lmericaa Journal of Computational Linguistics 
Microfiche 76 
FINITE STRING 
-$ THL 
NEWSLETTER OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS 
VOLUME 15 - NUMBER 3 
JUNE 1978 
AMERICAN JOURNAL OF COMPUTATIONAL LINGUISTICS is published by 
the Association for C~mputarional Linguistics. 
SECRETARY-TREASURER: Donald E. Walker, SRI Internationalr,, 
Menlo Park, California 94025 
EDLTOR: David G. Hays, 5048 Lakeshore Road, Hamburg, New 
York, 14075 
ASSOCIATE EDJTOR: George E. Heidorn, IBM Research Center, 
P.O. B'ox 218, Yorktown Heights, New York 10598 
EDITORIAL ASSISTANT: William Benzon 
Copyright 1978 
~ssociati on for Computational Linguistics 
American Journal of Computational Linguistics 
CONTENTS 
TINLAP-2: NOGRAM AND ABSTRACTS ............... 3 
DICTION'ARY bOFIETY OF NORTH AMERICAS SPECIAL MEETING ..... 37 
NCCt79 PER~NAL COMPUTING FESTIVAL ............... 38 
la79 NATIONAL CDMPUT~R CONFERENCE ............... 
39 
SHORT NOTICE OF UPCOMING CONFERENCES ............. 40 
RECOGNITION MEMORY (REMI: SEMIONICS ASSOCIATES . . . . 43 
SCREENSPLITTER ........................ 54 
THE TARGET PRIJECT'S INTERACTIVE COMPUTERIZED MULTJ'LINGUAL 
DICTIONARY. JohnBurge .................. .. 62 
American Journal of Computational Linguistics Microfiche 76: 3 
TINLAP-2: PROGRAIl AND ABSTRACTS 
TINLAP-2 will consist of six sequential sessions, each of which 
wi11 address questions of current theoretical interest and 
questions on long-range res-ea~ch directions. In each session 
researchers from artificial intelligence, linguistics, psych- 
ology, and philosophy will focus their points of view on a 
particular topic (see schedule below). 
Proceedings will be available before the meeting. Each author 
will give a 10-15 minute presentation (which may include s cri- 
tique of other papers, an amplification of points in the writr 
ten paper, etc .) followed by a 90 minute discussion period 
where questions and cfiments from the audience will be welcome. 
There will be other interesting evencs during and after the 
workshop, including the ACL Annual Meeting, a banquet, several 
opportunities for informal discussions, and events associated 
rlith the Linguistic Institute, to be held at the Universitv 
of Illinois this summer. The LSA (Linguistic Society of 
TINLAP-2 4 
America) aeeting will be held at the University of Illinois 
immediately aftet T'LNLAP-2, July 25-30. Ififormation ahout 
the LSA meetling can be obtainled from Proressor Braj Kachru, 
Department of Linguistics, University of Illinois. 
The program for TINLAP-2 is listed immediately below The frame 
number for the abstract is given in parentheses. ht:sentations for 
whhh no abstract was available arg designated with an asterisk. 
July 24 7:00 pm Reception and Registration at Levis 
9:00 pm Faculty center; Snacks 2nd Cash Bar 
July L3 9:OO am LANGUAGE REPRESENTATION AND PSYCHOLOGY 
11-45 am 
Chair: Dedre Gentner, BBN (7) 
David Pmelhar t , Univer si ty uf 
California , San Diego* 
Roger Schank, Yale>'< 
Leotlard Talmy, Neuropsychiatric 
In~titute of Los Angeles, U'CLA 
Terry Winograd, Stanford and Xerox PARC 
William TJoods, BBWk 
1:30 pm- 
LANGUAGE REPRESENTATION AND REFERENCE 
4': 15 pm Chair: Bpnnie Lynn Webber-, BBN (LO) 
Panelists : 
hn Anderson, Yale (11) 
TINLAP-2 
July 25 
July 26 
(Representatioh and ~eference) 
Herber\b Clark, Stanford (13) 
indrew Ortony, University of Illinois (14) 
3arbara Partee, University of 
Massachusetts (15) 
Zandace Sidner &$IT (16) 
Informal DiHscussion; Cash Bar and Snacks; 
Levis Faculty Center 
9:00 am- DLSCOURSE: SPEECH ACTS AND DIALOGUE 
11:45 em Chair : Barbara Grosz, SRI International (17 ) 
Pahelists: 
Joseph Grimes. Cornell (18) 
Jerry Morgan, University of Illinois (119) 
David Olson, Torohto (20) 
Raymond Pergault , Toronto* 
Andee Rubin, BBN (21) 
1 : 30 pm - LANGUAGE AND PERCEPTION 
4:15 pm Chair: David Waltz, University of Illinois (23) 
anelists : 
Ruzena Baj csy, University of Pennsylvania (24) 
Ray Jackendof f , Brandeis (26) 
Stephen Kosslyn, Harvard* 
Zenon Pylyshyn, University of 
Western Ohtario (27) 
Yorick Wilks, University of Essex (28) 
July 26 5~00 pm - ACL ANNUAI, MEETING 
6:00 pm 
6.30 pm Banquet (optional) 
Speaker JON ALLEN, M. I T 
July 27 9 00 am - INFERETJCE ~~ECHANI SMS 'IN NATURAL LANGUAGE 
11.45 am Chair. Aratind ~oshi, University of 
P enn s ylvania7Y7'~ 
Panelists : 
Eugmie Charniak, Yale (29) 
Allan Collins, Yale (30) 
Jerrold Raplan, Univer'sity of Pennsylvania (31) 
Raymond Reiter, University of 
British Columbia (22) 
Char1.e~ Rieger , University of Maryland (34) 
S,tuart Shapiro, SUNY Buffalo (35) 
Rand Spiro, University of Illinois (36) 
1 : 30 pm - 
COMPUTATIONAL I?ODELS AS 'A VEHICLE FOR 
4:15 pm THEORETICAL LINGUISTICS 
Chair. RonaAd Kaplan, Xerox PARC" 
Panelists. 
Joseph Grimes, cornellq 
Mark Libeman, Bell ~aboratories* 
Mitch Marcus, HITyc 
Tom Wasow, stanford* 
,I, 4- 
->\No paper to be presented. 
ABSTRACTS 
TESTING THE PSYCHOLOGICAL REALITY 
OF A REPRESENTATIOrjAL MODEL, 
Dedre Gentner 
Bolt Beranek and Newrnan, Inc. 
A research program is described in which a partlcular ze- 
presentational format for meaning is tested as broadly 3s pogsible. 
In this format, developed by the LNR research group at The hi- 
versity of California at San Diego, verbs are represented ab. inter- 
connected sets of subpredicates. These subpredicates may be 
thought of as the almost inevitable inferences that a listener 
makes when a verb is used in a sentence. They confer a meaning 
structure on the sentence in which the verb is used. To be 
psychologically valid, these representations should capture 
(at least): 
Similari 
The more 
the more 
ty of meaning : 
similar two verbs 
their representati 
seem 
ons 
. in meaning 
should over 
to peopl-e, 
lap. 
2. Confusability: 
The more confusable two verb meanings are for people, 
the more their representations overlap. 
3. Memory for sentences containing the verb: 
The sentence structures set up by the verb's meaning 
should in part determine the way in which sentences 
are remembered.. 
Seman Lic integration : 
The representations should allow for the integration 
of information from different sentences into discou~se 
structure . 
Acquisition patterns: 
The structural partitions in 
correspond to the structures 
are learning the meanings of 
the representations sh0uJ.d 
children acquire when they 
the verbs 
6. Patterns of extension: 
The representations should be extendable so as to reflect 
the ways in which people interpret verb meanings when the 
verbs are used outside their norinal context. 
7. Reaction times: 
The time taken to comprehend a sentence using a given 
verb should reflect the structural complexity of the 
verb meaning . 
Experiments concerned with predictions 1 - 5 are described 
here. The results are promisi~g for a general approach of repre- 
sentation of meaning in terms of interrelated subpredicates, but 
do not clearly distiaguish between several 
similar representations. 
For example, to test prediction (Z), I read people sentences con- 
taining verbs with similar meanings, and asked them to recall the 
sentences. The degree of overlap in the semantic structures was 
a good predictor of the number of confusions between sentences. 
In ahother sentence-memory experiment (predict ion (3) ) semantically 
complex verbs that provided more unde~lying ~dterconnections 
between the nouns in a sentence led to bettdr memory fof the nouns 
in the sentence than simple general verbs, or than gther complex 
verbs that did not provide such extra interconnections. To test 
prediction (5), I tested children's comprehension of a set of pos 
session verbs. Both the order of acquisition among the verbs and 
the kinds of errors fitted well with an account of the acquisition 
of verb meaning in terms of interconnected subpredicates 
This research illustrates a breadth-first approach to testing 
a representation. In the breadth-£ irst: approach, many different 
psychological predictions are made. Each different area of pre- 
diction requires a set of process assumptions, 
and in each case 
the process assumptions used are those that seem most plausible 
given previous research in the field. 
If one rspresentational 
fomat can make correct predictions about a number of differenr: 
kinds of psychological phenomena, then that representation stands 
a greater chance of being generally useful than one which was 
tested in only one depth-first way. 
Tne flelatton bf Grammar to Cognition 9 
Leonard Talmy 
Neurop~yc hiatric Inat ikut a, UCLA 
A aentence (or othe? p~~tlQt+ of diacourae) la ?men to evoke in 
the lL8tener 8 rns&niPlg complex, here called a "00@iti,~@ r@~r@88ntation" 
The Xexical elements of tba nentanca aaem, by snd large, to specify 
tHe aontent, or 8ubotance, of the cognitive reprbsentatlon, while the 
 ramm ma tical elements apecify it'a structure. Thua, lookine systemat- 
ically at the actual notions apecifiaa by &rammatlcal elaments cpn 
$i%e ua a handle for aacer%oining the very makeup of (linguistic-) 
co&ultive strl~cturtng. We accordingly examine a numbsy of ~rammaticolly 
specified notiono, obaarle the syatems or categorlea in which they 
pdtteqn, and speculate on broader cognitive connections. 
Some provisional findines have already emereed: Grammatical. 
apacifications for atru~~ure are preponderantly relativiatic or 
topological, and exclude tho f lxed or metrically Euclidean. The 
%gatems in which grammatiCa1 notion3 pattern include: 
plexlty (uniplexfmultiplex~ doves of axtenalonalitg 
stat a of boundadnean pattern of distribution 
atSte of dlvidcdncea axial charactariatios 
level of eynthesls 
perspectival oharaOteristics 
level of exemplarity 
ecaria~brcakup charrctaristioa 
drammrticcal apsc if iocltion of 8truoturin8 appears, in certain abstract 
charactaristlca, to be laomorphic wlth the structuring of viaual 
Des~ription Formation and Tliscourse Mode% Synthesis 
BOhnie Lynu Jebber 
Bolt Beranek and Newnlan Jnc. 
50 Moulton Street 
Cambridge, b1R 07138 
Researchers in li.nguistics, psychology and art if icii~l intcllfgence havc 
recently begun to abandon a *purely linguistic appruutlh to def initu ar~nphora 
(definitc pronouns and noun phrases. Instcad they posit the notion nf rcfcrcncc! 
into a modcl that a listcll~er/resdcr 1s sylzthcslzing fro* thc disronrcr: the 
referent of a definite anaphor is then not a linguistic object, but rather an 
entity in a model. Such a model has been called a ''world of discourse" 
[levin & Goldman, 19781";a "ur~ivcrs~ of discourse" [Lyons 19781, al'discoutse 
nodel" [Nash-Webber 1977; Webber 19781 and a "domL~irx uf itzterprctatJ.ont1 
[Stemling 19751, inter nlia, Its syntl~csis is what xnceresbs me. 
Discourse n~odel synthesis intuit ivel)rt seems to result fro111 interact ions 
between che listencr/rsadcrls expectations and various features of the text, 
That these interactions are is not clear, A discussiorr of haw the listener 
r-eader 's changing expectations can atiect iiscsurse model synthesis can be 
Found lm [Collins, Brown & Larkin, 19771. What I shall discuss here are some 
feata~res of the text that aFfect what entities appear In a disc ours^ model and 
how such entitles are described. In the course of presenting these features, 
I will argue that having an appropriate description for a discourse entity 
is critical to its successful reference later on. I will then argue thaL 
recognizing formal aspects of the text is critical to the formulatioh of 
appropriate descriptions. While this is not a sufficient cdhdition for 
successful reference, it is certainly a necessary one. 
Representation of Individuals in S~~aantic Nets 
dohn Anderson 
Yale University 
Abstract 
Reseatch is reported concerned with how subjects process 
multiple refesrkng expressions+ 
In one experiment, subjects 
learn sentences such as: 
The smrwrt Russian cursed the salesgirl 
The smart Russian rescued the kitten 
The tall lawyer adopted the child 
Thk tall lawyer Caused the accident 
and only later learn that the smart Russian is the same person as 
the tall lawyer.. How do subjects integrate the information about 
the smart Ruqsian atkh information abgut the tall lawyer? It is 
information subjects have set up two nodes in memory, 
one for 
each definite description. Upon learning of the identity of the 
two descriptions, they introduce into memory a proposition indicating 
the identity of the two individual nodes. They also start a 
process of copying infornation £rum one node to the other node. 
In effect, they choose to abandon one of the nodes. 
It is argued that a similar process occurs when subjects 
recognize the referent of a definite descriptibn--but on a much 
shorter time scale. So, suppose a subject hears: 
The first president of the United States was a bad husband, 
The proposal is that the subject creates a new node to represent 
the subject of that sentence, attaches to this node network structure 
to encode it is the first president of the United States, 
uses thfS 
network structure to guide a search of memory for the referent, finds 
a node corresponding to George Washington (GW), indicates that the 
new node and the GW node are the same, copies fromthe new node to 
the GW nbde the bad husband predicate, and abandons the nev node, 
Data is presented consistent with this process model for dealing 
with the referents of definite des,criptions. 
Reference Diaries 
Herbert Ti. Ciarb and Catherine Marshall 
Stanford University 
Standford CA 
Speakers and listenets are fotced to keep diaries about what they wow 
about each other because, to use or interpret a definite reference, they 
!I 
have to qssess the knowledge they 
ahare" with each other about the thing 
being referred to. More prec?iseJ.y, it can be shm that the speaker and 
listener have to assess what is technically called their mutual knowledge 
about "the referent. 'Nnis, however, raises a striking paradax. The assess- 
ment of muthal knowledne logically requires an infinity of separate tests, 
and if each test takes a finite amount of time, then people would take aq 
infinwe length of time to make or interpret'any definite refe~ence. 
As a solutim to this problem, we argue, people use the heuristic of search- 
ing their diaries for an event that satisfies a condl'tion we call triple 
co-presence. With such an event they can satisfy the infjhity of tests 
required by mutual knowledge in a single step. We discuss the kinds of 
events that satisfy tripble co-presence, dnd we provide experimental evi- 
dence that when people cannot find such an event they are open to error 
in their intPs~re&ation of definite reference, 
Some pra5rrat I c constro i nts on the ~onstruct ion 
Sna in?-erpretd?ion of definite descriptions. 
Andrew Or -tony 
'Iln iversit). of I l 1 i nois ot Ilrbana-Chof:lpoiqt; 
Goth ttie proauct ion and the cmprshensian of definite dascrip- 
tions requires that inferences be na4~- In mdny casw the infer- 
ences ars tr i v i % 1 and s"f i I e the~re-t i ca I. i mportarica or i n- 
Terost, M(lve~er, there is a class of definite descriptions that 
Fsve Tha charclcter isric fhat their re1 at ion to their * anfsccdents 
depends on prs~rnatic inferences ('contrastad \Y i th deduct ive.fy log- 
ics I inferences), In SL;C~I cases, Phe pred icate under lying b-::t: 
defYni-i-z descrYption cannot be taken to be true of the antecedent 
as a result of any enfcilrnent re~tions. Rather, the pred~cate is 
taken as being pra~aSilistica"lly rslzted, This psper exzmrnes 
shis clas& OF. Ifpracpatic definite descr ipi~ons~~ nore closely, 
pay ins particul ar attant ion to what sonstrai'ns *the set of cand I- 
cste descriptions that can be used to refer to the anteceden*. 
Lne of the results of thrs eYsMnzkion IS the postulzt.ion of a 
theory abocrt the extent to which an indirect speech actV can be 
a 
1neirec-f. 
Bound Variables and Other Anaphors 
Barbara H. Partee 
ABSTRACT 
The aim of the paper is to delimit a subset of pronominal anaphora 
for which the logicians notion of bound variables gives the best account. 
It can be argued that some cases of anaphora must be iewed this wav mrd 
some cases cannot be. The clearest cases of bound variable anaphor& involve 
antecedents like every man and - no man which are singuldr in form but da 
not refer to individuals. But even with an antecedent 1 i ke - John, an ana- 
phoric pronoun must sometimes ukviewed as a bound variable to dccount for 
one of the readings (the so-called "sloppy identity" reading) of TI): 
(1 ) 
John was sure he would wjb, and so was Bi 11. 
Bound variable anaphora wi 11 be cobtrasted with free "di scourse" anaphora; 
the differences between them suggest that the former is essentially a 
semantic phenomenon, the latter largely pragmatic. 
The Use of Focus as a Tool for 
Disainbiguat ion of Definite Noun Phrases 
Candy Sidner 
NIT A1 Lab 
This paper will center on a djscussion of the use of focus in 
the interpretation of anaphoria noun phrases in discourse. The 
need for focus will be discussed, and a description of focus shifting 
will be given. Focus provides a means of representing the centra1 
concept of a discourse. The ways in which a definite noun phrase, 
specific or generic, can be used are constrained by its relation to 
the focus and by the ways in which the focus can be shifted. The 
discussion of anaphoric defnps will present a taxonomy of cases, 
distinguished by the relation of the defnp to the focus, This taxonomy 
inclues several kinds of inference dependent cases. The paper will 
concentrate discussing on the process of understandins defnps, and 
will present rules governing the ways a defnp can be used so that the 
hearer/reader can understand its co-referent . This paper will also 
distinguish reference, co-reference and internal reference, and point 
out the need Eor these distinctions in natural language research. 
Barbara J, Grosz 
SRI Inter national 
Menlo Park , Cal i fo~ nia 
wnen cwo people talk they focus On only a mall portion of what 
each of them knows or believes. Both what gets said and how it gets 
interpreted depend on this narrowing of attention to a comnon 
highlighted portion of knowledge. One of the effects of understanding 
an utterance is to become focused on certain entities (relationships and 
 object^) and on particular views of those entities. A speaker provides 
a hearer with clues to what to lock at and how to look at it -- what to 
focus on, how to focus on it, and how wide or narrow that focus sholrld 
be. These clues may be 1 inguistic or they luay come from knowledge about 
the relationships among entities in the domain (the structure of the 
things being talked about) or from the environment in which the dialog 
occurs. Linguistic cues may be either explicit, given directly by 
certain words, or imp1 icit, deriving from sentential structure or fror 
rhetorical relationships between sentences. 
This paper examines focusing in dialog, discusses an initial 
representation in which focusing 
is based on domain structure cues, and 
examines from this perspective what other information and models are 
needed to extend the formalization of focusing to more general dialogs. 
The importance of focllsing is illustrated by considering 
its role in the 
processes of under standing and generating definite descr iptions. 
TOPIC LEVELS 
Joseph E. Grimes 
cornell Univetsuit 
Ithaca, N.Y. 
In order to interpret either a dialogue or a monologue,' some 
referential dements mqst be agreed on by the speaker and the hearer as 
a starting point. This is the topic in me sense proposed by Searle and 
Gundel. Even though the topic normally shifts away from its starting 
pint in the course of a text, whatever is being treated as topic in a 
particul ar part of the text receives special treatment in determining 
the expression to be used.. 
IeveJs of topic, global vd local, iri ~nglish conversation have 
been noted by Grosz. They imply difPerent strategies for establishing 
the reference of pronouns. It is useful to consider than in the light 
of two other languages,, ~onguda of Nigeria and ~acairi of Brazil, that 
distinguish topic from nontopic by their pronoun systems. 
Finally, there is some evidence from both Greek and ~nglish that 
there may be more than two topic levels operating simu~taneously in 
nonconver sational texts. 
ABSTRACT 
Toward a rational model of discourse comprehension 
J. L: Morgan 
Center for the Study of Reading 
and 
Department of Linguistics 
  university of Illinois 
Models of discourse br text often treat connected discourse in 
a manner analogous to the treatment of sentences in traditional and 
generative grammar; i.e. as a formal. object to be decoded by means 
of c~rtain formal operaaions. I point out in this paper that even 
where this view is not explicitly proposed, it is often implicit. 
Against this common view I argue that the only kind 05 discourse 
model that is likely to succeed is one that is built around two 
important hypotheses: first, that the key to d~scuurse comprehension 
is the attempt to infer the deta'ils of the plan that the speaker/ 
writer follows in constructing khe text; secondj that a large portion 
of the work of a discourse comprehension model should be derived from 
a theory of practical reasoning. I will sketch the outline of a model 
(or more accurately, a ~chema for a large class of possible models) 
that incorporates these suggestions, pointing out the rdle of 
practical reasoning processes, and arguing that notoriously confused 
nottons like "gicves/'new1' and I1&xpected inforrna~ion~~ can wly be made 
sense of in such a model. 
SOCIAL AND ~OGICAL GBECTS OF MEWING IN ?HE WAGE OF 
SCHOOL-AGED CHILDREN 
David R. Qlson 
ontarlo Institute for Studies in Mucatlon 
Toronto, Canada 
f t is conventional to treat the meaning of an utterace in a 
discourse in terns of two components! the ?ropositional component and 
the pragmatic or speech act component, the first indicating the meaning 
of the sentence, the secona r indicating its intended use. by the speaker. 
I shall present so111e arguments and evidence that these two systems are 
interdependent. Roughly, it appears that social considerations, 
prharily status. determine which a-cts of a proposition are 
lexicalized in the utterance. Thus, a child with high status relative 
to his interlocutor may use a conunand, "Give me "a block", while if he 
has low status relative to his interlocutor he may use a request, "May I 
have a block?" I£ he is an equal, a peer, (and perhaps only then) he 
will use 
an explicit true propsitian such as, "You have two more than 
me." Only in this third case is the propositional meaning explicit in 
the sentence per ae, and only in this case is an a££ irmative or negative 
response dependent strictly upon truth conditions (rather than 
compliance, for example) . 
*is conception of the social aspects of meaning will be examined 
through an analysis of what is said vs. what is meant in Gome child- 
ch jld and teacher-child conversations, 
Paper prepared for ~heoret'i2al Issuesb in Natural Language Processing 
(TINLAP). Urbana, University of fllinois, July 25-27, 1978. 
WHO XiJ1 I TALKING TO AM) CAN TlWY TAM BACK: 
THE! EFFET OF AUDIENCE AM) INTERACTION ON DISCOURSE MODELS 
Andee Rubin 
Bolt Beranek and Newman Inc. 
Cambr idqe , Mass, 
Communication among people occurs in a vast vhr iety of settings 
frm reading a book to partiaipatiw in a conversation, from Pistening 
to a tape to reading a transcript of a lecture. Most discussions of 
discour se , speech acts and d ialogue , however, consider a very par t icular 
kind of comnunicative situation: Eaae-to-f ace 3rd conversations between 
two participants in which there is a common spatial and temporal 
context. Dialogues between a computer system and a person differ from 
this model along at least two dimensions: the modality of the 
interaction (current computer-person d ialogues are wr itten) and the lack 
of spatial cmnonal ity , indicated by the impssibil ity of comunicatirq 
with gestures and facial expressions. The imp1 ications of these 
differehces for theories oE discourse are poorly understoodr Worse yet, 
they illusttate only a mall subset of the dimensions along which 
language experiences may vary, mat relevance do the theories we 
advance to account for these interchanges have for other communicative 
experiences such as listening to a lecture or reading a play? 
This paper will focus on two other aspecfts of language experience 
which have consequences for the dialogue models we build: audience and 
the degree of - 9nteraction. In both situations described abdve, tne 
audience is a single other person (or systen) and interaction between 
the participants - or even interrupti-on - is immediate. 
But in a book, 
for example, the audience is larae and not well defined and the  boo^ s 
reader must adopt new strategies to cornpensate for the fact that 
interaction is impossible. In a personal letter, on the other hand, the 
audience is a single other person, similar to the conversational 
situation. Interaction, however, is impossible or at least attenuated; 
the reader can obtain clarifying information, but the time lapse will be 
significant. - 
I will consider in this paper where various lahguage experiences 
lie along these two ddnenaions and what the impiications of these 
differences are for models of discourse and dialogue. 
On the Interdependence of Languagk and Perception 
David L. Waltz 
Coordinated Science Laboratory 
University of Illinois at ~rb@na/~hampaign 
Without a conneckion to the real world v&a perception, o language 
system cannot know what it is talking about. Similarly, a perceptual 
system must have ways of expressing its autputs via a language (spoken, 
written, gestural or sthef). The relationship between-perception and 
l~nguage is explored, with special attent lon to what implications 
results in language research have for our models of vision systems, 
and vice-versa. It is suggested that early language learning is an 
especially fertile area Ior this exploration. witkin this area, we 
argue that perceptual data is conceptualized prior to language acquisition 
according to largely innate strategies, tha~ this conceptualization is 
in terms of an internal, non-ambiguous "language," that language production 
from its beginnings to adulthood is a projection or the inte~nal Language 
which selects and highlights the most important portions of internal 
concepts, and that schemata produced in the sensory/motor world are 
evolved inso schemata to describe abstract worlds. Examples are provided 
which stress the importance of "gestalt" (figure-ground) relationships * 
and yrojecthon (3-D to 2-1/2 or 2-1) , conceptual to linguistic, ad 
linguistic ta conceptual); finally mechanisms for an integrated vision- 
language syshem are proposed, and some preliminary results are described 
The Problem of NilZnir~g Shapes : 
Vision-lar~epagc Intc~factl 
R. Bajcs-yfi 
and 
AX. Joshifi 
Canputer and Snfonnation Science 1kp;lrtm~n-t 
Vniversity of Pe~msylvmia 
Philadelphia, PA 1904 
JII 1ll.i:: ~,II>PI l, wc? w ill l)osrl rlwnl CIUCS ~.~OIID tlun plcstmt s.01 utions. K:'c ci~u~t to 
raise sor;,e questions in the context of the representation of shapes of 8-D objects 
One way to get a hCindle on this prob@.m is to investigate whether labels of shapes 
mind the* acquiGtion reveals any striicture of attributes or cmpnents of shepes 
that might be used for representation purposes. hother aspect of the puzzle of 
rcprescntation is the question whether the information is to be stored in analog 
or,rprapositionol form, and at what level this transformtion from analog to pm- 
psitional form takes place. 
In genepal, sha-pe of a 3-D cornpact object has tsJo aspects: the surface 
aspect, md the volume aspect. The surface aspect incLudes pmpsrties like con- 
covit+y, convc.sj ty , pl,mcuxi ty of surIacas , edges, and cornws. The volumz aspect 
di:; tingciishc:; ob jcc is wi th hol CIS fmiri those without (topo1oj;icctl properties) , and 
describes objects with respect to their syrrunetry plales and axes, rclatlve 
~mrti-ons, ctc. 
" 
'Rlis vork hcls been supported ~mdw NSF Grant d::CS76-19465 and NSF Grant BNCS7F 
19466. 
We b:ill discuss some questions pertinent to representation of a shape of a 
3-D compact object, without holes, for example : Is the surface aspect more im- 
portant than the volmc aspect?, Are there any shape p~imitivea? In what form 
are shape attributes stored?, e tc. We shall extensively draw from psychological 
and psycho-linguis tic li'twature , as we1.l as from the recent AT activities jn 
this area. 
An Argument Combining Linguistic,and Visual Evidence 
Ray Jackendoff 
Brandeis University 
ABSTRACT 
The notion from gestalt psychology of a "figure" emerging rrom 
a "backsround" will be shorm to be crucially involved in a complete 
descrtption of the successful cammunication of so-called 
"pragmatic anaphorta" - uses of pronouns without Einguisti,~ . 
antecedents such as that in (1) . 
(1) I bought that pointing last Saturday. 
A survey of types of pragmatic anaphora in English will then be 
used to show that the notion of "'figure" must encompass, a muck. 
wider range of perceptual entities than commonly assumed. Finally 
the implications for linguistic semantics, philosophy, percept la1 
t-heory, and cognitive theory will be discussed 
(very tentative abstract) 
Language and Percept ion 
Zenon Pylyshyn 
University of Western Ontario 
A language comprehension system without a perceptual component would, 
in an important sense, not know what it was talking about even if it could 
carry on a sensible dialogue. More significantly, a theory of comprehension 
would be seriously deficient if it did not relate linguistic representations 
to ones which derive from non-linguistic sources. This bridge is necessary 
in order to explain how terms refer as well as to explain how language is 
acquired. This paper will discuss and support the position that natural 
language learning is only possible because of the prior existence of 
menialese --a language-li ke system of representation for perceptual as we1 1 
as more abstract conceptual contents. How this comes into being cannot be 
given as an inforrnat3on processing explanation since it requires an account 
of the development of the underlying machine architecture--not of its 
langu3ge processing software ( e. interpreters). 
SEMANTIC PRIMITIVES IN LANGUAGE AND VISION 
Yor ick Wilks 
Department of Language and Linguistics 
University of Essex 
England' 
ABSTRACT 
An argumeht is presented that, on the basis of the evidence at 
present available, there is no reason to believe that the semantic 
primitives required by natural language understanding have any 
basis or grounding in vision. And, moreover, whatever may 
ultimately turn out to be the way we work, there is no reason 
to believe that trying to ground one sphere of A1 on the other 
language primitives on visual ones, would assise research in 
either area. A number of systems of primitives are examined 
briefly in order to strengthen the above argument. 
With a Spoon in my Hand this must be the Eating Erame 
Eugene Charniak 
Department of Computer Science 
Yale University 
ABSTRACT 
A language comprehension program using "Prames, " "scripts ,I1 etc . 
n~ust be able to decide which frames arc appropriate to the text. 0iten 
there will be explicit indication ("Fred was playing tennis" suggests 
the TENNIS frame) but it is not always so easy. ("The steering wheel 
was hot, but Jack had to be home by 3" suggests DRIVING, but how?) 
This paper will examine how a program might go about determining the 
appropriate frame in such cases. The basic idea will be taken over 
from Minsky (1975) in that it will be assumed that or~c usually has one 
or more cogtext frames, so that one only needs worry 2f information 
comes in which doe,s not fit them. As opposed to Minsky however the 
suggestions for new context frames will not come from the old ones, but 
rather directly from the conflict5ng. inEormation. A major portion of 
the paper then will. be conccrncd wick how wc will index context frames 
(em g., DRIVING) under the clues nhich suggest them (e. g., STEERING- 
WHEEL) . 
AUarz C~IL~ 
Bolt Beranek and Newman Inc, 
TItc papch ai~tLi~i~ a co)nl.utntioj~cLe tl~cohy 06 i~w~ia~i ptaubibLc ncn~o~~ij~g 
cons,&~~cted dhoa n11&!{6iA CJ~ pe.o)~cr ' n bl11~tYY~b to cvchyday queb tioris. L-ihc .Cog ic, 
,tltc the ohy .is cx)u~c,~nrd in a ca~~tc.~~ t-b~~dc-pcncfc.r~t 6ohsinei'~ln. UJI CiCe log ic, .the 
tl~eo~y ~pccidi~~ !lo10 di66c~~l:t i~i~o~in~a~o~t ill IIICJIIOCI~ n6bcct.h the cc,'l+tndlz.tg 06 tkc 
COMC Cl~biol~n, rih,n[~~~. TIIC .thcony conn cJs t~ o 6 a dialails ior~nli 2rd npnce o 6 did ~CACIL~ 
i~q(e,ter~ce tgpil.$ nrtd thcdr ccj~tairztc~ col~tlitiorln, i~ictttdi~~g it vahioty 06 lilctn- 
l&6~/i~nce $gpa idt c,t e ih c ill bftlcrzcc dcpc~zris ort tl~c p~ 5 ort ' 4 k~~o~Pedg e abo td hi., 
old11 brzodedg e. Tlie ptlo taco.&$ 6~om pCoj3tc' 6 ~JI~$L)I)'C/LA .to QLLCA ti0126 ~RP a~zatyzrd 
Indirect Responses to Loaded Questions 
S. Jerrold Kaplan 
University of Pennsylvania 
ABSTRACT : 
Casual USCFS .QS ~ntuul lalrn~a~c (NL) sys tans am -t-ypicalLy inexpert not 
only with regard to the i-ecl~nic;ll details of the unrlcr~lyin~ prol:l~;lmr,, but oftcn 
with regard to tile. structuin? NIC~/~P ~ontcll t of the dormi-n of ~~SCOUI~SC. 
bnsequer~tly, NIJ systems rnust lx. designed to irspond appmp-iately w11m they 
can dc4ect a misco~~ception on the part of thc user. SevcraT conventions e} ist- 
h cooperative convell. .,it-ioll that allow a speaker. to indirectly cncode their 
intentions and beli ~f s about: the domin i11t-o thcir 11 tte~~3nces, ( "load ing" t-he 
utterances) 
and allow (in fact, often rr~quire) a coopemtive mslmnder~t tci 
address those intentiolls and be3 iefs beyond a literc~l rgsponse. To be effective, 
NL computer systems must do the- same, 
This paper will explore several types of indircct responses to I4L questions, 
showbe that in the Data Base query domain ~cnqra-1 corny tat ional models exi st 
that cat] determine both when - an indirect response is required and d~at that 
response sbuld be. AII implementation of these ideas will be presented that 
demonstpates their hnediate practi~al valfie in I% sys tenjs . This paper will 
take the position that language r9lated inferences (i.e., infererices driven 
lkctly 5.m the phrasing of the question) are to a peat extent separable firm 
deeper reasoning and deductien processes, and are sufficient la proclwe a 
wide variety of useful and cooperative behavior. 
S.J. Kaplan 
Ray Reiter 
University of Brltish Columbia 
ABSTRACT 
I prowse to di scuss a nu.lllxr of principlcls for stY~icturlncl k~~otlrltxlclc, 
principles which arc n~tivated by td~c? need for cfficicnt deductive infcl-cncc in 
quest ion-at7st~~i-ing sys tents. The notion of stx-uc turc -trl~at I v1i.11 define is in 
son= sense, ortrJnqona1 to but hot a1titl2ctical to a ntm&r of current ic7ccls in 
Al: reg=-Ainq UICJ organi zation of knowlec7c~e. 
Intcns ional vs . Extc?nsional Rerx esentations of IWo~~lcdsc 
Given CI predicate P, we can rcpreser.t 141a.t we knqw about P cxtmsic)nally, 
or intmsionaJly, e.g. as a prcxed~re or a general axlom) or by som cornbinakion of 
. HOT sho~~ld xhis dccis~on be made? It turns out that if we represent 
appropriate predicates cxtgsXonally then 
(i) 
No lbfinitc deductive searches can arise. 
(ii) Certain inca~sional knowledge becomes irrelevant for 
deducti-on ar~d my ke di scarded. 
The Closed World Assunption (W 
In domins for, brhll=h we have perfect knowklge (e .q. blocks worlds) 
it 
is appropriate to mfie the CWA. Thls mems, roughly speakmg, that to establish 
a ncgativs fact, it is suffic"ient to fag1 to prow its psit~_ve counterpzt. The 
CE7A yields a Significant decrease in the camplexity of deductive reasoning. In 
addition, it induces a deco~sition of ale available lu~owledge into tvm cownents, 
one af which is used only for integrity, and the other only for deductive inference. 
Horn b-ta Bases 
-- 
It is wel-l. known that whgever the knowledge about a domain is 
representable by Horn formulae (i . e . fomlae of he f om P+ P+ . . .A P,,2Pn+1 where 
plf--tpn+l are positive) then consequent and/or antecedent reasoning is mlete 
for that domin. This result is not true for non Horn darni~ls - mre :opl~isticated 
reasoning, such as case analysis, may he required. Al~otl~er nice feature of 1Ior.11 
donuh~s is Chat the W does not lhd to any h~consistcncies. 'Not all dorruirla car 
~IQ ~cpresa~tccl by ~16m fo~milac. 
For soma such domains it is possibl e to render 
than "essential ly " Horn by cxtcnsionolly repepresenting certain appmpri otcly choser 
prcdicatcs, in wllich case all af the Virtues nf Horn donuins my be salvaged. 
pmpos ing 
1. If pssible, mkc the CWA. 
2. If the hcxilcdge base is non Horn, makc it "essentially" ~IQ.~II by 
extensionllly representing appropriate pmdi6dtes. 
. 
Elinhate infinite dd~~ction paths -by extu~sionally represent-bcj ccmin 
svi t&ly chosen yr-cdicates. 
4. Und~r 1, 2 and 3, certab mkensions will no longzr ke relevant for 
deduction. Reri~vz these. 
Inference ntld lJnrsin6: Arahit~c turc in GHIND- 1, 
a Eull-Scale Story ,Cornpr*chcnrlcr 
Chuck Hi~r.er 
Computcr Scic~cc Depsrt+mcnt 
U11ivc~rs1t;y of b;arylarld 
Colla,qc;? Park , ha i!;lantl 20'74% 
APSTRACT: The paper dcscrii~~s the inf'ercncc and parsin? co~l~ponents 
- 
of G-HIND-I, a full-scale story compr*chcr~sion pl-oj~ct ba:,(?d on a 
\\'a1t Disrley Pool.: of ttw I lontti Club hook ,* V"c I!a~ic (;rindcrIt . 
Topics incluclc: ( 1) thc .sense net,\.:orl.: parsrr and it in tci-ac t ion 
with iraJ'erenco, (2 1 chax4acter pcrso~~~~lity trait f~odelin~: via 
behavioral tags, (3) two-cl~aracter relationshi;, modclinr, and (4 ) 
plot veprt~scntation and plot le~c~~pre~liction. Thc main areas of 
etrlphasis will be en the representatior~ of inf'crence, and or, t3c 
various types of inference conditioning that stem from the 
character models and plot. 
Path-Based and Node-Based Inference in Semantic Mctworks>k 
Stuart C. Shapiro 
Department of Computer Science 
State University of New York at Buffalo 
Amherst?, New York 14226 
ABSTRACT 
TWO styles of performing inference in semantic networks are presented 
and compared, Path-based inference allows an arc or a path or arcs 
between-two given nodes ko be inferieed from the existence of another 
specified path between the same two nodes. Path-based inference rules 
may be written using a binary relational calculus notation. Node-based 
inference allows a structure of nodes to be knferred Erom the existence 
gf an instance of a pattern of node structu~es. Node-based inference 
rules can be constructed in a semantic network using a variant of a pred- 
icate calculus notation. Path-based inference is more' efficient, while 
node-based inference is more genera,l. A method is described of combining 
the two styles in a single system in order to take advantage of the 
strengths of each. Applications, of ~aeh-based inference rules to the 
representation of the extensional equivalence of intensional cohcepts, 
and to the explication of inheritance in hierarchies are sketched, 
- 
;'c 
Preliminary version of a paper to be presented at "Theoretical Issues in 
Natural Language Processing," the 1978 annual meeting of the Associatian for 
Computational Linguistics, UrbanalChampaign-, Illinois, Juw 25-27, 1978. 
Processing of Inierences 
Rand Spiro and Joseph Esp~slco 
University of Illinois 
Abstract 
The I~ypothesis that pragmatic inferences @presented in ~ext are taken for 
granted, superficially processed, and not stably or enduring1 y represented 
in nlemory was invcstlgatcd. Stories were read which in sorne contii t i~ns con- 
tained inforniation vitiating the iniplicational force of explicit1 inferences. 
The vit iat fng informat ion was presented eittrqr before of- after thc inferences. 
In Experiment I, errors ir memory for the inferences were prevalent in th~ 
"aftert' but not the "beforet1 condition. Two kinds of errors were made: 
saying the inference had not been presented in the story; or, if it was 
remembered as having been presented, alter~ng the specific content of the 
inference to produce the opposite f what was actually presented. The latter 
errons produced coherence with tii'e vitiating information, and subjects were 
not able to differeptiate these errors from correct responses. In Exp~riment~ 
11, the results of Experiment I were replicated, ar~d a "spontaneous correction" 
interpretat ion was rejected. The results of bath experimenlts combine to sup- 
pfct the hypothesis of superficial processing and unstable representat ion of 
expl ici t inferences. The resul ts proviae a l ink betkeen processes ~~curr ing 
at comprehension and recall in the State of Schema model of accomodative 
reconstruct ion. 
American Journal of Computational Linguistics 
DICTIONARY SOCIETY OF NORTH AllERICA 
SPECIAL MEETING 
ROOM 4078 LEVIS FACULTY CENTER 
CONTACT: Dr. Ladislav Zgusta Phone: 217 - 333-3563 
Department of Linguistics 
4085 Foreign Languages Building 
University of Illinois, Urbana 
KEYNOTE LECTURE: YAKOV MALKIEL, DEPARTMENT OF LINGUISTICS, 
UNIVERSITY OF CALIFORNIA, BERKELEY 
The Lexicographer as a Mediator Between 
~inguistics and Society 
Other lectures ificlude: 
FREDERIC G. CASSIDY. Computer Mapping of 
Lexical Variants for DARE 
JOHN J. NITTI, Computers and the Old Spanish 
Dictionary 
NEIL H. OLSEN, Computational Cexicography at 
the University of Ha8ai.i - Methods and 
Applications (participation tentative) 
GEORGE FARR (NEH), Funding Possibilities fbr 
Lexicographic Work (participation tea rative) 
HOUSING: Illini Union $15.00 single 
Univecsity of Illfnois $21.00 double 
Green .Street 
Urbana 61801 
Make reservations directly with 
the IJnion. 
American Journal of Computational Linguistics Microfiche 76: 38 
NCC '79 PERSONAL COAPUTING FESTIVAL 
PRESENT A PAPER 
GIVE A IALK 
ORGANIZE A PANEL 
DELIVER A TUTORIAL 
Is WHAT WORTH IT? Any and every aspect of personal computing 
is being questioned 
WHAT IS IT WOKTH? How is personal computing enriching our 
individual lives, the lives of our families, 
and improving the quality of life in general? 
1s IT WORTH WHAT? he maney, the time. the effort, the acquiring 
~f technical expertise, even the criticism. 
Potential participants should send a $'letter of intent" as soon 
as possible, but no later ~han February 1, I979 to Jay P. Lucas 
The letter should include an abstract and a brief biography 
PAPERS presented during the .Logram will be published. Potential 
authors will be mailed a Festival Author's Kit with instructions 
and materials. Papers must be received by March 15, 1979 in 
the specified camera-ready format. Authors will be notified by 
Kay 1, 1979. 
P&?ELS, TUTORIALS' AND TALKS: Session leaders should submit a 
brief abstrac: describing either the scope of the proposed ses- 
sion or the tentative title of the presentation by February 1, 
1979. The prospective organizer should submit a list of pro- 
posed participants, their affiliations, and a brief biography of 
each, 
FESTIVAL CHAIRMAN 
Richard Kuzmack 
1435 Layman Street 
McLean, VA 221,01 
JOINT PROGRAM CHAIRMEN 
Russell Adams Jay P. Lucas 
3008 Mosby Street 
3409 Saylor Place 
Alexandria , VA Alexandria , VA 
22305 22304- 
701 548-8261 (home) 703 751-3332 (home) 
American Journal of Computatiqal Linguistics Hicrofiche 76: 39 
1979 NATIONAL COMPUTER CONFERENCE 
JUNE 4 - 7, NEW YORK CITY 
How TO PARTICIPATE: Write a paper. 
Propose a technic%k or panel session 
Volunteer to be a panelist. 
Send ideas for topics, 
Suggest special activities. 
SUGGESTED AREAS FOR PART FC I PAT ION : 
MANAGElflENT SCIENCE AND TECHNOLOGY 
APPLICATIONS SOCIAL IlPLICATIONS 
GUIDELINES: 
PAPERS: Should be previous1 y unpublished. Must be in 
final form with quality figures and tables. 411 papers 
will be r~efereed. 2500 words to 5000 words. S,ix copies 
of the paper should be submitted along with six copies of 
a title page containing a title, 150 word abstaact, 4 to 
6 keywords, authr's affiliation, telephone number and 
mailing address. 
TECHNICAL OR PANEL SESSIONSr Proposals should include 
a topic description, suggested session chairpersons and 
presenters, panelists, and indication or importance of 
session and anticFpated audience. 
SEND SUBMISSIONS BY NOVEMBER 1, 1978 TO THE PROGRAM CHAIRMAN. 
CONFERENCE CHAIRMANz Merlin G. Smith PROGRAM CHAIRMAN 
T.J. Watson Research Richard E. Herwin 
P.O. Box 218 Box 32222 
Yorktown Heights, Washington, DC 20007 
New York 10598 
American Journal of Computatiol~al Linguistics 
Microfiche 76; 40 
FOURTH JOINT INTERNATIONAL CONFEMNCE ON PATTERN RECOGNITION 
November 7 - 10, 1978 
Kyoto, Japan 
Sponsor: IEEE 
Codtac t : Professor Makato Nagao 
Department of ~lectiiical Engineer irlg 
Kyoto University 
3akyo, Kyoto 606 JAPAN 
ACM '78 
December 4 - 6, 1978 
Washingf on, D . C 
Contact: Richard Austing 
Department of Computer Science 
University of Maryland 
College Park, MD 
COPPUTER ELEMENTS WORKSHOP ON PUTTTING A MATURIkJc TECHNOLOGY 
TO WORK 
December 11 - 14, 1978 
Mesa, Arizona 
Sponsor: IEEE - CS 
Contact: S.M. Neville 
Bell Labs, Room 2B438 
Naperville, IL 60540 
CONFERENCES 
145th ANNUAL MEETING OF AAAS 
January 3 - 8, 1979 
Chicago, Illinois 
Contact4: Dr. Arthur Herschman 
1776 Massachusetrss Avenue, NW 
Washington, DC 20005 
Exbibi ts Dr. Edward Ruffing 
Only : Scherago Associates 
1515 Broadway 
New York, NY 10036 
ANNUAL MEETING THE COMPUTER ASSOCIATION 
March 4, 1979 
Washington, D.C. 
Contact: Michael lwursriaw 
Suite 1100 
1776 K St., N.W. 
Washington, DC 20006 
NFAIS 21st AmUAL CONFErnNCE, 
March 6 - 7, 1979 
Arlington, VA 
Contact: Toni Carbo Bearrnan 
NFAIS 
3-401 Market St. 
Philadelphia, PA 19101 
ANNUAL CONFERENCE OF THE CANADIm LIBRARY ASSOCIATION 
June 14 - 20 
Ottawa, Ontario, Canada 
antact: Business Manager 
Canadian Library Association 
151 Sparks Street, 9th Floor 
Ottawa, Ontario KIP 5E3 
Canada 
CONFERENCES 
N-ERICAN LIBRARY ASSOCIATION ANNUAL CONFERENCE 
June 24 - 30, 1979 
Dallas, Texas 
Contact : American Library Assoc ~ar~on 
50 East Huron Street 
ChiCago, IL 66611 
SIXTH INTERNATIONAL JOINT CONFEWNCE ON ARTIFICIAL INTELLIGENCE 
(I JCAI - 7 9) 
Tokyo, Japan 
Prof. Bruce Buchanan , Program 
Computer Science Departaent 
Stanford University 
Stanford, CA 94305- 
American Journal of Computational Linguistics Microfiche 76 : 41 
RE C 0 G N 1.T-I OIJ I1 E I10 RY 
SEMIONICS ASSOGIATES: TECHNICAL NOTE 
R~L ognition Memory (REM) 
Assoc~at ive hler~~oiy 
C't~i;~l~iitcr 111ctnc11 ici rii,~y bc dl\ ldt>J 1tlrc3 11s o 
~..I\Ic I> ~CS, 011~' of ~~111ih 11.i~ l)ttcn .tl~ll~ct U17- 
bno\\;n lip to now. c\r'cpt .inl\>ng \pz~i,lllgs In 
tI1c \+Y.II-LIIL~\\*II 11 pr ~tort~il it lbt~s Jrr :it L*L\\w~ 
by ilic.ins 'of tllrblr is tl~c :~tlJr~*ss 
I>LBIII~ 3 IIL~II~\\L*~ tI1;it tti~h11 t tfirs 111~ ,IOL tli~tl~l 111 
w ll1k.h the Ilcni 1b 5t r)ft*d 1'111s i\ tllc type of- 
IIIL~~IIOQ u\t-d 111 all thv \i i*ll-Lt~~ni~~l t t3111put tUrs, 
tlirc~ughout the Illstor> s r t'-ft-~tro~~~c cl.it,~ pro- 
. . 
c.t*\<lng I l~t. rt~i,lti\ cl) 11 t tlc-Lno\+ n t > pe 11,is 
6~*tt1 cl~cc~~rs~rl ,111d .~TL',I~III*~ :I~O\I t b? \ c~r~oll~ 
c't3i11l3utcr \c'icnli\ti for sollie ! t-.lrs LICIH , LATI- 
cltbr thtb cJt51;igliatigns A\sosi,ili\*r hic.~~lc>ry ,II~ 
Co11tc.n t- -4ddrt~~s~blc hlcn~ory (CAM). With 
tilt\ t) pc of ineiilor_\p lttlrn ltlay h~ ;YCCYSS~~ 
I 1 11 I if it i\ prccc*llt, tl~o\tl 
lo\ ,rt~un\ u 111~I-1 h,i\c it rc~ c~g!~itc. it ;ilni rc- 
\j-o~id In 111\ rcqL cBt1 t 1x)oA on tl~c* ~tl'l~r-ctl l'ko- 
f~ \\or I 011 l:o\t cr ~L-L rl\v\ t iil- t11 f f~-1 ~~nct- 
bL8t AL>L~~I IIIL*~ t \4 o t> pes of 1llt*t11{>ri~~< 
L [1,11p,i'r:\~)r1 \+ 1111 .I \it II~I~IOII 111 1\1+1~li ~r I~*~r.?~t~r 
\$*1111\ to l.~ltl\\ \41~1~11 3fllL~:~llt. 111 111.1 Ll,!\\flll 
.ill> . ?;.ii tm ,I ~JJI i~c 111~1 1)c)oL If lit* L\;\L*I .IIL*\ 11 LC 
7 ,t~~1i~pl~1i~r \$ 1111 t~-~c~r~+~*i~tio~l,~l SIIL*IIIO~>* (*III~! 
.I\\LIIII~TI~ 110 prit3r clut~\ i:i.~t \$t>~iliI t~li1111!1,1te 
t~r1,lin ~>:,rtlor~\ of thr c~,I\\ f'ro~n ~011~1iicr~l- 
11or-t) lit* \s~)111cl stLirt ,\s 1t11 111~~ t<r\t \:,it 111 t11c 
fi i i I I I I I 1.1 1 I)o, 
J ot~ iiLi\ t 1~0oL'~" I~c \vt>111tj ~IIL*II r tbpcL~t 
1111\ <tcp for 1l1t~ \cL~>nd \t*ht, ,incI \u f~)i;th, 
thro\~gli cicrj \cLif 111 111~ ~l,i\\ If IIC op.br,tied 
IILL* LIII ;4\~1~1.1t I\ .> n~;*li~or> . 011 t 111- oill~~r 11~11d 
1 1 11pl s 1 It \ 1 I1 \5 111 
211 ~hc \tutIcnts u 110 l~.i\t* t111s 14ooA plt~~sc. 
rdlw tl~c~r II~\II~S-" As I hi\ ill~islrL~ti~>~~ sug- 
I I 
MODES OF OPERATION 
I 
t 
it 
HEM, 
I 
REM Functruns 
il,lt;i 121-11 :11a0 I;,ir rccclgnition funt-t ions in- 
~~ol\ 1112 L]\I.I[I 1 i!:ttiite c~c~i~~~~~iri~oris~ ( 1 > gr-c.i:~~r 
th,i~ i\tl\~.iI 40 (2) It>\< tIiii11 or C~L~LI:I~ '1'11~ 
tIlrc>th \.i~it~t~~s of r~~c-c\griitii~i art1 F~lilt ilifc~ flit 
I 1 ti. I~L*~ILI~"s l~~it~g iI\,~l~Icidir c'ctly, 
tli~*y pro\ lde- thc Il,i\is for \*,ttrculs atlti~tlotlal 
f'~~~i~f,t~~~~s tj>c'clfi.i\~1y 15)' -soft\\ drc, \\Ic+~ 35 (I ) 
not t4q\l;il to, (1) _crt*,rttBr 111,111, (3) Ic~s tll,in, 
(4) ~L*~\ICYI~ (I e., gt.-.tttsr tl~ir~ lcnt~~r Iin~it hut 
I:*\\ t 11,1t> ti ppt*r li1111i). 
>!t,rc*ui vr, (l~Sf:~~~*t~f ~~:JI~~I~IIS. it~~~lt~tl~llg no 
SLI 11~ tlor~, ~:ln bi* Ilr\rrirl ~ncd trri ~ilff~br tbn t por- 
t ions of Kt 11 t8~jtrit.l; "Yo fullc-~~on'' rnrdtls 
th,tt .i ~~+rt.iin purilon Zlf ttir RE!? cntry is 
I I d ia 1i.i t~>\ 'r h.tp.,pc:~s t u Fe there 
I$ .~i~~*pt~d 
T\ic 11~ oil tl~c~t. 1 .irlouk t 6~~ul~il~t~c~:~~ pt'rh~lps 
\)tb\t ll~ltfi'~ktO~d \I lfh L~L' Ilclp of ,111 t \.illlp]e 
Lct ~1\ \~ry~;'tsw that \Le l!~ic KE3i sj stcni 
1 ~,~titvl \i 1111 fjlc of cnirltb, tsd~;l~ clttr! con- 
\l\t I~I$ of III~CIIIII~~ !lo11 ,i170~lf Lin indrt ]dual. 
I-llt. cnrr> IS for-~;l,ittt.d. 1t.t us ~y. to cunsiqt 
of tT1c f0110\\ ill$ 
(1 1 3\t ri;tllie 
(2) First IiLirnc and ~nltl;il (or in~tl.ll and nilddle 
1~dlllt2, or . . .I 
(-3) Strzct .idtircs< 
(4 ) ('1 t>, 
(5 Sl,ll.~.,i?lcf /1p ciKic 
(0 1 1 <'IL ph011~ li\i!l;1~t2r 
(7) OLL ~4p.itron 
( 8) 41111\i.il IJICCJ~IIC 
((11 1st 
zo~rld tli~sn. ~f dtwrt)d c>t>t,~ln J II.;~ of ,111 
111c p~~rLons ( 1 ) 11i lng in ,4r1zo11a (2) 110 Inore 
tlldn -75 \ c.<jrc old. (3) ltIi ct~~n~~dl IJILOIII?< of 
520'000 ur iil$~er YOIIC: t1i~1 d~fft r~~nt rciog- 
I-IIIILIIT ~IIIIL-T I(II;S (L~LIJI to lcsc t 1i.ir1 or eqlli~l 
grc.i1tar t11~t~i clr ~*q~i~il t~) ~-~111 1~ p~>r~~)rnicd 
upon d~f'fc'rc'nt portions OF the t'i~~r~ (inc1:~d- 
111s no function dt :dl for the irrcli;\,~nt por- 
lions of tkr t31itr)~) .;21<0, qix~ce the systcn~ is 
pt'ifthctly fleublc as to \!{hat it docs with the 
rcwlt E of 1 lie r~~c.c)gnltio~~ opcrat~on, we m:ly 
a~k it to rc.,id out ji~st a pc~rtion of (rather 
th,in tlw ~lholt. of) the q-\!alif>+ing entries (e g , 
just the nanlc dnd the Jclepllo~lc number) or 
ue m,~y altt.rndtivvly sprcify that some infor- 
mation is tb be K rittcn into some part of- the 
q~~al~fying entr~es, Qy illearls of the mi)lti\vrite 
oiwration Or, one field of the entry 111ig11t 
contiiin an ;iddress to a loiation in a disk flle 
wllcre Inore exteiisive information about the 
~t~di\~~d~iaI is-stored. 
Note that fqi~ctions such as those just de- 
scribed can be performed by ordindry compu- 
ters; hut they ~tould be required to perforni 
starching opcrd tions in place of recognition, 
and ;i seriesaf ~nd~viduak write instr~~ctions in 
place of ~nulti-~i~te. The time of optmition is 
considtrably longer, and it increases sharply 
as the size of the file grows By contrast, a 
recopntze or multi-urjte operatlon can be per- 
for~n~d througl~out KEM about as fast as a 
read or write operatlon, and the time does not 
~ncrertse wrtn the size of the rneniory. 
hioreover, for elaborate specifications such as 
that in the above illustration, tt~e~oftware can 
get qu~te complex in systems uith ord~nary 
(KASI) rl?ernones And rnore cornillex softwdre 
requires not only Inore human tirlle f+r its 
creation, but also marc nicnlory \pace It is pE 
course for just such reawns that wInr 
have been wlllihg to bulld,,and others- 
to buy, ;issoziati\tA procc54ors even a'y 
high prices Chat hdve prciailed untim 
Masking 
A nla\Fc may he applied to any of the REM- 
functions, even to the ordinary location- 
acceqwd read :ind % r~tc operations The mdsk 
1 Sr 
has the functior~ of hJocking out certain bits, 
so that they hre 1111affv~ted by the operation 
Any pittern of I 5 and 0's can be usedl as a 
niask the bit positions for which the mask 
has I partic~patr in the operation while those 
for: which the nurk has 0 arc n~asked out. 'The 
5i7e of the 111a$k is that of tile co~nputer word, 
\&*h~c'h has been set at one byte (eight bits) in 
the first Iil,$I s~~stcrns bcing made available 
by SEXIIONICS. ,4s an exami~lc, the Inask 
I0000000 ~+lould cause all b~t pos~tlons but 
the Ieftnlost to he ignored by \+thatever ~t is 
used wlth. N'i th the 111ulti-write operation, 
this mask I+ ill allow dafd (0 or I ) to be I+ rit ten 
in tlie leftmost bit position leaving the other 
positions uncliangrd Such an operation might 
be used to flag all records whjch have ultisf~ed 
a preceding recognition upelation. 
Effective masking of byic-sized units is also 
prov~ded for, but wlthout the need for okert 
mrtsks Since tj~e C~nf ral Proce~sing Un~t 
(CPU) operdtes upon only one computer 
word at a time. ~t sir~lply or-~tlts cons~dtr~tion 
of those ~vlllch are to be et'fe~tivcly masked 
This slnlple pr,lctice is followed in the abo9e 
~llustration. in wh~ch certain entire fields 
(e g Last name'.) are ignored in specifling 
the recosnltlon criteria. At smaller lei els, 
bitn to the individual byte, it is just as easy 
to us? such recognition criteria as, for ex- 
anipleb (1) last n,me beglnning \wth B (all 
other letters dlsrcgrrd rd), (2) telcplione area 
cede 303, (3) first d~git of the zip code greater 
than or equal to 7 (4) fast nanlc Anders9n 
 here ? indlcatcs uv~ld cl~aracter--i e , byte to 
be ignored) 
Complex Functions 
Syctemr \+bich have t-he ~~iipdb~l~ties dr\crlhed 
above arc ~alled Cont~nt- 4ddrcssablc Parallel 
Procc.s;ors" (CAPP's) by Foster Such ma- 
ch~ner arc qul te po~ er ful B)' ~II(C~H caving 
rtbcugwit~on 'ind multi-\+ r itc ~pc'r~itions, w~th 
appropriate use of bit-rndqhit~g, a CAPP is ahle 
to ric-hicte speeds, flcr;lbi11 t y, d11d pr~r~ril~t~rning 
easc well beyond t hc rlinze uf CL en. 1 ery l~rgc 
and cxpcn\ivc conlptitcrs of the con\ent~onal 
k111d. 
Thus a recogmion opcrat ion, as already men- 
It 1c tlii~r c\,isy to IIILILI~IC ~*ltll2r*or t~nd~tlotls 
111 the rccog~i~l~on c~itcria In the atjo\e cl- 
a~iipie, i~ist~~,~il of :1\L11ig for tilt' rt~orl1s of ril! 
~L~T~OIIS li1111g in Ir~idr~a, OIIC c~j~~ld \ptbcify 
.-Iri~ona, New +l?\ico, or Colr>r,ido l'lic' 3p- 
propri~tc REV s) 4tc.111 s~il~rout i~le c ,in tllerl 
rn~tlti-\\~ite a fl:ig 111 :>I1 'rc.cords \trth Ar~/uria, 
then try Nciv ~IC\ICO rind tli~~l.ti-\+ rift3 rl flag 11-1 
tllc cnlnc position of III~SL? rc~p~~lditlg r~~~)rtls, 
then lihc\\\ i\c for Colorddo; after \ililch those 
records nith the fl,ig  ire the OIIGS s~tlsf! inp 
the ci~~jt~nctl\ cr~tenon 
Other conlplew opcrLitions made povible by 
the ~h111kcs of REV hcl~~de (1 1 incr~~~~~cntlng 
the count flcld of ,ill rc~orrfS rnect~ngsapeclf~t~d 
rCco::n;t ton ~rit.cr13, (2) 1711 b) i)~t coIIiImriso11 
of 317 i~lptif pit tcrn with stored patterns, (3) 
1 
,o,ating the rthcord Ii,i\ 111g tIie rnaxin~urv \?slue 
for a hpec~ii~d byte por~tlon or field (e g the 
count field), (4) like (3) for nlinirnurn idli~e 
(~l\t.f\~l. In rll~hc~l~~t,~~ wrting sincc dlpl~dbctlc 
order cc)rr ~kspon~is to numeric 111 stdnd,ird b.1- 
n,lry ~.odt.s f'or ,~ll~li,iht~tic clr;ir,tcicrs, (5) find- 
ing best f~t 111 p,it tern rcLc~gnit~on s~t\~,itiotis in 
hlLli tl~clre 1s not I~h:.ly to 1~ ,l perfect ~lidtch, 
1 i 1 1, 2 .III~ (3) CG) pr~tlt 1ng 
(1~11 ~1.1 o~Jc.rt~d 1i\t (t~.i\t~I on cilpl~Lll~c.t~c or nu. 
lncric c,r~lcr of' ~l~~~citt~~d fi~bld) of ,111 rz~ords 
IL~I IC I or crlttbrla, (7) 
[no\ 1112 r~iform,it~on fro111 one flsld to ,inother 
\\ ithn :ill rtbcordc or ;ill r ctords nlcct~rjg \pee- 
1f1t.lci r~~coyllt~on ceil tcrlrl. (8) L~d~l~~~g n conctdnt 
o (or \~~l~tr,tc ting frorn) ,311 records h:i\llng 
ptb~~f~~d plcspcrt~~s, (9) ~(l(1111g 1 iilo ficldl; (or 
t~\)tr,i~.ting, c~ric fi ym the other) with it^ :tll rec- 
\rc!q hrl\ III~ ipccif1~4 pr~pc'rtics; (1 0) ar ious 
0gi~,i1 0pcr~11mis 1iL)on fldgs, tbtc, 
The REM Data Systern 
Kr!lF <.'in bc p?~cAagt\d \s~tli \,trious types or 
equ~p~iit.tit htlhides the CPU lor ~llffcrcnt appli- 
cdt~ons 111 the typi~al qtstrirq, the CPU is also 
contiectvd to some R.qhl (for progrdms and 
otllcr itifc~~~ndti:)~~ not rt-q\lirl+~lg the somewhat 
norc c\pcnclvc KC!!) and tn various pvripllcr- 
sl Jt.\'~crs, such as d h~)bo,trd for input, CRT 
~.~id/or pri~iter for output, ~i~id e~tcrndl stordge 
on 1,ipe dnd or disk X t\.pii~l KEhl Data Sys- 
tvin rn,i>+ be dj:~grcil~~i~lrd as follc?ws 
TERMINAL 
-3-3- 
-- - -- - - - -- J ---- 
. --7 __ __--_-._pp- 
Recogri~tion ICSernory (R EM) 
111 g~111.ra1, 1<k-VYls ~1~f:il \\ ~ICIL\ cr \:.~I~~'~III: 
C' 
is rcqulrzd 1 I coll~pu t crs ,ind 
\j Iicrt'\cir orCl!~~dry ~(jlnpl11er ~rlft\\ ,ire s!*ktc.liis 
.ire being 11-t.d for ISI~PXIII~ or oi1lc.r ~I~;III\ of 
hecp~i~g trdcL of \\lidre r1,lld:iq slurtd Its :id- 
\ .~rit.!g~s 11e 111~1niy 111 sr~ ,it11 lncrcJ\i t! q~:t~d 
of jvo~c\cins and In SI;I~~~I~~L-,I;IOII of \oft- 
\+.'ire. In aJditlor?, the ~nulfl-\\r~fe ~~ip(lbiiit) 
rind the ope.:a-tkil~ it rrl;ihes :I\ L~~lL~k~le. WLII 35 
pl~~allt.I ail-ilrnetlc. open up n:~ \I\~J< in 
coinputc'r ctp;71~c ~t~ons \+II:zII ~IC.ISI~III!II~TS 
and ~!~>tr.n~ dt'slgi~crs \s 111 'IIC ~\rIorin$ Tor 
nl'in) 1 cars to cortle 
SII~~ IiEll c'in 2'0 e\er\tl~~ng that c)rcl~ndr)r 
RX3I zdn do In r~dditlon to j~ir~ill~l prcl~css- 
In=, ,rnd since it\ coct 15 onlj 111odcr~tx'l) Itlgh- 
er tl~ln 11131 of TI431 ~'l,ina\ I~~~on-rc ~dc- 
1) 1!~2d ;I\ FZ 4J1 tf~inng- tllc 11cx t ID to IS 
j cdrs 1'111~ proypt'c t yccnl\ t.q3~+:1~!11! liht.l\ In 
IICM of :11c fCic-t 11~1 llL1rilu  IT^^ .oyt, <ire LUII- 
tlnulng tq ;l~>~lln< \il111i. tht. ~l?\i of ic~ft ,ire, 
\~I~IcI~ rtx1~11rti4 p1o_~'raiill11c'r\ colji III!IC< 10 In- 
crt~~ise" Thus tllc \i~p=rf~~~'~ll! grc..ltcJ co\t of 
S) ql.2lll~ c31l Ilt' *r:l~~id IiiLill f lfil~\ O\ t'r 
In ~JLI~~S of progi.trl;r~~~~ig LC~~IS The potinn- 
1131 ~;~,irLet-for RE31 I\ thu\ no 1244 :11,.111 r*:~<t 
and ~t Jpp2nrs pos~iblc tlldt RI.>I \ill1 IL,\ olu- 
t~oni;lc t lit- c-clrnputer 111di14fr) 
0 OJlt' trriII POT~~C~ ,lII tilt. ll~it'dc's L'IO~JII~L':I~L 
R 1-11 111 l~rlng after ~JIC~!JrdJ1l~l~~'T\ c~~~~l L :)III- 
puter \cit'l~f~st~ pet a ~l~dnce to uorh \iltll it 
But cral arc35 arc irnn~c.tll,~t qly 4py.iii.nr ln 
hlcl~ a RE)! syqtcn.1 hac ~1~ar clcliPc~r~t:~ges 
oicr orcIin~ry cornputen SL~nle nf the ;LI-L~~S 
Applications and Nlarkets for R EM 
1. Pattern Recoynitron 
1t4z/! YIICY~~ J, 6drtlc/, >.i~l> 77 '-5 in Jw,e'c , (1 ~'~~.~;er 
7)7:Lnl In ~t\ 1CSt \lc.;t 1s i~'d1X~'Ij 
SIL'W, 1.h 3,: 16 
]IC)\~I:, to CC):~l;~drC ~)ilt' cd-~c'it'~ pNT)t3 r\ll]l li\* 
b~se 'bf I7,OtrCl ~rn~i~r~al prints D~rt \+i!~c~~ th: s) .t~m 
IS co,nplejc. Ir v.iU sTdn !he enrlre filr In v!erd 
n3nutcs '" 
111 I~WIV~III~ 111f ~LI LBII; !oIJ)~\ !II~~~J l!ioli re- 
I 11~'~ .11 Illdl dc'~(\lllll f-01 Il?Ol L' tll.111 11~111 Of ,111 
~ol:~l?utt*r LI~\,!;L* t0i1.Ij1 I llcb I\L* J~I tj 01 J<f 11 
f'0r ELlllll~l!J .I~ii.'44 10 lllf 01 liltlfll~fl I< 1ll.lf If 
IS tllc LI~ \111;111y !'~l::iC I! I;J~ 1)~ 
\I .~t\t\, t 11~ r~*~oril ii 111.1l \.l!i\!i~*\ t1;c IL*,~LIL~\~ 
o~I\I~c\, 111 ciic-~ t, t 1\31 11 !),I\ ~RL*L~~I ,~\hc*d 
101 13) ~i)~ilr,i,t, <adl1,1.11\~ *:t)!Fii>\il~Y' L,,~II ~c~t 
~LI~II .~II~L*~ t ;IL LC\\ ti3 111Vor1!?.:11~~,1 OIII> l?) 
II~~~,III\ 01' fllc ,111411~>4\ tilt' ~oc,il~~~~i \illclC 11 
I\ 4tCOrc~1 \\'I)L~II Ill:> 11.11 c lo j11/J \~IIIIL il1111.;~ 
they it cltl~cr \~.1rcl1 t 111 ~LI::II 1.11111 ~p!~ 10- 
ioi \L~TI.III> or i a of I 1 I ilhc 
I OI 11 1\11 :oiI 1112 I$ rlii itc IIIIIIIL*~~ ill 
I!\ II\C~L:~IIL$S. 110'bi il~r, ;II.II~II> 1 I;~~,:~I\C it 
lLlLL$ ~1L~\l~~lIlt)~ It \AL)IL\ Olll> 11. t!l< <.1:1,2 
b L 
\L*b~r~l~ h~l). i\ III~II yc1t5 11:i\l1: LI" 11ito ;~II &id- 
JrCqs or ,tp111 O\IIII~I~L~ .t~lLIrc~\ I\ :iI~i:~y\ \~\r~i 
For ~\~i:l~plc. 111 a I L>\L~ )OII \> \I t.111 1 or '1 
p~rt ~~ill~r L~LIIV 1111~. r~ i>\\ 10 .I ~tbcortl rc- 
t]LJIrJCS ( 1 ~JJ~\c!l~csl \ llL~~lit.'. ~illd ( ?) Ifl]> I]L~I?I- 
hcr If ~II III~IIII 1112 p~,~~i~cr ~IC)C\II'I I CIII~!II- 
iwr 111s fr~p nurnbcr Ilc 13 out of luck l~otll 
it~b~:l\ of (1~1~ are U-C~ for tile 11~\ll11is. 
-' L,.,ll \~;~1.. ,i;.~.ji:~ 111k' L~~o~cc of ,i 10\\-.0\1 
, , I I ;I L I t c rliql. fur 
],liL.I !I\.< 11 1.1 ~~~~~\~,t,i~-\~..~~' ril.'\ ~'$11 
I i. ~I,)~II \I~LI; L\~;L I 11~11 bt~jl OIIC l>'~~~Il at 
I! 1 L o t I I ~ii ,I \ A 
1.11;:; JIG 111~~ L)II I!]: ~)iIt~r ~I~ITI~. \l~oul~i 
t! p~~,!ll! -I>< j~ro\r~lltl if ~tli ;I !':I 11 IIII~CX. LOII- 
t :I jl~\f fllLit IIII-;~I~ fro111 c.;lcll 
ICLJJLI I!] 13 l1\.~1> to 11~ IIL~L~~L~~I f01 rcti~:~,il 
~,II~~L'~,.*~ 5!1*1\ Lll. 11111~~~ \I!lL< 11 P Ill l<l 31 I\ 
,l'l!o:!l,lt lL.lll> ll~~l~ti-l~ll?lc~~~~oll:il 
u.11) to Lcj!;kiLl:r l!uu p,n. c 1!1s ~ihii of 
I,c:?~L:,:t't" - 
L 14 In 11t11~1an I~fc to ;ct ;T !'111t of the 
I~\I ,~ljpc t)f j3~~i~~~t131 ~ppj~L.qt IC>:~\ ,*!1tl prod- 
11~ t \ ijl,it' \, II~ GIIIC~C ;IS l.id1d ~JIO~~LYIII~ 
:;,,tL lll:~e\ ,tr: 13: (1, JJCJ \i ,fIi rilor: JJI~ I,IUI e 
~t~yl;~,tic, 1~x1 1lvg bl\t~c LLi~dl\~lAt~eq 
111~Jcr 1111\ I~L.ILI~II~. L*\ .IIII~>~~\ of 1t1~u-c c CJ:II- 
6 Cornp~tet Science 
Col rt.l:~t lo115 arnoiig diffcrcnt fr.,itures 
of q ~~t;ic't~c 4ru;ture 
7. S~rnulation and Modelling 
PROGRAMMSNC REA 
is easy! !!?henyour REM system is operating in REV mode - that is, 
executing a recognize or multi-write instruction - the CPU thinks 
it is dolng a write operation. The RE14 board converts it to the 
desired REM action For a recognize, the data written on the data 
bus by the CPU is a cornparand - the Ctem to be recognized. Further 
details axe given in the RE, Programmer's Manual, a copy of which 
is included at no charge with each order of 1 or more REV boards 
It takes only minutes to learn how to add REM operations to your 
repertoire if you are accustomed to programming machine language, 
assembly language, Bbsic, or what-have-you. You can use any 
assembler with REM, since all the CPU is doing is executing a 
write instruction. 
To make thing 
REM subroutin 
functions as 
s even 
.es for 
the £01 
easier, 
complex 
lowing 
SEMIO1lICS has mitt 
REM operations, inc 
(all of which operat 
en a package 
luding such 
e in paralle 
all REM records) : 
- Identify all records which have a specified character 
sequence 
- Identify all records with zero (or all 1's) in specified 
byte position. 
- Compare (=,Z , or s) specified character string to 
specified field of all records, with responders Flagged. 
- Locate first responder (useful if there are multiple 
responders to a comparison) 
- Erase specified byte of bit position in all records, or 
in all tagged records. 
- Count number of bits in given byte position which match 
specified byte (for all records in parallel). 
- Identify the record having maximum (or minimum) value in 
speci fied field. (Very useful in sorting, or for reading 
out responders in desired logical sequence.) 
- IJrite specified data (1 to 255 bytes long) in all flagged 
records. (Very useful in sorting, of for reading out 
responders in desired logical sequence.) 
- Increment (or decrement) specified field of all flagged 
records. 
- Add (or subtract) specified binary number (1 or 2 bytes) to 
(from) specified field of all flagged records. 
- Add two fields (or subtract) within all flagged records. 
- Move data from one field to another within all £Lagged 
records. 
- Boolean operation on flages within all records 
The subroutine package for the 2-80 occupies 2K bytes of memory. 
An 8080 version (which requires 3K bytes) is planned for release 
in July, and this will be followed by an Alpha Micro version if 
there is sufficient demand. A higher level package for general 
applicati ~n in in£ ormation management is also in preparation. 
With or without the subroutine package, yoq will find that program- 
ming for a REM system is far easier in most applications than for 
pre-REM computers. Table loop-up becomes trivial, sorting and 
pattern recognition become easy. You get more programming done in 
less time, and the programs qccupy less memory space and run 
faster. 
And you are able to do more. Information systems can be morc 
flexible. REM makes it feasible to bring new sophistication 
into such areas as data base management, language data pr~cessing, 
and artificial intelligence. 
SEPlIONICS ASSOCIATES, 41 Tunnel Road, Berkeley, CA 94705 41F34U-L4UU 
- 
G % -5 + -2; @ -=~.=71%1 c,ir;ti~:;a: eu9zutd for all Pi,?. resJ.-as 
-ad A'- 
- 
4 d 
* * L.,. G. hq -- -- 
.. - !- :*a- =-k. i:1,3*1. ?'Lr,OG Cbim? i5 4 ,k,f~~p5C?35:3d~~ YdO ='ttter how 
-+.-,* "-W ; -. ." jz ;-r's 
a& -%r .--. r , 4 L.: LZ- -it drscs ~ot 2ncr~sst3 F;iLh &r.c=c,tsfSd zmoqi 
--.C-. 
EL=. K3cb st..;: T'.. - . - T?:~SS~+;! in p~~ral-lel.c~~-+i~ay, tt-ylt",pl eF?~?! 
- t 
-\-. - .9* 
're- _r4slfr.~r c . , ""3 - -. - s 3; :,cii. , ,5a -.;sk--f;.-~ de!Zred es:bizatlor 
sf E. '..I;: -- L I= i~-~: - -- L *n-tt,*~c-,.,?L-.z-~e or :~\lt.f--~~te, i3-~t Also vith 
-" 
-,I 8 l' 
- 
* 
-, 2 I 
La .-. -:-;- . : . 4 . . 2. zc: a_i:s-i* ::CS~T.-,GT,S~ :\l,i-ri? t5e .--.SK 5:s 9 
. . 
r;-+ :> -- rl) - 
Ainerican Journal of Computational Linguistics 
Microfiche 76: 54 
from Micro Divcrslons 
RoheAs Information Scnices, Inc. (ROBINS) 
6.305-G hlcrrlfield Ahe 
Fairfax, Virklnt~ 22030 
Phortc ('03)SW '8438 
SCREENSYLITTZR is a complete text-oriented TV display system 
for S-100 cojnpatible hobbyist, small business and OEM 
microcomputer s st,eas, , SCREENSPLITTER represents a significant 
advance over ot er TV text display systems in three important 
respects: 
h 
* The q~antity of displayed text is large. SCREENSPLITTER 
displa s 40 86-character Lines of ,upper-lcwer case text 
on a s f andard (10 mhz or better) 'kV monitor; 
* 
Onboard softhare is provided. One kilobyte of Jispla 
driver software, the Window Phcka e, is provided in {he 
baslc system directly on ,the boar8 ~n a 1K 2708 EPROM. 
This 3oftn.ar.e clef ines 20 ,user-callable functions that 
permit the definition and control of any number of 
~nput/output tlwindowsn on the display screen; 
* The character generator is user-reproerarnmable. Srnce 
SCREENSPLITTER s character generator 1s a 2708 EPROM, the 
user may design and lmpleinent his own character set. 
The SCREENSPLITTER hardware is mounted on a s~ngle, high 
quallty S-100 corn atlhle PC board. The kit comes complete with 
sockets far all 4g ICf3, all dl screte components-, and 8 feet of 
coaxlal cable with connect,or. 
SCRLYNSPLlTTER occupies one slot of the host S-109 buss, 
re ulring agproxi~ately 1-5 smps at +8 volts, 100. ma at +18 volts, 
an! 100 ma at -18 volts (all unregulated). Virtually all 7480 
serles loglc is low ower Schottky, and the board presents no more 
e than one standard TT load to the host buss. SCREEIJSPLI'I"rEH1s 
output daka buss llr~es are capable of drlvlng 20 standard TTL 
loads. Table 1 llsts all S-100 lines used by SCREE'NSPLITTER. 
The text dlsplay buffer 1s 4K b tes of statlc RAM (8-2114's). 
K 
Onboard softuare resides in a 2'108 E ROM with another 2708 EPROM 
servlng as tne character generator; The IK byte software 2708 mav 
be up radea to 2716 (f-or 2K of onboard software) via jumper 
recan !? ~guratlon. 
TV sync 1s generated by the Industry. standard I415320 sync 
generator, using an onboard 10 mhz,crystal tlme base. This, In 
conjunction w~th the use of s ncbronous counters In the dls,play 
loglc,,results in a very stab ! e and crisp display. 
For extenslbilit and corn atiblllty with other TV display 
systems SCHEENSPLIT'I' 8 R nakes ! he 14 si~nals shown in Table 2 
available to the outslde via a 16 pln D'IP socket at the top of the 
PC board. 
* 
Table 2. 
- 
5-100 buss lines used by SCREENSPLITTER. 
Mnemonic Function S-100 Pin 
A15 address line 15 
A14 address line 14 
A 1 
A11 
address line 1 
address line 1 3 
Bq 
A1 1 address line 11 
address line 10 
$7 
'lo address line, 
i! address line 
address llne 
I 
A aodresa line a 
ii 
8 3 
82 
S 
address line 
3 
29 
address line 
A3 address line 3 
A2 address line 2 
A 1 address ?~m 1 
A0 address llne 0 
IS 
El 
dztz-in llne 7 
data-In 1;ne 6 
i! 
DI 
DI3 
data-ln llne 
data-in llne 3 
; 2 
DI3 data-in line 3 
D I2 data-in llne 2 
2: 
4 1 
DI 1 data-in llne 1 
DIO data-in llne 0 
data-out line 
8: 
El data-out 1 lne 
DO data-out llne 
I 2: 
D 0% data-out line 
D 03 
'I 
data-out line 3 
D 02 data-out line 2 
DO1 data-out l$ne 1 
i! 
DOO data-out llne 1 
SVEMR status, memory read 
KWHITE memory krlte 
%I 
PbFIN poker data b~ss in 
6g 
PSYNC power sync 
PHI-2 second phase clock 
5: 
SWO-NOT status, meaw or outp 
SOU T status, outp 
P RDY processor ready 
ii 
+a 
7 2 
power supply 1,51 
+18 poder supply 2 
- 18 po~er supply 
GND pcwer supply 220,100 
Table 2. 
P 
External signals genenated by SCFSZKSPLITTER. 
PIN 1 10 3hz c10ck 
PiN 2 10 mhz clock, lcverted 
2 PIN Cmpo3it.e Sync, inv5rLed 
PIN H~rlzontal E:ank.ing 
Pm a R~rizontal Elenkilg, inverted 
Vertlcal Elaqking 
Vertical Elanklng, imerted 
FIN 9 Flnal Vldeo TTLy,l~vel (less sync) 
PIN 10 Co=po~ite video, 2 volts 
PIN 11 Chkracter Ceaerator, bit 
PIN 12 13 aracter Ger~erator, bit 
Chzracter Winver 
g:: :$ Yldeo Suppressor 
tp 
i 
Logical 
SCKEENSPI,I'T'~'P;R 1s designed to be lo~ated In t t~c \lost syst c.~' 3 
addt-css space on ari even 8K hau~~da~-y, ds s~luwn in Flgure 1 
Addt-ess lines A15 A14 3nd A13 h1.e joltper s~~lcctable tlo allow user 
sclt?ctlon of the 8K b~~tlddf'y. 
The dlsplay buffer qycar-s to the )lost CPU as a 4K block of 
450 ns (or faster) st.atic RAM. 01.rani~at-ion of the dle lay hilffer 
is ns a 40 by 88 b te m-13ay. The ?wst 86 b tes of cat! line 
ond LO ri.lKBc -PI>~I II ~l:~~:-,ctvri, WI e n tile 87th b te 
l lne ident lficat-isn-, etc. 
r unuc,?b """"'"~ e, and i he 8 th byCe :~vallabI e to the user for ~n ,crl-ial 
To write a ch3:-~c.ter onto the scl>trcn the host s stem stores 
r 
an ASCII code Ln the eppruprlate array cell. Hence, a though the 
h'lndow Paelage 1s ordlnarlly used for a11 dls *ay control, the 4K 
dlsplay buffer 1s dzr>ectly- accessible to the P lost ' sy t em. If 
dvslred, the host can actually use SCREEXSPl.i'lrPER's %K ststlc HAM 
dlsplciy buffer for corlputing during pe!llocis for wbich a tn~an~ngful 
displdy 1s not req~lired. 
The Xlndow Packngc softxal-e 3pglears to the host CPU as a 1.K 
block of 450 ns EPRCIM, sltuated ln the first 1K of the 4~ block of 
oenory i~~n~diately below t,he dlsplay buffer In the host's address 
space. Thls 1K is fully decoded, so that the user can ma ?K 
bytes of hls own RAM in the unused portion of the lo~er 4R 6locic. 
Users ~ho prefer to situate the Wlndcw Packege elsewhere In memory 
can do so ,(after softhare relocation) b disabling the-onboard 
2708 EPROM. In thls case, SCREENSPLITT E R occuwes only 4K of the 
host system's address space (st111 on an even 8~ boundary). 
Eecause the entlre dlsplay buffer 1s mapped lnto the host's 
a5dress space, relatively b1 h s eed dls la tr~nsactlons can 
occur (see the Product Spec1 !! ~ca ! lons). b' ~RL~!~SPLITTER'S clrcultry 
1s deslgned to permit the host CPU to run at full speed wlth no 
walt states, and has loglc for suppressing the whlte-on-black 
l1snowV often associated with such unmpeded buffer access. A 
jumper perrnlts optlonal introduction of one walt stste for memory 
accesses, making SCREENSPLITTER fully cornpatlble with 2 and 4 mhz 
280-based systems. as well as 8080-based systems. 
Physical 
SCREENSPLITTERts Window P22kage software module 1s su plied 
{p a prepro-ri~~d 2708 EiROM, orlglned at any requqted 8E 
boundary. klnce the klndow PscAa 708 1s j~~~er-d~g,ra?able to a 
2K byte 2716, the' user can extend onhoard softhare by &not her 
1K bytes. (Kicro Dlversl~ns wlll relcaslng a 7K byte 
2ge-oriented text editor wblch c he po~erful Window 
lackage funct~ons in the basic 1K 
Full object code (wlth svmbol table) and thorough1 
co-c.nenied source code llst lrlgs for the 4 byte k'lldow Xhckage, 
as well 3s the 'dlrrdcw F-lzkage User's 1:2~~1?1 and Appllcatlons 
idates, are provided with both Llts and ass~mbled 11n1t.s. , 
Addltlonall hex llstlngs and lctorial plots of the character 
generator EBROM~~ are provlded o facillt ate user develop~ent of 
custom character sets. 
F 
--- 
at 
Dizplay buffer 
- 
-p----- 
3K 
Unused 
------------- 
1 K 
bi7dau Pachase 
Y-3;ry zap or SZPr&hSPLI'IfER 
in host add-e?s srace. 
- 
::le ctr~?rhrd iilnljow Packrige deflr~es 20 uxr-callable functions 
f~r ca:l:?-nlliqg :ind wrl t ing to the ,$lfipla 
buffer (see Table 3). 
Tibe ttislc un:t or c~nt,l-ol 1 s the wln?~. tlrtual ly sny number of 
I (rectsr~gular subregions ofthe screen) can be defined and 
~~~.it-;r,,?ept ly controlled. A11 status lnformntion for each hlndow 
:s in~llt ?lr,.d in the wlndowts wil~dow cj~-s$_r_1p-b-r Ueck -(luDB), an I1 
tytp blilck of' hAM supplied by ale user dt fie Elrnexe window 1s 
i,jt.:,,?d (ilggr-e 2). In all tt.rar~sactions with ~indcw fl~nctions, the 
L,rt?r ic7dd~ the addfess of soinc kDB into the HL re ~ster, loads any 
a ;:::*lin'?tirs into rc$~sters B, C, D and b, then GALLs the 
Logical 
~c.s:red hlrl\j~\w fur~et Ion. Window FacAclge fttnc tlons generally 
dt.sLr oy all r-2glster-s. 
Kith the exsoption of two references to the display buffertz 
l1:122t1on In the host address space 
the Wi~ljcw Package 1s 
I -c~nt,:ant, 2nd ~-equir*$s less tnan 64 bytes of system sLack to run 
(A spi.'clal, fully reentrant verelon of tk~e iiindow Packcige that 
cont -013 nult lple SZFIEEVSPL~'~"~'ER dlsplays from a single 1K EPROM 
1s akailable,) 
+C.----, 
W 1 Sf I ntstunb~tu 
.---..- 
1 Sl 7 ,:rrali rclw or u4nd0w top 
*------) 
f a I cu-30- lfnn ulthln ulndow 
7-Xi-7 last +qnc!t,v line rinhr 
*-----4 
I SC I bcprcn CO~IM of v!rdow ICTt ell@ 
*----..- 
I CC I cursor cclw ulthln wlqdow 
*------ 
I LC I :a~t u: >dnu rr: w nm&t 
*----- 
I a 1 cc-r ,r cbwactor 
c-----. 
I SP : -o~l raarfinur 
i--iL-l usar hid Fnrr,wr mjcraso, lor ~YL. 
I HH I user hold prwa-wr add-csa, hi@ byte 
.------+ 
+-------- ---- ----- ----- -.------------ 
STATUS BIT1 I It I fr : rc : cm I ho I ob I no 1 na I 
*---------------------------- ---------, 
fg #'el rcund bit 
rr ri8.r aft 
rc v'~lb1a ru-sor bit 
a cu--0- node blt 
ho held blt 
ob nulpu' bu-st bit" 
nc Ll-rad 
Each window deflnes a rectangular region of the ais lay 
scrzen from size 1 by 1 up to 40 by 86. Each vlqdcw can &e 
canlpuiated Independently from all other wlndohs. Overla ylng 
E wlndows are perrnltted, but operations on a w;ndow take p ace 
wlthout re~ard for posslble effects on any overlapping windows, 
Each wlndow has its own set of pnr.^v:let ers gaernlng: 
1. whether the window's tl urdground 1s reversed (black 
text on whlte bsckgroun 8 
2. aiizther the wlndow bas a frarrie (iislble bgrder) 
3. whether the window's cursor 1s Vl-srble 
4. what character is to be used 3s the kinSodVs cursor 
5. hhether the cursor displaying technique 1s to prlnt the 
cursor character at the cursor location, or slmply to 
reverse the f~gur>e/grounu of the character at the curtsor 
loci t I on 
6, whether the wlndow 1s to be held (by callin a 
user-specified "hold processoru) at scroll finie during 
output bursts 
7. the type and degree of scrolling (pop-up or wraparound) 
the wlndow wlll perform at wrndow-full tlme. 
Lpdir.idnmr Rackage Functions 
All the following functions perfoa error checkin&; detection 
of an er31*oneous request occur-s before any alterat ions o window 
descr-lptor blocks or the vislble dlsplay &re made. 
INIT(ch) 
-- 
Tpears the entire display buffer. to char-act,er CH (typically 
ASCII bTank). 
OPcN(wdb,x,dx,y,dy) 
- --- 
Tpens a window of size DX rows by DY columns with top-left 
carner at screen row X, screen colfimn Y; initializes the 
nindowb window descriptor block, WDB; does not clear the 
reg ion. 
CLEAR ( ud b) 
mars the window and resets the cursor to the top left, 
FRAME(wdb ,hc ,vc ,cc) 
Tzrres the Mlndow, uging HC for the top an8 bottom borders, VC 
for the left and right borders and CC for the-four corners; 
reduces tne window's interior region by two characters in each 
dimension, and clears the winaow. 
UNFRAME(wdb) 
Hernoves the window's frame (if any); if removed, increases the 
window's interlor b$ two characters in each dimension and 
cl~ars the window. 
LABEL(wdb,str,len) 
the window has a frame, prints the st-ping pointed to by STR, 
of length LEN, centered on the window's top border. 
LABELS(W~~, str) 
dehaves identically to LABEL, except that the string is 
terfinated bt the string terminator character (octal 3771, so 
tha a lengt parameter 1s not required. 
CURSORCH(wdb ,ch) 
~efices CH to be the window's cursor character; autoaatically 
cosp1eir;ents CH ~f window's flgure/ground is reversdd. 
SCROLL(wbd ,n) 
yxs the ~lndow's scroll p3rarneter to N (0 $or wraparoclnd 
greatzr than zero for Ksllne pop-up). 
PRINT(wdb,str,len) 
-Tints the string pointed to by STR, of length LEN, to the 
window, -starting at the curr6nt cursor osation; LEN Kay be 
from 0 (no characters prlqted) to 65,53g; the cursor is forced 
to the beginning of a fresh line aft"er ~ne print. 
PRIN(wdb,str,len) 
Fehaves identlcall to PRINT, except that the cursor is not 
forced to a fresh' 5 ine after the print, 
PhI%TS(wdb ,str) 
- 
henaves ~dentlcally to PRINT, except that 'the strln is 
terrninztpd by the string tcrrrinator clJa:-acf er (act a? 377), sc 
that a l2ngth par+a:ncter 1s nat required, 
PRINS(wdb,str) 
-maves identically to PRIN, except that the string is 
terminated by the string terminator character (octal 377), so 
tk,at a length parhmcter 1s not required. 
F - - a~'-Y!,l ___ 'tZ(i~3b) _ 
rL)r-ct.s the ~~YI~OW~: cur-sor to the bYglnn;ng of a fresh line lf 
CLEkRL,TGE(udb) -&--- - 
~~1~21-s  he window's cursor llne, rc>scttlnrg the cursor to the 
flrst ~haracter of the llne. 
59CKSFACE(wdb) 
- 
?%zkS-ttie wlndowls cursor up one c?,ar-actcr if not already at 
the k.lndowt s leftmost col~~~n; f--: asps one chal-hcter as ~t does 
SO. 
COI+.WEI?-TNT( db) 
"h%Vfis the flgure/3round of the wir~dow's interior. 
FL~SH(W~~) 
A 
_T_ 
neverses the figure/ground of the, wlndowts interior momentarily 
(aSout a thlrd of a second for 2 mhz CPU). 
PLOT(wdb,x,y ,ch) 
- 
rrints character CH at window llne X, column Y; does not affect 
the cursor yositlon, and c5nnot invoke scrolling. 
MOVEWIWDOW(wdbrx ,y ,ch) 
-'-~~3~s~he wlndow so that the window's top left corner is on 
screen row X, screen column Y; 
IllS anf 
vacated reglon of the 
scrsen with zharacter CH (typically ASC I blank). 
------ - - -- - - -- 
r 
at& 3 V:r4or Pa:. pee functi~n s--hry. 
1 
IN11 
UPEN 
CLEAR 
F=L% 
LLIFPAKE 
LABa 
LAEELS 
CURS~RCB 
SCROtL 
PELKT 
PRIW 
PRIhTS 
PRl NS 
f F,ESYLINL 
rSLEARLINE 
EACrSPACE 
C~~PLPLNT 
FLIS 
PLOT 
W3\ EVIhinm 
str 
f QJ 
CC 
len 
1 en 
len 
SCREENSPLITTER1s character senerator represents chzracters as 
5 bs 8 plxel grids. Two standard char%cter sets (Graphics and 
Selcntlflc) are available, each defiq~ng 32 useful sq~bols In the 
otherwise, non-printbing first 32 ASCII codes.# rln APL chdracter 
generator 1s availzble opt,ionally. 
The standard font 1s an aksthstic upper/lower case design, 
wlth underhang for the lower ca$e cnaracters whlch require it. Any 
character can be deflned as an 1nh3rently w~nklng character by 
progreuinlng high bit 6 of each of the chsracterls scan 11~~s In 
the 2708 character generator. In faat, slrlce each scan line has 
~ts om wlnk bit, it is possible to generate partially wink~n 
characters. (There are two in the Gra hlcs charact,rr set.) 5l&s 7 
and 8 of each scan llne in the chhrac f e@r generator, not used by 
SCBECNSPLITTER, are made aJallable externally for user extensions 
(Table 2). 
The high-order blt of each byte in the 4K TV dlsplay buffer 
1s interpreted b the dl splay loglc as the c:hrdrclcterVs 
f~ gur3e/grounA. T K us each of the 314140 vl slble ~crnc-n ~I;ar2c, tc3rs1 
f lgure/gruund is ~nde~sr~drntly conlrol lnble . 
1. As an lt~terface to a hlgh level Ihngusqe, such as BASIC 
In thls application, each irr~porthnt subroutlne and function 1s s~signad ~ts 
own prlvate wlndow. As ~t runs, ~ach subroutlne or function ca? output 
trace ~nforrnation , status irldlcatlons or user prorrpts through ~ts own 
ulndow. The visual effect would be fiurrles of actlvit 
from window ta 
wvldow, cach lndep ndently scrolling, flashing etc. T K 1s provides a very 
effective Kay to see "ln two dir,cnslcnsn exactiy wljatTs golrlg on lnside 
your pro rams. User tyyeln can be dlrected througn indlvlduzl wlndows, 
giving t E e illusion of rr.aklng it posslble for the user to coriv?rse wlth 
components of a large system lr~depc-ndently. 
2. As a Debugging Dlsplay System 
In thls application relevant status information concerning the executlo~ 
of a program (as output by a debugglrlg packege) 1s dls layed through 
P 
nux3rous wlndows. One wlndow, for example, could dlsp ay the rc lster and 
e 
accunulator contents; another could display the top N items on t ,e s~stem 
call stack; another could flash up ~nterru ts as they were serviced; 
another could dlspla a selected portior. oP central nrmory, such as a 
F critlcal array; anot ~er could dlsplay a real-time/rur,-tlme clock. You 
could even do all these thlngs lr~dependent~ly for a nuvber of subroutines, 
presentin and recording, say, the re ~srers and accumulator as they uere 
at subrou f ine exlt time, You could a y so arrange to have the nornsl 1/0 of 
your program appear in one wlndow, with debugging rnformatlon popping up 
through another wlndow beside 1t when requested. 
3. As a Easls for Controlling Several Keyboards 
In thls application, there are several keyboards, each with ~ts own-wlndow, 
as might be useful ~n multiple player computer games. Each player h.ould 
have hls own area on the screen, into which all lnput typed b hlm wo~ld be 
il 
echoed, and to whlch all output dlrected at him would be writ en. Undbp 
certaln circumstances, a player might be given access to another player's 
window or "partyn wlndous mlght be established to comblne the inputs from 
several users into a sangle window. 
4. As a Easls for Advanced Page-Orlented Text Edltlng 
In thls zppllcation, entlre pages of text are displayed on the scrcen In 
one relatively large wlndow As ou decide to nlove paragrzphs o,r lines 
around, ou issue corr,cands to plcg up a paragraph or line aqd plice lt in e 
sifiall~r Koldlng wlndow. As ou rurr.maSe through the rfialn Klncow looking for 
the >pot at hnlch to lnsert fhe llne or paragraph, the holding wlndow 
rcdaalns flxed. Flnall you issue a corr~~znd to Insert the contents of the 
hoidlng window at a srY4cted point In the main one. The ~dit or would 
perhaps be capable of dl-rectlng ~ts pdltln powers at the text In any 
window, so that you could also modlfy the fine or paragraph in the holding 
wlndow before reinserting it intc the maln wlndow. Plso, u~ing the 
M0VEWIbII)OW functlcn, you could actually lay out h screen of text (as it, is 
about to be pr~ nted on a-hard-copy device) by ~~ovlng 2;lragraph-slze ch~nks 
of text 5round. Mcanwh~le, of course thcre*cc~uld be a very :mall wlr,dow 
up In the korr~er containing a real time cl9ck tlcklng ahay! 
5. As a Easls for Networking and Concurrent. Processing 
In this appllcatlon, you might wlsh to be dolng local cur11 utln , but rcrcaln 
connected to some external computer (over the phone lines7 or &o a ~~rtg~icr 
n~twork of other ersonal rcachlncs qzmllar to yours. One l*lndow would then 
be reserved for a !? 1 1/0 ~n your local compuiatlon, wlth add~t,~onal wind~ws 
through whlch cc~rrIrnunlcat,son wlth the external cocput,ers could occur. You 
TI ht allocate one window for each other ~ersonsl co~puter attached to the 
f i nc work. Thls would enable you to keep a 1 the 110 to the various 
cmrnunlcating co~puters separate from each other, and L,r arate from your 
" 'F' 
local cclrnfbtatlons. Again, party wlndows coil1 d be cctab I shed. 
PRODUCT TECHNICAL SPEClFlCA rlONS 
K,;sinal bcbrd size 5.0" x 10.0" (1.r~ eC~e card fir1g?r+s) 
kcrc~nal' t card t.ei 
(YI Ch cu rp;r,rn&k) 0.75" 
C?--x!al cable ;y pe ?>59 
Cr ?rial c able -rrigth 8' 
rc-xial cdb?e corj: yctor style P' 259 
hJ,tlar of IC1s 
IC pocket style 
4 8- 
icw prof!le, solder 
kard co-struction G-1C c~terlal, solder r?sk both sides 
1.5 +xps @ +8 bolts (~~~rc~ulated) 
150 ma @ +I8 volts ur,rrgulated 
100 rra @ -18 volts t uVir rpulated I 
O9-53dsd pc-er supply 
filtration 
IC; ty pea 
Lo~d~qg of 4ost 
buss logic lices 
Bass drib ~ng ca dbility, 
data output' Pine. 
Timing and* logical 
Video 
Software ( Logical) 
Software (Timing) 
Read w-ste c cle time, 
d:sP~~Y RA~ 
Read cycle time, 
k~?dou Facdage Software 
Nuber of user-seiectable 
halt states 
LocatYon ln hodt 
address space 
Vlsible display size 
Mlnlrcm Frequency rzting; 
TV m~qitor 
Foco,ie?ded monitdr size 
Rrco-~cnded m~nltor phosphor, 
 OW ablent l~ght~ng 
conartlons 
2e2c- -ecded ncnltor hcsphor, 
hl&t- .?blenL lightlng 
rorid i ticns 
Output ievel, 
c:,:p~5lte video 
Pu' ?\it 1-,c~ 
h:*f .ng ~~b-acttr rate 
Cf,,r,?ter ~rld 5?ze 
In',e--co? .--I Y, a?~ng 
Iqter-rm s2acrng 
Fig,~e/gro~.nd COT trol 
si zs 
h, -3er of ucnr-czl?able 
f ~ntt ~orls 
AS%? 1~5~ed location 
hi3ber of !#~ic.al dzndows 
Fcr-.;at of dp~lied 
ob~cct Ylc. lngs 
x! ndqW sif01111~ ~2f lDRS 
k,ricu cc~t-ols 
(All troes are for 
a ?nhz host CPU; all 
tlws dre ap>roxisate.) 
Full qLreen clear 
Full screen cur-p:~tenL 
hindow fl9sh (c~?ay laop) 
h71c print rate 
to a wivldou 
Window nave 
(20 by 40 window) 
1 ami d -, volts 
50 na @ -12 volt8 
10 nicrof~rad elertrolyti (+8 volts 
6.8 ricrrlarad t*;talarn iri, +I&, 
1.O.nicrof~rad t~~t2lum +5, t12, 37 
0.01 ~lcrc, nrad eisc (+5 bcspiklng) 
2114 4K ststic RAH (dlsplay buffer) 
20 s nz gtnrrator 
Ya8 LF~M 
7 00 Zerles Low Fcher Schottky 
7400 Series Staqzird Rdffera 
no more than 1 stsrkard TTL load 
20 s:&~dard TTL l~ada 
500 ns (non,nal) 
500 ns (nrJmlna1) 
0, 1 
any 8h bo~ndary ( JL-per sel~ctable)~ 
40 lines, 86 coluzx~s 
10 mhz 
13" or largep (9" are ac~e~table) 
daroal {fast decay) 
P-39 or slnilar (?low decay) 
2 volts ?batz--rak (qn~iqal) 
75 c.5-- frxlna~)' 
Z/aec (s;[-ox.) 
5 by 0 p:xel3 
1 pixel 
4 +can lines 
high-order bit of each 
dl-play buffer byte 
L U 
any 6K boundary (uscr-spec1 fied) 
(no limit - hi?d~~z cay ?be-lap!) 
both octal and I-e~, 
-rdpiir.>,~pd and h'-*,ne pcp-up 
rwnd, c J- so-, f~?~e, 
LfC;'' At, cut put torbt hold 
American Journal of Computational Linguistics 
The - TARGET - - -- ---- Project's ------- lnter -- -- active ------ Computerized --- - ------ MuIIilin~.ua1 Dictionary 
John Burge 
[bpar tmcnts of Moclern Languages and Computer Science 
Carnbgie -Met lon University, Pit lsburgh, Pa. 152 13 
summary: 
This documerrt is a brief introdilct ion to Carnegie-Mellon University's 
il~teractive con~p~terized m~rltilitrg~ral dictionary. It describes the use of 
Rhis cfictionary both by translatots in the course of their vjork and by the 
irr~rinologcs~s responsible for updating and maintaining it. This discussion 
is placed in the context of the overall effort (known as the target Project) 
to provide attls to translators. A final seclion presents the solution to the 
problem of representation bf term equivalence adopted in Target. 
1. The - - - -- - Target - 
-- ---- Project -- - ---- 
Tarfict -- is an intcrdi~~ciplinary research project undertaken jointly by the Translation 
Center and the Department of Computer Science at Carnegie-Mellon University to 
invcstlgate ana develop computer aids for langua~e translation. 
Sincc high quality automatic trdnslat ion does not seem to be immediately realizable, 
our efforts at introducing computerizahon into the translation task have been directed 
towards providing practical aids for translators. Workin6 with the assumptioil that each 
t ran5lator ca'n be provided with a standard video terminal connected by a dial-up line 
to a remote c~ornputing facility,2 we are exploring primarily two aids. They are (1) an 
interactive multilineual dic tidary. and (2) an environment consisting of (1) plus text 
rnar\iptrlation facilities within a windowed page editing environment. The latter 
research will be described in a future AJCL paper; this document justifies and 
descrihrs only the former, how it ii accessed and how it is built up and maintained. 
' It 1s nn nrnondnd vsrslon of .n informal deecrtptlon of ihs TARGET and TLRMIN protrams d~rnonstrstsd 84 the 
Fere~gn Rroadc~st Information Servlc~~ eemlner on Alda 10 T~RTIB~R~O~B, Wsihingfon, DC, Mayq 1970 
The rfinf~furalron in dally uee by the Tranelnl~orr C~tlter ri CRrnea~s-Mellan Unrvere~ly involves r Lerr Siogler 
ADM-3 t~rminal connected by a 300'baud d~al up !me \a e POP-10 run under the TOPS-I0 operaling aystsm by 
the Computer Sclencfi Department ~nd shared e~multaneouely by user8 worklnt on many d~ffstent ptojectr 
The prima y hotivation for the iriterilctive dictionary is that a technical translator 
may sperid up to 60% of his or her time simply looking up terms. This may include 
unsi~crcssful searches in several tfictionaries, partly because these dictionaries are out 
of date by ttie time ttiey are publisli~d. An interactive computerized dictionary would 
provide effectively '%mmediatcw access to entries and moreover could be kept 
constantly updated at the central computing' facll~ty. 
Currctitly the diclionary contalris spccial~zcd terminology in English, French and 
German in a number of fields Spec~al~nc! terminology was chosen because this is 
often most helpful in practice to the professional translator and also because this is 
wl~ertt the benefits of st ar~dardization could be most immediately apparent The 
lang~ia~cr, were chosen because they are the most immediately useful in the Local 
en\lironnicnt, as were the fielcls (mainly finance, business and iron, steel 
and mining 
fechnoloey). 
The next section shows in some deta~l how a translalor would access the dictionary 
and ck~termine a correct eqirivalcnt The section after that describes the facilities used 
to maititain and augment ttic d~ctionary The interface to the dictionary described in 
the nc.xt two sections reprr~~ents the fruits of continual close cooperatioh over an 
extenclcd period of trme brtween researchers from the Computer Science Department 
and from the Department of Modrr6n Languages Such coopc"ration, while it presents 
many problems initially, is a sine qttn nor1 of success in a venture such as Target. 
Wliile perfornilng initial sttrd~er, for the represcntat~on of equivalence between terms 
tho most central relation in a rnt~ltiline,ual dictionary -- we have departed from the 
common practice of using an alingcral set of concepts realized differently in different 
1angt~ap.c~ Close examination qtiow~d tlint 'ltii5 could not accommodate some nuances of 
meariiriy, in disparate langilagcs ancf warb* not precise enough, for making inferences 
wlicnl a particular equivaler~ce was not already present in the dictionary. Moreover, it 
was found to be less efficient than another method which was investigated and 
ult inlatcly adopted. Some arg~rrncnts proposed for adopting this different method are 
set fortli In the final section of this document 
2. The. --- TARGET -- - --- Program - --- 
TARMT ----- -- is also the name of tlie program used by translators to access the entries in 
the dic tionary while doing their tr arirlat ~on work This section describes how it is used. 
Thc ilttlstrations are exact traces of the interaction between the program (in a roman 
font) atld the translator (in an italic font). 
We are first asked for the term names and the languages we wish to translate From 
(6.e. the source languap,e) and To (i.e, the target language): 
Termt bond 
From Laaquaget en 
To Language! fr 
Now, if there is only one equivalent for that term between those languages, we shall 
get that equivalent directly. In this case Me have a choice to make: 
Term; hond 
From Lanqtragei en 
To Langiraget fr 
bond Chemistryi Theore t lca l Ch~mlstl yj (ch4) 
The N~rclear Itirl~cstryr Nuclear. Enet.qql (at61 
Financial Flffairs - Taratlo11-Customs~ ((1) 
Select Codei 
Let us say the article we are trarislalitlg is in Chemistry. Then we just rype rne 
appropiate code. TIIC~C are in parcnthcscs in the example and are the same codes as 
used in the EEC's Euradicautom sy5tem. Here we select a code: 
Tat mt bond 
From Lsnqu~qe! en 
To Langtraqst Ir 
bond Chemls try! Thooret lcal Chamistryi (ch4) 
The Nuclear Industryt Ntrclear Energy! (a t6) 
Financial lllfalrs - Taxation - Customsj (fl) 
Select Cods: ch4 
and we shall get the appropriate fiche: 
Select Coder ch4 
bond liaison (FR) 
Chemis tt-y: Theoret lcal Chemistry; 
Tlie Nuc l enr Indits try! Nt~c l ear Energy! 
Reference Terms! bond l ng enerqy, 
Term: 
We lia\le been told that the -,amp cql~ivalcnt is used for both chemical, and nuclear 
bonding Had ttierc been furtl~br information, such as a usage sample, a definition or a 
note, we would have been askcd wl~etl~cr we wanted to see it with the question More? 
Answering yes would show th~ information to us. 
After this first use, Target arlsumcs ttfat we are translating from English to French 
Notice t tiat it does not ask us t lie From arid To qtrestions: 
Term i bond 
hond Chemlslryr Theorot lcnl Ch~mistryl (ch4) 
The Nuc tear Industryt llazardsj (at81 
Flnanclal Qffalrs - Taxation -Customs; (11) 
SB lsct Code: fI 
bond emprunt (FR) 
Reference Termar qovrrnmnnt hond 
Termr 
We can override the assumption by lyp~np. all on onp lrne fhe term name, the sourr* 
language and the rarget language. t4cre we check on the equivalent just obtained: 
1 ermt emprunt fr en 
emprunt bond (EN) 
Flnanclal Flffalrs - Taxat ion - Customsl 
Fngl~sl~ and French now become the new anticipated source and target languages, 
respet t lvely. 
3. The - - --- - TERMlN - - ----- - -+- Program -- 
TERMIN ----- --- is the narne of the procram used by tcr ntinologists to augment and maintain 
the di'r ttonar~es. Clearly the facllit~es in TERMJN must be more varied than the simple 
retrtrval fac~lity which is TAHGCT; the facilitrcs in TERMlN are accessed through a set 
of cor~~n~arzds: 
119 1 p 
Create Entry 
Ewtt 
Retrlove Entry 
Edit Entry 
Oslete Entry 
List Contents 
Tat qe t 
f7sr.01 d Transact ion 
Tsrm Hardcopy 
Dictionary Hardcopy 
Rqqenerate 
Rero~er Space 
b 0 
Slat t 
Slop 
(10 qst instruct Ion) 
(to enter a form) 
(to leabe TERttINI 
(to qst term and term Information) 
(10 revise or attgmenf entry) 
(to delele arr enjt y) 
( to l is t lerms ~n one 1 anqi~age) 
(to get tat-get language equivalent) 
(to reco1.d s~5s ion) 
(to mate hardcopy of a term) 
(to mabe hardcopy of a whole dlctlonrry) 
(to correct faully dictionary) 
(to recober space used by deleted and updated entries) 
(to ex'ecuto a command f i lo) 
(to start using an Uption) 
(to stop using sn Option) 
Ttie TARGET program essentially just repcats the TERMIN command of the sape name 
I hf3 rommancis are tllc larti,rt furicttonal utiitr, in trr tns of which we interact ~ifh thr 
clic tionnry. Wc use them when cntrrinu new terms (C~scate Entry), editinp, existing terms 
(Edrt Entry), printine dictioriarirs? (Dictronary Hardcopy), etc. -- a-nd in fact for 
everytliing we do with thd d~ctionary. A short dc~~riptlon of the use of each command 
follows 
An example pale ie reproduced ir) an Appendix, exactly ~IJ il wan ~r~nlmd by our Xerox Grmphica@rlnter 
3 1 Help 
This provicfes on-line access to written comnients on various aspects of the use of 
the TFRMIN program by terminologists All of the twenty or so tekts which may be 
acccsscd in this way were writtcr~ hy the termtnologists (who also ordered the list of 
cornmat~ds as above and provided the brief description beside each one). 
>lie I p 
He lp ant altorlng 
ALfERING is a uay to cot'rect errors ui tl~o~t typing the whole 1 ine 
again. You can use It hy using the Option RLTFRING before you use 
th~ Command EDITr 
>start altsrlnq 
~ed l l 
nLlER1NC ulll contitiu~ t~ntil you eqit from the proqram. 14 you uant 
to go back to the regular way of edi tlnq before than, do 
>stop a1 toring 
This le a silmmarg of RLTFR mode commandsr 
cSPflCF RnR, : addanre the cursor by one character 
control-14: bark crp ona character (same as rubout or B~J 
D r Oalete the nevf charactat- 
J t Join this llne uith the next, ie. delete the next 
carriage-rettrtn In the text 
H r Tqpe this help tevf 
L t List - type tli~ uhole strlnq 
Q 1 Qult edltlnq the strlng ~nd iynore all the changes mnde so 
far 
P : Print string, ptrttlnq tho cursor back to the same posltlon 
T I Trat~spose the n~r t charlrc ~FI- u MI1 tlie one af ter i t 
V r fnvort the ca.;@ of 811 the characters in the current 
uord, startlrig ultl~ tlm nnrt one 
FSC> t escape from inqsrt mode (-<ALT>) 
A c <RETURN : lormlnate edtting ef the string 
Flx r Fldd thn nert charart~t to he typed, ia, 'x' here 
Cx t Change the ne~t ch~rsrtar into 'Y' 
Yx t Y l l l (le. dplete) all characters until the next 'K' 
Sx t Skip to the n~xt occlrrrpnre of character 'x' 
1 r Insert all chat actors t~jpsd in, startlng at the currant 
positlon, ontll an .ESC> Is presscsd 
Hx r Hunch up chnractnrs till the next 'up, then go into 
inser t mods 
R t Replace - dslete tho noxt charactel- and than start 
Inser t ing. 
X t Ewtond - go 1110 end of thn string and start Inserting 
Most of thc tents pro\~ictccl in tt~~s way glvc hints and reminders in an informal manner, 
rat her than a detailed seqiience of it~r~tructions on how 'to use the command, 
3 2 Create Entry 
Tllic, command is i~cd to crcatr tirw rntrics in the dictionary for terms which have 
rjol. yrfQbeen entered Let us quppor.c llrat we have prepared a fiche specifying the 
equival~nt in French of the Encl~qh t~rm bot~d appropriate for the field of Chemistry. 
Space coristra~nts prevent a dct ailccl cic~cription of the interactions lead~ng to the 
entry of this term into the dictionary, but the follnw~ng is a trace of the process: 
>create 
Term: bond- 
Language: en 
Field Classl 1 lcat Ion! ch4 
Equlvalentsr llalson 
L atigtraqe t Ir 
Grammatical Cateqorlesl nou 
Select Usage Sample 
Definition 
Reference Terms 
Note 
Synonyms? reference 
Re f erencs 1 er rns : "boridlng energy" 
Reference Terms: bonding energy 
nou 
I New ferm "I lslson" belng entered In "FR".: 
TET?MIN does wt~at it can to save 11s effort; here it has eerierated a small fiche to 
conta~n each equivalent which docs not already exist. The minima! fiche it generated 
for l~crlson is this: 
l i a i son bond (EN) 
Chemls try! Theoret lcal Ch~rnlslryi 
This will usually need to he aup.rncnt~d w~lh other information, such as the ~ramrnatical 
cate~or y 1hiS is done by uc.ing thp Edtl Entry command (see below). 
3 3 Exit 
  his is how we leave TERMIN: 
T OX/^ 
EXIT 
~ne rrwr IS now no longer using the TERMIN program, but is using the TOPS-10 
monitor. 
3.4 Retrieve Entry 
Tlds command is used to print wliole entries at our terminal. Here is bond: 
>ref rlove 
Tsrm: bond 
L nnqyage: en 
bond lialson IFul 
Chemlstryr Theoretical Chehistryj 
R~ferencs Terms: bond i nq enerqy 
7 here iq otily orie Equivalctit, onr Ficlri Code and one Reference I erm for bond at this 
po~r~t. More conlplcx cntricc,, wit11 tnorc of ttie optional term information, will display 
all thaf irifor~ation ac well Rctr,rvo Entry shows all the information there is for bond; 
only portions may be displayed w~th the Target and Ed~t Entry commands. 
3 5 Edit Entry 
Th~rc is a number of rcasolis why ~t may be necessary to edit an entry. Perhaps 
the perr,on who cntercd it mad a typo, pcrt~ap~ it is ncccsqary to extend an entry to 
incl\lcjr~, ..ay, a Usage Satnple (or per liaps to replace the old one with a better one) or 
perhap., a new ftct~o mirst be rr~lcred for an exlstlng term. 
Clcdotc _tl7~-Cltl~ocie. S~~ppor~e that further investigation of the term bond has 
r~vcalecf that the same Frcnch cak~~valcnt 15 also used for the "bond" holding the 
nirclcirs of atoms together as for the clectronlc bond which keeps different atoms 
together. Had this been known when the orig~nal fiche was created, it would have 
contained two field codes, cli4 and at8, We need to update the original entry. 
Add o. New f-~chc: The term bortd is also irsed in financial and commercial circles 
4 
(among others ), but here the e'quivalents in French are not liaison 
So we can 
enter a new fiche for this term. 
LIelottrtg --- -- - o Fichg: It is occar,ionally necessary to delete a llche within a term, but not 
ttie whole term. 
Ed~t Entry is used to accomplish all these functions. It is the major tool used in 
maintainhe the dictionaries. 
3 6 Delete Entry 
This is used to delete eptire entries: 
zdslete 
Term: "blast off valve" 
Canquaqet en 
> 
It will no longer be accessible 
3 7 List Contents 
This gives an alphabetrcal l~st of terms in a specifled language. We can limit the list 
by spccifiying the f~rst one or two letters of the f~rst and last term. It may be aborted 
by typing m at any time, and will return us to the command level immediately. 
Let 11s get a list of the English terms from 60 through c: 
tanquaqe r on 
From la t ter: 60 
To Letter: c 
'body centawed pubic 
'bold' 
'bond' 
'bonding energy' 
'boom' 
'bore' 
'hor inq barp 
'boundary pos i t l on' 
'bracb e 1,' 
Nofa lhnt Ihere is nothmg to slop olrlsrpnl rtcrtes for the same !err sharmt f~sld codma A d~ffersnf fiche im 
needed whenever and only when, the equtvalente are dtet~nct 
'breakdown* 
'breaking' 
'bridge connection' 
'broker' 
'brokerage' 
'buffer' 
'bur den' 
'bi~sines~' 
'business prof it' 
'butterfly valve' 
'by means of' 
'by-pa.,., valve' 
'by" x-ray diff rac tion' 
'capacity' 
'c api t al' 
'capital and reserves* 
'cap1 t al gain' 
'capital goods' 
'capi tal-intensive' 
Targrt is the command (or program) we use to find equivalents for a rerm The 
TERMIN command Target works icleritically to the TARGET program (described above) 
with tttc exception that in TARGtT if an abort character cm is typed in answer to the 
Tcrnt: protnpt, the progam exits; in TERMIN, an r;~ at this point gets us back to the 
commanrl prompt > 
3 9 Record Transaction 
This command is 115cd to keep a record of the intcract~on between the program and 
the rlwr It is ~rsccl for t,tt~c!yirre, how ur,ers interact with the system in order that it 
may l~ecorne better ta~lorcd to tlirir ticcdr, (All of the examples in this document were 
drawn tlirectly from rccortlt, madr in cxactly ttiis manner). Each interaction between 
ttic syr.lem and the user may a1r.o bc t~med in the record by using the Tinung option. 
This pro\/itl~s an extra tool for st~~tlyine haw the system is used in practice. It is also 
usefill for some purposes to he able to annotate a record while it is being produced. 
TEHMIN will ignore afiy line beginning with a semi-colon: 
>I Thls shows that commonts gnt Into the record 
3 10 Term Hardcopy 
Thir. command it, used to pr~nt a spcc~f~c term In the same format as that of the 
terms in the Appcnchx 
Tl~cl file j11st g~ncrated conta~ns formatting information AS well as the term itself. It 
must be conzprleci by the PUB document formatter 
3 1 1 Dictionary Hardcopy 
Th~s 15 like Ternr Hardcopy, but thr! program w~ll select the terms for us and put 
thetn in alphabct~cal order. We can choose the language, the initial two letters of the 
first ~ntl la5t terms, atid can also rcstrtct the f~elds for the terms, by specifying a set 
of ficlcf codes -This allows us to make selective microglossaries, choosing perhaps just 
tl~ose terms relevarit to to the Petrolccrm Industry or Medicine. Here we illustrate 
obtaint~~p, an entire dictionary One pae,e of it is reproduced as an Appendix. 
dlct lonary 
Source Canquaqsr tr 
From Letter! a 
To Lettw: z 
Restricted Fields? no 
.............................. 
Please sae TE HP:FRENCH. PUR 
This file must be formatted with the PUP document compiler. The title page of the 
hardcopy will describe any limits we may have imposed on its contents. 
3 12 Regenerate 
This command is *used to re-c:.tablish the links between the various files which 
coot a~n t t~e dictiot~flry. They can hccorne incorrect when the computer crashes while 
certain operations are being pcrformcd, or when there are problems with the system. 
3 13 Recover Space 
Wt~rtn a term i~, stored after tlav~nc been Edtted, a new entry is made for the new 
version of the term Tlic old entry IS, however, still there, and hence takes up space. 
The qatne is true in tli~ CWIP of the Drlrt~ command. When a term is delefed, it actually 
otdy t~ccotncs ~narrcs~~iblc! -- arid r,a ~t 1s taking ilp spatp. Every so often this "wasted" 
space (wl~ic h actually provides Ihe potential for some backup). is recovered using this 
command; it compacts the dictioriary 
3,14 Use of Commands 
As can be secn from the forcegoing, to use a command we type its name (e.g. help). 
llpppr rase anrl lowcr caw ar~ ~q~~ally ac~~ptaI>le. The program will ttien begin to 
prompt 11s for the further spccificationr; necccssary to carry out the command. We may 
Typr /Ihcnd the rcc,ponseq to thc~,e qtrcstions, in which case the prompt Is not given 
Help is u~ually obta~nahle by typtng ? and any command can be aborted by typing @ it, 
re.,ponc,e to any prornpt 
3 I 5 Conveniences of Interactions wlth User 
One convenience in TERMIN 15 that ~t is oftm not necessary to type the whole of 
the rr5pollse to a prornpt In fact, all we need to type is an unambiguous abbreviation, 
t huq: 
rot r 
for R.ctrl~ue Entry, for instance. If we type an ambiguous abbrevation, the system can 
help us out: 
7 re is amhlguous: 
Retr leve Entry (to qst term and term Inlormat Ion) 
Record Transact Ion (to record session) 
Regenerate (to correct faulty dict lonary) 
Recover Space (to recover space from deleted terms) 
>roc0 
7 reco Is amb lquous; 
Record Tran~actlon (to record s~sslon) 
Recover Space (to recover space from deleted tsrmsl 
We can also be assisted when we make typ~ng mistakes: 
. Islt 
... did you mean LIST CONTENTS (TO LISl TERtlS IN ONE LQNCUflCE) ?yes 
As it happens, users of the syqtern often rlnu rhis hind of help confusing inilially and so 
therc ir* an Optmt, callecl Hclpfiil, to control it. Initially this option is turned off, but it 
can be turned on simply by typ~ng: 
st art helpful 
TERMIN can often anticipate the answer to one of its questions. For instance, if we 
have jilst Rctrtrucd a term and then we issue an Edit command, the chances ere that 
the trrnl we just retrieved is Itre term we want to edil. The Defaulting Option makes 
similar assirmptions To use thifiption, we type: 
>start daf aultlng 
and ttien we can utilize it: 
- > retr llalson f r 
l la i son bond (EN) 
The Nuc l ear. Industry* Nur l ear* tnerqyl 
Chemistry! Theoretlcrl Chualstryl 
Re f erence Tm+-ms r I'enerqle de lirlson 
not) mas 
> adtt 
term: t I f e isonl 
Lahquaqe: IFRl 
(Note that we have accepted the cicfault by simply striking the <RETURN> key.) 
When a_a:?lpp, terms, we need not specify accents (unless they distinguish 
between two terms) * Th15 iq a cot~vcnirnce to be used when accessing- and fheir 
coritcntr;: when we type the tcrt in a fiche we must use the correct cases and accents. 
4. Some _ _ _ _#_ _ _ ___. Theoretical _ _ ___ lssucs 
Whilt. clcqigning the representation of terms and their interronnections wilhin the 
dictionaries, researchers at the largct Project disc-overed some difficulties with sDme 
of tt1(? methods adoptrd by okcr terminology banks. This section is a brief 
prcr,entation of some of them. 
Wlicn a scnse in onr Iangt~age is tr~n~~lat~d by a sense in another, they are said to 
be aqiitr/alcnts of each otl~er This rs what is crucial to the translation task, and what 
1s t~nclr-r dlr,cusc,ion In th~s sect~on 1s the reprerentation of cqrtrualcnced between scnses. 
Two m~t hods for do~rig thi.; will, hc comparcd. In one, called the Inter-mediate 
Corrccnt Spaw R~prrrrntclt~otz (ICSF4), tticrc is hrld to he a language-incjepcnd~nt set 
of caiir cpts whic ti are real17~~I in clif frring Iangiiagm cac 11 with the appropriate term. 
Figitre I shows come rquivalt~nl~ t)clwflrti Scl~aufrl (German), Aube (Fr~.nch) and Vane 
(Engile+ti), whlch are appropriatc in thc fkid of Astronai~t~cs. It must be noted that 
I* 
what WP t~avc c_alled "r,eri.,cc, al~ovc arP rcpreqented by their termsnames only both in 
the figurer, and in Ihr trwl; this rl~vicc is u~d merely for clarity of exposltian. 
lnlarmedlate 
Concep t i 
Schru t m l 
1 I 
Sch8kfm I \(,be Vane flube y------- - --+ Vans 
Flqt~re It 1CS Reptesentrtion of 
-- - Ftqure 21 UE Representation of 
VaneeSchau f a l , SC~I~LJ l e l tflt~be 8 fliibs=Vans VanesSchaufei, Schaufelzflube 8 ~uba!?Yahe 
In an altcrnat ~ve mctliocl, the Drrcct Eqii~vnlcnt Repre~ent'nfsrarl (OER), lhere is no need 
for c,trct~ an .iriterrr\rcjiate cotltcpt space. Each setrsc acresues its equivalents directly, 
as c,howrr in Figure 2. 
lCSR is attractivtr hrca~~~~c it offcrr,, a conccptijal elcgancc absent from DER -- there 
i~, a iltjlvcrse of ot~jectr,, each of wl\~ch I~as a cl~ffcretit lir~guistic rcpresentatlon in each 
Ianp;cra~,c. This is a liypothcsir. ahout the nati~re of languagc which- is known ta be a 
rni~llc~~rl~r~p, oversimpl~f~cat~on for ~vrryclay \rcap,p, but its proponents presumably hope 
that it coirld turn oirt to be wrfficie'ntly trire for the more restricted clornains of 
c,prc i~l17ed termiriolop,y 
Tho two major waw of comparltie, thrqc two allcrnatives-are (I) in terms of the 
cornpr~tcr spacr Iab,c.n In liolclinp, tlicm atid timc taken in retrieving them, arid (2) in 
tcrms of their adequacy wl~~n thr dictronary mtr.~t he rnadrfied The former ind~cates 
that irncler some circc~mstancrs ICSf? can be cheaper irr Terms of space, but the latter 
show*, that DFR is rc90tindir1~ly more adequate for the task of representing 
equ~valcncc, and fhus was choc,cn for Target. 
4.1 Space and Time Analysis 
Lach of the lines in Figcrrcq 1 ancl 2, wlictl~rr bctwral, an Intermediate Concept and 
a sctise (Flg11t.e 1) or bctwcct~ two wnr-es (Frgt~rc 2), repreqents what is called in 
con~p~rtc.r parlancr a porrzttr. Pnintrrs nerd <pare in tlic conip~rkr arrd -- pcrbps 
)I 
more irtjportarllly -- take procesqlng t~mc whr.n crscd. Thus to gel from vane" in 
Engli~,li to its Gr?rman rqtrivalrnt (Schattfcl) requlrcs ttie use of two pointers in 
Figirrc 1 (ICSR), but only one in Figtrrc 2 (DFR) This ereater efficiency of DER is true 
for all pair% of eqirivalrnts. 
tl~ff~rc*nccs hetywcn [If-17 atld ICXR .,o Iar art .,pace is qoncerned depend upon the 
ittlnl\)r!r of Inn~rla[:r~c it1 the t~ltlltll~~lr,rral (IIc!IoII;!~~. If tIl(?re arc N lanfiila~~s attached to 
an Intrtnrrcliatc Cr~rrt ~pt, tltrbrc w~ll he N port~trts, orir to each fn the worst caw for 
DrR, v;ir h of the N will hav~ a po~ntrr lo all of t lir (N-1) otlicrs, requiring {NI (N-111/2 
poit~ttlr*, Since tlirbrr arc ttircc Ianp,lrng~s in Fietrreq I and 2 (Englkh, F-renrh and 
Grrman), N is 3 and tlcncfi., tticrc if* no arlvantap,c for c~ther DtR or ICSR The larger N 
t)cc orncr* aimvr 3, lfir* grcatrr is tlic arlvantage for ICSR; for irislance, if N is 5; then in 
flic worst car.c DER rcqrrlrer. tw~rc ac. many po~ntcrs as ICSR and If N is 7, DER 
reqllit rr. 3 timrs a% many In tt~c worst carac Thts worst carqc occurs when equivalents 
are prc>scnt in thc dict~onary for every language, which may not be so in practice, 
cspccially while a dict~onary is being cornpllcd In the most favorable! case, !J=2 and 
DtR tin., tllc advantage by a factor of 2. Dlclionaries prepared for American use may 
oftrn t,c Fnglisli'*X, so that N-? and DER has a space advsdage as well as the time 
advant ,~gc dcmonstr ated above. 
4.2 Modifiability 
Irroc~prctivr of these con5itl~raltnnc, a dlrttoriary must remain functional while it is 
irrtorI~plvtr. To be reaiistic, it i~ pt ol~ably ur~rommon for a dictionary to be "finished", 
anti all ailtomated rlicllotiarir~ rnl~~.t t~c butlt incrementally, equivalent by equivalent. 
Thrrc arr in~portar~t tlifferrt~cc~ het wcvn ICSR arid DER, hoth in' processing when 
rqc~ivalt~ncrs are eritcred and wl~rrl at1 incomplete dictionary. 
We r~ced only consitl~r as strt1pl~ a carbe a5 t tie rtruclures of Figures 1 and 2. Let US 
supposr that none of t IIC rqilivalcnccs vnnr=Schnr~fel, vonp-nitbe and Schnufel-otrbe 
t,a;e yi.7 been inwrted in the tlata-l~af.c and they must be inserted. After entering the 
eqr~ivalrrjc*e vqtw Schalcfel, ICSR will look-like Figure 3a and DER like Figure 3b: 
koncsp t 
rP 
Vane 
- "I 
Schau f e l 
Schaufa l 
(Notr that thiq is a case ~\~ere N--3 ~n thp spare analysis above, and so -- at this point 
-- IC!;F; has two pointers and DFR only one.) 
Now the cquivalcncr vanc-(~ithc is to bo inserted. Wifh ICSR, the terminologist has 
no chvirc hut to tfetcrmine whcti~cr acthe 1.3 cqil~valenl to Schaufel If they are, then 
Figurr! 1 is ohtainrd. Rut si~ppos~ that they are cob !hen Figure 4a would be obtained: 
ntarmodlnte Schaufr l 
V.wvr \ 
flubs -D Vans 
Dgt- be! V~ne,-Schau fs I, Vane-nuhe, -- F Igure 4br V~nezSchrufe I, VanerRtrbs, 
but Schrru f e l #Vatm but SchaufeltVane 
Notc t11rlt for KSR, the tct minolof:ict ir, forced to check every cristcng cquit/aicrtt of a 
tcrm wlirr~ arlrlina arlothcr, a prncc~d~~r~ wl\or9c. cotnplcxity increaqe5 expancritially with 
ttir nirt~tbcr of IXn#tap,~q, Thc tonlpctcrlcc of tt~c tcrm1nolagis1 must extend to all the 
lanp,~~acr.s in the tlataba~~e Wtth IIi R, on ttir contrary, no more need bc done than 
simply arldtng the new eqllivalrt~t c as in Figure 4b. Ovk - - if th~re is a kttt~uj 
~nlrtt/olr~t~rc betw~rn Schnrtf~t arrtl alde will the situation shown in Figure 2 be 
obtarrlcd 
A tempting, but incorrect, solut~on to this problem for ICSR is to assunle that 
artb~-Scharlfel, producing Figure 7 whe(her or not it is actually appropriate. This is a 
kind of risky and i~ncontrollod infrr~t~ce wl*ich ICSR mn naturaiiy force upon the user. 
T hcrc rnay be subtle d~ffcrcnrrr it] nlrat~ing bct w~en languages, yet ICSR forces 
tratir1i!ivity of t rrlation of crj(~ivr71~n~e betwren all langaugec,. There is an 
altcrn;ltlve approach within ICSR, in which th~s is not the assumption, but this will lead 
to pre( iqrly the p~olifcrntion of r~rnr,rs which ICSR was designed to avoid. 
Fur! tirr rnore, t tse simplif iciit lor1 of int crmccliate concepts which are found to be 
recloricJ,rt\t will be a cornpl~catccl prorcdtrre. 
111 qtrmmary, the point so far I.$ that thc addition of an equivalence is a drastically 
more conlplcx procedk~r~ it1 ICSR than In DER, and sccondly that ICSR requires the 
tcrrn~rirtloy,ic,t to- be as multilin~~~al ah; the tlataba:,e, while DER does not. A further point 
may bc rnarle whir h cor~ccrns L~~~CICIICII~~. 
*I 
Infcrct~cing" rncarls ~I~IC~IIIC, a nrar eqkrivalent when an actual equivalent is not 
irnrnrrli.~t~ly obtaitinlllc 0f courc.p, t is to bc l~opcd that ar) automated dictionary will 
~~c~t~ally t~avc an ~rrtt~lrdlcrtc answcr to a uso -r~qurqt for an equivalent, in the setfse 
tl~al t r~cll~f~tcd ~II (la5 prev~o~~~ly been entcrcd explicitly. However, 
qittratiot~~ will ocr\rr wllrrc an inwnc\rfi;ltc anqwcr is not available In that case, 
forrri of LIL fcr wzrirtg may hrlp. W~th ICSI?, that infcrcr~tinp, has already. been done in 
ti t~p tt10 ir~t~rrn~d~atc r.~t~r~~ /ly trwan~. of ~IIC a~si~tnpt~on above, and Hhi~s the 
rtlforrt?al Ion that it is an infcrc3t1cr ic, lo~~t at rrtrlcval time. W~tli DER, the pointers must 
br followcd !lit ough rxpl~c~tly c7r1cl tlltlcr thr s~T~~PI)~ car1 rcport to the user tke extent 
of tlrt\ tc.ntatrvenc*.ci of tllc cl~riv~cl ncar ~q~livalcnt 
I lie cl~c,atl\/at~taf:r, fur tllr! lntertnr\c!i;~le Conr~pt Space r?epresentation, then, is that 
I t t~c orlc hand fincl~ng an cqi~ivalcnt always t akcs two pointers, while Direct 
Fsl li\/alct\t 
R~pre.~cntat~on n~c.ck. ordy one, and on the other -- more importantly -- 
DFH i' nlare able to rcprcsr~~t nuatlcrs of mcanlrle, acror3s languages and incomplete 
qtates of the dtct~onary FIer>ce Target uses the Dlrect Equivalence Representation for 
term eqi17valcnce. 
1 7- Mny 7t3 
CMU TARGET French ~ltctlonary 
frafsc a fflater thread mill (EN) 
1 ron Steel Industries: Machines and Apparatus; 
11 0 11 
fra lse can iqrre counters Ink (FN) 
I rot1 Stee 1 Industr les , Machjnes end Apparatus; 
no 11 
frajse-roars hob (EN) 
I rot1 Steel Industriesr Machlncs and Apparatus; 
nou fern 
frajsos mil Tins cutters (EN) 
t4echanical Enqlnoerln(l Mact~lnos for Movlnq al~d Processinq Materials; 
t ran Steel 1 ndustrles: Maclllnes and Apparatus ; 
nou f em 11 111 
fraf s- I fxes f fxcd cysts (EN) 
Economics; 
Rrfrrcnce ---- -- Terms. frais 
no!) ply 
llsatlr Sctlnplc . . .line not~vel le a~~rjrnrntat<on des f rals f lxes . . . LKre43761 
..- 
fra fS gcncraux ovrrl~ead cllarrl~s (EN) 
Tccl~noloqy and 1 ndustry In General ; 
Financial Affairs - Taxatlon - Customs; 
r~op 
Usagr_Salnpk . . - IPS frals qcneraux (frals admlnfstratifs, de personnel et de 
. - -- 
qrst Ion des pol ices t~ asst~rance). [SGB6/77] 
frtttago frittlnrl (EN) 
Iron Stcel Indu~trles. Ply Iron Production; 
tIi111nq Prcpar-atlon arid Ref ir~lnq of Raw Materiels from Hines; 
noif ma5 
[Icf dl1 l t loll: 
------- -- roast 111q process In stcelmnklnq [fr78] 
frottemcnt frlct ion (EN) 
Iron Stcel Indi~s tr ies : Stress-re1 levlnq Oeformatlon; 
Gcrleral Termlnoloqy ; 
nou mils 

References
Talmy, L. Rubber-Sheet Copition In Laneuagc. In: Papers from 
the 13th Regional Meeting, Chicago Linguistlo Society,
W. Beach, at. rl., ada. Univeraity of Chicago. 1977. 

Col Lins, A*, Brown, J.S, & Larkin, K. (1977 j 
Inference in Text Understanding. Technical Report No. 40, Center 
for the Study of Reading, University of Illinois and Bolt Bcraneh 
& Newman Inc. December 1977. 

Levin J. 6 Goodman, N, (1978) 
Process Models of Reference, Unpublished MS., Information Sciences 
Institute, Marina Re1 Rey CA. 

Lyons, uJ. (1977) 
Semantics. England: Urnbridge University Press, 1977. 

Nash-Webber , F. L. (1977) 
Inference in an Approach to Discourse Anaphora. Technical Report No, 77, 
Center for the Study of Reading, Universlty of Illinois and Bolt Bsranek 
& Newman Inc. December 1977. 

Stenning, K. (1975) 
Understanding English Articles and Quagt if iers. Unpublishted Doctoral 
Dissertation, the Rockefeller Universlty , 1975. 

Webber, E.L. (1978) 
A Formal Approach to Discourse Anaphora. Technical Report No, 2761, 
Bolt Beranek & Newman Inc., Cambridge MA April 1978. 
