Ame6~tU1 Journal of Computationd Lu~gUitti~ Hi ~icr~fi che 32

P R O C E E D l N G S

1 3 T H A N N U A L M E E T I N G

ASSOCIATION FOR COMPUTATIONK LINGUI ST1 CS

Timothy C. Diller, Editor

Sperry-Univac S-t. Paul, Minnesota 56101

Copyright @ 1975 by the Association for romputational Lf nguf $tic@

PREFACE

The 13th annual ACL meeting was held at Boston. Massa-

chusetts, October 30 - November 1, 1975, in conjunction with the 38th meeting of the American Society for Information

Science. The ACL thanks the ASIS for i t s assistance in pub-

licizing the conference and in handling registration.

This and the fallowing four microfiehe8 contain 27 of

the 30 papers presented at the meeting. The breadth o f the

oonference is evident in (a) the modes of communication in-

vestFgated (speech, sign language, and written text), (b) the

styles of communication (monologues, dialogues, and note

making) , and (c) the uses envisioned for @he processing of

language data ( e . g . , theoretical modeling, data collection and

retrieval, game playing, story generation, idiolect charac-

terization, and automatic indexing).

Topics considered include the development of language

understanding systems, the integration and utilization of

specific components of language, specifically syntax and

semantics, the representation and use of discourse structure

and general world knowledge, and the construction of text

processing eystems.

The program committee was so le ly responsible for select-

ing the t a l k s to be given, and hence the papers to be pub-

lished hereln. (Reg~etfully, nearly half of those submitted

could not be accepted for lack of program t i m e . ) Members of

the program committee w e r e Jonathan Allen, Joyce Friedman, .. Bonnie Nash-Webber, and Chuck Rieger. A special word of ap-

preciation is due Jonathan Allen, who a l so served as Local

Arrangements Chairman. Working with h i m were B e t t y Brociner

and Skip McAfee of the M I S . Aravind Joshi, president of

ACL, provided guidance in all areas of preparation.

The AJCL kindly provided advance publication of the

accepted abstracts and now makes possible the publication of

the entire proceedings. David Hays, e d i t o r of AJCL, provided

guidance in publication format and each author provided final

copy in accordance with requested s p e c i f i c a t i o n s . The Center

for Appl ied Ltnguistics (in p a r t i c u l a r , David Hoffman and

Nancy J~kovich with guidance from Hood Roberts) contributed

in a variety of ways, most notab ly in the preparation of

meeting handbooks.

Tkis microfiche contains the papers as submitted by

their authors for ffve of the s ix talkb touching on Language

Understanding Systems. The paper detailing "Conceptua1

Grammartt by William Mattin was too long f o r inclusion in

&baa microfiche and will appear elsewhere. My thanks to

Yorick Wilks for chairing the session.

--Timothy C. Diller

Program Committee Chairman

TABLE O f , CONTENTS

Program Schedule

PEDAGLDT and Unlderstanding Natural Language Processing

. . . . . . . . . . . . . . . . . . . . . . William Fa.bens 9

A General System f o r Semantic Analysis of English and ilta Use in Drawing Maps from Directions ~ e r r y , ~ . H Q ~ S . 21

Arl Adaptive Natural Language Parser P a r r y L. Miller . , . 42

Conceptual Gramar (abstract only) W i l l i a m A . Martin . , 57

Semantic-based Parsing and a Natural-language In ter face f o r Iaterac tive Data Management Ja'm F , Burger , Antonio

Leal, and Arie Shoshani . . . . . . . . . . . . . . . . . 58 PHLIQA 1: Multilevel Semantics in Question Answering

P. Medema, W , J. Bromenberg, H. C . Bunt, S. P. J , Landsbergen,

R . J. H I ScM, W. J. Schoenmakers, and E l P. C. van Utteren . . . 72

THIRTEENTH ANNUAL MEETING THE AS$OCIATtON FOR COMPUTATIONAL UNGUISTICS

Sheraton Boston Hotel Boston, Massachusef t s

October $0-November 1, 1975

Thursday, October 30, 197.5

S&SSION 1: i,AhrCUAGE C/IVn ERSTAArDlllFG SYSTElhf S Session Chairman: Yorick Wdks - h r v e r s l t y of Edtnburgh

990 A.M. Greetings and Irrtroductory Rernarks

9: 15 A.M. PE'DfiGI,OT and Underrl art d i n g Natural 1,nnguagr Procr t sing Willtarn Fabens - Rutgers University

9:40 AM A Syrtcm /or Gencral Scmankc Analysis And l t , q Use I n Drawirt g A l apa from Dircctiotts

Jerry R. t.lobbs - The C ~ t y College of CUNY

tO:OS A.M An Adaprivo Nutural Lartguago Parser Ferry t. M~ller - M I T.

1030 AM. COFFEE & D O N U S

t 1 :30 A.M. Semantic-Based Parsing Arzd A Natural-Langun /ar, Irttr,r farr, Far Intcrraciittc! Data Af artngctnrrlt

John F. Burger, Antonio Leal, and A r ~ e Shoshanl - System Development Carporation

L2a0 NOaN Pl iL lQA I : Mulrilrucl Srrmaniic~ i n Qrrrlrtiort Ancu:crina P. hkderna, st. a1 - Phil~ps Research Laboratorres, The Netheviand$

1290 P.M. LUNCHEON BREAK

S E S S I O N 2: LANGUAGE G E N E R A T I O N S Y S T E h I S Sess~on Chairman: Martin K a y - Xerox Corporation

2:OO P.M. A Framework for W r i t i f t g Ga~~erat io tr Crurnrnnrs for Ints tact ivc Cornputr?t Progrnrnn

Dav~d McDonald - M I T.

2:30 P.M.

3:OO P.M,

3:30 P.M.

4:OO P.M.

4:30 P.M.

5:30 P.M.

8:00 P.M.

Incrarn~ntnl Sonlenco P r o c ~ s ~ i ~ t g Rodger Knaus - Bureau of the Census

A IJoricnl Proces~ Model of IVomit~crl Compounding In English

J.R. Rhyne - University of Houston

COFFEE & OONUTS

Gsnerat in~r ns Parsing horn A Natzrtork irtto a Idheat Str ing

Stuart C Shap~ro - Indiana University

Speech Gcnrrntior~ frdm Scmnr~i ir Nctr Jonathan Slocum - Stanford Research lnst i lute

Using Plnrming Structurra to C~rzcratc Stories Jim Meehan - Yale University

DINNER BREAK

WINE, CHEESE & COMPUTER DEMONSTRATIONS

SESSION 3 : P A R S I N G , S Y N T A X , flND SEhl ANTICS Session Chairman: Joyce Friedrnan - Stanford Research Institute

9:00 AM. Synrucric Procanrirta in tho B R N Spcaclr Urtdcrstnrtdi~tjy System

Madeline Bates- Bolt, Beranek & Newman, lnc

9:30 A.M. Sygtnrn latonration and Coi~iml for $prcr h Uttdrrrrnrzdin/r Wllimrn H, Paxtoln and Ann E. Robinson Stanford Research I ~ s t i t u t e

10a0 A.M. A Tuneahlo Porfarrnancc Grnrnmnr Jane J'. P~binson - Stanford Rosearch Institute

10:30 A.M. COFFEE & DONUTS

lla0 A.M. Scmdrriic Processing f ~ r Speech Underxianding Gary G Hendrix - Stanford Research inst i tute

11:30 A.M.. SPS: A Fortnalim f o r * Scmani ic Itrterlrrr~ation artd Its Use irr Processing P r c p s s i t i o n ~ that Rcj'rrettcc Spacc

Norman K Sondhetrner - Ohro State University

12:OO NOON Tho Nature and Computational Use of n f i lcnning Reproscrrlntion Tor Pard Conccprz

Nick Cercone - University of Alberta

1 2:30 P.M LUNCHEON BREAK

SESSION 4: MODELING DISCOURSB A M EBOR1,D KN111171,1CIIGE I Session Chairman: Carl H e w ~ t t - MIT

2a0 P.M. Ertabliahirrg Conrcxt irt Task-Oricnicd Dinlogs Barbara G, Oeutsch - Stanford Research Institute

290 P.M. Dircoursc Modclx and Language Cotnpr~lzerl cion Bertram C Bruce - Bolt, Bersnek & Newman, Inc

300 P.M. Judging ihc Coherertcy o/ Disrourse (and Some Observations About Frtzrnc?s/Scrip t s)

Brian Ph~il~ps - University of Illinois at Chicago Circle

3:30 P.M. COFFEE & DONUTS

4f i0 P.M. Art Approach to r hc Orgariizatiort of Murtdnrtc 117orZd K r t o ~ l e d ~ a : tho Carterarioir and fifariaacrrterrt of Scripts

R.E. Cuflingford - Yale University

490 P.M. Tho &ncaptuaZ II~h;cTi~t ion of P hysicnl Ar l i . t i i t i~s Norman Badler - University of Pennsylvan~a

5:OO P.M. f i Frarno Artalysit 01 Arn~rican S i g n l,nr~gunae h d y Keg1 (MIT) and Nancy Ch~ncho r (U. of Mass )

5:30 P.M. ACL BUSINESS MEETING AND ELECTION OF OFFICERS

DlNNER: ACL BANQUET

Saturday, November 1, 1975

S E S S I O N $A: AIOl)Ef,liVG DISCOURSE & WORI,D KNOlVI,ItIIGlI: / I Session Chairman: Georgette Silva - System Development Corporation

9:00 A.M. Cross-Sct~t~tztinE R ~ f i r ~ n r c Rr~olutiorz David Klappholz and Abe Lockman - Colutnbia Uhiversity

9:30 A.M. Ilaia Doc$ a Systcrn Knou~ TVhclz to Stop l ~ t f ~ r c r ~ r i n g ? Stan Rosensche~n - University of Pennsylvahla

10:00 A.M. COFFEE Rt DONUTS

S E S S I O N 5i?: TI5XT AiYflLYSIS

1 1 :00 A.M.

1 1 :30 A.M.

D;c?ucZoping n Cornputcr Syatrm for Ilanrlling Iltizcrclrttly Vtzrinhlc I , i i ~ ~ u i s ~ i c Data

D a v ~ d 8eckles, Lawrence Carrrngton, and Gemma Warner - The unrversity of the West lndies

A Nururul I1a~tguagc Proecs.sirtg Pnckrcgc, David Brill and Beairlee T Oshika - Speech Communications Reseatch Laboratory

On the K o l c of W o r d s and Phrases irt Autornntir T e r t

Analy,& and Cornru~niiort Gerard Salton - Cornell University

12:OQ NOON Crnmrnnlicul Comprrssiolz in N o t ~ s and Rcrards: A~talysib and Cornl~v tntiolt

Barbara Anderson (University of New Brunswick), Irwin Bross (Roswell Park Memorial Institute), a ~ d Naomi Sager (New 'fork University)

American Journal of Computational Linguistics H i c r o f i c h e 32 : 9

CGmputer Scf ence Department

R u t g e r s Uni versi ty

Few B r m i c k , New Jersey 08903

ABSTRACT

PEDAGLOT is a programmable parser, a 'meta-parser . ' To program i t , one

describes not j u s t syntax and some semantics, but also-- independent ly-- i ts

modes of behavior. The PEDAGLOT formulation of such modes o f behavior follows

a ca tegor iza t ion of pars ing processes i n t o a t t e n t i o n - c o n t r o l , d iscovery, pre-

d i c t i o n and cons t ruc t ion . Within these o v e r a l l types o f - a c t i v i t i e s , cont ro l

can be s p e c i f l e d covering a number of syntax-processing and semantics-process-

ing operat ions . While it i s not t h e only p o s s i b l e way of programing a meta-

parse r , t h e PEDAGLOT mode-specification technique i s suggest ive i n i t s e l f of

var ious new approaches t o modeling and understanding same language processing

a c t i v i t i e s besides parsing, such as generation and inference ,

7% i s wotk was sponsored by through NIH Grant #RR643.

I t i s well known t h a t t o process n a t u r a l language, one needs both a

syntactic desc r ip t ion of poss ib le sentences, blended i n some way with a semantic

desc r ip t ion bf a c e r t a i n domain of discourse, and a r a t h e r d e t a i l e d desc r ip t ion

of t h e ac tua l processes used i n hearing o r producing sentences.

An augmented t r a n s i t i o n network (Woods, 1970) i s qn example of t h e blending

of s y n t a c t i c and quasi-semantic desc r ip t ions , Here r e g i s t e r s would be repos i -

tor ies o f , o r po in te r s t o , semantics. When used i n conjunction with a semantic

nqtwork, an ATN can be used 60 parse o r t o generate (Simmons and Slocum, 1912)

sentences. The i s s u e o f changing the des'cription of t h e actual processes used

i n such systems has been touched on by Woods ( i n using a 'generation modet), t o

some extent by Gimmons and Slo~um (usi~g decis ion funct ions t o control s t y l e of

generat ion) , and t o a l a r g e r ex tent by Kaplari (19751, i n h i s General Syntac t ic

Procdssor, GSP. GSP indeed is one example of a system i n which syntax, semantics

and t o some extent processes can each be u s e f u l l y defined.

If we look at syntax, semantics and processes a s t h r e e descr ibable components,

these systems j u s t mentioned i l l u s t r a t e how thoroughly intertwined they can become--

t o the extent t h a t t h e o r i s t s from time t o time deny the exis tence o r a t l e a s t t h e

importance of some one of them. Ignoring t h a t d ispute , I would- l ike t o concentrate

on the quest ion of being a b l e t o comprehensively descr ibe one ' s theory of language

i n terms of its syntax, semantics and processes i n a way t h a t allows fo r t h e i r

necessary and extensive in ter twining connections, bu t a t t h e sane time allows one

t o describe them independently.

I came t a t h e need for doing t h i s while designing a Tre laxa t ion p a r s e r , ' a

parser which can make grammatical r e l axa t ions i f i t i s given an Ill-formed s t r i n g ,

so as t o arrive a t a k l o s e s t t poss ib le parse f o r the' s t r i n g . This probl-em involved

descr ib ing a k o r r e c t t grammar and then (in some way) descr ibing a space of deviat ions

az that night be allowed by the paxser. Thus the syntax would be fixed and the way

the parser uses it would separately have to be described. I t was soon noticed

that efficiency could be greatly enhanced i f some rudimentary notion of semantic

p laus ib i l i ty could also be used. I t would have t o be described i n a way related

t o the cbrrect syntax but st i l l be usable by the parser. Thus, for my purposes,

the descriptions had t o be independent of one another.

One feature of a relaxation parser i s tha t it can ' f i l l i n the gaps' of a

string tha t is missing various words. If one could, which my re laxat ion parser

did not, specify the semantic context of a sentence, the generated sentence might

be semantically rather plausible. In any case, the relaxation parser operates i n

various respects l ike an actual parser or like a generator, and it was t h i s re la -

tionship between parsing and generating that became of in teres t ,

Out of the design of the relaxation parser, the notation (independent of syn-

tax) which t o some extent describes various processes and choices of al ternate ways

of processing was developed. Thus, one may take a s e t of syntax and semantic de-

scriptions and then through describing the processing 'modest involved, define a

processor which uses the par t icular algorithm t h a t the individual processes together

define, One may c a l l the parser that i s programmabLe i n i t s processes a meta-parser,

of which various existing qarse r s and generators appear t o be special cases,

A closer examination of the parser I have developed (called PEDAGLOT*) may show

some such aspects of meta-parsing, especially as regards the relat ionship between

parsing and generating. I will describe the syntactic and semantic parts o f t h e

parser first: by noting i t s resemblances to t h e parser of J . Earley (1970) and the

ATN system of Woods. Then I w i l l describe t he process-type specifications t h a t are

available, and the use of meta-parsers as a basis f o r defining general language be-

haviors. Purther detail can be found in the PEDAGMT manual (Fabens, 1972 and 1973) .

*for pe&~ogic polyglot

1. The Core of the Parqer -1

The fundamental operation of t h e parser i s very s imi lar t o the operation

of Earleyvs parser , with augmentations f o r recording t h e r e s u l t s of parses

(e ,g , , their t re s t ruc tu re , and various of t h e i r a t t r i b u t e s , which I c a l l

ftags'). It is given a grammar a s a s e t of context-free rules with various

extensions, most i m p ~ r t a n t of which a r e t h a t LISP functions may be used as

predicates instead of terminals, and thay each rule may be followed by opera-

t ions t h a t are defined i n tbnns of t h e syntac t ic elements a f the r u l e i n question,

An example of t h i s notation i s as follows:

S -+ NP VP

=> [AGREE [REF NP] [VB VP] ]

[SUM = [REF NP] ] [OW = [REF VP] [VB = [VB VP] ]

S -* NP [BE] [VPASS] BY NP

=> [AGREE [REF NP] [VB [BE] ]I

[SUM = [REF NP I ] ] [OBJ = REF NP] ] [VB = [VB [VPASS] ] ]

NP -+ [DET] [N]

=> [REF = [N]]

W + [VINP

=> [VB = [V]] [REF = [REF NP]]

Here, each bracketed symbol i s the name of a recognition predicate ( e .g.,

IN] recognizes nouns, [BE] recognizes f o n s o f h t o b e 1 ) , Following t h e => are

the post -recognit ion functions. For instance [AGREE [REF NP] [VB VP] ] specifies

a ca l l t o the AGREE function which i s given, as arguments, the REF a t t r i b u t e (tag)

of the sub-parse involved in tha t rule and the VB a t t r i b u t e of t he VP part of

t h e rule.

Following i s a parse tree fo r 'The Man Bites t h e D o g l and values o f tags

after the parse.

The Dog

The general flow of the parser is from top-down, and as the lowest compo-

nents (symbols i n the s t r ing) are found, the post-recognition functions tha t are

associated with t h e ru le tha t recognized them a r e applied. Tags become associated

with sub-parses when the post-recognition operation uses the form [x = y ] ( i n which

the value referenced by y i s stored as the x t ag of t h e sub-parse). In the example,

[DET] and [N] recognize 'The Manf and 'Manf i s used as the REF a t t r i b u t e ofi t h e

first NP. In t h e second S ru le , t h e operation of [SUM = [REF NP']] would be t o

retrieve the REF tag of the second NP (thus the prime), and t o s tore t h a t as t h e

SUM tag of the final p a n e .

As i n most top-down parses, t h i s parser begins with S and i ts two ru les ,

s ince S i s non-terminal. S is expanded into t h e two sequences of matches it should

perform. This expansion r e s u l t s i n various (in t h i s case, two) predict isns of what

t o f ind next, When t h e i n i t i a l symbol i n some r u l e i s a terminal o r a predicate,

a discovery is cal led fo r (in which a match is pexformed, possibly involving the

known values of the tags). When some complete sequence of elements is found (here,

for instance, when NP -+ [DET] IN] has matched t h e [N] ) . Construction invokes the

post-reoognit ion operat ions and then usual lyt completes some e a r l i e r part of a r u l e

(here, the 'NPi ~f S + NP VP) So fur ther predictions (involving VP) or discoveries

are then specified.

1 have broken up the parsing process into t\he$e three parts so as to simi4arly

catalpg t he 'parsing modes,' turn ing this parser into a meta-parser. Before doing

so, f should note tbat th i s parser stores each zesult under construction in a

'chart' as is done by Kaplan i n his GSP, so that, for instance, the NP ' testt

w i l l only have to be evaluated once for each place one i s wanted i n the string.

[ N l [;I 1 [ N l 5

The T

Man Bites The $

Dog

I1 lustrat ion of PEDAGLOT ' s Parsing Chart

Simple Arrows indicate 'Predictions.'

Double Head Arrows indicate iDiscoveries,

Dotted Arrows indicate tlConstruction.

Also, for various well known reasons of efficiency, Earley's concept of

independent processing of syntactic events i s used (combined conceptually with

the chart), SO that a main controller can evaluate the individual syntactic ' t e s t s 1

i n almost any order, and not just in a backtracking sense (cf. Woods, 1975). Thb

efficiency i s realized here since many 'partial parses (partially recognized forns)

15

are effectively abandoned if other results can complete t h e parse, o r a sub-

parse, first . 2 . Meta-Parsing Modes

One can see that, except f o r the nota t iona l i ne f f i c i enc i e s o f the context,

free formalism (as opposed to the augmented t r a n s i t i o n network form), t h i s parser

is very much like other standard parsers (especially ATN s) . I t differs i n t h a t

there is a waytof specifying how t o proceed. Currently, this system has approxi-

mately a dozen toodesr and I will present some of them here. Each mode spec i f i e s

how t o handle a certain part of the parsing process. They can be classified i n t o

four categories: attention control, prediction, discovery and construct ion.

a, Attention Control W e s :

Since the parser operates on a chart of independent events ( 'parsing

questions1), one must give t h e parse r a method of sequencing through them.

Thus, one may specify 'breadth-first1 or 'depth-first1 and the appropriate

~echanism will be invoked {this merely involves t h e way the processor stacks

i t s jobs). A 'best-first ' option i s -under development, which, when given an

evaluation function to be applied to the set of currently a c t i v e part ial

parses, allows the system to operate on the 'best1 problem next , Experi-

Bents with this mode have so far been inconclusive.

One also can speci fy when t o stop (i.e., at the first complete parse,

or t o wait until a l l other ambiguous parses have been discovered). The d i s -

it~gbiguation routine (which i s described as a part o f t h e construct ion modes)

defines which parse is %est l , Further, one may specify a left-to-right or

right-to-left mode of how to progress along the s t r i n g .

b. Discovery Modes:

The starting point of building a relaxation parser is to specify what

t o do when an exact match i s not made. If the parser i s expecting one word

and finds another it can look arowd the indicated place in the s t r i n g t o f ind

what- it i s looking f o r , o r it can i n c e r t a i n other circumstances simply

i n s e r t t h e expected word i n t o t'he s t r i n g . Thus, under discovery.modes,

there are vaxious options: e i t h e r t h e parser i s allowed t o attempt matches

in out-of-sequence p a r t s of t he string, or n a t , And i f not , or i f no such

match i s found, t h e parser may or may not be allowed to make an inser t ion .

So i n PEDAGLOT, t h e r e i s an INSERT mode (and various r e s t r i c t e d versions

o f f t ) and a 'where t o look1 mode which i s used t o control t h e degree t o which

the parser can t r y t o f i n d out-of-place matches, There a re tags associated

wi th- these two spec i f ica t ions , t he INSERT t a g and t h e OMIT t a g , which a r e

associated with the parses involving inser t ions and omissions t b a t contain

the number of insertions made and the number of input symbols omitted i n

building t h e parse.

There i s also a rearrangenient mode. mus, given ce r t a in cons t ra in ts ,

the parser could be givep 'The Bites M a n Dogt and produce a parse f o r *The

Man Bites t h e Dogt s ince it would have found 'Man,' by temporarily omitting

'Bites,' but then it looks f o r and finds 'Bitest and f i n a l l y , f inding no se-

cond lthe,fthe,l i n s e r t s one [or some other determiner because of t h e [DET] func-

tion]) and f inds 'Dog. In a similar way it would t r y t o produce a passive

form [i.e., the Man Is Bitten By the Dog) but since t h i s involves more inser-

t i o n s , etc . it would not be chosen.

These h e u r i s t i c s a r e control led by recording numerical summary t a g s

with each sub-parse that p a r t i c i p a t e in , and are judged1 by the disambiguatic;~~

rout ines . Similar ideas are used by Lyon (1974).

c. Predict ion Modes:

As Woods (1975) has pointed out , t he extent t o which a parser ' s prediction

increases efficiency varies with the quality of the expected input. This f a c t

affects greatly our discavBry procedures, since, if inser t ions are to be made,

one aught t o be rather s u m of one's p ~ e d i c t i o n s , o r risk a combinatorial ex-

plosion. In PEDAGLOT, t h e r e i s a programmable choice ' funct ion tha t - con-

t r o l s predlc t ions . Spec i f i ca l ly , when the parser encounters a non-terminal

symbol, t h a t symbol is t h e left-hand s i d e of various r u l e s . An uncontrolled

pkediction (used by a canonical top-down parser ) i s t o s e l e c t each such r u l e

as the expansion. I n t u i t i v e l y , however, people do n o t seem t o do t h i s . In-

stead, as i n an A'I??, they t r y one and only i f t h a t f a i l s , go In to the next .

In PEDAGLOT, t h e choice of which r u l e t o t r y can be defined as the r e s u l t of

the c a l l t o a 'choosef funct ion (or it can be l e f t uncontrolled] , We have

des&ned various approaches t o such predic t ions (e.g., a l imi ted key-word

scan of the incoming s t r i n g , and the use of 'language s t a t i s t i c s such as the

s e t of rules which can generate the next symbol i n the s t r i n g as t h e i r l e f t

most symbol).

The predic t ion is cur ren t ly made once f o r any given choice poin t ; its

outcomes are expected t o be an ordered s e t of r u l e s t o t r y next.

d . Construct ion Modes :

The phase of parsing i n which t h e p a r t s of t h e parse t r e e and associated

tag values are formed, is a p lace where most of t h e non-syntactic information

(tags] about the s t r i n g being parsed can come i n t o play.

In t h e first place, new t a g s can be formed as funct ions of lower l eve l

parse tags tbough a process called melding, Thus, 'nonsense1 can be discovered

d pronoun references can sometimes be t i e d down, In t h e second place, it i s

a r e s u l t of construct ion t h a t ambiguity i s discovered and dea l t w i t h ,

Since these fea tures of parsing deal pr imari ly with semantics (and s ince ,

i f anyrcthere, sttsntantic representa t ions of the s t r i n g r e s i d e i n the t a g s ) , most

of t b PEDAGLOT construct ion modes involve tags .

One play e x p l i c i t l y meld t a g values by using post-recognit ion operators , o r

one nay def ine an 'implicit' melding rou t ine that i s associated with t h e tag

names themselves ins tead of with indiv idual rules. I n our example we use

this device t o i m p l i c i t l y form a simple l i s t of t h e two REF t a g s t h a t be-

come associa ted with t h e S ru le . This implicit melding operat ion can a l s o

include a blocking function, o r some reference t o a d a t a base. The t ags

t h a t contain INSERT and OMIT information are used i n t h i s way t o keep running

t o t a l s o f , and t o minimize the munber o f such h e u r i s t i c s i n t h e r e l axa t ion

pars ing modes. One may a l s o a s s o c i a t e a LIFT funct ion which, when t h e par-

t i a l parse becomes complete, s p e c i f i e s a transformation of t h a t t a g t o be

used as The tag of the next higher level parse.

Ambiguity i s discovered when two parses from the same symbol, cbvering

t he same s t r i ng segment axe found. For t h i s case, an AMBIG funct ion i s asso-

c i a t e d with t ag names, and it makes a 'value judgement1 of which t a g i s ' b e t t e r ,

hence which i n t e r p r e t a t i o n t o use. (Other types of c r i t e r i a can a l s o come i n t o

p l a y here such as u s e r in te rac t ion , (cf. Kay, 1973).

3. The Uses of Meta-Parsers

I ha-re just catalogued some of t h e parsing modes ava i l ab le i n PEDAGLOT. Others,

such as Bottom-Up ( instead of Top-Down) o r Inside-Out ( ins tead of Left-to-Right, e tc . ) ,

are envisionedlbut not implemented. Since PEDAGLOT is an i n t e r a c t i v e program, the

u s e r can change modes a t w i l l , j u s t a s he can change syntax o r introduce new t ags ,

Thus, the obvious first use a f meta-parsers i s t ha t one may use them t o des isn

language processors without having t o t i e oneself down from t h e s t a r t t o say, a

d e p t h - f i r s t pa r se r ,

Meta-parsers a l s o have a c e r t a i n amount of t r a c t i b i l i t y t h a t parsers t h a t

blend a l l . a c t i v i t i e s i n t o one huge network may not . Ono may sea a t a r a the r high l eve l

what is going t o be happening ( i . e , , a l l t a g s of a c e r t a i n name w i l l meld together

i n a c e r t a i n way, unless t h e grammar s p e c i f i e s otherwise) , If one, however, wants

c e r t a i n foms of local behavior, one may use predica tes o r funct ions on individuaQ

r u l e s . Further, i f one wants t o change t h e order i n which predict ions a r e evaluated,

one can program a tchoosel function which w i l l make t h a t global change. To a

large extent, the language designer may specify mch of t h e processor in broad

ternas and s t i l l be able t o cont ro l local events where necessary.

In a more general sense, a meta-parser allows one t o understand and build

higher order theories about how people might represent and process language.

For instance, while it may be true that generating is t h e inverse of parsing,

there is more than one way t o do such inve r t i ng . One could s tart from a senantic

network, using the choose function along with t h e INSERT mode t o restrict means o f

expression consistent with the intendea message, and using AMBIG functions to weed

out a l l but reasonable messages from m n g the many the parser may produce o r one

might simply t a k e from the semantic network a simple str ing o f meaningful words,

and then we a less t i g h t l y programmed 'relaxation parser' t o rearrange these words

to be syntactically correct. We are now considering using a crude 'backwardsT mode

which begins with the operati~n part of a ru l e and, by using predicates (e .g . , AGREE)

to yield inverses, specifies what the context-free pat tern must produce. Thus there

are many variations of how t o generate using a meta-parser.

In the area of language inference, t o take another example of language processing,

PEDAGLOT suggests various differing ways of approaching the problem. First, ofie may

use it a5 a 'relaxation-parser, the 'parse t ree1 can be pattern-matched aga ins t

the new sentence, and hypotheses can be famed. Or, one could place a more rudimentary

inference systw on the 'prediction' part of the processor i t se l f , and using other

controls, the predictions that are successful could be rewritten as a new gramar.

These two learning paradigms could each be strengthened by way of t h e use of tags

t o contain (in a sense) t h e meaning of t h e sentelzces t o be learned, Each of these

paradips can be modeled using a meta-parser like PEDAGLM. Thus, a meta-parser can

raise [and be prepared to answer) a nlrmbor of interesting questions.

References

Earley, J. (1970), llAn Efficient Context-Free Parsing Algorithm,I1 Comm. ACM 13, number 2, (February 1970)) pp, 94-102.

Fabens, W, (1972), PEDAGLOT Users Manual, Rutgers University CBM-TR-12, kt. 19722,

Fabens, W. (1973), PEDAGLOT Users Manual : Part 11, Rutgers University CBM-TR-23, Nov. 1973.

Kaplan, R.M. (1973), "A General Syntactic Proce~sor,~~ in R . Rustin (ed.) Natural Language Processing, New York: Algorithmics Press, (1973), pp. 193-242.

Kay, M. (1973), llThe MIND Systemjfl in R. Rustin (ed.) Natural Language Processing, New York: Algorithmics Press, (1973). pp. 155-188.

Lyon, G. (1974)) "Syntax-Directed Least-Errors Analysis for Context-Free Languages: A Practical Approach.lr Comm. ACM 17, number 1, (January 1974), pp. 3-13.

Simmons, R. and Slocum, J. (1972), "Generating English Discourse from Semantic Networks,''l Comm. ACM 15, number 10, (October 1972), pp. 891-905,

Woods, W.A. (1970), "Transition Network Grammars for Natural Language Analysis," Comm, ACkl 13, number 10, (October 1970)) pp. 591-606,

Woods, W.A., [1975), Syntax, Semantics, and Speech, BBN Report No. 3067, A.I . Report No, 27. Bolt Beranek and Newman Inc , , t o appear in D, R Reddy (ed ,) - - ~ S e c h Recognition, Academic Press (1975) .

American Journal of Compatationd Linguistics Microfiche 32 : 2 1

Department of Computer Science The C i t y College of the

C i t y University of New York Convent Avenue at 140th Street Hew York, N e w York 10031

ABSTRACT

We describe a semantic processor we are constructing which is

i n t e n d e d to be of general applicability. It is designed around

semantic operations which work on a s t r u c t u r e d data base of world

knowledge to draw the appropriate i n f e r e n c e s and to identify the

same entities i n d i f f e r e n t parts of t h e t e x t . The semantic oper-

ations capitalize on the high degree of redundancy e x h i b i t e d by

all texts. Described are the operations for interpreting higher

predicates, f o r de t ec t ing some intersententialqrelations, and in

particular detail, for f i n d i n g the an tece6en t s of definite noun

phrases. The processor is applied to the problem of drawing maps

from direct ions . We describe a l a t t i c e - l i k e representation

intermediate between the linguistic representation of directions

and the visual representation of maps.

OVERVIEW 1,2

We are trying to cons t ruc t a semantic processor of some 7 A This research was supported by the Research Foundation of the City University of New York under F a c u l t y G r a n t No. 11233. The author would like to express h i s indebtedness to Harry Elam for many insights i n t o the problems discussed here.

2 2

generality. We are using as our data base a set of f a c t s involv-

i n g spat ia l terms i n English. To test the processor and to s t u d y

the interfacing of semantic and task components, we are building

a system which takes as i n p u t directions in E n g l i s h of how to get

from one place to another and outputs a map, a map such as one

might sketch for an unfamiliar region, hearing the directions

over the phone.

A typical input might be the text

"Upon leaving thi,s building, turn right and follow

Washington Street three blocks. Make a left, The

l ib ra ry is an t h e r i g h t side of the s t ree t before

the next coxner."

The ou tpu t would be t h e map I

L i b r a r y I

To bypass syntactic problems, we are us ing a s our input the

o u t p u t of t h e Linguistic String Project's transformational pro-

A I

Washington Street

gram (Grishman et al. 1973, Hobbs & Grishman), which is very

.

close to a predicate-like natation. The semantic component is

. 1

This Building

designed around general semantic operations which work on a

r

s t r u c t u r e d data base of world knowledge to draw the appropriate

N

inferences and to identify phrases in different p a r t s of the t e x t

which refer to t h e same e p t i t y . The text, augmented and i n t e r -

related in t h i s way, is then passed over to the task component,

which makes arbitrary decisions when the map requires information

not given by the directions and produces the map.

ORGANIZATION OF TEXT AND WORLD KNOWLEDGE

The kwp problems of semantic analysis are to f i n d , o u t of a

p o t e n t i a l l y enormous collection of inferences, the appropriate

i n f e r ences , and t o f i n d them quickly . Our s o l u t i o n t o t h e first

is i n our semantic o p e r a t i o n s described below. Our approach t o

the second problem is in the organization of the data base.

The d a t a i n the semantic coptponent is of two sorts:

1. The Text: the information which is explicitly in t h e

t e x t , I n the course of semantic processing t h i s is augmented by

i n fo rma t ion which is only implicit i n the text. The text con-

sists of the set of entities X1,X2, ..., e x p l i c i t l y and i m p l i c i t l y

referred to in the text, and s t r u c t u r e s of $he form p (X1,X2) rep-

resenting the statements m#de or implied about t h e s e e n t i t i e s , e . g .

walk (XI) = X1 walks,

building (XZ) = X is a building, 2

door ( X 3 , X2) = X is a &or of X2. 3 2 . The World Knowledge or the Lexicon: the system's knowl-

edge of words and the world. Words are the boundary between the

Text and the LexPcon. A word is viewed as a key indexing a large

body of facts (Holzman, 1 9 7 1 ) .

Associated with each word are a number of facts or i n f e r e n c e s

which can be drawn from the occurrknce of p(X1, ..., X,) in the

Text. The facts are expressed in terms of p ' s s e t of parameters

Y l f ,Ykt and a s e t of other l ex ica l variables z l , . . , , z m'

stanaing for entities whose existence i s also implied. A fact

consists of enabling c o n d i t i o n s and conclusions. When p ( X 1 , ... X,)

occurs i n t h e Text and the semantic operations determine a

24

particular inference appropriate, its enabling conditions are

checked. If they hold, the conclusions are instantiated by

c r e a t i n g a copy of them in t he Text with the lexical variables

rep laced by Text entities.

Clusters. One way td state the "frames" problem (Minsky

1974) is "How should the data base be organized to guide, confine,

and make e f f i c i e n t t h e searches which the semantic opera t ions

require?" W e approach this by dividing the sets of inferences

i n t o clusters according to topic and salience in the particular

application. In the searches, the clusters are probed in order

of their salience. In our application, the top-level cluster

concerns the one-dimensional aspects of objects and actions. For

example, the fact about a block that it is the distance between

two intersections i s in the cluster. If "around the block" is

encountered, less salient clusters will have to be accessed to

f i n d i n fo rma t io ,~ about the two-dimensional nature of blocks, The

mast important fact about an apartment building is that it is a

building, to be represented by a square on the map. But if the

d i r e c t i o n s take us inside the building, up the elevator, and

along the hallway, the cluster of facts about the interiors of

buildings must be accessed,

A self-organizing list (Knath 1973) of the clusters is main-

tained--when a fact in a cluster i s used, it becqmes t h e top-

level cluster--on the ,assumption that t h e t e x t will continue to

talk about the same thing.

The ''<Truth Status" of Inferences. In natural language,

unlike mathematics, one is no t always free to draw cer ta in

inferehces. We t a g our i n f e r e n c e s always, normally, o r sometimes.

These notions are d e f i n e d o p e r a t i o n a l l y . An a lways i n f e r e n c e i s

one we are always f r e e t o draw, such as that a street i s a p a t h

through space. A normal ly i n f e r e n c e i s one w e c an draw if it is

not explicitly c o n t r a d i c t e d e l sewhere , such as that b u i l d i n g s

have windows. A sometimes inference may be drawn i f r e i n f o r c e d

elsewhere, such as the f a c t used below t h a t a b u i l d i n g i s by a

street. This c l a s s i f i c a t i o n of i n f e r e n c e s c u t s across t h e cluster

structure of the Lexicon.

Lattices. A large number of statements i n any natural lan-

guage t e x t , especially t h e texts this system analyzes, involve a

transitive relation, or e q u i v a l e n t l y , say something about an

underlying scale. For example, the word "walk" i n d i c a t e s a

change of location along a p a t h through space, o r a distance

scale; " tu rn" indicates a change along a scale of a n g u l a r orie,n--

t a t i o n .

I n any p a r t i c u l a r t y p e of t e x t there are scales o r t r a n s i t i v e

relations which are important enough t o deserve a more economical

r e p r e d e n t a t i o n than predicate n o t a t i o n . I n this particulak task,

the impor tan t scales are a distance scale, a s u b s c a l e of t h b i s

indicating the path "you" $ill travel, and a scale representing

angular orientation. This is the principal information used in

constructing the map. For these scales w e t r a n s l a t e i n t o a

directed graph or l a t t i c e - l i k e representation (Hobbs 1 9 7 4 ) .

Some of the things which can be said about t h e structure of

a scale are mat some p o i n t i s on t h e scale, t h a t of t w o p o i n t s - on the scale one is closer t o t h e positive end tHan the o t h e r ,

26

and t h a t a scale i s a part of another s c a l e . If a point B i s

closer to the positive end of the s c a l e than point A , this *fact

is represented by

A-B

If po in t C l i e s i n t h e interval from A t o B the representat ion i s

The diagram

mean& the scale from C to D is part of the scale from A to B, It

is possible to represent incompleteness of information. For exam-

ple, if it i s known that points A and B both lie in a region R

of a scale bu t their re la t ive positions are n o t known and if it

is known about C only thati,tprecedes B t h i s i s represented by

The lattice for the distance scale for t e x t (1) is as follows:

Washington St. The Second St. the

cross

st.

Library

The lattices are intermediate between the linguistic repre-

s e n t a t i o n of the directions and t h e v i s u a l representation of the

maps. They are used at several po in t s in the semantic and t a s k

27

processes. They can be constructed f o r any transitive relation,

and could be very u s e f u l , f o r example, in representing causal and

enabling r e l a t i o n s in a system translating descriptions of algo-

rithms into flowcharts OE programs.

SEMANTIC OPERATIONS

Basic Principle of Semantic Analysis. We bedieve the key to

t=he first problem of semantic a n a l y s i s , that of finding which

inferences are appropriate, is Joos ' Semantic A x i o m N u m b e r One

(Joos 1972), or what I w i l l call the Principle of knitting.

Restated, this is, "The important facts in a text w i l l be repeat-

ed, explicitly or implicity." That is, we capitalize on the very

high degree of redundancy that characterizes a11 texts. Consi i fer ,

for example, the simple sentenced "Walk out the door of this

building." "Walk" implies motion from one pLace to another.

"Out" implies motion from inside something to the ou t s ide . "Door"

i s something which permits motion from inside something to the

outside or from the outside to the inside, or if closed, prevents

this motion. "Building" is something whose, purpose is for people

to be in. Thus, all four c o n t e n t words of t h e sen tence repeated-

ly key the same facts. Those inferences which should be drawn

are those which are keyed by more than one element in t h e text.

This p r i n c i p l e i s used both formally and informally by the

semantic operations. It is used formally in the interpretation.

of higher predicates and in finding antecedents. It is used more

informally for deciding among competing p l a u s i b l e an t eceden t s ,

resolving ambiguities, d e t e c t i n g intersentential relations, and

knitting the text together in some minimal way. Here it isd

p r i m a r i l y the formal uses that w i l l be desc r ibed .

X n t e r p r e t a t i o n . o f Higher P r e d i c a t e s . I n "walk o u t " , "walk

s lwoly" , and "pleasant walk" , t h e h i g h e r p r e d i c a t e s "out", " s l o w "

and ' ' p leasant" a11 apply t o "walk", b u t t hey narrow i n on d i f f e r -

e n t aspects of walking. That is , each demands t h a t a d i f f e r e n t

inference be drawn from t h e s t a t e m e n t t h a t "X walks". "Out" and

"slow" demand t h e i r arguments be motion from one place t o

another., f o r c i n g us t o infe ' r f r o m " X walks'' t h a t "X goes from A

t o B " . "Out" then adds in format ion about t h e l o c a t i o n s s f A and

B, whi le "slow" says something abou t t h e speed of t h i s motion.

"Pleasant", on the other hand, r e q u i r e s i t s argument t o be an

awareness, so we must i n f e r from "X walks" t h a t "X engages i n a

b o d i l y a c t i v i t y he i s aware of" .

Stored i n t h e Lexicon w i t h each h i g h e r predicate is t h e

i n f e r e n c e which must be drawn from i t s argument and t h e informa-

11 t i o n it adds t o t h i s i n f e r e n c e . For example, go (z l , z2 , z3 )" must

be inferred from t h e argument of "out" . When t h e s ta tement

"out(waDk(X1))" i s encountered i n t h e Text, t h e higher predicate

o p e r a t i o n makes e f f o r t s t o f i n d a proof of 1 1 g o ( z l , ~ 1 , ~ 3 ) I1 from

" w a l k ( X L ) " . The search for t h i s i n f e r e n c e is s i m i l a r t d t h e

search procedure described below f o r f i n d i n g antecefienes. T h e

f a c t s in the resulting c h a i n of inference are i n s t a n t i a t e d

t o g e t h e r w i t h the in fo rma t ion added by the h ighe r p r e d i c a t e , and

t h e y are subsequent ly treated as though p a r t of- the e x p l i c i t Text .

I t i s u s u a l for them t o be u s e f u l in f u r t h e r p rocess ing , u n l e s s

the mod i f i e r i s simply g r a t u i t o u s in format ion .

Note t h a t t h i s o p e r a t i o n a l lows c o n s i d e r a b l e compression i n

29

the number of senses that must be s tored for each word* It

ellows us, f o r example, to define "slow" a s something like "Find

the most salient associated motion. Find t h e most specific speed

Scale for the object X of this motion. X ' s speed i s on t h e lower

end of t h i s scale". This definition is adequate for such phrases

as "walk slowlyn (the most salient motion is the forward motion

of the walk ing ) , "slow race" [the forward motion of the competi-

tors), "slow horsew (its running at f u l l speed, usually in a

race), and "slow personw. This last case is highly dependent on

context, and could mean the person's physical acts in general,

h i s mental processes, o r the act h e is engaged in at the moment.

This operation has a default f e a tu r e , If a proof of t h e

required inference can't be found, it is assumed anyway. This

allows a t e x t to be understood even if all the words aren't

known. Suppose, for example, "veer rightw is encountered, and

the word "veern isn't known, i . e . no inferences can be drawn f r o m

it. Since "rightn requires a change i n angular o r i e n t a t i o n a s

its argument, it is assumed this is w h a t "veer" means. Only the

information that the change is small is lost.

FIND ANTECEDENTS OF DEFINITE NOUN PHRASES

~ n t i t i e s referred to in a text may be arranged in a hierarchy

according to t h e i r degree of specification:

1. proper names, including "you" and "I"

2 . other noun phrases, inc luding those w i t h definite,

indefinite, and demofistrative articles

3 . khird person pronouns

4 . zeroed arguments am5 implied entities.

3 0

So far our work has concerned p r i m a r i l y definite noun phrases ,

but it is expected that many f e a t u r e s of t h e d e f i n i t e noun phrase

algorithm w i l l carry over t o other cases,

The d e f i n i t e noun phrase a lgor i thm consists of fou r steps.

First, "uniquent2~s cond i t i onsn are checked t o determine whether

an antecedent i s requ i red . If so, t h e Text and Lexicon are

searched for p l a u s i b l e anteceaents . Third, cons i s t ency checks

are made on these. F i n a l l y i f more than one p l a u s i b l e antecedent

remains the Principle of Kn i t t i ng is app l i ed t o decide between

them.

Vniqueness Condi t ions , I n t h e phrase "the end of the block",

we know we must look back i n the t e x t for an e x p l i c i t l y o r impl i -

citly mentioned "block" ( the search case), b u t we do fiat neqes-

s a r i l y look for a previously meptioned "end" (the no-search case) . Given a d e f i n i t e noun phrase t he a lgor i thm first tries t o deter-

mine whether it b e l o n g s t o t h e search or no-search case. This i s

done by checking two broad cr i ter ia . (These criteria were moti-

vated by a large number of examples no t only from s e t s of direc-

tions but a l s o from t e c h n i c a l and news ar t ic les , )

These criteria are checked by sea rch ing t h e Lexicon for

c e r t a i n f e a t u r e s . However these searches are generally very

shallow, i n c o n t r a s t t o the p o t e n t i a l l y much deeper searches in

the riext s t e p of the algorithm. S i n c s by far the majority of

d e f i n i t e noun phrases are i n t h e no-search case, checking unique-

nes s cond i t i ons can r e s u l t i n g r e a t savings .

A caveat is in order. W e state the c r i t e r i a at a very high

level of abstraction, We feel i n f a c t t h a t t h e a lgor i thm can

work at that level of abstraction if the ex icon is proper ly

constructed. But how to construct a large exi icon properly is

a problem we have not yet tackled in detail. In any event, we

give examples f o r each case, and the examples themselves form a

reasonably exhaustive classification.

1. A d e f i n i t e entity is in the no-search case i f it can be

located precisely w i t h respect to some framework. n his includes

me following conditions. a. Objects which are located with r e s p e c t to some identi-

f i e d point in space: "the building on the corner".

b, Plurals and mass nouns which are restricted to some

identified region sf space: "the trees in the park", " the water

in the swimming pool". Here "the" indicates a l l such objects or

substance.

c.. Points and intervals in time khich are fixed with

respect to some identified event: "the minute you arrive", "the

hour since you left".

d. Events in which at least some of the participants are

identified and which can be recognized as occurring at a specific

time: nthe ride you took through the park yesterday1';

e, P o i n t s or intervqls on more abstract scales: "the end

of the block", "the size of t h e bui ld ing". The end is a specific

poin t on the distance scale defined by the block. The size of

the building is a specific point on the general s i z e scale for

objects , i . e. the volume scale.

f. Superlatives, ordinals, and related terms: " the largest

house on the block", "the second house on the block", " the only

house on the block". If the set of comparison is identified,

the superlative or ordinal indicates the scale oE comparison and

the place on that scale of t h e e n t i t y it describes. This is a

subcase of (e) . A l l of these c o n d i t i o n s can be checked in one operation if

the facts in the Lexicon are expressed in terms of suitably

abstract operators relating entities t o scales. We simply ask if

the definite entity is on or part of a scale or a t a p o i n t on or - - along an +interval of a scale, where the scale can be identified.

However this r e q u i r e s that w e t a k e very seriously m y suggestion

in Hobbs (1974) t h a t the lexicon for the entire language be built,

insofar as possible, along the lines of a spa t i a l metaphor. We

have no t yet had to f a c e these problems since our only scales are

physica l -- our " a t " and "on" are the locative " a t " and "on".

Also checking this c r i t e r i o n presupposes a very sophisticated

s y n t a c t i c and semantic analysis. For example, [d) assumes that

the times of events mentioned in tenseless constructions can be

recovered.

2. A definite entity is in the no-search case i f it i s the

dominant entity of t h a t description. This d i v i d e s i n t o two sub-

cr i t e r i a :

a , Those e n t i t i e s which are unique or dominant by virtue

of the properties which describe them: " t h e sun1', "the wind". If

t h e p roper t ies p1 (X) ,pZ (X), ..., are known about the d e f i n i t e

entity X, the definitions o f p1,p2, ..., are probed f o r the f ac t

that the entity does not normally occur in the plural. Included

under this heading are proper names beginning with "the", like

"the Empire State Buildingff, and appositives, like "the city of

Bos tonr' . b. Those entities which are unique by virtue of t h e prop-

erties of an entity with which they are grammatically related:

"the door of the building", "the Hudson River valley". "The door

of the buildingn is represented in t h e Text a s "xl 1 door'(^^,^^ 1 building{X2))' i.e. "the Xl such that XI i s t h e door of X2 which

is a building". The uniqueness or dominance of XI is not a prop-

e r t y of "door" but of "building". Stored w i t h "building" is the

fact that a building has in its front surface a main door which

does not normally occur i n t h e p l u r a l . "The door of t h e bu i ld ing r '

is interpreted as this dominant dosr.

If the tvliqueness conditions succeed, a poin te r is s e t from

t h e dominant lexical variable to the corresponding e n t i t y . If

subsequently the same definite noun phrase occurs, the uniqueness

check will discover t h i s pointer and correctly identify the ante-

cedent. Thus, we can handle the example

"Walk up to the door of t h e building. Go through

the door of the building."

Here the uniqueness check gives us a s h o r t c u t around the n e x t

step in the algorithm.

The Search for Plausible Antecedents. To illustrate the

search for an antecedent, consider

"Walk out the door of this bu i l8 ing . Turn right.

Walk to the end of the block. "

What block? From "block" W e follow a back p o i n t e r t o the f a c t

stored with "streetn *that "streets consist of blocks", and from

34

"street1' the fact with "buildingt' that "Buildings are by streets"

Since a building is mentioned, we assume it is "the block of the

street the b u i l d i n g is on". T h e facts in the chain of inference

leading to this are instantiated, An entity is introduced i n t o

the t e x t fo r t h e "street" and the Text is augmented by the state-

ments that "the bu i ld ing is on the street" and "the block is part

of the street". This information turns out to be required for

the map. Note that t he Eact that a building is on a street is a

sometimes f a c t and that we are free to d'raw it only because "the

blockn occurs*

To conduct the search of the Lexicon, ideally we would like

to send out a pulse from the word "block" which travels faster

over more salient paths, and look for the first entity which the

ptXlse reaches. The saliency is simulated by the cluster

structure descrihea above, The parallel process of the spreading

signal is simulated by interleafing deeper pfobes from salient

clusters with shallower probes from less salient clusters. For

example, i f "streets consist of blocks" is a c l u s t e r 1 fac t , t h e n

we might probe for a cluster 1 fac t involving syreets and a

cluster 2 Eact involving blocks at roughly the same time, After

one plaus ib le antecedent is found in this way, t h e search is

continued for possible antecedents which are n e a r l y as plausible.

If after a time no plausible antecedents are found, the search

is discontinued.

Searches for antecedents are conducted not only for entities

but also for definite noun phrases that the nominalization trans-

formations of t h e syntactic component have turned into statements

3 5

- -e .g . "The walk was t i r i n g " . Here we look back for a statement

whose predicate is "walk" or from which a statement involving

"walkn can be i n f e r r e d . There are cases in which the required

inference is in f a c t a summary o f an entire paragraph--e.g.

"These actions surprised. , . "--although of course we cannot

handle these cases.

Consistencv. Each of the plausible antecedents is checked

for consistency. Suppose X1 is the definite entity which prompt-

ed the search and its properties are

and X2 is the proposed antecedent with properties

We must cycle through the q ' s and the r ' s to ensure they are con-

sistent properties. Of course, to prove t w o properties q(X) and

r(X) inconsistent can be an indefinitely long process with no

assurance of termination. One admittedly ad hoc way we get

around this is by placing into a special cluster those f a c t s we

feel are likely to lead quickly to a contradiction. The second

tool we use for deriving inconsistencies may t u r n out to be

qui te significant.

In the course of processing, the lattice described abave is

constructed for several predicates. They c o n t a i n in fo rma t ion

which can be useful i n deriving a n inconsistency. Suppose we

have a t ex t in which "the block" occurs explicitly several times.

Toward the end of it, we encounter

"Turn right on to Adarnii Street. The library

fs at the end of t h e block" .

The search algorithm looks first for explicit mentions of "blockl"

and finds them. Yet none of these entities is the one we want.

Intuitively, the reason we know this is our almost visual feeling

that we are already beyond those points.

The lattice consistency check corresponds precisely to this

feeling. If a definite entity X1 is a point or interval in a

lattice or at a point or along an interval, we ask if the propos-

ed antecedent X2 is or can be related to a portion of the lattice.

If so, then s i n c e the lattice represents a transitive relation,

we need only ask i f there is a path in the lattice from X2 to XI.

If there is, they cannot be the same entity.

Many cases which pass for applications of the supposed

recency principle--"Pick the most recent plausible antecedentn--

are in reality examples of this consistency check. The earlier

plausible antecedent is rejected because of lattice considera-

tions.

As the text is processed, the whole structure of the

discourse is built up. When a definite noun phrase is encounter-

ed, this discourse structure is known and it is this knowledge

that is used to determine the antecedent rather than the linear

ordering of the words on the page.

Competition among Remaining Plausible Antecedents. Even

after the consistency checks, several plausible antecedents may

remain, forcing us to decide among them on less certain criteria.

To do this, we appeal to the Principle of Knitting again and make

the choice that will maximize the redundancy in the simplest

possible way.

A probe is s e n t out f r o m the definite entity and from each

plausible antecedent. Each plausible antecedent i s searched for

properties it has in comon with the definite entity. Common

properties Count most if they are already in the Text, an8 with-

in the Lexicon, comon properties count more if they are within

more salient clusters or they result from shorter chains of

inference.

Default. Like the higher predicate algorithm, the definite

noun phrase algorithm has a default feature. If the uniqueness

conditions fail and the search turns up no antecedent, we simply

introduce a new e n t i t y . In fact, in the direct ians texts there

are a disproportionately large number of default cases, for "the

object" may simply be the object you will see when you reach

that point in following the directions.

Other Anaphora. We have not y e t implemented rou t ines for

handling other anaphora. However, we believe they a re very

similar to the definite noun phrase rou t ine , w i t h c e r t a i n d i f f e r -

ences. For entities tagged with demonstrative a r t i c l e s , we do

not check uniqueness conditions, and the search will be narrower

since the antecedent must be an entity or statement actually

occurr ing in t he text. For pronouns also, no uniqueness cond i -

t i o n s are checked. The search will turn up more consistent

plausible antecedents, and a correspondingly greater burden will

be placed on the competition routine.

INTERSENTENTIAL CONNECTIVES

We de tec t unstated inter-sentence connectives by matching two

successive sentences S1 S2 with a small number of common

38

patterns. In the directions texts the patterns are usually few

and simple. The most common are

1. S1 asserts a change whose final state is asserted or

presupposed by S 2 .

2. S1 asserts or presupposes a state which is the initial

state of a change asserted by S2.

(These are likely very common patterns in all narratives ,) For

example, in the text

"Walk out the door of this building. Turn right.

Walk to the end of the black",

pattern(1) j o i n s the first two sentences, where the state is

"You at X", Pattern(2') joins the last two sentences, where

again the state is "You at X-". Note moreover that the sentences

axe interlocked by n second application of the two pa t te rns : The

first sentence assumes an angular orientation which is the

initial state of the change asserted in the second sentence.

The final state of this change is assumed by the third sentence.

In addition to providing the discourse with structure, this

operation i s one of t h e - p r i n c l i p a l means by which implied entities

in one sentence, like X above, are identified with those in

another.

When pqttern (2) is applied, we delete the independent occur-

rence of the s t a t e in the Text, so that subsequently it ex i s t s

only as one intermediate state ih a la rger event. Changes across

time are handled in this way.

TASK PERF-ORMANCE COMPONENT

Arbitrary Decisians, The semantic operations are quite

39

g e n e r a l and can be used for any application. The augmented and

i n t e r r e l a t e d Text i s t h e n handed aver to the task performance

component, which of course is specific to the a p p l i c a t i o n .

Our task component first makes arbitrary decisions r e q u i r e d

by the map but not given in the text. Both natural language

d i r e c t i o n s and ske tched maps allow information to be incomplete

and imprecise, but in different ways. Far example, in

nTurn right at the third street or the second stoplight".

we must decide whether to put the first stoplight at the first

or second street,

The l a t t i ce representing the p a t h "your' take must be complete

i n the sense t h a t it i s continuous, begins at the initial loca-

tion, and ends at the desired goal, and that the relative loca-

tions of all points on the path are known. The lattide is

complete if and only if there is a directed path passing through

every point in the lattice at least once. If it is not complete,

it is completed by supplying t h e fewest possible new links.

Gsometr-izing the Lattices. The second task operation is to

c o n v e r t the topological lattice representation into the geometric

r e p r e s e n t a t i o n required by the maps. First we assign d i r e c t i o n s

to all t h e points in the angular orientation lattice. In the

simplest case we may have something like

where "a - b" means direction b results from a clockwise rotation of d i r e c t i o n a. If no explicit directional information

4 (0

is present, we simply assume a, c, and e are the same direction,

and b and d are the same, and then assume the two directions are

at right angles, Then in the distance lattice, contiguous or

overlapping paths which share the same orientation are assumed

to be parts of the same path and are mapped into a straight line.

Information about names is accessed and assigned to the streets

and buildings and the map is drawn,

Specific Systems with a General Semantic Component. We are

aiming not so much at the construction o f a general natural

language processing system, which still seems reasonably f a r o f f

b u t a t an easier way of constructing specific systems. The case

of syntax is instructive. It would be foolish for one who is

building a natural language processing system to build his

syntactic component from scratch. Large general grammars and

parsers for them exist (e.g. Grishman et al. 1973, Sager &

Grishrnan 1975). It is easier by several orders of magnitude to

begin with a genera l grammar and specialize it, by weeding out

the rules for constructions that don't occur in the texts one is

dealing with, and by adding a few rules f o r constructions and

constraints peculiar to orre's application.

We are trying to make a similar facility available for the

most common kinds of semantic processing. Specializing the

general semantic component would consist of several relatively

easy steps. First the Lexicon would be organized into a

cluster structure appropriate to the task. At worst, this would

mean specifying the necessary knowledge in a fairly simple format.

If a very large Lexicon were available, this could mean no more

than designating for each fact the cluster it should appear i n .

Cer ta in inferences could be made obligatory while others which

are irrelevant t o the task could be l e f t out of the special Lexi-

con altogether. Second a Task Component would be built which

would take, as ours does, the semantically processed Text, and

use it t o perform t h e task. W e are demonstrating the usefulness

of this approach in performing a task involv ing a v i s u a l repre-

sentation. It is likely to be useful in other sorts of tasks also.

BIBLIOGRAPHY

Grishman, R., Sager, N., Raze, C., & Bookchin,B.,"~he ~inguistic

String Parser," Proc. NCC, M I P S Press, Montvale, N . J . 1973.

Hobbs, J e t "A Model for Natural Language Semantics, Part I: The

Model," Yale Univ. Dept. Comp. Sci. Res. Rep. 36, Nov. 1 9 7 4 .

Hobbs, J., and Grishman, R., "The Automatic Transformational

Analysis of Engl jsh Sentences: An Implementation,"

Submitted to International Journal of Computer Mathematics. -- -

Holzman, M., "Ellipsis in Discourse: Implications for Linguistic

Analysis by Computer, The C h i l d ' s Acquisition of Language, and

Semantic ~heory," Language and Speech (1971, 86-98.

Joos, M., "Semantic Axiom Number One," Language (1972) 257-265 .

Hnuth, D. The Art of Computer Programming, - 3 , Addison-Wesley,

Reading, Mass., 1973.

Minsky, M., "A Framework for Representing Knowledge," MIT A1 Memo

306, June 1974.

Sager, N., and Grishman-, R., "The Restriction Language for Compu-

ter Grammars of Natural Language," CACM 18, 7 (7/75) 390-400, -

American Journal of Computational Linguistics Microfiche 32: 42

PERRY t. MILLER Massachusetts Inst i tute of Technology Cambridge, Massachusetts 02139

ABSTRACT

\Jheh a user interacts w i t h a natural language system, he may well use words and expressions which were not anticipated by the system designers. This paper describes a system which can play TIC-TAC-TOE, and discuss the game while it is in progress. I f the system encounters new words, new expressions, or inadvertent ungrammaticalities, it attempts to understand what was meant, through contextual inference, and by asking i h t e l i i g e n t c larifying questions of the user. The system then records the meaning of any ne9 words or expressions, thus augmenting its 1inguist;lic knowledge i n the course of user interaction,

A number of systems tire being developed which communicate with users i n a natural language such as English. The u l t imate purpose of such systems is t o provide easy computer access to a technically Onsophisticated pepon. When such a person interacts with a natural language systemr, however, he is quite l ikely t o use words and expressions which were not anticipated. To provide truly natural interaction, the system should be able t o respond intell igently when this happens.

Most current systems, such as those of Winograd [ l o ] and Woods I l l ] , are not designed t o ;ope wi th such "l i igu is t i c i n p u t uncertainty." Their parsers f a i l completely i f an i n p u t sentence does not use a s p e c i f i c , b u i l t - i n syntax and vocabulary. A t the other extreme, systems l i k e ELIZB [93 and PARRY [ Z ] allow the user to type anything, but make no attempt t o fully understand the sentence. The present work explores the tnlddle ground between these extremes: developing a sys.t;em which has a great deal of knowledge about a particular subject area, and which can use this knowledge to make language interaction a flexible, adaptive, learning medium.

In pursuing t h i s goal, the present work is most closely related t o work being dona i n the various speech recognition efforts [5 , 7, 8, 121 which ara studying how l ingu i s t i c and semantic constraints can h e l p deal w i t h the ACOUSTIC error and uncertainty of speech. The adaptive system, however, is designed t o deal with a much mors LINGUISTIC type of uncertainty.

When people use unfamiliar words or expressions in conversation, we can usually deduce from context what is meant, and i f not, we can a t least ask inte l l igent clarifying q u ~ s t i o n s . To allow the machine to do the same, there must be a very flexible interaction of syntax and $emantics i n the parsing/understanding process, There must be a

dif ferent parser organization, and a more f l e x i b l e use oP l i n g u i s t i c and semantic constraints , than i s p f e s e n t in current natural language sys tern .

The adaptive system is a step towards t h i s goal Tha cu r ren t implementation i s a prototype, design'ed to i l l u s t r a t e many o f these ideas, and t o t i e them together in a restricted system t h a t is complete but not t o o complex. The system's domain o f discourse is TIC-TAC-TOE: it. plays a game and discusses the game in progress. It has a ( c u r r e n t l y limited) se t o f semantic concepts, and attempts t o interpret the u s e r ' s sentences ih terms of these. The set of semantic concepts is FIXED. The system does not attempt t o learn new concepts, merely new ways of describing concepts that are already famil iar.

When t h e system encounters a new word or a new phrasing, it uses contextual information t o detername the p o s s i b l e meaning, and to ask clarifying questions which a re as i n t e l l i g e n t as p o s s i b l e . It then remembers the meaning of the new e&pressfon, and w i l l use it i tse lF *en it talks.

2. SAHPLE SESSION

The best way t o introduce the system is t o show i t in o p e r a t i a n . In the sample session that follows, user input is pteced~rd by 'U:", machine output by and comments added by the author arg in small type, enclosed in brackets,

(HACHINE: I PLACE AM X IN TClE TOP LEFT SQUARE)

[The machine makes its opening move. ]

(USER: PLAY A ZERO IN THE CENTER WARE)

(H: 1s ZERO A NOW HEANING 0 )

(U: YES) (H: OK)

[The new word *zeroa is added t o the machine's vocabu1ary.l

(R: I PLACE AN X IN THE BOTTOM RIGHT WARE)

(O: SLAP Ab4 0 IN THE BOTTOV! LEFT CORNER)

(PI: IS (TO SLAP SOMETHING SOREWHERE) A WAY OF SAYING (70 PLACE SOMEmING SOMEWHERE))

(U: YES) (19: OK)

[ A new surface framer (as described in section 4 ) I s added to the set sf surface b r a e s which express the concept of making a move.]

( m : 1 SLAP AN X IN TWE TOP RIGHT SQUARE)

f n t h i s sample sass ion , the machine uses i t s newly acquired vocabulary a t every opportunity.)

HOU CORE YOU PLACED AN X IN THE TOP RIGHT SQUARE)

(H: IS (HOW CQHE MMEONE DOES SOHETHING) AN EXPRESSION ISEANIMG ( l!lOES SOMEONE 00 SQMETHING))

(U: YES) (M: OR)

(H: TO-BEAT YOU)

[ A new surface frme i s added t o those sxpssssing the concept ' I

Syntactic featuras are associated w i t h each such frauge i n d i c a t i n g such things as Wether subject-verb invsrsion is ds be DerBomed.1

(U: CAN YOU CURRENTLY BEAT M E )

m: 1s CURRENTLY AN ADVERB REFERRING TO THE TInE OF A RWE)

[These are the only adverbs the system knows abou t . In f ac t , a t present, the system ignores adverbs in its sewantic processing. I % asks the question because it wants t o account for all t h e words in the sentence.

(U: YES) '(M: OK)

(11: YES)

( M : r s WEST AN ADJECTIVE ~ E A N I N G EITHER (1: CORNER) ( 2 : LEFT)

(U: 2) (n: or;^

(H: NO)

[Notice that t w o top squares contain an x . However, only the l e f t onel was played while the bottom r ight square was Free. Therefore the systcszn concluded t h a t the top l e f t square was being described. This example i l lustrates how contextual information can be used t o ask as intelligent a cBarifying question as p a s s i b l a . ]

(U: *yOW COME YOU PLACED AJ4 X IN THE TOP EAST SQUARE)

(!I: IS EAST AN l@JECTLVE NEAWXNG) EITHER (1: RIGHT) (2: WEST)

( 0 : l p (U: on)

M IS (TO LAY SONETHING SOHEWHERE) A WAY OF SAYING (TO SLAP SOflETHING S0UEWERE) )

(U: Y E S ) (M: OK)

(H: IS (TO POLISH SOMEONE OFF) A WAY OF SAYING) E I THER ( 1 : ( TO BEAT SOMEONE) ) (2: (TO LOSE TO SOMEONE)) ( 3 : (TO TIE SOMEONE))

(U: I ) (?'¶: OK)

(R: BY LAYING AN X ON WE TOP CENTER CELL)

[ I f t h i s same session were subsequently repeated, sf course, the system would not need t o ask any questions of t h e ushJr. 1

2.1 Comments on C u r r e n t Lirnitatians

There are a number of l i m r t a t i o n s to the a d a p t i v e system as it now s tands . Some of these may be apparent in the smple session, bud an in t roduc t ian t o the system is not complete without discuss ing them explicitly.

(1) The number of concepts ava i lab le t o the system a t present is very small. T h i s , in fact, is why the system's first guess is usually the correct one. I f the sentence is at a l l with in the systea's comprehension, t h e options as to its meaning a re currently q u i t e limited.

( 2 ) The range of expressive devices presently recognized is q u i t s limited as well. For instance, the system does n o t recognaze relative clauses, con junctions, o r pronouns (except f o r 1 and you).

( 3 ) The system currently d e a l s only wi th TOTALLY U N F M I L I A R words and expressions in this adaptive fashion, It w i l l not correctly handle familiar words which are used in new ways (such as a noun used eas a varb, as i n wzero the center squaren) .

( 4 ) The system tr ies to map the meaning o f new wards and expressiuns into i t s speci f ied s e t of underlying concepts. It then displays its hypotheses t o the user, g iv ing h i m only the option of saying yas or nu. The user cann-ot say "no, not qui te , it meahs . . .". (Thus concepts like V h e 'northeast1 square" o r "the 'topmost' squarew would ba confusing and not correctly understood.)

The present simple system h a s been developed w i t h two goals in mind: (1) to explore the techniques required t o achieve adaptive behavior, and ( 2 ) t o h e l p fornulate the issues which will have t o be faced when incorporating these techniques in to a much broader natural language system.

3 . OVERVIEW

Fig. 1 shows ths various stages that the Adaptive System gees through in understanding a sentence. In this sectian, we s h a l l watch while t h e system processes the sentence "Mow came you placed an x in the top right ~ q u a r e . ~

( 1 ) Local Syntact ic Processing: In this f i r s t stage, the system scans the entire sentence look ing f o r local cons t i tuents . These i n c l u d e Hsimplem noun phrases (NPs) and prepositional phrases (PPs), ("simplen meaning 'up to the head noun but not including any modifying clauses or phrases"), and verb groups (VGs) consisting o f verbs together with any adjoining rnodals, auxilliaries, and adverbs. In t h i s instance, the system Finds the t w o N P s , "youe and "an xm, the PP "in the top r ight squarem, and the VG nplacedw.

( 2 ) Semantic Clustering: A t t h i s s tage , the c lause- leve l processing s tar t s . U n l i k e most systems, this clause- level processing is driven by SEMANTIC r s la t i onsh ig s , rath-er than by syntactic form. It uses a semantics-first kclustssinsg*, with a sscondary use of syntax for cormnents and confirmation+ In t h i s example, a l l t h e l o c a l constituents found can be clustered i n t o s description of e single concept: t h a t o f making a nave, Section 4 describes the mechanics of this stage in more detail.

( 3 ) Cluster Expansion and Connection: During t h i s stage an attempt I s mada t o account Psr each word in t h e sentence by expanding the concept c lus ters , and i f there i s more thaw one, by j o i n i n g them together t o form an e n t i r e multicXausa1 sentence- In t h i s case, ths concept c luster rnlght b s axpanded I n two ways. a ) One possiblllty night be t h a t I t i s a "MOW" type q u e s t i o n , and t h a t wcornc.tn is some sort of adverb, However this possibility v io la t s f a semantic constraiet, since the system is not s e t up t o answer haw a move is made; only how t o win, how t o prevent sorneons From winning, e t c . There fore this possibility is ignored. b) The other p o s s i b i l i t y f r; t h a t "how come" i s a new way of describing soma other clause f u n e t t o n .

(4) Contextual Inference; Clarification; and Response: During t h i s f i n a l staga, any c o n t e x t u a l inf~rrnatfsn avai lable is brought t o bear on araas of uncertainty, any necessary clarifying questions are asked, and the system responds t o the sentencs. In this example, the only uncertainty is the meaning of "how comew. Since t h i s i s the main

sentence

1

Xocal constituents

concept clusters

complete sentence hypothesf s

system responds t o sentence

Fig. 1: Adaptive System Overview

clause of the sentence, the possibility of its b e i n g an Wn or *aftsra clause are discarded. The remaining p o s s i b i l i t i e s are n i m p e r a t i v s w , "hown, m ~ h y n , and "canw. The system does n o t answer %own and "canw quest ions i n relation t o making moves. Similarly, "imperativen does n o t make sense since the action described is a previously made move. Therefore the system asks i f "How come someone does somethingw means Vhy does someone do somethingn. The user answers "yesn, so the system stores t h i s new way of asking "whyn, and proceeds t o answer the question.

4 . SEMANTICS-FIRST CLAUSE-LEVEL PROCESSING

One of the major differences between t h i s approach t o parsing and tha t of a top-down, syntax-driven system (such a s Moods' or Winograd's) is the order i n which s y n t a c t i c and semantic processing is done a t the clause level.

In a top-dom system, a sentence must exactly match t h e b u i l t - i n syntax before semantics can even be cal led and given the various const i tuents o f a clause, T h i s IS clearly undesirable when one i s dealing with i n p u t uncertainty, since one cannot be sure exact ly how the user will phrase his sentence. One would prefer to Bet semantics opera%@ First on any local consituents present, so that i t can make a reasonable grgss as to what is being discussed.

As semantically-rslated clusters of local constf tuents are found, syntax can be consulted and asked to comment. on the rslative grmmaticality of the various c lus ters . If there are two competing semantlc inte~pretations of one part of a sentence, and syntax l i k e s one much better than the other , then the "syntactically pleasing" interpretation can be pursued f i r s t . Later, i f this does not pan out, the syntactically irregular possibility can be looked at as wsP1. In t h i s way, syntax can he lp guide the system, but is not placed in a totally controlling p o s i t i o n .

A by-product advantage o f t h i s s e m a n t i c s - f i r s t approach i s that the system can handle mildly ungrammatical input without any ex t ra work, In addit ion, t h e semantics-first c lus tar ing approach lends i t s e l f q u i t e naturally t o handling sentence fragments.

I n the remainder of t h k s s e c t i o n , we describe how the adaptive system organizes i d s linguistic knowledge t o implement this semantics- f i r s t approach. As we s h a l l s e e , there are three componeflts o f this knowledge.

( a ) Ths local racognizars which initially find local constituents. recognizers are represented t n Augmented Transition Network [ I l l f o m , are q u i t s s i m p l e , and are not described further i n t h i s paper . (b) Clause-level knowledge sf how actions and clause-functions are described. This knowledge is expressed i n a descriptiva fash ion which makes it msily manipulabla, and easy to add to. ( c ) Clause-level syntac t ic knowladge which is sxprssred ira a domain- indebpendent fom.

4 . 1 Knowledge of how A c t i o n s are Described

Figure 2 i l l u s t r a t e s how t h e system s t o r e s i t s knowledge sf how act ions ( o r events) are described. This knowledge is stored a t two l eve l s : the conceptual l e v e l , and t h e surface (or expressive) l e v e l

As shown in F i g . 2, the concept PLACE represents the a c t o f making a TIC-TAC-TOE wove. ( a ) On the CONCEPTUAL l e v e l , there are three "conceptual s lo t s ' i n d i c a t i n g the actors which are involved in the ac t l on : a player, a @ark, and a square. (b) On the SURFACE, or expressive, level there is a list sf surface frames each indicating one poss ib l e way t h a t t h e concept can be expressed. Each surface frame conslsts o f a verb p l u s a set of s y n t a c t i s case frames to be f i l l e d by t h e ac tors . (Notice that neither the conceptual slots nar the sur face frames i n d i c a t e explicitly t h e order in which the varlous constituents are to appear Fw a sentence.)

When the system processes a sentence, it fills t h e concsp tua l shots w i t h local constituents found rn the sentence I f i t h a s f o u n d a f m i l i a r verb, then i t a l s o gets any surface e ( s ) associated w i t h that verb. A t this p o i n t i t c a l l s syntax, a s k i n g for c s m e n t s .

For instance, i f the input sentence is "1 place an x in the corner", t h e n all the conceptual slots of #PLACE would be f i l l e d , and the system would pass the following string to syntax wagen% verb o b j ppw . As a result, clause-level syntax does not see t h e a c t u a l constituents of the sentence, only t h e l a b e l s specifled I n the surface case frame, plus information indicating number, tense, etc .

An interest ing aspect o f this approach is t h a t t h e clause-level syntax is entirely domain-independent. I t knows no thing about TIC-TAC- TOE, o r even about the words used t o talk about TIC-TAC-TOE. Tke surface frames allow semantics t o t a l k t o syntax purely in t e rms o f syntact ic labels. As a result, one could write a single syntact ic module, and t h a n insert i t unchanged in to many domains.

4.1 .1 Using t h i s Information

In t h i s s e c t i o n , we descr ibe i n more d e t a i l how this knowledge can be used when processing a sentence.

(1) I f the verb and constituents a re familiar: I f t h e r e i s no uncertainty i n a c lause , then each const i tuent can

be put into one of Ghe conceptual s lots , and any surface frames associated w i t h the verb can be examined The frame ~ n d i c a t e s the csse (agent, object, etc. ) associated with each c o n s t i t u e n t whon that verb is used. The frame is used t a create a string of case l a b e l s t h a t a r e s e n t t o syntax for coments .

For instance, iF the sentence is "1 place an x i n the center

CONCEPT: PLACE

CONCEPTUAL SLOTS:

P: player H: mark S: square

SURFACE FRAMES:

VERB: place (as in: AGENT: P m I place an x i n the centera) OW: PI in: S

VERB: play (as in: AGENT: P sf play an x in the centers) ow: H In: S

VERB: play (as in: AGENT: P w X play the center") 00J: S

FLg . 2 : Linguis t i c KnowleMge about Actions

square", the string passed to syntax is "agent verb obj pp". Syntax replies t h a t t h e sentence follows normal order. Had the string been "verb obj pp" syntax would reply t h a t the subjec t had been deleted . I f the s t r i n g was @'do agent verb obj ppn, syntax would reply that subject- verb inversion had taken p l a c e . Given "gent obj verb ppn, syntax would reply that t h e object was out of position.

Thus syntax i s se t up to notice both g~irnmcatical and u n g r m a t f cal permutations i n constituent order, and t o comment appropriately. The system must then decide how t o interpret these comments.

For instance, if syntax repl ies t h a t the object is out of position i n the clause, or t h a t there is incorrect agreement in number between subject and verb, the system may decide that t h e user has made a minor grammatical error, and allow the sentence t o be processed anyway, especially if there i s no better interpretation of the sentence. In this way, clause-level syntax plays an a s s i s t i n g role rather than a castrolling r o l e i n t h e analysis of a sentence.

( 2 ) If a constituent is unknown: If an unknown constituent is p r e s e n t , then both the frame and

slot information can be used to h e l p resolve its meaning. For ins tance , suppose the sentence is " I place a c r o s s in the canter squarew, and the, word ~ c r o s s u is unfamiliar,

Here, during t h e semantic clustering, t h e conceptual s l o t s for a player and a square can bs f i l l e d by "Iu and "in the center square", b u t the slot for a mark is u n f i l l e d . I n a d d i t i q , there is the unknown constituent "a crossg.

A natural hypothesis, therefore, is t h a t the unknown constituent refers t o a type of mark. Since the verb is familia~, a surface frme is avaflable. Next, assumtag the unknown constituent is a mark, the s t r ing "agent verb ob j ppw can be passed to syntax. Men syntax approves, this offers addi t iona l confirmation t h a t the hypothesis is probably right.

Subsequent evaluation of this hypothesis indicates t h a t the sentence makes sense only if the mark referred to is Etn x , so the system asks i f "crossu is a noun meaning

( 3 ) I f the verb is unknown: I f an unfamiliar verb is used, then there i s no sur face fsme

availabls t o h e l p guide the analysis. Instead, syntax must ba used in a different mode t o propose what the surface frame should be.

Suppose the sentence is "I p lunk an x in the center squareM. Here, a l l the constituants can be clustered into the concept #PLACE, but t b r e is an unknown word, and no verb. Ths loglcrrl hypothasis is t h a t t h e new word i s a verb. A special syntactic module i s therefore passad t h e followfag s t r i n g "NP(P) verb(p1unk) NP(M) PP(in,S)# This module examines the string and produces tn new Frame:

VERB: plunk AGENT: P OW: R in: 8

The system can then ask if "to plunk something somewherew means " t o place something somewheren, and upon getting an affirmative reply, can add t h e new frame to those associated w i t h the concept PLACE.

Since the system uses the surface frames to generate its o m replies, it can now-use this new frame i t se l f when it talks. When the system wants to generate a c lause , it passes a selected frame, the constituents, and a list of syntactic features to a clause generator which o u t p u t s the specified form. (Thus, c l aus s - l eve l syntax can be used by the system i n three different modes: (1) to comment on the g r m a t i c a l i t y of a s t r i n g of case markers, (2) t o constrbct a new surface frame, and ( 3 ) t o generate clauscas when t h a system itself replies . )

4.2 Knowledge of' how Clause-Functions are Described

As i l lustrated i n Fig. 3, knowledge of how clause-function concepts are described i s also expressed as two Lexals.

CONCEPT: #WHY

CONCEPTUAL SLOTS:

ACTION: #PLACE

SURFACE F

Why ACTIQN(SV1NV) ( a s in: *Why does someone do somsthkng")

flow come ACTION() ( as i n : "Now come someone does something")

Fig. 3 : Linguist ic h o w l edge about Clause Functions

Each clause function has a conceptual s l o t indicat ing what types of action can be used w i t h t ha t clause type ( i n t h i s case, the ac t ion #PLACE), and a list of surface frames ind ica t ing di f ferent ways i n which t h e cancspt can be expressed.

A clause-type frame currently includes any special words which introduce the c lause ( i e . "whyn or "how comen), together w i t h a list sf syntactic proparties which should be present in the clauss. This list of syntactic properties might include SVIMV, nsubjec$-verb inversionw (as in "why does someone do something"), ar 9 u b ject deletionH, 'ING fomm, and "use of a particular preposition* (as i n "from doing somethingw).

These syntactic features, however, need not bs inflexible rules. Sentence understanding can still psocaed wen i f tha syntact ic features

found by syntax do not exactly match those spec i f ied by the clause- function frame. Thus, an inadvertent ungrammaticality cam readily be recognized as such, and processing can cont inue .

4.2 .1 Using the Clause Function Knowledge

In this section we examine how this clause function knowledge can be used.

(1) With no uncertainty: I f the i n p u t sentence is "Why d l d you place an x in the center

squarew, then during the semantic clustering the s tr ing Rdo agent verb obj ppu i s passed t o syntax, which repl ies t h a t subject-verb inversion has taken place.

When exarninlng t h e whole clause, the system sees t h a t it e x a c t l y matches one of the surface frames for a #WHY-type question, since it starts with the word n ~ h y V i n d contams subject-verb inverslbon,

Suppose, however, the sentence had been "Why you place an x IR the center squaren, or "How come d i d you place an x i n the center square*. Each o f these sentences matches a surface frame for a MY-type question, except that i n both cases subject-verb inversion i s incorrect. In such a case, the system can, if it chooses, decide t h a t the user has made a minor error, and allow the sentence t o be processed anway. The locally-driven semantics-first approach Lets this happen i n a natural way.

( 2 ) A new surface frame: Another problem arises when a new clause introducer is

encountered, as i n : "Wherefore d i d you place an x i n the center squareM. Here, as described i n section 3 , the system hypothesizes that this may be a new way of asking a #WHY-type question. Since syntax reports that subject-verb inversion has taken place, the system can therefore create a new surface frame:

Wherefore ACTIOM(SV1NV)

t o be added t o the frames associated w i t h #WHY.

B In summary, the adaptive -5ys tern stores i t s l inguis t i c knowledge i n a very accessible form. I t is not embedded in the parsing l o g i c . howledge of how actions and clause-functions are described is represented i n a descriptive, manipulable format. Syntax is domain independent, and is used only t o make cornants, with semantics playing the guiding role. This organization allows the parsinglunderstanding process t o proceed kn a f lexible fashion,

5 . CONCLUSION

Language communication is an i n h e r e n t l y a d a p t i v e medium. One sees t h i s c l ea r ly ~f one takes a problem t o a lawyer and spends time trying t o assimilate t h e r e l a t e d " l e g a l e s e n . One a l s o sees i t i n any conversation where a persron is t r y i n g t o convey a complicated idea , expressed i n his own mental te rms, t o someone else. The l i s t e n e r must t r y t o r e l a t e t h e words he Rears to h i s own set of concepts . Language has , presumably, evolved t o f a c i l i t a t e t h i s s o r t of i n t e r a c t i o n . Therefore it is reasonable t o expect t h a t a good deal of the structure of language is i n some s e n s e s e t u p t o assist i n this adap t ive process. By t h e same t o k e n , studying language from an adap t ive standpoint shou ld p r o v i d e a f resh p e r s p e c t i v e on how t h e va r ious levsls of l i n g u i s t i c structure i n t e r a c t .

REFERENCES

[ l ] D a v i e s , Q.J.M., and Isard, S.D., 'Ut terances as Programs, "resented a t t h e 7 t h I n t e r n a t i o n a l Machine I n t e l l i g e n c e Workshop, Edinburg, J u n e 1972. [2] Enea, H . , and Colby, K , M . , ' Ideolectic Language Analysis f o r Understanding Doctor -Pa t i en t D i a l o g s ' , Proceedings o f t h e 3rd IJCAI, Stanford, August 1973. [3 ] Fillmore, C.J. , 'The Case for Case' , i n 'Universals i n L i n g u i s t i c Theory', Bach and Warms (Eds. ), Wolt, Rinehar t , and Winston, I n c . , Chicago 1968. [ 4 ] Joshi, A . K . , and Weischedel, R.M., 'Some F r i l l s far the Hodaf TIC- TAC-TOE of Isard and Davies: Semantics of Predicate Complement Constructions,' Proceedings of t h e 3 rd IJCAI, Stanford, August 1973. 5 ] e , P .L., 'A Locally Organized P a r s e r f o r Spoken I n p u t ' , Corn. ACM 17, 11 -(Nov, 19741, 621-63@. 163 Miller, P.L., 'An Adaptive System: f o r Natura l Language Understanding and Assimilation', RLE Na tu ra l Language memo No. 25, H I T , February 1974. [ 7 ] Reddy, D . R . , Erman, L . D . , Fenne l l , R.B., and Nealey, R . B . , 'The HEARSAY Speech Understanding Systemt, Proceedings of' the 3rd HJCAZ, Stanford, August 1973. [ a ] Walker, D.E., 'Speech Understanding through Syntactic and Semantic Analysis', Proceed ings of t h e 3 rd IJCAI, Stanford, August 1973. [ 9 3 Weizenbaum, J . , 'Eliza- a Computer Program f o r the S tudy of Natural Comunicatian between Man and Machine', CACM 9 , 1972. [ l o ] Winograd, T. Procedures as a Representation of Knowledge Fw a Computer Program Tqr Understanding Natural Language, MAC-TR-84, P r o j e c t MAC, MIT, Cambridge, Mass., February 1971. [ l l ] Woods, W.A., and Kaplan, R . N . , 'The Lunar Sciences Natural Language Information System" BBN Report No. 2265, Bol t , Beranek, and Neman Xnc. September 1971, [12] Woods-, W.A., and MakhsuP, J . , 'Ovlechanical In fe rence Problems i n Continuous Speech Understanding , Proceedings of t h e 3rd HJCAB, Stanford,

1973.

1575 ACL Mcetlng

CONCEPTUAL GRAMMAR

W I L L I A M A , M A R T I N Kassachusetts Insti t u t e of Tech~ology

In OWL, an implementation o f conceptual grammar, t h e two

types o f data items are symbols and concepts and the two bas ic

data composition operat ions are specialization and restriction.

A symbol is an alphanumeric s t r i n g headed by ". Symbols

correspond to words, suffixes, pre f ixe s , and word stens in

Znglish and the programer can introduce them a t willm

OWL concepts correspond t o t h e meanings of EEglish words

and phrases. They are constructed using the specialization ope-

ration, comparable t o CONS i n LISP* (A B) is t he specialization

of A , a concept, by B, a concept o r symbol. OWL f o r m a branch-

ing tree under specialization, with SOMETHING a t the t o p .

Concepts are given properties by restriction, which puts a

concept on the reference list of another concept (compare proper ty l ists and S-expressions in LISP). A / B is the r e s t r i c t i on of A

by B. The categories in the specialization tree are semantic, but

we use them also f o r the purposes usually assigned to syntact ic

dategories.

A predication is a double specification of 2 model such as

present tense or can. Examples are

The pool is full of water. ((PRES-TNS (BE (FULL 94TER)) J POOL/THE)

The cookie can be in t h e j a f . ( (CAN (BE (IN JAR/TIIE))) COOKIE/THE)

aob is the fa ther o f Sam. ( (PRES -TKS (BE (FATHE: SAM) ITHE) ) BOB) 3ob hits the b a l l . ((PRES-TNS (HIT BALLITHE)) Boa) Bob is hitting the b a l l . ((PRES-TNS (BE (-ING (HIT BALL/THE))))BOB)

Starting from t h i s base we will discuss a number of issues

buch as n~minalization incorporat ion, and deep vs surface cases.

American Journal of Computational Linguistics ~ i c r o f f c h e 32 : 58

JOHN F. BURGER, ANTONIO LEAL, AND A R I E SHOSHANI

System Development Corporation Santa Monica, California 90406

m c T

We describe a natural-language recognition system having both applied and

theoretical relevance. A t the applications level, the prwram w i l l give a

natural ccmmunications interface facility to users of existing interact ive

data management systems. A t the theoretical level, our work shows that the

useful infoxmation i n a natural-language expression (its "meaning") can be

obtained by an algorithm tha t uses no formal description of synt-. The

construction of the parsing tree is cont ro l led primarily by semantics i n the

form of an abstraction of the nmicxo-world" of the DMS's func t iona l capabil-

ities and the organizat~on and semantic relations of the data base content

material. A prototype is current ly implemented in LTSP 1.5 on tho IBM

370/145 computsr at System Development Corporation.

In a recent article in Scient i f ic , American, Dr. Alphonse Chapanis says, "Tf

t r u l y interactive computer ( ; y s t m are ever to be created, they will ~omehow

have to cope w i t h the... errors and vio la t ions of format tha t are the rule

rather than the exception in normal human ccmmunication" [1] . An example

dialogue produced by t w a persons interacting w i t h each other by teletype-

writer to solve a problem as~igned to them by experimenters showed that :not

one grernaaatfcally correct sentence appears in the entire protocol. tl

Many existing language pmcessors (woods , Kellogg , Thcmpson , etc. ) [ 2,3,4)

are limited to what Chapanis calls "Irmnaculate prose," that i s , "the sen-

tences that are fed into the computer are parsed in one way or another so

that the m e a n i n g of the ensemble can be inferred frm conventional rules of

syntax," which are a £0- descr ip t ion of the language. In effect, users

are required to in teract w i t h these s y s t e m in sme formal language, or at

least i n a language that has a formal representation i n the computer system

that a user's expression must conform to (we are t h i n k i n g , in t he latter

instance, of Vhampsonls REL, which has an extensible formal representation

facility). In addi t ion , most natural-language question-answering systems,

including all referenced above, require that a user's data be restruct-wedl

and reorganized acwraing t o the pa r t i cu la r data base requirements of the

natural-language system to be used.

A t the level of a r t i f i c i a l in te l l igence research [ti ,6 ,?'I , Mere is same

interest in systems that recognize meaning i n natural-language expressions

by methods that dd not m i r e compiler-like syntactic analysi~ of an

expression prior to asmantic interpretation. We believe it is possible,

practical, and feasible, using new lingufstic processing strategies, to

design a natural-language interface system that will permit flexible, intu-

itive coaansmicatiba w i t h information management systems and other computer

programs already in existence. This interface is open-ended in that it has

no prejudice about t h e user's system funckians and can be joined to almost

any such system with relatively l i t t l e effort. I t i s , i n addition, able to

infer t h e meaning of free-form English expressions, as they pertain to the

host system, without requiring any formal description or representation of

English.

THE SEMANTIC INTEREACE ALTERNATIVE

The syntactic inflexibiiity of existing natural-language processors limits

their usefulness i n interactive man-madine tasks. O u r approach does not

use a collection of syntax rules or equations as they are normally defined.

Instead, we construct a dictionary in which w e define words in terms of their

possible meanings with respect to the particular data base and data manage-

ment system (DMS) we want to use and according to the possible relations

tha t can exist between data-base and I3MS elements ( e . g . , an averaging func-

t i o n on a group CKE numbers) i n the limited "micro-world" of this precisely

organized data collection. Words appearing in a user's expression t ha t are

not explicitly defined are ignored by the system i n processing the expres-

sion; an example would be the word "the," which is usually not meaningful in

a data management environment. Wa thus avoid the expressive rigidity that

formal syntactic methods hposa on tha user and the excesaivcs time and

resource consumption tha t results from the catibinatorial explosions usually

produced by such rnethade.

We distinguish in their def ini t ions beween two types of words: content

words m d function w o r b (or "operatore"). Content words are wads whoae

'meaningsw are the objects, events, and concepts that make up the subjects

being referred t o by users, More precise ly , for data axetnagernent systems,

these meanings (or "concepts") are the f i e l d names and entz'y i d e n t i f i e r s f o r

*e data b-e and the names for available IHS operations such as averaging,

s d n g , sorting, comparing, etc. Function words serve as connectors of

content words. Their use i n natural language i s to indicate khe manner in

which neighboring conltent words ar'e intended to relate to one another. In

the example "the salary of the secretary ," used belaw, "salary" and

"secretary," are content words, and "of" is a function word used to connect

theta.

Many cmntent wor& are context sensitive, In a particular data base, fo r

btmcm, the ward "salary" may refer t o the data-base f i e l d name SECSAL if

the saXW frs "of a secretary," but may also indicate the f i e l d name CLKSAL

if it is a *salary of a clerk." In recpgnition of this we therefore def ine

eaah aontent word by a set of one or more pairs of the form

( ( X I Y l ) (X2 Y2) . . . (Xn Yn)) where the Xi a d Y i are " o o n c e p ~ " (that is, f i e l d names, etc.) as described

above. This expression may be interpreted as, "if the word so defined i r j t

contactually related in a sehtance to Xl, its particular meaning in this

centact is Y1, if it i s r eo related b X2, it meme Y 2 , m d ao forth." This

particular oontextual mnaranfng af the word is callad its sense. Two content

warm are consrid=& to bls artmantically related i f the in te rsec t ion of the

X i ' a fmtn the definition of one wort! w i t h the Yi's from the d e f i n i t i o n of

U1Q other ira not empty.

To get a more i n t u i t i v e understanding of this process, suppose, again, t ha t

a data base contains ent r ies for both secretaries and clerks w i t h salaries

fox each. Suppose "Suzi&' is an instance of a secretary and om" is an

instance of a clerk. We then have three words defined as follms:

Suzie ( (SUZIE SECY) )

Torn ( (TOM C-LK) )

Salary ( ( sECY SECSAL) (CLK CLKSAL) )

Processing me phrase "Suzie ' s salary" would i n t e r s e c t the Y i ( " (SECY) " )

from t h e def in i t ion of "Suzie" w i t h t h e Xi's ("SECY" and "CLK") from t h e

definition of "salary." The intersection is nan-empty ("(SECY)") , and, i n

discovering the semantic relationship the sense "SECSALI-' is assigned t o the

word "salary." Similarly, "Tan's salary" assigns the sense "CLKSAL" t o

"salary. !I

A particular bplmentation of the natural-language interface processor

operates for a par t i cu la r DMS/data-base t a r g e t system. It contains a

particular &&ion- created for t h a t t a r g e t system. For a par t icu lar dic-

tionary, the s e t of a21 l is ts 05 pa i r s as described above, therefore,

consti tutes the equivalent of a ~ a n c c p t q ~ a p h ox network for the part icular

data b a a malogous to those U R Q ~ hy many of the more conventj-onall, parsers

Pox semantic analysis folluwing (or during) the syntactic phase of parsing.

In the analysis of a particular input by our system, two words i n context

are t e ~ t e d using t h e "intersection" method described abave and, if they are

found to be semantically r e l a t e d , they are considered candidates fo r

"connection" as descrrLbed below. Two words so connected £ o m a phrase.

Function words are defined as operators or processors t h a t perform this

semantic test . The defini t ion of one function word dif fers fm that of

another according to its slope (see belaw) and also in that t h e operational

definition of a function word can reject a connection even though t h e two

words may be samntically related. In the operational def in i t ion of t h e

function word may be a list of acceptable concepts or a rejection list of

unacceptable concepts. In most conceivable data bases, the phrase "salary

in the secretary" would be thus rejected by the function word "in. n

As the analysis of an input expression proceeds, a "clumpifig" of word and

phr as e meanings more and more explicitly normally,

processing of the entire sentence r e s u l t s in a tree structure made up of the

connected senses of a l l the content words fran the sentence. This result we

term the sentence qraph even though the input expression may not be a

grammatically cmplete sentence. This sentence graph will be t ransla ted

in to statement.

We recognize t ha t the linear ordering of the words in an input expression

is not entirely randm and t h a t certain aspects of me function of syntax

must be taken into accorunt. This is done by means of a new and pwerful

azgorithm b k d on what we cal l the syntactic-semantic slope. Linguists

generally recognize that whenever two units of meaning are combined, one is

semantically domfnant and t h e other subordinate, as a modifier is sub-

ordinate to the modified word. A f t e r coenbinatfon, the d d n a n t word may be

wed in m o s t cases to refar to the canjoined pair. Thus, a "red herring"

18 a "herring" (not a "red") , and the "salary of t h e secretary" is a

"salary." If this relationship of dominance i s represented vertically on a

ltrectangular graph (i.e., dominance on the Y-axis), and if t&e l i n e a r order-

ing of the words in the expression is represented on the X-axis in n o w 1

left---right: order, then the connection of an adjacent pair of content

words or phrases will describe a linear slope on the graph. The slope is

positive eir negative as the dominating sub-unit is, respectively, to t h e

right or to the left of the subordinate sub-unit. For example, the phrase

"red herring" makes a positive slope, thus:

HERRING

/ RED

and "the salary of the secre=" makes a negative slope:

S;71LARY

Thus, the ~ p e r a ~ o n a l meanings of fqnctian words operate on the meanings of

nearby content words. Dominance is assigned, semantic relationships are

verified, and the relationships so discovered are accepted or rejected. If

accepted, the two word-meanings are connected, and the acceptable sense is

assigned to the dumllnant word.

Eunction words may connect content words in "positive," "negative ," or

"peak" connections. me follming are examples of each mannax of connection:

1. "Of" is a negative operator, as in " the salary of the

SALARY

2. " ' 8 " is a positive operator, as in "the secretary 's salary":

3. "And" is a peak operator, as in "Atlantic and Pacific. " In

contrast w i t h positive and negative operators, peak operators add

a representation of their m semantics i n t o the structures they

build ;

AND

\ A-IC PACIFIC

4. Between any two adjacent content words there is an implicit "empty"

operator t h a t is a positive operator, as in "red herring":

RED

In general, all prepositions are defined as negative operators. This is

equivalent Go the rule

used by syntactic processors. The positive empty operator is equivalent to

the rule

N P + A x x r P 3 P

and athew, while vexbe and conjunctions are defined as peak operators,

giving our atatemcnt o f rules such errs

s+NPvE'NP

MP + NP CONJ NP.

Each operator has the faci l i ty to accept or reject any semantic rejlation

accordin9 to the precise def in i t ion of the function word for the host data

management system.

Progressive connection of word meanings and previously connected groups or

"phrase meanings" results in a tree graph t h a t we ca l l the sentence qraph.

For example, the question "What is ;t;he surface displacement of U S . diesel

submarines?" could, f o r a particular data base, produce from the dictionary

a string of content-word and funeion-word definitions that might be rep-

resented typographically l i k e this:

( (SUB SURE-DISC) ) <OF> ( (U . S. LOC) ( (DIESEL TYPE) ) ( (LOC SUBS)

(TYPE SUBS)

As a xesult of processing, these will assemble into a tree structured (using

the senseg of the words) l i k e this:

WHAT / sUm-D=sP P

LOC AsuBs TYPE

U , S . DIESEL

Even though this tree, or sentence graph, i s created as a result o f semantic

relationships instead of Eonnal r u l e s of grammar, it still. closely resembles

the "parse t ree" produced by m o ~ t conventional syntactic language processors.

With respect t o the user's target data management system, t h e sentence graph

is preci~e and unambiguous and contains enough information for a

straightforward translation into the formal query language of the EMS. In

SDCrs DS/3 lanwage, f o r example, the above question would be expressed as

PRINT SURF-DISP WHERE TYPE EQ DIESEL AND lXXl EQ U.S.

The response to the usex's question will thus be the response frclrn h i s DMS

t o the formal query statement.

The user's input in this hypothetical example i s proper i n fom and grammar.

However , it need not have been. The request

OBTAIN SURFACE DISP FOR US SUBS SUCH AS HAS TYPE EQ DIE=.

would produce exactly the same sentence graph and thexefore, exactly t h e

same f o m l query statement with the same response f r o m the DMS.

It is not l ike ly t ha t a syntax-based parser would have anticipated the odd

laxxguage-use and grammar of this last request. Without a syntax rule t h a t

would alluw for the phrase "such as has" such a parser would not look at the

semantics involved and would be unable t o interpret the request. Our syntax

algorithm gets the same results that would be expected f m m the application

of syntax rules without the need t o anticipate each grammatical construct

expected from the user.

In overview, the parsing algorithm makes a series of positive, negative, and

peak connections based on the operational meanings of the function wards

(including the "empty" aperator) and on the relations between meanings of the

content wort%?. The algoridt-Xlm adheres to the following rules:

e 1 Connections between content words are possible only if

the result of the intez'sectfon t e s t described & m e is non-empty

and i f this result i s not rejected by the operation of the function

word p e r f o d n g t h e test. The function word d e f i n i t i o n also deter-

m i n e s which w o r d supplies its X ' s and which its Y's for the t e s t ,

It thus controls which w o r d has its sense d e t e d n e d if t h e t e s t

ia successful. Most of ten (though there are exceptions) , posit ive

operators use the X's f r o m the w o r d to the r i g h t and the Y ' s from

the word to the left of . b e operator. Positive operators, these-

fore, determine the sense of the word t o the right. This is

i l lustrated using, again, the secretaxy and her salary, Consider

the defini t ion of "Suzie" and "salary" as shown on page 5 , The

phrase "Suzie's salazy" has two content w o r d s , "Suzie" and

"salary, " separated by the function word , " s , " This function

word is a positive operator and, hence, applies the intersection

t e s t t o the X i from the definition of "salary" w i t h the Yi from

the definition of " ~ u z i e . " These values are, xespactively,

'I (SECY CLK) " and " (km) . " The intersection yields " (SECY) , "

which is acceptable to the " ' s " operator, and the connection is

made with "salary" as the dominant word. The sense of "salary"

is the Y i associated with "SECY" in t h e def in i t ion of "salary,"

hence, "SECSAL." T h i s selection process is reversed f o r negative

aperators, while peak operators employ both kinds of t e s t s , one

on each s i d e of the peak.

Rule 2: N o node i n a sentence graph may have m o r e .than one dominating

node. That is to say, a l l connections m u s t r e s u l t i n trees, This

I s a canmon asswnptLon consistent with conventional syntax-driven

parsers.

Rule 3: Given a subtree, a const i tuent on its left has the poss ib i l i ty

of conneation only to nodes of the subtree's positive adjacent

slope, and a const i tuent on the r i g h t can connect onLy t o the nodes

i n the adjacent negative slope. In tu i t ive ly , this means that if

the nodes of a subtree are connected by "lines" that are "opaque

b a r i e r s r n then a constituent on either side of t h e subtree may

connect to it only on those nodes that it can rlsee.r' I t may not

connect t o nodes on the "inside" or the "fax s ide" of the subtree.

This i s a powerful h e u r i s t i c rule that eliminates t h e need t o t ry

connections to many syntactically impossible portions of the sub-

tree. In effect this one rule, together w i t h the definitions of

the function words, replaces all the syntax rules used by most

conventional parsers.

Rule 4: In order t o minimize disconnection of existing subtree

structures (badcup) and s t i l l consider a l l possible connections,

the system should, whenever possible, constrztct,subtrees s t a r t i n g

from the top and make new connections from belaw. This rule leads

to the following algorithm: Scan the consUtuents from left t o

right making negative connections, then scan from right to left

making positive connections. S c a n thus back and forth unti l no

more connections can be made. Then make any poasible peak aonnec-

t ions and repeat the algorithm. Continue t h i s process u n t i l a l l

const i tuents have been connected i n t o a single tree,

We have observed t h a t if ambiguities exist under these conditions, they w i l l

be semantic and, in all probability. not resolvable by any further processing

or analysis of the expression. Therefore. there is no need to carry along

temporary multiple construction poss ibi l i t ies , The algorithm may eirher

query the user at this point for disambiguation or W d w t the pxocesging and

inf o m reason,

I. Chapanis, Alphonse. Interactive human cammunlcation, Scientific

American, May, 1975.

2. Woods, W. A, Trahsition network gr-ars for natural language analysis.

Cozmnunications of the ACM, October 13, 1970,

3. Kellogg, C. H,, et al , The CONVEXGE natural language data management

system: current status and plans. ACM Sym~osium on Information Storaqe

and Ratrieval, University of Maryland, 1971.

4, Thompson, F, B . ; 'Lockman, P. C.; Dostert, B.; Deverill, R, S. REL:

a rapidly extensible language. Proceedings of 24th National Conference,

ACM, New York, 1969, 399-417, - Riesbeck, C, K. Computational understanding. Theoretical Issues i n

Natural Langu~ge Processinq: Proceedinqs of an Interdisciplinary

Workshop in Canputat icmal ~inguist&cs, Psychology, Linguistics and

Artificial Intelligence. Cambridge, Massachuastts, June 10-13, l975.

6, Waltz, D. L. On understanding poetry, Theoretical Issues i n Natural

Langtmgs Processing, Proceedings of an Interdisciplinary Workshop in

Camputational Linguistics, Psychology, Limuistics and ~rtificial

Intelligence. Cambridge, Massachuset-, June 10-13, 1975,

7 . Sdhank, Roger, and Tesler, L. G. A Conceptual Parser for N a t u r a l .

Language. Stanford Artificial InteUigence Project. Memo No. AI-76,

Januaq, 1969.

American Journal of Computational Linguis ties Microfiche 32 : 7 2

P. MEDEMA, W . J. BRONNENBERG, H. C. BUNT. 5. P. J. LANDSBERGEN,

R , J. H. SCHA, W . J. SCHOENMAKERS, AND E . P . c. V A N UTTEREN

Philips Research L a b o r a t o r i e s E indhoven , The Netherlands

ABSTRACT

This paper outlinee a recently implemented que~tion answering system , called

PHLIQA 1 , which answers English questions about a data base . Unlike other existing aysteme , that directly tramlate a syntactic deep structure

into a program to be executed, PHLIQA 1 leads a question through several

intermediate etages of semantic analysis . In every stage the question is repre-

sented a0 an expression of a formal language, The paper describes aome features

of the Languages that are &uc~essivelg used during the analyeis process : the

English-oriented Formal Language , the World Model Language and the Data Base

Language . Next , we ahow the separate conversion steps that can be distinguished

in the process. We indicate the problems that are handled by these conversions , and that are often neglected in other systems.

1. Introduction

PHLIQA 1 is an experimental ~ y e t e m for answering isolated English questions

about a data base . We have singled this out as the central problem of queation

anawerlng , and therefore postponed the treatment of declaratives and imperrt

tives , as well aa the analyak of discourse untll a later vereion of the system . The data baee is about computer installations in Europe and their users . At

the moment, it is small and resides in core- but its structure and content

are those of a realistic Codagyl format data base on disk ( CODASYL Data

Base Task Group [ 1971 'J )

Only one module of the system , the wevaluation componenVT , would have to be

chmqpd in order to handle a lha l t f data base .

2, PELIQA 1 ' e top level design

Like other recent QA systems ( e,g, Petrick 1 1973 ] , Plath 1 1973 ] , Winograd 1 1972 ] , Woo& [ 1972 ] ) , the PHLIQA 1 system can , on the

most global level , be divided into 3 parts ( aee fig. 1 ) :

-- Underetandtng the question : Translating the question into a formal expree-

sion which represents its meaning with respect to the world model of the

- Computing the answer : Elaborating this expreseion , thereby finding the

answer, it is repreeented in the system' s internal formalism.

-- Formulating the answer : Translating this answer into a form that can be

more readily under8 toad .

questlon in English

I formal expression , representing the meaning of the question

I Answer Computation

I answer In internal format

Answer Formulation

answer in external format

Fig . 1. Global subdivision of PHLIQA 1,

The interface between the Question understanding component and the Answer

Computation component 1s a formal language , called the World Model Language

( WML) . Expressions of this language represent the meaning of questions with

respect to the world model of th@ system. Its conrrtants correspond to the concepts

that canstitute the universe of discourse . The language is independent of the input

language that ie udled ( in this case English) , and also independent of the storage

structure of the data base.

If we now look at a further subdivierion of the component& , the difference between

PHLIQA 1 and other systems becornea apparent . Both above and below the World

Model level, there is an intermediate stage of analysis , characterized by a

formal language , resp r

- The Engliaboriented Formal Language ( EFL) , which containa constant^ that

correspond to the terms of English, This language is wed to represent the

semantic deep structure of the question , That divides the Question U n d e ~

standing component into two succes~ive subcomponents I

a. Constructing an EFL expression . using only linguistic knowledge . b, Translating the EFL expression into a WML expression, by taking

knowledge about the structuf.e of the world into account.

- The Data Base Language ( DBL ) , which contains conatants that correspond

to data base primitives . ( The World Model constants do not correspond to

daW base primitives , because we want to handle a realfs tic " data base :

one that was designed to be stored efficiently , rather than to reflect neatly the

structure of the world . ) This splits the Answer Computation component into two successive subcomp*

nenta :

a. Translating a WML expression into a DBL expression taking knowledge

abut the data base structure into account,

b. Evaluating the DBL expre~sion . The aebup of the system that one arrives at in this way, is shown in fig, 2.

In section 3 , we gay eamething more about PHLIQAq s formal languagqs in

general . How the three succeesive translation modules are further divided into

smaller modules , c a U d ftconvertorsw , is dfscu~sed fn the sections 4 , 5 and 6,

Section 7 treats the evaluation component . The Answer Formulation component

is very primitive , and will not be considered further .

question in English I

Question Under0 tanding

Answer Computation

expreabion of Englisboriented Formal Langua$te

I ( Semantic Deep Structure )

EFL- WML - - - - - owledge of tsanslation - - - - - World Structure

expre $ sion of World Model Language I

- WML- DBL - - t - -

translation f - - -

[ expredsion of Data Base Language 1

I answer in internal format

Formulation

anrswer in external format

Fie 2, PHLIQA 1 main components .

3. PHLIQA 1' B formal laxlguages

3. 1, sylitax

The three PHLIQA languages ( the English-oriented Formal Language , the World Model Language and the Data Base Language) have largely identfcal

syntactic definitions . A s pointed out already, their moat important difference

is in the constants they contain . T h y share most , but not all , syntactic

COIlJ3 t~C!tf~Ils ,

PHLIQA expresgions are rt trees TT that conaists of terminal nodes ( conetants

and variables) and syntactic constructions . A syntact'ic construction is an

unordered collection of labeled branches , departing from one node . The branches of a PHLIQA fl tree " can converge to a common subtree . Using a system of semantic types , the syntax of a PHLIQA language defines

how expressions c m be combined to form a larger expressfan. For every

syntactic conetruetion, there ie a rule which specffies :

- What the semantic types of it8 Immediate sub-expressions are allowed to be . ( There is never a restriction on the syntactic form of the sub-expressions , )

- How the semantic type of the remitting expression is derived from the

semantic types of the immediate sub-expressions . Given the types of the elementary expressions ( the constants and variables ) , this def'lnes the language, ( Sources of inspiration f o r the syntax of our formal

languages were the Vienna Definition Language- ( Wegner [ 1972 ] ) , and a

formulation of Higher - Order Lo@c by J.A. Robinson [ 1969 ] . ) Some ~imple examples of semantic types are the foXlowing :

A comtant reprersenting a single object has a simple type . E.g, , 6 has

the type " integer " , A c6nstant representing a collection of objedta of type oc

has a type of the form <d> . E,g. , companies has the type "(company)

" intagera has the type "(integer) .

A constant representing a function that can have arguments of type and

values of type ('3 has the type + . E.g. , the function

Tt IL-cornpany-sites TI has the type ?? company* &il%y: the function &sum " has the type t v (integer) integerw.

The syntactic rule for the construction function - application t' could state

that the emreasion

is well -- formed if T is a well-formed expre~lsion of type and T i s a 2 1

well - formed expression of type 6 -+ /3 , where oC and may be

any type ; the whole expression then has the type P The PHLIQA languages contaln a wide variety of syntactic constructions , e,g.

constructions for different kinds of quantification , for selecting elements from

a list, for reordering a list, etc ,

3. 2, Semantics

The PaIQA language8 have a formal semantics which recursively defines the

values of the expressions, This definition assumes as primitive nations the

denotatian~ of the conetants of the language : function - constants denote

procedures , and the other canstants denoh value - expressions , This means

that if we know the denotations of the constants occurring in an expreesion , the

value of the expression fs defined by the semantic rules of the language , For

t b Data Base Language , we indeed know the denotations of the constants ; what

we call the data base is nothing but the implementation of the " primitive

procedure8 ", t e. : the procedures corresponding to DBL functions , and

the procedures for finding the value - expres~ions of the other DBL constants .

Therefore , the DBL expressione are actually evaluable . For the World Model Language and the English-orientad Formal Language , such a data base does not exiat , but one could be imagined . We express thls by saying

t4&t the WML and EFL expressions are * evaluable with respect to a virtual data

base

4, Constraction of the semantic deep structure of a question.

A s we have seen, the EnglfsMriented Formal L m a g e differ8 from the other

tfttu, languagee in two respect8 :

1, It has different constants , of'whieh the most important are t

a names of sets corresponding to noune ( e.g. * computers ") , to verbs ( " buy - sitrtatiane * ) and to ssme of the prepoeitions

( in - place - situations ) . b. grammatical functions t subject, object, etc .

2, It Borne different constructione . Here the most striking difference is that

EFL conekuctinns contain eemantic and syntactic featurea . The semantic

features influence the formal semagtfca of the constructlorn ( e,g, the definite-

nees or indefiniteness of a noun phrase influences the choice of the kfnd of

quantification for that noun phrase ) . The syntactic features only play a role

during the tranaiormatian process from English to EFL . T t should be noted that Ln general two eynonymoue eenteqes need not be represented

by tho same semantic deep structure in EFL . For example , the synonymy of

A buys B from C and C sells B to A is not accounted for at tbia level . Hwever ,at the level of the World Model Language synonymous sentences are

mapped onto equivalent ( not necesaarilg identical ) WML emrerssr iom . The construction of the semantic deep structure in EFL consists of three main

phanes r

phase 1: a lexicon , providing for each word one o r more interpretations , represented by pairs ( CATi, SEM \ , where CAT I s a syntactic category

i i and SEM an EFL expression .

i

phase 2: a set of rules that enables to combine the sequence of pairs ( CAT SEM1) , i t corresponding to the original sequence of words , into higher level categories and

more complex structures , until we have ultimately the pair ( SENTENCE , SEM ) , S

where SEM is the EFL expression for the bomplete sentence . S

A rule of phase 2 is a combination of a context free rule and a set of rules on EFL

expressions , that show when and how a sequence of pairs

can be reduced fo a pair ( CAT , SEMR) . R The general format of theae rules i s :

- context free reduction rule :

........ CATl +. + CATk -> CAT R

- EFL rules :

The C O N D ~ ' s are conditions on the EFL expressions SEM . . , , , 1'

SEMk . The ACTION ' s ahow how a new EFL expression SEM can be constructed with the

i R helpofSEM .....

I' SEMk . The rule i s applicable if at least one of the

conditions COND is true . Then SEM ia constructed according to ACTION and I a i

the aequence of pairs i s reduced to ( CAT SEM ) . If more than one of the R' R

COND is true , we have a local ambiguity. i

phase 3: transformation rules that transform the semantic surface structure into

an EFL expression that I s called the semantic deep structure . ~ h e e e t r & m f ~ r

mation rules handle aspecte of meaning that could not be resolved locally , during

phase 2. This applies for Instance to anaphoric references and elllptic clauses

in comparative cons-ctlons . A ~impler example is the specification of the subject in a clauae like ' to uee a computer ', The eemantic surface structure of this clause means: there is a

usesituation , with ~ a m e computer as its object , and an unspecified subject . Phase 2 can be said to ' disambiguate ' thi@ expression in a context like

' when did Shell start to q e a computer 3 . A transformation specifies the subject of the use-situation as Shell '. This

transformation would not apply if we had the verb propose instead of start ' .

The condition8 of phase 2 and phase 3 contain a rkhortcuV' to the world model1

the semantic types of the world model interpretations of the EFL congtants are

inspected in order to avoid the construction of semantic deep e tructures that

have no interpretation in the world model . This blocks many unfruitful parsing

paths.

5 . Translation from semantic deep structure to unambiguous World Model

Language expression

The translation from a semantic deep structure ( EFL expraseion ) into an un-

arnbiguoua World Model Language expmsarion proceeds in 3 phases1

phase 1s Translation from EFL expression Into ambiguous WML expression.

b tbls phase , traneformations are applied which replace expressions containing

EFL conetants by expreiseiolu containing WML canatants . Their most conspip

uow effect is the elimination of "situations" and rTgrarnrnatical functionst1. It is

important to note that the resulting expreseion often contains several "ambig-

uous constantsW, These ariae from polyeemous brms in English r words that

have a "range1? of posaible meanings . Such terms lead now to expressions with

ambiguous constants8 constants that stand for a whole class of possible "insta*

cesT' . An expression containing such constants , stands for the class of wellr

formed expressions that can be generated by 'Ymtantlating" the ambiguous c o w

stants .

phase 2% Disambiguation of quantification^ . Many sentences are ambiguous with respect to quantification ,

E .g . Were the largest 3 computers bought by 2 French companies ? can either

ask whether there are 2 French companies such that they both bought each of

these computers , o r , perhaps more plausibly , it can ask whether there are 2

French companies such that together they bought these computers . Until thie stage in the process , the representation of such questions contains

constructions which stand for both interpretatiow at once . But now that the

system' 8 assumptions about the structure s f the world are reflected In the ex-

pression, some such interpretations may be ruled out as implausible , because

they would lead to the same answer , independent of what the atate of affairs in

the world is . E ,g ., the first interpretation of the above example question

has the value 'YalseW , independently of the values of the constants in the ex-

preaeion . ( Because the assumption that a computer can only be bought by one

company wapJ Introduced by a previous traneformatfon ) . Therefore , the second

interpretation is chosen,

phase 32 Di~arnbiguation of WML conestants . The ambiguous WML constants can be instantiated in a very efficient manner by

using the semantic type system: The possible interpretations of an ambiguous

comtant are severely restricted by the semantic types of the other constants

that appear in it8 context,

6. Tramlation from World Model L a n w g e expression to Data Base

Laqpage expression -

In the World Model Language , constants correspond to the concepts of the universe

of discourse, In the Data Base Language, conatants correspond to primitive

logical and arithmetical procedures and to primitives of the data base . The choice

of these primitives was governed by coneiderations of efficiency, rather than by

the wish to represent neatly the structure of the univeree of discourse. Therefore , WML and DB conb fn different conatants . The translation from a WML expression to the DBL expression that will be evalu-

ated, proceeb in three stages :

1, Paraphrase of the WML expression, in order to eliminate * infinite notions ". WML contains conrrtanb representing infinite sets or infinite continua , like

integer8 * , * moaey~amounts and ?' time ' l . Such comtants can not be

directly or hidirectly represented in the data base , and hence have no D B b

tramlation. By paraphrasing the expression, the infinite notions can of*n

be elirntnated . 2, Translation of expressions conklning WML constants into expressions con-

&ining DBL cow tanh ,

This tranalatlon is required by phenomena like the following :

- it Ls poasible that a class of objects is not represented explicitly in the data

baee , while propertlee of ib elementa are represented indirectly, as

properties of other , related objects , ( E.g. , cities do not occur in the

PHLIC&Il data base , but their names are represented as the ciwnarnes

of sites . ) A special case of this phenomenon ie the representation of a continuum by a

class of diacrete objects ( E.g. , core ie represented by rr core

memories ") t

-- objects may be represented more than once in the data base. E.g. , in the

PHLIQA 1 database, the flle of computer users and the file of manufacturers

can contain records that represent one and the same f i rm.

-- the data baee is more limited than the world model . Some questions that

can be expreased in WML can be answered only partially or not a t all r

the WML expresrition has no DBL translation. The present convertor detects

such expressions and can generate a message which specifies what informa-

tion ia lacking . Examples of this caae are r the se t '' integers '* ( if the attempt of the previous

convertor to eliminate it has been umuccesr~ful ) , and the date-ottaking-

o u t - - o w e ?* of a computer ( which happens to be not in the data base ) . 3. Paraphrase of the DBL exprenr~ion , in order to improve the efficiency of its

evaluation . The DBL expression produced by the previous convertor can already be evalu-

ated, but i t may be possible to paraphrase it in such a way, that the evaluaii~n

of the paraphrase expression is more efficient, This conversion is worthwhile

because , even with our small data base , the evaluation is often the most

time-consuming part of the whole process ; compared to thie , the time that

transformations take is negligible .

7. The evaluation of a Data Base Language expression

The value of a Data Base Language expression is completely defined by the sernaxl-

tic rules of the Data Base Language ( see section 3 . 2 . ) , and one could cohceive

of an algorithm that corresponds exactly to these rules . For reasons of efficiency,

the actual algorithm differs from such an qlgorithm in some major respects r

- in evaluating quantlficatiom over sets , it does not evaluate more element0 of

the sat than ie necessary for determining the value of the quantification . - if ( e-g. during the evaluation of a quantification) , a variable assumes a new

value , this doe8 not cause the, re-evaluation of any subexpressions that don* t

contain this variable . Currently , evaluation occurs with respeet to a small data base in Core , To handle

a real data base on dierk , only the evaluation of constantn would have to change .

8, PELIQA I ' s Control Smckrrc3

The sections 4 thmugh 7 sketched what the basic modulea of the system ( the

convertors ") do . W e shall now make some very general rernarh about the

way they were implemented . These r e m a r k apply to all convertors except the

parser, whioh is described in some detail by Medema [ 1975 ] . The convertors can be viewed as functiong which map an input expression into a set

of zero or more output expressions . Such a function fa defined by a collection, of

transformations , acting on subexpresslons of the input expression . Each tr&aa-

formation wnrrists of a condition and an action , The action ie applied to a sub-

expression if the condition holde for it . The action can either be a procedure

transformfngra subexpression to its * lower level equivalent '' or it can be the

decbian this subexpressfon cannot be translated to the next lower level '' , "I1 convertore are implemented as procedures which operate on the tree that

repregents the whole f~uestion . The procedures cooperate in a " deptb-first ?'

m m r : a conversion procedure finds suc~es s ive ly all interpretations that the input

expression haa on the next lower level . Far each of theae Interpretations , as soon

as it is found, the next convertbr ie called. If no interpretation can be found, a

message Bving the reason for this dead end is buffered , and control fe returned

to the calling convertor ,

If the answer fs found, it is displayed. If requested, the ayatem can continue its

search for more interpretatlorn . If the answer level is not reached , it displays

the buffered message from the " lowest " convertor that was reached ,

Colophon

The PHLIQA 1 program was written in SPL ( a PL/1 dialect) , and runs under the MDS time sharing system on the Philips Pl.400 computer of the Philips Research Laboratories a t Eindhoven . The quantfflcatio~i~lambiguation ghaae of the EFG-WML translation, the effi- ciency-conVersion ( step 3 ) in the WML-DBL translation , as well a s some parts of the grammar , are not yet part of the running system , though the convertors are complekly coded and the grammar is elaborately specified. During the design of PHLIQA 1 , the PHLIQA project was coordinated by Piet Medema . He and Eric van Utteren deaigned the algorithmic structure of the aye- tern and made decisions about many general aspectxi of implsrnentatlon . The formal languages and related transformation rules were designed by Harry Bunt . Jan Landabergen and Remko Scha . Wijnand Schoenmakera deaigned the evalu- ation component. Jan Landsbergen wrote a grammar for an extensive subset of English A l l author6 were involved in the implementation of the system . During the design of PHLIQA 1 , exteneiva discussione with members of the SRI Speech Understanding team have helped us in making our ideasl more explicit,

References

CODASYL Data Base Task Group April 71 report. A C M , New York, 1971 .

P. Medema A control structure for a question answering sys tern . Proceedings of the 4th Inte~national Joint C~nferen~ce on Artificial Intelligence . Tbilisi , USSR , 1975. Vol. 2 .

S,RPetrick SemanticInterpretaticmintheREQUESTsystem. Proceedings of the International Conference on Computational Linguistice , VoL 1 , Pisa , 1973 .

W, J. Plath Transformational Grammar and Transformational Pars fng in the REQUEST system, Proceedings of the International. Conference on Computational Linguistics , Vol. 2 , Pisa , 1973 .

J. A. Robinson Mechanizing HighexLQrdelr Logic , In : B, Meltzer and D. Michie ( eds. ) , Machine Intelligence 4 , Edinburgh University Pres~l , 1969.

P. Wegner The Vienna Definition Language . Computing Surveys , Vol, 4 , no. 1 , 1972 .

T, Winograd Understanding Natural Language . Cognitive Psychology , VoL 3 , no. 1 , 1972 ,

W. A, Woode , R. M. Kaplan and B. Nash-Webber The Lunar Sciences Natural Language Information System : Final Report . BBN , Cambridge , Masa, 1972 .