue 
Ottawa, Ontario X2G 1V8 
Cbnada 
American Journal of Computational Linguistics 
Microfiche 29 : 19 
SECOND INTERNATIONAL CONFERENCE UN 
HISTORICAL LINGUISTICS 
Registration deadline: December 26, 1975 
Fees : $112 private room 
$ 77 twin bed 
$ 73 each of two in double bed 
Fee include room, ground transport, and conference activities. 
Address : William M. Christie 
Department of English 
University of Arizona 
Tucson 85721 
American Journal of Computational binguirtics Microfiche 29 : 2 
MEETING BRIEFS 
SHORT ANNOUNCEMENTS AND REMI NDERS 
COLING 76 
The 1976 International Conference on Computational Linguistics 
will be held at Ottawa, June 28 - July 2, 1976. Details on 
Microfiche 17; registration form on Card 63. Deadline for 
abstracts of contributions : December 1. 
National Computer Conference, New York, June 7-10, 1976. Call 
for papers on Card 62; deadline January 5. Registration $60; 
write to AFIPS, 210 Summit Avenue, Montvale, New Jersey 07645. 
LINGUISTIC SOCIETY OF AMERICA 
Hyatt Regency, San Francisco, December 27-30, 1975. A session 
on computational linguistics is in preparation. 
IIIFORMATION THEORY 
1976 International Symposium, Ronneby, Sweden, June 21-24, 1976 
Information from Jack Salz, Bell Laboratories, Room 1G-509, 
Holmdel, New Jersey 07733. 
DIGITAL COnMUNICATION 
1976 Zurich Seminar, March 9-11. Information from A. Kundig, 
Technisches Zentrum PTTlV907, CH-3000 Bern 29, Switzerlan6. 
MORE 
MEETING BRIEFS 
APL 76 
Fundamental issues and techniques in practical use pf the APL 
programming language. Ottawa, September 22-24, 1976. Informa- 
tion from C. A. Wogrin, University Computing Center, University 
of Massachusetts, Amherst 01002. 
PROGRAMFIING SRALL PROCESSORS 
SZGMINI and SIGPLAN. Delta Towers Hotel, New Orleans, March 
4-6, 1976. Information from Lamence J. Schutte, Room 6B-302, 
Bell Telephone Laboratories, Naperville, Illinois 60540. 
PATTERN RECOG?~ITIO~~ 
Third International Joint Conference, Coronado, California, 
November 8-11, 1976. Information from Allen Klinger, UCLA, 
3531-C Boelter Hall, Los Angeles 90024. Deadline for submis- 
sion: lurch 1, 1976. 
Information--Dilemmas, Decisions, Directions. Theme sessions 
on such topics as changing patterns of primary sources, document 
delivery, user education, indexing systems and current research 
related to abstracting and indexing-. Program Chairman is John 
E. Creps, Jr., Engineering Index, Inc. Christopher Inn, Columbus 
Ohio, liarch 9-10, 1976. 
Binghamton, New York, June 14-16, 1976. Papers invited on actual 
experience: concrete results, specific materials, problems, pro- 
grams, md measures of success or accomplishment. Deadline for 
full papers: Janleary 15, 1976. Information from CCUCI7, Compu- 
ter Center, SUNY, Binghamton 13901. 
American Journal of Computational Linguistics 
Microfiche 29 : 22 
SOV STUD 
Grants will be offered subject to refunding, for research in the 
social sciences and humanities relating to Revolutionary Russia 
and the USSR; sponsored jointly by the ACLS and the Social Science 
Research Council. Emphasis is placed on interdisciplinary studies 
and on applications which bring to Soviet studies insights of 
sociology, social psychology, cultural anthropology, economics, 
law, and geography. Only in exceptional cases will grants be 
made in support of travel for brief visits abroad or. to relieve 
scholars of the necessity of teaching beyond the conventional 
academic year. Grants will rarely exceed $8,500. The Ph:D. or 
its equivalent is required. /~eadli~e : December 31, 1975. 
EAST EUROPEAII LANSUAGES 
Grants will be offered subject to refunding, for schobars and 
graduate students (who have completed at least one year of gradu- 
ate study at the program deadline) for intensive, intermediate 
level atudy abtoad of the languages of Albania, Bulgaria, Czecho- 
slovakia, Greece, Hungary, Poland, Romania, and Yugoslavia; 'spon- 
sored jointly by ACLS and SSRC. Grants are also offered for en- 
rollment in language courses on all levels in the US if such 
courses are not available in the regular program of the home 
institution. $300 to $1000. Deadline: February 1, 1976. 
American Journal of Computational Linguir tics 
SPECIAL LIBRARIES ASSOCIATION 
S-CHOLARSHIP PROGRAM 
1976177 
Three $2,500 scholarships will be awarded by Special Libraries Association for the aca- 
demic year 1976/77 The awards, to be granted in May 1976, are for graduate study 
lead~ng to a master's degree at a recognized school of library or information science in 
the United States or Canada Preference will be given to those applicants interested in 
pursuing a career in special librarianship. Awards are made withoutregard to race, sex, 
age, religion, or ethnic background. 
Special Libraries provide research arid information services to business, industry and 
government Special Librarians are men and women trained in the theory and practice of 
librwy or information science as well as in the fundamentals of a particular subject field. 
Specialists are needed in many organizations, among which are research institutes, news- 
papers, insurance companies, banks, law f~rrns, hospitals, and gwernmental agencies. 
Subject specializations may include the social sciences, economics, the fine arts, engineer- 
ing, and the physical and biological sciences. 
ELJGIBILITY: College graduates or college seniors with an interest in spec~al 
librarianship. Work experience in a special library is helpfltl 
Citizens of the United States or Canada 
QUALl FImTIONS: 
Definite interest and aptitude for spkial library work 
Good academic record 
Financial need. 
APPLICATIONS: May be requested by writ~nq to 
Special Libraries Association 
Scholarship Committee 
235 Park Avenue South 
New York, N. Y. 10003 
Applications must be completbd and returned by 
January 15, 1976 
America Journal of Computational Linguirtics 
Microfiche 29 : 24 
The following text was inadvertently omitted from the revised 
manuscript submitted by Allen Klinger and published in AJCL, 
Microfiche 21, 2- 25 : 
This research was sponsored by the Air Force Off ice of 
Scientific Research, Air Force Systems Command, USAF, under 
Grant No, AFOSR-72-2384. The United States Government is 
authorized to reproduce and distribute repririts for Govermental 
purposes notwithstanding any copyright notation hereon. 
The author would like to express his appreciation for this 
support . 
In addition, the author is also grateful for the partial 
support of Defenre Advanced Research Projects Agency. 
Work 
presented here began under Contract No.. DAHCl5-73-C-0181. 
American Journal of Compatational Linguir ties 
Microfiche 29 : 25 
COMPUTER GENERATION OF SENTENCES 
BY SYSTEMIC GRAMMAR 
Department of Information Science- 
University ef Melbourne 
ParkviZle, Victoria, Australia 3052 
ABSTRACT 
The paper describes a colnnuter mdel oC svstemic grammar. 
a penerative prammar for natural lanpuppe. P propram is 
explained which piven the features OF an fterr, determines the 
structure of that item accordinp to a svsteric Eramar s~ecifiec 
as 8ata. The vrograr thus deaonstrates the ~rinciples of 
systemic grammar, a brief sursarv of the vechanicg of which is 
also included. 
Some imvlications of the proarm for systemic 
grammar itself are discussed. In ~articular, jt is shown that 
~revious definitions of the oneration of qtructure-huildfnp 
rules require modification. 
1. ?ntroduction 
This paver describes a computer model of s~stsmic grammar, a 
prammar for natural languages developed by Halliday and 
colleagues at Universitv College, London (Halliday , 1961, 197fl). 
Svstemic grammar has recentlv been of interest to comoutational 
grammarians, arimarily as a result of the imn~essive work .of 
Winograd. ( 1972 ) , who develoaed a natural lanpua~e understandinp 
system one component of which was stronglv influenced bv the 
~rinciples of systemic grammar. More recently, Power (1974) has 
also investigated how svstemic pramnar can be used to ,analvse 
natural language. There have, however, been no attempts to use a 
computer to investi.gate systemic grammar itself. As Friedman 
(1971) says, in introducing her computer model of 
transformational grammar, adequate natural lanquage grammars are 
bound to be so complex that some mechanical aid in investigating 
their ~ro~erties will be mandatory. 
The aims, then, of develo~ing a computer model of svsteric 
prammar are threefold. First, the model enables the Rrammar to 
be tested, i.e. it enables contradictions, amhip,uities and 
incomnletenesses in the grammas to be found. Pecondlv, the model 
enables systemic grammar itself to he imnroved, slnce the 
consequences of adfustinp parameters and rules can be more easilv 
followed. And, thirdly, the model serves as a demonstration of 
how systemic grammar 'works ' 
Earlier descri~tions of systemic grammar were somewhat 
ihcomplete, but that of Hudson (1971) seems sufficientlv ~recise 
to encourage the feeling that a computer program could be based 
upon it. The progpam described below generates (in the 
linguistic sense) natural language sentences, i .e. "assigns 
structural daacri~tions to sentences" (Chomskv, 1965 1. It is 
not concerned directly with understanding or producing sentences : 
2. The Mechanics of Svstemic Grammap 
This section brieflu describes the ~enerative apnaratus of 
systemic grammar - for a fuller discussion, and for linguistic 
justifications of the procgsses, the reader is referred to 
Hudson (1971), from which tfrc? example gripmrnar and generations 
given later are taken. 
In systemic grgmmar, 'lstructures are entirely medictable 
from features: given a11 of an item's features, we can predict 
exactly what its structure. will beff (Hudson. pg 87). In 
general t'erms, an jtemTs fearures or classes are those 
categories to which it belongs irres~ective of the  articular 
sentknce to which the item belongs; an item's functions are 
- 
those categories to which it belongs as a result of its role in 
a sentence. For example, 
"mustN has the features MODAL-VERB , FINITE-VERB (among others 
and in the sentence 
"Must it grow darker?" has the functions !PRE-SUBJECT, 
! MOOD-FOCUS (among others 
(A preceding ! will be used to distinguish functions from 
features.) An item's structure is defined by its immediate 
constituents ' functions and the sequence in which they occur, 
~iven all the features that an item has, tKe item's 
structure may be determined, according to systemic hnammar, by 
the sequential application of rules of four kinds : 
(1) feature-realisation rules 
In the simplest case, these rules are of the form "if item 
has feature x then its structure will contain function yv - y is 
said to be the realisation pf x. Some rules ape conditional in 
that the realis~tion only holds if certain other features are or 
are not present. Also, some rules specify that tktb functions 
must be conflated, i.e. bath functions apply to the same immediate 
constituent. (Further details of these an? the following rules 
ape given later when the program is discussed.) The application 
of the feature-realisation rules provides an unordered sot of 
functions, some of which may be conflated. 
(2) structure-buildinp rules 
These rules expand and arder th?a set of fwnctions to provide 
the structure of the item. Structure-building rules are themselves 
of four kinds, which in the simplest case are of the following form: 
(-a) addition rule3: "if function y (or some combination of 
functions) is present, then so must be function z (oassibly 
conflated with other Functions 1". 
(b) conflation rules: "if some condition expressed in terms 
- ~ 
of functions is satisfied then some function must be conflated 
with certain other functions". 
(c) sequence rules: "if two functicns y and z are  resent 
then y must be canflated with, orecede or not follow z" . 
(dl compatibility rules: "functiops v and z must not be 
conf lated" . 
Addition and conflation rules are onlv applicable if the resultant 
structure does not conflict with a sequence or comatibility rule. 
Structure-building rules are not extrinsically ordered in any wav. 
After applying these rules, we have a complete. specification of 
the item's atructum , in that we have specified function- 
"bundlesw, each of which consists of the functions of one sf ?he 
immediate constituents of the item. 
( 3 function-realisation rules 
These rules specify which features are implied bv an item's 
functions. They are of the form "if a structure contains 
function y the corresponding item must have feature xl'. 
Applied 
to the function-bundles obtained from (21, these rules help to 
determine the features possessed by the immediate constituents. 
( 4) systems 
System networks specify which features are implied by other 
features. These networks are equivalent to rules of the form 
"if feature x is present then so is one (or all) of a set of 
features, and conversely". These rules expand a set of feature$ 
(possibly the result of applying (3) , not necessarily into a 
complete set, since some features mav be freely selected. "he 
feature-realisation rules may then be recursivelv applied to 
this ser of featuren, if reQuF~ed+ 
3, The Promam 
The program reads in a definition of a systeric gramwar 
(provided as data so that it may be changed without necessitating 
major modifications to the program) and generates a structure 
from a specified list of featu~es. The interested reader should 
have no difficulty in relating the rules given below to thb 
grammar given by Hudson (p~. 53-101). The rules are shown in the 
form in which thev are presented to the propam, and are numbered 
to ease explanation and understanding of the program's execution. 
In order to enable the reader to follow the computer generations 
given later, an English interpretation of selected rules follm~ : 
(1) feature-realisation rules 
Rule 1'3 (below) means "if an item has the feature 
INTERROGATIVE then, provided it also is DEPENDENT, its structure 
contains the function !QUESTION (which is thereby introduced if 
not already present) conflated with !BINDER1'. 
Similarly, rule 32 
NO FEATURE REAL1 SATION CONDITIO~~ 
1 (CLAUSE 
2 (PHRASE 
3 (WORD 
4 ( 1 NDEPENDENT 
5 (DEPENDENT 
6 (DEPENDENT 
7 (DEPENDENT 
8 (IMPERATIVE 
9 (INDICATIVE 
10 (INDICATIVE 
11 (DECLARATIVE 
12 INTERROGATIVE 
13 ( INTERROGATIVE 
14 (POLAR 
15 (NO??-POLAR 
16 (W)I 
17 (ALTEIUVATIVE 
18 (SUBJECT-FOCUS 
19 (SUBJECT-FOCUS 
20 [MODAL 
21 (NON-SUM"ECT-FOCUS 
22 (NON4'mDAL 
2 3 (INTRANSITIVE 
24 CTMSXTIVE 
25 (TRANSITIVE 
(TRANSITIVE 
ATTRIBWT. VFi 
(ATTRIBUTIVE 
( ATTRIBUTL VE 
(NO#-ATTRIBUTIVE 
(NON-ATTRIBUTIVE 
I[ ACTIVE 
tPAsSIVE 
C"PSSIVE 
(ACTOR-SPECIf LED 
( ACTOR-UNSPECIFIED 
!PROCESS ) 
!HEAD) 
!STEM) 
-1 
!BINDER) 
!SUBJECT) 
! FINITE 
-) 
!SUBJECT) 
! PINTTE) 
-1 
! MOOD-FOCUS TNDEPENDENT) 
(+ !QUEG'PION = !BINDER) 
DEPENDENT 
-1 
-1 
(+ !QUESTION = !MOOD-FOCUS ) 
INDEPENDENT) 
?ALTERNATIVE 
(!OUESTION = !SUBJECT) WH) 
( !ALTERNATIVE = ! SUBJECT) 
ALTERNATIVE: 
! MODAL 
-1 
-1 
-1 
!GOAL) 
?ACTOR (NOT ACTOR- 
UNSPECIFIED) ) 
!TRANSITIVE) 
+ ! ATTRI BUANT = !SUBJECT) 
! ATTRIBUTE) 
! COPULAR) 
(+ TACTOR = !SUBJECT)) 
! INTRANS ) 
(!ACTOR = !SUBJECT)) 
(!GOAL = !SUBJECT)) 
?PASSI~FL) 
(+ !AGENT = !ACTOR)) 
-1 
may conflate !ACTOR and !SUBJECT, but only if both are already 
present. (There are ~omplications , explained later, when the 
funeion to be canflated with, e .g. !BINDER and !SUBJECT above, 
is abeent. ) 
( 2 structu+-bui lding mles 
(a) Addition rule 2 means "?MODAL and !PASSIVE, if present, 
must be conf lated with ! POST-SUBJECT , added if necessary". 
Rule 1 means "if !MOOD-FOCUS is present but not conflated with 
!SUBJECT then !PRE-SUBJECT must be added if not already prc .~t" 
NO ADDITION RULE CONDITION 
(b) Conflation rule 1 means "if !MOOD-FOCUS is not 
conflated with !OUESTION then !PRE-SUBJECT and !MOOD-FOCUS 
must be conflated, if present". 
NO CONFLATION RULE CONDITION 
1 ( ( IPRE-SUBJECT = ? MOOD-FOCUS 1 
(?MOOD-FOCUS # !QUESTION)) 
2 ((!PROCESS = !COPULAR !TRANSITIVE !INTRANS)) 
(c) Sequence rule 1 means "whichever of !MOOD-FOCUS or 
!BINDER is present, if either, will precede or be canflated 
uith the first of !PRE-SUBJECT and !SUBJECT, if present, which, 
if both are present, will be in the specified orde~, and 
!POST-SUBJECP, if present, will follow the last of these functions, 
if any, and ?PROCESS, if present, will follow or be conflated with 
the last of these functions if any, and !POST-VERB, if present, 
will follow the last of these functions, if any". 
NO SEQUENCE RULE 
1 (.(!HOOPFOCUS OR !BINDER) 
=> ( IPRE-SUBJECT -> !SUBJECT) 
-> !POST-SUBJECT 
=, !PROCESS 
- > !POST-VERB) 
2 ( !FINITE 
= ( ! PRE-SUBJECT 
= (!MODAL -> !PASSIVE) 
-> !PROCESS)) 
(dl Compatibility rule 1 means " !POST-SUBJECT must not 
be conflated with ! PRE-SUBJECT" 
NO COMPATIBILITY RULE: 
( 3 1 function-realisation - rules 
Rule 12 meas "if an item has none of the functions 
!SUBJECT, !GOAL, !ATTRIBUTE or !AGENT then if it has !BINDER, 
it has the feature CONJUNCTIONf' 
1 (!COPULAR 
2 (!EN 
3 ( !FINITE 
4 ( !INTRANS 
5 ( !MODAL 
6 (!PASSIVE 
7 (!PROCESS 
8 ( !TRANSITIVE 
9 ( !AGENT 
10 (!ALTERNATIVE 
11 ( ! ATTRIBUTIVE 
12 ( !BINDER 
CONDITION 
COPULAR-VERB ) 
BIN-FORM) 
FINITE-VERB) 
INTRANSITIVE-VERP) 
MODAL-VERB) 
BE ) 
EXICAL-VERB 1 
TRANSITIVE-'VERB 
PREPOSITIONAL) 
DISJUNCTIVE) 
(OR ADJECTIVAL NOMINAL PREPOSITIONAL) 
CON JUNCTION (NOT (OR 
! SUBJECT ! GOAL 
! ATTRIBUTE !AGENT) 1 ) 
( OR NOMINAL DEPENDENT 1 1 
QUESTIONING) 
( OR NOMINAL DEPENDENT ) ) 
* 
Rule 10 means "an item with feature INTMISITIVE also has 
one of the features ATTRIBUTIVE and NO?!-ATTRIBUTIVE , and also 
has the features naming its supersystems, i.e. 9, 23 and 1, 
i.e. CLAUSE and ITEM". 
The *OR in the subsystems column indicates 
NO NAME (IF ANY) SUPERSYSTEM SUBSYSTEMS 
1 (ITEM I (*OR 23 24 15)) 
23 (CLAUSE 1 c a2 91, 
2 (- 2 3 h l? OR 3 25)) 
3 ( INDEPENDENT 2 (OR IMPERATIVE 26)) 
2 5 ( DEPENDENT 2 27) 
26 (INDICATIVE 3 27) 
27 (- (OR 25 26) (AND 4 8) 
4 (- 
2 7 (OR DECLARATIVE 5)) 
5 ( INTERROGATI VE 4 (OR POLAR 2811 
2 8 (NON-POLAR 5 (AND 6 7)) 
6 (- 
2 8 (OR WH 4LTERNATIVE)) 
7 (0 
2 8 (OR SUBJECT-FOCUS 
NON-SUBJECT-FOCUS ) ) 
8 (- 2 7- (OR MODAL NON-MODAL) ) 
9 (- 23 (OR 10 11)) 
10 (INTRANSITIVE 9 (OR ATTRIBUTIVE NOH-ATTRIBUTIVE ) ) 
11 (TRANSITIVE 9 (OR ACTIVE 12)) 
12 (PASSIVE 11 (OR ACTOR-SPECIFIED 
ACTOR-UNSPECIFIED) 
24 (PHRASE I (AND 13 14)) 
13 (- 24 ( OR NOMINAL ADJECTIVAL 
ADVERBIAL PREPOSIT13NAL) 
14 (- 24 ( *OR NON-QUESTIONING 
QUESTIONING) ) 
15 (WORD 1 (OR 29 CONJUNCTEON 1 
29 (VERB 15 (AND 16 19)) 
16 (- 29 (*OR 17 18)) 
17 (NON-FINITE-VERB 16 
(*OR FORM-0 EN-FORM ING-FORM)) 
18 (FINITE-WRB 16 
(OR PAST-VERB PRESENT-VERB) ) 
19 (- 29 
(*OR 20 22)) 
20 ~GRAHMATI@AL-VERB 19 (*OR 21 MODAL-VERB)) 
21 (NON-MODAL-VERB 20 (&OR DO BE HAVE)) 
22 (LEXICAL-VERB 19 ( OR COPULAR-VERB 
INTRANSITIVE-VERB 
TRANSITIVE-VERB) ) 
the 8ystem are so numbered ta correspond with Hudson's 
labellings Cpg. 71). 
36 
that the first name& feature or rule is the "defaultv oation, 
taken unless there are environmental reasons for selecting 
another, 
The rules of the grammar are in fact input and stored in 
the form of McCarthy lists (McCarthy , 1965) , and the program 
is written in a list-processing extension of BCPL (Self, 1975). 
It is important to realise that the Tules are not 
extrinsically ordered in any way, and that the program may 
(concepMly) execute the rules in any order, with the 
objective of finding a structure consistent with all rules. 
Hence, rules are executed recursively, with backtracking when 
inconsistenices become apparent. 
The generation of the structure of a sentence with the 
features CLAUSE, INDEPENDENT, INDICATIVE, INTERROGATIVE, 
NON-POLAR, WH , SUBJECT-FOCUS , NON-MODAL, TRANSITIVE, PASSIVE 
and ACTOR-UNSPECIFIED, e . g . "Which of the tents were errected?" 
(Audson , pg. 100 1 ,@ is shown below. Each piece of outpw is 
preceded by an indication of the rhle that has been executed, 
e . g. FR 1 indicates the first feature-realisation rule. Tn 
the printout of structures, 
indicates that A, B , . . (which may be functions or structures 
are conflated. Similarly, 
indicate, respectively, that A precedes B , that A precedes or 
is conflated with B, and that the order of A and B is undetermined. 
When A, B, . . 
are all functions, then these appear as, e. g. 
(= A B **I, 
(GENERATE (CLAUSE INDEPENDENT INDICATIVE INTERROGATIVE 
NON-POLAR WH SUBJECT-FOCUS 
NON-MODAL TRANSITIVE PASSIVE ACTOR-UNSPECIFIED)) 
FR 1 (!PROW 
FR 9 ( !SUBJECT !PROCESS) 
FR 10 (!FINITE ?SUBJECT !PROCESS) 
FR 12 (!MOOD-FOCUS !FINITE !SUBJECT !PROCESS) 
FR 16 (!FINITE !SUBJECT !PROCESS (= IOUESTION !MOOD-FOCUS)) 
FR 18 ( ! FINITE !PROCESS (= !OUESTION !MOOD-FOCUS !SUBJECT) ) 
FR 24 (!GOAL !FINITE !PROCESS (= !QUESTION !MOOD-FOCUS 
!SUBJECT) ) 
FR 26 ( !TRANSITIVE !GOAL !FINITE !PROCESS = !QUESTION 
!MOOD-FOCUS !SUBJECT)) 
FR 33 (!TRANSITIVE !FINITE !PROCESS (= ?GOAL ?QUESTION 
!MOOD-FOCUS !SUBJECT)) 
FR 34 (!PASSIVE !TRANSITIVE !FINITE !PROCESS (= !GOAL 
!QUESTION !MOOD-FOCUS ! SUBJECT) ) 
SQ 1 
3 
!PASSIVE 
!TRANSITIVE 
! FIN1 TE 
e > 
(= !GOAL !QUESTION ! MOOD-FOCUS ! SUBJECT) 
!PROCESS 
- > 
3 
(= !FINITE IPASSIVE) 
( !GOAL !QUESTION !MOOD-FOCUS !SUBJECT) 
!PROCESS 
ADDN 2 
3 
!TRANSITIVE 
-> 
? 
( !POST-SUBJECT !FINITE !PASSIVE) 
(= !GOAL !OUESTION ! MOOD-FOCUS !SUBJECT) 
!PROCESS 
( f GOAL !QUESTION !MOOD-FOCUS !SUBJECT) 
(: !POST-SUBJECT ! FINITE !PASSIVE) 
ADDN 4 
? 
!TRANSITIVE 
-3 
(= !GOAL !QUESTION !MOOD-FOCUS ?SUBJECT) 
(= !POST-SUBJECT !FINITE !PASSIVE) 
(= !EN !PROCESS) 
CQNF 2 
3 
-> 
( !GOAL I QUESTION !MOOD-FOCUS I SUBJECT) 
( !POST-SUBJECT !FINITE IPASSIVE) 
(= !TRANSITIVE !EN ! PROCESS 
ADDN 3 
? 
-> 
( = !POST-VERB ?GOAL !QUESTION !MOOD-FOCUS !SUBJECT) 
(= !POST-SUBJFCT !FINITE IPASSIVE) 
( = ?TRANSITIVE !EN !PROCESS 
COW 2 
? 
- > 
( = ? GOAL f QUESTION !MOOD-FOCUS 1 SUBJECT) 
(= 1POST-SUBJECT ! FINITE IPASSIVE) 
= !TRANSITIVE IEN !PROCESS9 
( ! GOAL !QIESTION ! MOOD-FOCUS ! SUBJECT) 
FMR 13 ( (OR NOMINA& DEPENDENT) 1 
flJR 14 (QUESTIONING (OR NOMINAL DEPENDENT) 1 
SY 14 (PHRASE QUESTIONING ( OR NOMINAL DEPENDENT ) 1 
SY 0 (PHRASE QUESTIONING NOMINAL) 
FNR 3 (FINITE-VERB) 
FNR 6 [BE FINITE-VERB) 
SY 21 (WORD VERB GRAHMA'SICAL-VERB NON-MODAL-VERB BE 
FINITE-VERB ), 
FNR 8 (TRANSITIVE-VERB) 
FNR 2 (EN-FORM TRANSITIVE-VERB) 
FNR 7 (LEXICAL-VERB EN-FORM TRANSITIVE-VERB) 
SY 22 (WORD VERB LEXICAL-VERB EN-FORM TRANSITIVE-VERB I 
SY 17 (NON-FINITE-VERB WORD VERB LEXICAL-VERB EN-FORM 
TRANSITIVE-VERB) 
Thus, the structure generated is 
"Which of the tents were erected?" 
! GOAL !POST-SUBJECT !TRANSITIVE 
!QUESTION ! FINITE ! EN 
! IjOOD- FOCUS !PASSIVE ! PROCES~ 
! SU3JECT 
"Which of the tents were rn @ erected?' 
The progrdm may then generate the features .of the immediate 
constituents , using the function-realisation rules and sys terns , 
and then repeat the above process to detemnine the structure of 
the immediate constituents. The first stage of this is indicated 
above. 
Of course, this is a particularly simple sentence and structure 
designed to make it easy to see what the pTogram does, and it 
should be clear that considerably more complicated grammars can 
also be handled. The generative process i8self will not usually 
Dnvolve such s traightfarward intermediate s,tructbres or proceed 
so immediately to the final structure. For example, the 
generation of the structure of a sentence such as. "Must it gr& 
darker?", requiring five 'loops' of the structure-building rules 
before a structure compatible with all rules is obtained, is as 
follows : 
( GmERATE ( CLAUSE INDEPENDENT IN DIC RTIVE INTERROGATIVE POLAR 
MODAL INTRANSITIVE AT TRTBVTIVE ) 
FR 1 (!PROCESS) 
FR 9 (!SUBJECT !PROCESS) 
FR 10 (!FINITE SUBJECT !PROCESS) 
FR 12 ( ! MOOD-FOCUS ! FINTTE !SUBJECT !PROCESS) 
FR 20 (!MODAL- !MOOD-FOCUS !FINITE !SUBJECT !PROCESS) 
FR 27 (!MODAL ?MOOD-FOCUS !FINITE !PROCESS (= !ATTRIBUANT 
!SUBJECT) 
FR 28 ( !ATTRIBUTE !MODAL !MOOD-FOCUS !FINITE !PROCESS 
(= !ATTRIBUANT !SUBJECT)) 
FR 29 ( ! COPULAR !ATTRIBUTE !MODAL !MOOD-FOCUS ! FINITE 
!PROCESS (= lATTRIBUANT !SUBJECT)) 
! COPULAR 
!ATTRIBUTE 
1 MODAL 
! FINITE 
=> 
! MOOD- FOCUS 
-> 
(= !ATTRIBUANT !SUBJECT) 
!PROCESS 
! COPULAR 
! ATTRIBUTE 
- ? 
? 
(= !FINITE !MODAL) 
*> 
! MOOD-FOCUS 
(= !ATTRXBUANT !SU&J'ECT) 
!PROCESS 
ADDN 1 
? 
!PRE-SUBJECI 
! COPULAR 
! ATTRIBUTE 
-> 
? 
(= !FINITE !MODAL) 
= > 
! MOOD- FOCUS 
(= IATTRIBUANT !SUBJECT) 
!PROCESS 
SQ 1 
? 
! COPULAR 
! ATTRIBUTE 
! COPULAR 
! ATTRIBUTE 
ADlZN 2 
3 
! COP WLAR 
!ATTRIBUTE 
-> 
= > 
! MOOD-FOCUS 
(= !POST-SUBJECT !FINITE !MODAL IPRE-SUBJECT) 
(= !ATTRIBUA)dT !SUBJECT) 
!PROCESS 
COW 1 
? 
! COPULAR 
!ATTRIBUTE 
- > 
= > 
1 MOOD-FO CUS 
C= !FINITE !MODAL !PRE-SUBJECT) 
(= !ATTRIBUANT !SUBJECT) 
!PROCESS 
ADDN 3 
! COPULAR 
=> 
!MOOD-FOCUS 
(= ! FINITE !MODAL JPRE-SUBJECT) 
(= !ATTRIBUANT !SUBJECT) 
!PROCESS 
( = !POST-VERB !ATTRIBUTE) 
! COPULAR 
-> 
=> 
! MOOD- FO CUS 
(= !FINITE !MODAL !PRE-SUBJECT) 
(= !ATTRIBUANT !SUBJECT) 
!PROCESS 
( = ! POST-VERB !ATTRIBUTE) 
CONF 1 
? 
! COPULAR 
-> 
(= !MOOD-FOCUS !FINITE !MODAL ?PE-SUBJECT) 
(= tATTRIBUANT !SUBJECT) 
1 PROCESS 
(= !POST-VERB !ATTRIBUTE) 
CONF 2 
3 
ADDN 2 
3 
-> 
(3: !POST-SUBJECT !MOOD-FOCUS !FINITE !MODAL 
!PRE-SUBJECT) 
(= !ATTRSBUANT !SUBJECT) 
(= !COPULAR !PROCESS) 
( ! POST-VERB ! ATTRf BUTE ) 
-> 
(= !MOOD-FOCUS IFINITE ?MODAL !PRE-SUBJECT) 
( = I ATTRIBUANT ! SUIJJECT 1 
(= !COPULAR IPROCEGS) 
(= !POST-VERB !ATTRIBUTE) 
4 Conclusion 
'he obvious canelusion - that the mechanics of svstemic 
grammar (as described by Hudson) are sufficiently well-defined 
to form the basis of a computer vodel - 
is, for linguistic 
descriptions , a significant one. 
However, the program also 
demonstratks that some pule-descriptions require clarif ieation . 
For examle, feature-realisation and function-realisation rules 
are impIicitly unordered (since features and functions are 
unordered) , m, more precisely, the rules are to be considered 
to apply simultaneously. This causes problems with those rules 
which prevent features being introduced (e.g. rules such as 
Feat-ure-realisation rule 32, which means "if !SUBLTECT is ~resent 
the realisation 1s as stated, otherwise the first function, i .e. 
!ACTOR, must not be introduced by any other feature-realis ation 
rulew). The solution seems to be to reapply the rules, 
recursively, until a structura is produced which is combatible 
with all the rules. 
More seriously, the expectatiop that the structure obtained 
is independent of the order of a~plication of structure-building 
rules is not realised, at least for the grammar specified. 
Polr 
example, considering t+re second of the above generations and 
applying rules SQ 1, SQ 2, ADDN 2, SO 1, ADDN 1 (in that order) 
to the set of functions obtained by the feature-realisation 
rules, we generate 
3 
!PIG-SUBJECT 
! COPULAR 
! ATTRIBUTE 
-> 
- 
-> 
! MOOD-FOCUS 
(= !SU&TEm IATTRIBUANT) 
(2 !FINITE !MODAL !POST-SUBJECT) 
!PROCESS 
This strmctuH cannot satis fv both sequence rules , 
i .e. the 
generative process is blocked. Clearly, either the grammar 
requires modification or it does matrer which order the 
s tructure-bailding rules are applied. Hudson (personal 
communi"cation) has concluded that there, are linguistic grounds 
for ordering e tructure-building rules , s.o thqt ' abnormalT cases 
precede 'normal' ones, with the latter only applying if the 
former had not already been applied. Whether it ie possible to 
do sc consistently requires further experimentation. 
Further possible extensions to the work could involve 
trying to specify a lexicon so that the generative process ends 
up with a structure with words as leaves, and one could also 
attempt to apply the rules in reverse, i.e. To start with a 
string of lords and produce a etructurdl description. Both 
problems are, of course, very d3fficult ones. 
Acknowledgement 
The author is veIy grateful to Dr. R.A. Hudson University 
College, London) for his comme~ts on this work. 
Chomskv, N. (1965). Aspects of the theory of svntax. 
Carnbrfdge , Mass. : MIT Press. 
Friedman, J. (1971). A computer model of transformational 
grmar. New York: American Elsevier. 
Halliday, M.A.K. (1961). Categories of the theorv of grarw . 
Word 17, 
- 
241-292. 
Halliday , M. A=. K. (1979 1. Language structure and languape 
function. In J. Lyons (ed. ) , New horizons in linguistics, 
London : Pelican, 
Hudson, R.A. (1971). English complex sentences. London : 
North-Holland . 
McCarthy , J. (1965). Lis~ 1.5 programmer's manual. Cambridge, 
Mass.: MIT Press. 
Power, R. 19 74. A comuter model of conversation. Ph . Dm 
thesis, University of Edinburgh. 
Self, J.A. (1975). CBLP - a computer based learning ~rogramrning 
, J 
system. J. of. Inst. of Computer Sciences, in Dress, 
Winograd, T. (19 72 1. Understanding natural lanpuage . N~H York : 
Academic Press. 
American Journal )f Computational ~mpuistics 
Micr'bfiche 29 : 46 
INTERPRsETATION 8 INTEGRATION 
OF SENTENCES INTO A C-NET0' 
Groupe dlEY'udes pour la Traduction Automatique 
~nive~sit& Scienti fique et ~6dical e de Grenoble 
ABSTRACT 
A dstailed proposal for the represe~tation 16 
manipulation of C-nets (suitah15 for computer 
programming) in interpreting pronominal ref er~nczr,. 
It is shown how this th~ory accounts for the 
4 isambiguat ion a£ pronominal re fsrtnce, & th2 
determinatiiorr of focus E comment, rn'ora coapl~tily 
than ar~y existing semantic or syntactic thaory. 
The theory of C-nets appsars.to bs the iaost ad3quat.e 
linguistic theory for semantic ana,lysi.s of cortsnt [sy 
TREES 1.. Ha explore hers thc po;sibiIity of automating this 
analysis to aid in automatic translation. Tramlation 
involves analysis of cantent, without which it can only bc a 
matching of lc~xicdl & syntactic structu.res Letuwn languag~s., 
Such ?natching has been shown inddequate by many researcher?. 
Besides being necessary for automatic translation, an 
automated arialygis of c~ntsct is cqcsssary for - other tasks 
such as constt:'uct;isi~ (~ensrdl qnzstion-answering systems, 
v3ic~-writers, dutomatic indkxing 6 ebstractinq, pgo;iagjand.t 
measurnmcnt E txplicaticn, fallacy findirul, I. others. 
Interpretation E Integration 
In addition to being useful, C in the final analysis, 
necessary for practical problems in computer understanding or 
dacoding of bulaaa lasguage, this theory [see my DESC~IPTIOYS] 
is shown to allow disambiguation of procouns vhich are left 
as aabiquous by contemporary theories of semantics or syntax, 
but vhich in fact as2 no+ ambiguous. As will be note3 but- 
not eiplored, this t,heory dlso allows the consistect 
deteraination of the focus of a sentence from its contoxt, 
i.e. not using positiw or prosod-ic features. 
The thaory of C-cets is in rapid svolutiono2. The 
versioc described here is chosen partly- because it matches 
the desp structures obtainzd by the present Engllsh graamsr 
of the TkUl project at the Universite de Montreal [ TAUfl]. Ir. 
that system, a sentence is reducsd to its deep sttluctu~e t?) a 
Q-systemoa grammar. This deep structure would then be 
converted into a D-net, a t-nrporary C-net, which is then 
interpreted lexically (i-.e. lexical items are replakcti by 
their concepts) E integrated into an overall Cpnet. This C- 
nat represents the integrated meaning of that E prpvious 
sentences. In this theory, successful integration models the 
comptehensidn of the senteace, as described in my LCGC]. The 
dnterpretatiori E integration of D-nets are described here, 
with a demonstration of how later sentencos ars disambiguatad 
ill terms of the colnprehension of the earlier ones. 
3x1 continuations (a) & (b) belov, the syntactic 
structures are not significantly different. Yet both 
pronouns -OIL she refer unambigwausly to different people; to Mary 
ir, (a) G to Susie in (b). 
Nary told Sahn that Sush was corning, 
(a) but shs said it! softly. 
(b) but she didn't arrive for an houj. 
Any nondintr!grative semantic theory is forced to cdaim that 
these -.- shews both refer to the same person, or that they are 
both ambiguous. Either alk.ernative is wrong; these pronouns 
rofsr unarnbigu~usly to different people. A sentence oftsn 
picks up rn~aning from its linguistic context. Ths same 
sentence may express different things ir. different contexts. 
The mechanism by which this can happen is sxplained below.. 
The only way to determint; the ~orrect referents of 
these pronouns is to taks account ot the relationships 
expressed by the verbs ---- told t ---...f said be-h comiw E --I-I- arrive. An 
adequate way of doing this is to build a C-ncr't £ the 1st 
sentence & then integrate tha  continuation^ in,to it, as 
outlined in my [ IXTEGRATIVE]. 
A C-det is a directed, labelled graph with orders3 
arcs. We can represent a C-net by a list of iteps, 1 for 
each node. Each item then consists of: an index number for 
the node, a label which indicates its rn~~ning, 6 an ordered 
list of the indices of the nodss it dominates. The 
particular index numbers asslgn~d to the nodes ar~ 
Interpretation 6 Integrethoa 
irrelevant, except thirt B H2* ate reserv-ed to designate 
the speaker E addressee, respectively. Indeed, the index 
numbers are not part of the C-net. They are noriled ooly to 
nake a linear representation of a C-net, such as is corled 
for computer manipulation. Aim it is custoaary to 
capitalize labnls of nodes. (i.o. semantic atoms). G far 
simplicity, we take larp(r) . . as the meaning of Ew. Thus, 3 
sentence ; 
kry told John that ~usie was coming. 
has a C-net-: 
3 Harp (4) 
4 P. 
5 Tell (4,,6,7) 
6 P. 
7 Prog (8) 
8 Come(9) 
9 P. 
10 SusFe(9) 
11 John (6) 
12 P.ret (5) 05 
This list of aades with con~ections to dominated nod+s 
represents the graph below. In this & Eollovinq graphs, 
Pret 
I 
Tell 
Bary 
'I\ 
\ f Prog 
1 
1 
John I Susie 
\. 
Come 
The "referential pointn (V." iu the list natation & 

Interptetation & Intogration 
r i a(k,n)Q8 
,,,) 
1 n sa-y(jrm) 
i Tell (j,k,m) ]n+l Anim(j) 
t3+2 HUE (j) 
The qffect of the 1st 2 rulrr,s is to add tn~ features 
Hum (x) & g (x) or 6 (I) onto the points domi~ated by thr--. pzopnr 
Sosie !la=, -y, John. 
naBeS# .----L -- They express the linguistic 
fact that theso nalnas are human names (unlike e.g. Pirlo), t; 
are used for fenales & males respectively. In the 3rd rule, 
the atoms HUP(X) 6 Anifa(8) are iccluded in the laeaning of 
Tell (x, y, 2). to restrict collocational possibilities. Thry 
thus describe selectional restrictions, & will sometizes 
disambiguate an athetwise aabiguous senttince.09 Bacauss the 
predicates on any point must be %on-contradictory, the use of 
a non-human sub jet* for ---- tell ll(jk,l causes s 
contradiction around node j. The predicate place (x) occurs 
in the 4th rule for the same roason., 
There is also a aeneral sat of rules by which 
f htgtprstation E Integration 
duplicate nod3s are removed; 
 redact ion RJ& gqhema 
i Pn (k) 1 ----> i Pn (k] 
3 PQ(~) J &j:=i 
vhere j := i effects the r~placernent of all occilrr%ncr?s 
of j in the C-net by i, & Pu is a vciable ranging over 
' node labels. 
Rogt genstally, the lexical interpretation rules may 
be applied in any order E as many times as needed; then this 
duplicate-reduction rule remove-s any duplicate noses 
introduced by the lex-ical rules. To our present knowledge, 
Kowever, it appears that ths lexical rules can be constrained 
to operate only once on a C-net (at all appiicstule places 
sinulataneously) R this constraint will likely requirp soms 
ordering between the culeslo. My working assumptions are 
that lsxicsl rules can be oril~red so that c4ach can apply only 
once (everywhere) & that the duplicate-reduction rule 
operates last. 
By this process, the 1st graph above for lazy tpd.4 
Jphp 
thq,t sqg& 9s -. is converted into its 
final form 
- 
below. Since thero are no previous sentences, there is no 
integration to be done. This D-net is therefore also the C- 
net for the discourse up to this poj 
10 Susie (9) 
11 Jdn(6) 
12 Fret(51 
13 Hn~(4) 
or the graph; 
Pr ~3 
I 
John i3 
Huq ,' \ 
I/ vary Say 
- PI/ \ 
I 
e 
/ 
d a 
IL /I 
k /i 
Place # / I Susie 
The 1st' cOnMn.uation, {a. but she said kg garr&y 
yields the D-net below after the lexical interpretation rules 
& the aupli.~ation rebuction rule have applied, A D-net is 
analogous to a deep structure. It revrescnts the. meanirtg of 
a single sentence prior to integration, i.e. in isolation 
from the text it is a part of. It is integrated into the 
existing C-net to form a new C-net. 
Interpretation & Integration 
24 Soft(25) 
25 Say (27,26) 
26 P. L 
-27 P. L 
28 Hum(27) 
29 p (271 
30 Pret (25) 
or; 
Pret Soft 
\ 1 
/ 
sap' 
Hum / 
Q f/ 
\ 
\. L '. L 
The points in a D-net may bs annotated by an "LV'for 
definite reference. This "LU is peculiar to D-nets, E 
indicatss that th3 addressee should be ahla to find something 
it refers to, eithar in the C-net already existing, or in the 
context. When integrated, the "LN disappsazs because the' L- 
point is either identified with a poi-nt in the C-net, or it 
refers to some extra-cognit iye structure. This "I," derives 
from definite exprsssiocs such as fb& th&Sr kg, 
Ec. 
This b-net is intsgrated in the only possible way, 
namely with 
(i.e. she = Hary) 
(i.2. it = that Susie is coning) 
(i.e. say 3 tall) 
Aith the duplicates removed, the only addition is 24 
Soft (25) 12. These remaining additions thus obtained 
necessariiy include the focus (cornme3 t) & na w 
Interpretation & fnt+gration 
presuppositioas. The nmeaning" of this continnation ir a 
loose sense (what iaforrnation does it transeit vhich is not 
already kncrm) is simply the predicatg So£ t (1 5) . All the 
rest of the sentence is r2dundant in the sense that it, does 
n6t contain anythicg new, It: was nacessary, however, to 
allow the listener to determine what is soft. 
Pret 
1 Soft 
a /' 
John / \\ / 
"I' 
Say 
Bum \ / aary / \ 
'a PI/ Prog 
Place 1 / j/~um 
The refetonts of tho- pronouns are determined as. a by- 
product of tha integration. Because this continuation cannot 
he integrated in any other way, must refer to Hary, a 
feminine antecedent uhich is not th2 closest- one. 
There are several possible stratogles for integrating a D- 
net, o, which results from a cdnt'inaation sentence into the 
C-net, C, resulting from accretions from all the previous 
sentences. Basically, some of the nodes in D are found in C, 
& the D-net is traced -out in C. In general, the D-net wifl 
contain some nodes not in. C, & certain types13 of the codas 
Interpretation 8 ~nteqration 
in D may be missing in the C-net. The 1st D nodes to be 
sought in C should hav~ a high information content to make 
this procedure more effective 5 for this reason wj have 
chosen to start with the highest nodes in D: those nodes in 
D which are not dominathd by any other nodes. In the examplr 
above, nodes 29, 28, 30, 24, E 31 are all without domination, 
but 29, 28, & 31 are all directly predicating on points, 
which leaves 24 & 30 as highest*. Nodes with thss~ labels, 
Soft (p) & Pcet (g), are sought in C; Ar. equivalent for 24 is 
not found, hut 30 matches 21. Then, followinq domination. 
lines downward from 30 & 21, everything matches sxcept for 5 
B (6,15) . 8 (x, v) is one of the nodss which can be skipped in 
intsgrationl4. If the highest nodes do not yield an 
intsgration, the next higbsf are usqd until intsgration is 
possible. Integration is accornpllsh$d by a set of nod? 
equivalences such as explained ahove. Once these ar? 
effected, the duplicate reduction ~ule will rmove all th? 
nodes deriving from D which were already in C. Ths only 
nodes of D rsmaininq are thos~ which were not already in C. 
The pronoun she in this cpntinuation' is rsf srmtially 
ambiguous by any method of analysis dhich handles sentenc~s 
in isolation. Lt is rendered unarnnggnous, DU~ wrongly SO, if 
it is assumed to refer tc the nearest preceding fsrninln= 
antecedent. No English speaker can mistake that its ~efe rent 
is Hary; in reality this 889 is not ambiguous. But th.9 only 
way of obtaining the correct referexi€ for it is to utillzt: 
fntarpretation & Integration 
the information contaiaed in the verbs _sgx & BLL. This 
cannot be dong sim~ly by the similarity in the aoanicga of 
these verbs. la a different continubtion, 
she woqldn't be able to tell him aoything until 
arriving, 
the no longar refers to nary but to Susie, sron though 
tM main verb of the ccntiauatdon is idzntical to +,he verb of 
the initial sentence. 
In that contiauatioo, or in the less coslplex one (b) : 
she didn't arrive for an hour, 
the pronoun refers to Susie, E not to flary, primarily because 
of the verb azzivg This conthuation results in a C-nzt; 
24 Hbur(25) 
25 P. 
26 Duriag (25,27) 
27 leg (28) 
Hour Dur 
\./ \ 
neg 
J 
g Arrive 
This .is converted into: 
28 Arrive (29j 
29 P. L 
30 9U9) 
24 Hour(25) 27 Nebg(2R) 
25 P. 28 Finish (31) 
26 Dur(25,27) 29 P, L 
30 g(29) 
31 &3 (132) 
32 a(33,29) 
33 P. 
34 Place (33) 
Interpretation 6 Integration 
Hour Dur 
\' ' 
Ne3 
I 
Finish 
I 
8 
I 
Place d g 
by the lexical conversion rules 
i Finish (n) 
I n P (n+V 
i. ArriVe(j) ---> In+1 a(n+2, j) 
in+2 P. 
Ln+3 Place (n+2) 
i 8) (bm) 3 --- > f same as above ] 
a Arrive(j) A & n+2 := k 
It is inteq~ated unambiguously into the C-net as; 
Pret 
I 
6 d 
Hum I / \ Hour Dur 
\I/ 
1. 
\ 
9 S~Y \I' . Meg 
H:um 1 \ / 
nary \I/ 6rog Finish 
'-. \ / 
8 
I 
d 
1 /I 
Place1 / I 
1 Hum 
'. ' 
with the equations, 
Interpretation 6 Integration 
29 := 9 (i.e. she = Susi.) 
33 := 17 
30 := 21 
31 := 8. (i.e, arfiva core) 
32 := 16 
34 := 18 
Phen the duplicat5s are removed, the additiop is hat Susi9.3 
cotling was go& gi.g*s?~& for gg horn, the italicize3 porti.9~ 
being the comment. 
In the more complex cotitinuation rentlonc-fl abov-, -.)- sn= 
youldntt -- -I be -- able &_o &el-& gh anyrgig wil&gg, the & 
rsfers to Susib, lven thovgh the subject 6 the Is? vcrb 
significant for disarabiguation poist to #ary as fLe ref$r'$nt. 
This desirable rasult followS from the fact that the s~bjact 
of arrive is the same point as the subject of tell. With ths 
integration strategy usc.3. hers, both ell x) € Arriv~ (x) 
are on the same level in the D-net, but the sub-netuork 
dominated by Arrive (x) is, a perfect match, while the sub- 
network around Tell (x,p,l) has no match at all (because of 
Tell's diif erent 3rd actaet). Hence the sbg refers 
unambiguously to Susi?. Intoresting ~naqgh, if we insc.rt tha 
word gp~fj, gh 12-3 kg g,b& f;g fgLJ q&g anythi~g 
unt&J ~g~g, the 1st shy refers to H.aiy, 5 the 2~3 to 
Susi em The cause of this is that the interpretation of 
inserts an extra Tl(x,y,) into tho D-net, which will match 
the el x,) in the C-net from the 1st seatonce, but th;: 
only arrivhl remains Susiefs. 
Interpretaeion & Integration 
We have given a detailed description of the proclss by 
which different coritirruations of the sama sentsncs havr t'h?ir 
subject pronouns interpretad ds a function of their maic 
v 2r bs. The fntegratiori Frocess, which places iritsrprstatians 
on pror-ouns, also irlterprets duf inite occ,urrences of 
'nrmodifiad E nofi-specific touns as rsp3titions of i~revioutly 
mentionad mcrz specific (or more modified) nouns, fills FR 
dslated ~o~itals on thA basis of prior context & iatcrpr~ts 
gareric or unaodifid verbs as r~potitions of more specific 
verbs previously aenticnod. It must also handle ogrb pllras= 
dtlcltion, occurrencss of go sg & qapping. Th4 thfiory of 
which this is base3 is dscribed in my [Cl;C] t; 
[ PFSC31 PTf ONS G its relation to othar ~imant,Lc thcorics ir: 
discussed in my [ TREZS], [ ?JPi?:SFNWTfCIN], 6 in Paillct 
PROELERES 1. 
The essarit~a.1 contriblltion of this paper has b~r: 3 
foroai description of the+ pror~ss bf intet.jrat~on. ms ,nat 
process is central to any irdegrntiva semantic theory, if 
which the C-net theory is only 1 (setd. ruy- [ APPpqACfi 1) . I'h s 
f ornalizgtion of intrgrdtion presen%/td here is uudoubt~dly 
wrong in some aspects, C hquirss further r2soarch fqr 
improv~ilbr~t 6 vezification. A$ stated hare, it is apparently 
a3equats for most casqs of pronouns (but see not? 11, b 
excluding dsictic uses of pronouns & the yariphrastic jt) . 
The integration of the o-net dertved from a sbctnfic? 
La a modal of the user's I comptt-hspsion of ths snttrnc-1, 
Interpret &ti00 6 Integration 
vithout, ~f course, mod~ilinj his rvaluatio~ of the trutt ?f 
the,sentence ot thz motive behin3 its use. Tkss~ &re usually 
dcpond~rt oa the uchrerso of interpreta%ion E oaservatio~ ?P 
the sp+aker. Int3yra+,ior. prorldss a possible n=xt arta for 
rssearc~ toward fully aotoaa+ed translatioc b~causa it 
promises to prorids a comprehezsiv+ 3z-scription of tka 
content of tht. patagr aph E th; cor,t;;butioa c s+r,tccc 
aakes to tL3t COB~~C~. 
Pres?ntly concsivtd as an ddjur.ct to aL autmafe? 
translatian systeu, (it provides full infor~atior. 43 YO 
deletions. amphora of dofinite articlw G iJroriouns, 5 
interpretations of wotds) , the C-net conld provide the t~tal 
input into a target language rhetoric. This system will make 
~versionsm, trawlat ions wherein *he cant ent , but not 
necosaarily the wotds, syntactic canstr uctions or even the 
order of exposition are preserved. 
As opposed to a translation, a "versionN canzot be 
made with present thqoiy, bscause that r~qires at* a+--. ~~~atc 
theory of rhrtordc; how ttc material for a srnttscz. IS 
selected out of a conplut C-net, what cor.strair.ts thrra arc 
for selection of topics, coanc.nt.c;, focua~c, Gc. Nuch of thit: 
is unknown at present, thauga the rlFxt s5vcral ymrs may nt3w 
a great expansion of our kr.oulv?gr. (S~E T. Bpllc8rt -for 4 
direct attaCk-on this problem.) 
Ordinary translations in, houwc:r, a matching of 
Interpretation G Integration 
syntactic & lexical structures - as Om far -I as eossible ------- 8 without 
modifying the meaning. C-nsts provide a means to do this. 
With an adoquate representation of meaning, syntactic G 
lgxical matching can b done. The result can then be tcsted 
for chanqes in meaning by buildins a C-ket from the tarq?t 
language expressidn for co~parison with the original C-net. 
Hodification of the target expression can thsn be hads to 
make the input & output C-nets match to ar,y desir-.d degre-e of 
accuracy. A simpler means for good translation is found in 
my [ TRANSLATION I. 
01 This is an extended version of ac article "Icteryretation 
S Integration of Sentences into a CAnetwotkt9 which was 
written at Groupe de recherche snr la traductioc auto~atijur, 
nniversite de Montreal in the summer of 1971 & appeatjd ir. 
Rittredge [ ETUDES J. The tarminoiogy & rotation has also bear. 
revised to be consisterit with more recect work on C-nets. 
ot Compare, e.9- iity [PREBS'] C [JUDGING]. 
03 iha Q-system is a high leva1 programming ringda~s, for 
string nanipulation. SEO A. Colill~rauer, (LBS systBm~s-~ 013 1 
forpalisme pour analpser & syothBtiser des phrasas sur 
l'brdinateur', in -- TBUh - 473 (1971) Groupe de rech~rcha sur la 
traduction auotcaatiyue, UniversitC! de Hontrgal. 
0s Pray (#) C Pret (I) are abbrcv4 ations for ths meanings of 
the English fornatives for the progressive aspect, & the 
preterite or past tense. 
06 In contrast to this, Jqba tol$wHary rfi& susi.2 coning 
has a C~nat, 
Pret 
I 
Tell 
'I \ 
John I 
\ * -' i 
Prog 
I \ 
/ Svsie Come 
Hary t \ 1 
The difference in the list notation is that node 5 is 
Tell (6,4,7) instead of Tell (4,6,7) . 
07 See discussion in my [JUDGING]. 
On ~(~I-JI) stands for ar abstract locative atom of mea~ing 
which is npt realized exactly by any word in English. French 
B is closer to b(x,y). Recently discovered evidence leads to 
the beJief that Tell ( j,) has been incorrilctly analyqed 
here, E that this iB(r,q) does not occur at all. A bottar 
analysis is "to cause him to come to knov it by saying itw. 
The point of this payer, & ths d~monstrations rcmair 
unchan jed under this modlficatior., althouqh moat of tne C- 
nets are re~ised, 
o 9 T ha TAUH syntax presmtly attempts this sort of 
disambiguation by checking the appropriate NP for having thr 
necessary features, as proposed in Chomsky [ASPECTS]. 
HcCawley [POLE] has argu~d (p. 132f) that this canriot ba don;? 
generally by syntactic maas, b5cause the critical words may 
ba indafititely distant, c. f. : 
I will eat whatever Marid bakes. 
*I wili eat whatever laria believss, 
I will eat whataver Maria believes Toom to have bakdd. 
*I will eat whatzvhr Haria believes $om tb have drr:3rne1!., 
10 That is, that the rul~s can be or.dereu extrlrsically & 
applisd in sequence; after 1 tulc: has beer. applied throughout 
the network, it cannot be reapplied. This wili likaly 
requira the linguistician to arrange rules in an extrsnsic 
or&er a; some rul~s u4ad the output of others in an extrensic 
order to work, 
I1 The point '1' introduced Esom the analysis of come could 
now be re-interpreted as tho subject (4) or indirect objsct 
(6) of Say (4.7). It is not dope here for clarity of 
erpbsition. With the t~Visian tntntioned in note 08, thsqs is 
a simpls y?nsralizatior to id~ntify the points which ' 1' may 
be re-interpreted as. Th2rct is a list of preBicat?s 
(including ~3~1 bow -I &c) whose 1st actants may zep13cn a '11 
dominated by their 2nd actants. 
12 NoFe that the information by which hq$ contrasts with qc3 -- 
is lost in dntegratior~, consistcnt with the ptinciplo ~f 
intagrativo semantics that only the information transmitt1-d, 
not its order nor the speakr:rts reaction to it, i3 
represented. Conjunctions or succossiva sentences are simply 
integrated together: - but indicates some surprise or plrhaps 
inappropriateness. If it turns out desirable or evmt 
necessary to include such emotive information, semantic atoms 
for such may be defided E will be predicated on the eleman-t 
which is focussed, by a slightly sore saphisticated system. 
Here, it would result in an additional node Surprise (24). 
Phis would allow the reconstruction of the 2-clause syntactic 
form & the dse of $qt - from the intcgratcd ri4twork. 
Interpretation & Integration 
13 It is not yet known which elements may be missing from a 
D-net without blocking an integration. I assume that there 
is a finite list of such efelnents, including the perforraativc 
atoms marking the type of illocutionary act (statemect, 
question, Gc) . 
I* i B(j, k) can be substituted for by i= j during integration. 
See not5 33. 
Interpretation & Integration 

REFERENCES
Chomsky, N. (1965) Aspects of the theory of syntax. Cambridge, Mass. (M.I.T. Prsss). 

J. Friedman, 1971. A computer model of transformational grammar. NY, Elsevier.

M.A.K. Halliday, 1961. Categories of the theory of grammar. Word 17.

M.A.K. Halliday, 1970. Language structure and language function. In J. Lyons (ed) New horizons in linguistics, London.

R. A. Hudson, 1971. English complex sentences. London.

J. McCarthy, 1965. LISP 1.5 programmer's manual. Cambridge Mass., MIT Press.

R. Power, 1974. A computer model of conversation. PhD thesis, University of Edinburgh.

J. A. Self. 1975. CBLP - a computer based learning programming system. Journal of Institute of Computer Sciences. in press.

T. Winograd, 1972. Understanding natural language. NY Academic Press.

Chomsky, N. (1965) Aspects of the theory of syntax. Cambridge, Mass. (M.I.T. Prsss). 

Thomas R. Hofmann. An integrative semantic approach to intersentential phenomenon. 1975. Proc. of 1st Application of Computers and Mathematical models to linguistics. Sofia.

Thomas R. Hofmann. Comprehension and concepts in semantics, 1972. unpublished

Thomas R. Hofmann. Integrative semantics, 1973. Cahiers Linguistiques.

Thomas R. Hofmann. Verbs of judging: an exercise in further semantic description, 1973. Cahiers linguistiques.

Thomas R. Hofmann. Semantic representations and the theory of language, 1973. Language Sciences. 

Thomas R. Hofmann. C-nets in translation of natural language. In Kittredge.

Thomas R. Hofmann. Meaning doesn't grow on trees, 1973. Language Sciences.

J. McCawley. The role of semantics in grammar, 1968. In Bach and Harms , eds., Universals in linguistic theory, NY.

J.-P. Paillet. Prerequis pour l'analyse semantique, 1973. Cahiers Linguistiques.

J.-P. Paillet. Problemes de notation pour la semantique, 1974. Languages.
