The FINITE STRING, Volume 13, Number 6

American

Journal of

Cornputatimal Linguistics

THE

FINITE

STRING

NEWSLETTER OF

THE ASSOCIATION

FOR

COMPUTATIONAL. LINGU I

ST1 CS

VOLUME

13

-

NUMBER

6

SEPTEMBER

1976

Bath t h e Current Bibliography and t h e promised

index t o

AJCL

are absept from the present packet,

because of the difficulties inherent in o p e r e t i ~ n s

based on voluntary effort.

American, Journal.of Computational Linguistics is published by the Center for Applied Linguistics

for

the Assaciation for Computational Linguistics.

E d i t o r : David G. Hays Professor of Linguistics and of Computer Science,

S t a t e University of New York, Buffalo 14261

Editorial Assistant William Benzon

E d i t o r i a l Address 5048 Lake Shore Road, Hamburg, New York 14075 Managing Editor:

A.

Hood Roberts

Assistant. James S. _Megginson

Production and Subscription Address. 1611 North Kent S t r e e t , Arlington, Virginia 22209

Copyrrght & 1976

A C L t , 1 4 ~ ~ 1 ANNUAL MEETING

-

PROGRAM 5. A B S T R A C T S

.

3

A C M I EMPLOYMENT REGISTER AT COMPUTER SCIENCE CONFERENCE 39

N S F : ADVISORY PANEL FOR LINGUISTICS

. .

,

.

,

. .

4 0

EUROPEAN CQNGRESS ON INFORMATION SYSTEMS & N E T W O R K S

. . .

41

NEW J O U R N A L * TRANSACTIONS ON D A T A BASE SYSTEMS

.

,

.

, , , 42

APPLIED LINGUISTICS: 5 T H ~ N T E R N A T I O N A L CONGRESS

. . . .

43

A F 1 P S : O F F I C E S AND HONORS

. .

, ,

. . .

44

LATSEC: CONGRESSMAN QUESTIONS P A Y M E N T

. . .

. . .

46

THE FUTURE OF MT

.

.

.

. . .

.

.

Rudolph C. T r ~ i k e 47

M T : Moscow INTERNATIONAL SEMINAR

. .

I. I. Qubine 50

B O O K R E V I E W : E S S A Y S ON L E X I C A L SEMANTICS, V O L . I

E d i t e d by V J u . Rozencvejg

.

Reviewed by Raymond D . Gumb 68

American

Journal

of

Computational

Linguistics

Microfiche 5 1 : 3

REG

I STRATI ON :

$15

FOR MEMBERS

$20

FOR NONMEMBERS

$10

FOR STUDENTS

PROGRAM

:

FRIDAY

OCTOBER

8

8.30 Registration

TECHN I Q U E S I N LANGUAGE PROCESS ING

9:00 AN E A S I L Y COMPUTED M E T R I C FOR RANKING ALTERNATIVE P A R S E S

George E . Heidorn, IBM Researcn

. . .

Abstract. Frame 5

9.30 MEDIAN SPLIT TREES: A F A S T LOOKUP TECHNIQUE FOR

FREQUENTLY O C C U R R I N G KEYS

B. A . Sheil, Harvard University

. . .

Abstract, Frame 7

10.00 TRANSLATING ' W E L L - W R I T T E N ' ALGORITHM D E S C P I P T I O N S I N T O

CODE

Jerry R . Hobbs, City College of CUNY

.

. Abstract, Frame 9

10.30 G E N E R A T I N G N A T U R A L LANGUAGE EXPLANATIONS OF COMPUTER

PROGRAMS

George E. Heidorn, IBM Researdh

. . .

Abstract, Frame 1 3

11:0@ THE SEMANTIC I N T E R P R E T A T I O N OF MASS NOUN EXPRESSIONS I N THE P H L I Q A 1 SYSTEM

Harry C. Bunt, P h i l i p s Research Laboratories

. .

.

Frame 1 5

1 1 . 3 0 EMPIRICAL S T U D I E S OF NOUN MEANING

1 : 30 PANEL: EVALUAT !ON BF NATURAL LANGUAGE SYS JEMS

. . .

l*r.mth 2 1 Chair : Joyce B . Friedman, University o f Michigan Participants: William A . mar ti^, M.I.T.

lJilliam

H.

Paxton, Stanford Research Inst Stanley P e t r i c k , IBM Research

Naomi Sager New York University

Eric van Utteren. Philips Research Labs

Terry Winograd, Xerox Research

MACRO MODELING OF NATURAL D I S C O U R S E

8 30 METHODS FOR MODELING DIAL-OGUE ( S P E C I A L GROUP P R E S E N T A T I O N )

William C , Mann, James A Levin, and James A Moore Information Sciences Institute, University of Southern Calif 0lrni.a

. . .

Abstract, Franc 22

1 0 . 0 0 A PROCESS MODEL FOR SPEECH ACT UNDERSTANDING

Allen Munro, University of California, San Diego . . . rrCipx. :

1 0 . 3 0 NATURAL LANGUAGE UNDERSTANDING BY C O G N I T I V E N E T W O R K S

Richard Fritzson, SUNY at Buffalo

...

Abstract, Framt? 32

1 1 - 0 0 HOW TO REPRESENT AND USE KNOWLEDGE ABOUT C A U S A L I T Y

Charles J Rieger , University cf Maryland

.

. .

Abstract, Frame 3 3

1 1 . 3 0 A HEURISTIC S E A R C H A P P R O A C H T O D I S C O U R S E A N A L Y S I S

14th ACL Meeting

AN E A S I L Y COMPUTED M E T R I C FOR R A N K I N G u A L T E R N A T I V E PARSES

GEORGE E. H E I D O R N I B M RESEARCH

The idea of attaching a modifier t o its nearest potential

modificand i s o f t e n s t a t e d as a r e a s o n a b l e heuristic in n a t u -

r a l language p a r s i n g . ( T h i s i s e s s e n t i a l l y K i m b a l l ' s P r i n c i -

p l e Number S o * , "Terminal s p b o l s o p t i m a l l y a s s o c i a t e t o t h e l o w e s t n o n t e r m i n a l node." Although Kimball c a l l s this p r i n -

c i p l e " r i g h t a s s o c i a t i o n " and i l l u s t r a t e s i t w i t h r i g h t -

b r a n c h i n g examples, i t can a p p l y e q u a l l y w e l l t o l e f t - b r a n c h i n g

structures.)

I n t h e s e n t e n c e , "Are t h o s e i n v o i c e s produced from t h e

o r d e r s p r o c e s s e d by t h e system i n New York?", t h e p r e p o s i -

t i o n a l phxase " i n New York" c o u l d p o t e h t i a l l y modify "system",

''processed1', " o r d e r s " , "produced" and " i n v o i c e s " , b o t h s y n t a c -

t i c a l l y and s e m a n t i c a l l y . S i m i l a r l y , "by t h e system" has f o u r

possibilities and "processed" has two. According t o t h e

h e u r i s t i c s t a t e d a b o v e , t h e p r e f e r r e d a n a l y s i s i s : " i n N e w

York" m o d i f i e s "system", "by t h e system. .

.

I ' m o d i f i e s "pro-

cessed", and " p r o c e s s e d .

.

.

"

m o d i f i e s " o r d e r s " .

A " p o t e n t i a l m o d i f i c a n d r t must be a c c e p t a b l e p r a g m a t i c a l l ' y

as w e l l as s y n t a c t i c a l l y and s e m a n t i c a l l y . T h i s means t h a t

i f t h e system r e f e r r e d t o i n t h e above example were s p r e a d

o u t over the e n t i r e country and j u s t had one of its processing

stations in New York, the p r e f e r r e d analysis would have had

" i n New York" modifying "processed1' instead of "system".

Similarly, dialogue context can have an influence on what

qualifies as a potential modificand.

This paper describes an e a s i l y computed m e t r i c that in-

v o l v e s a t t a c h i n g a number to each syntactic unit as it is

formed, t o rank the possible analyses f o r any p o r t i o n of an

utterance according to the above-stated heuristic. The

technique is illustrated by its use in a system which supports

a . n a t u r a 1 language d i a l o g u e f o r automatic business programming,

This system u t i l i z e s syntax, semantics, pragmatics, and con-

1 4 t h

ACL

Meeting

M E D I A N S P L I T TREES:

A FAST LOOKUP TECHNIQUE F O R FREQUENTLY O C C U R R I N G KEYS

B.

A. SHEIL

H A R V A R D UNIVERSITY

Median split (MS) trees are a new technique f o r s e a r c h i n g

s e t s of k e y s with highly skewed frequency d i s t r i b u t i o n s (such

a s the lexemes of a n a t u r a l language). ¶?he basic idea i s t o

m o d e y frequency ordered binary search (FOBS) trees t o con-

t a i n two key values--a node value which i d e n t i f i e s t h e key

which resides a t t h a t node (as i n a conventional binary search

tree), and a s p l i t value which gives t h e l a r g e s t key value t o

be found i n t h e l e f t s u b t r e e . ( I f the keys a r e expensive t o

s t o r e in t h e t r e e nodes, or c o s t l y t o compare, they may be

hashed i n t o integers without a f f e c t i n g t h e algorithm. )

Searching an MS t r e e proceeds as i n a binary t r e e except that

the decision t o go l e f t o r r i g h t from a node whose node v a l u e

does not match t h e current key i s made by comparing t h e c u r -

rent key t o t h e s p l i t , r a t h e r than t o the n o d e , value. The

use of t w o different values allows one t o prevent t h e search

t r e e from _being unbalanced as FOBS t r e e s become when a high

frequency key has an.extreme key v a l u e . I n f a c t , by s e l e c t i n g

t h e median of t h e key values of a node's descendants a s i t s

split value, one can force the search t r e e t o be p e r f e c t l y

balancedo-which both allows a highly space e f f i c i e n t r e p r e -

The average cost per search in a n MS t r e e , like a FOBS tree, depend's o n b o t h t h e frequency distribution of the k e y s ,

and t h e o r d e r i n g relabion on them. Performance analyses f o r

s p e c E i c f r e q u e n c y distributions and key orderings of i n t e r e s t

are presented. and it is shown that, unlike FOBS trees, PIS

t r e e search t i m e i s l o g n bounded and is very stable around a value that c a n be shown t o be o p t i m a l f o r any binary tree

search p r o c e d u r e . F u r t h e r m o r e , i n addition t o being substan-

t i a l l y f a s t e r t h a n "optimum" b i n a r y t r e e search, a n EIS tree

c a n b e b u i l d for a given ordering r e l a t i o n and s e t of frequen- c i e s i n time n log n , as o p p o s e d t o n 2

A discussion of the a p p l i c a t i o n of MS trees to dictionary lookup for English is p r e s e n t e d , and the perfbrmance obtained.

1 4 t h ACL Meeting

TRANSLATING 'WELL-WRITTEN' ALGORITHM DESCRIPTIONS INTO CODE

JERRY R . HOBBS

CITY COLLEGE O F CUNY

In this paper we describe a system for the semantic ana-

lysis of "well-written" algorithm descriptions in English,

such as one finds in Knuth ( 3 ) , and t h e translation of t h e

resulting structures into a program in a higher-level pro-

gramming language, such as P L f I . There are two a p p r o a c h e s

one might take toward the sublanguage o f algorithm descrip-

t i o n s . One may t r e a t t.he E n g l i s h as a d e v i a n t p r o g r a m i n g

language and try to smooth o u t t h e d e v i a t i o n s . Or one may

v i e w t h e English as a well-behaved, h i g h l y regular form of

natural language, and a p p l y to the t e x t the sophisticated

t e c h n i q u e s of analysis that have been developed for less

t r a c t a b l e sorts of t e x t s . We have taken the l a t t e r approach,

Semantic analysis is done by a s y s t e m t h a t has been d e -

veloped for and is being tested on a variety of natural

language texts (1). I t takes as i n p u t the t e x t i n a p r e d i c a t e

calculus-like n o t a t i o n produced by a syntactic front end (4,

2). I t a p p l i e s s e m a n t i c operations which a c c e s s a d a t a base

of "world knowledge" inferences that are drawn selectively,

depending on c o n t e x t and i n response t o the o p e r a t i o n s . The

operations are described below w i t h examples that show their

i m p o r t a n c e in analyzing a l g o r i t h m d e s c r i p t i o n s :

1. Predicate interpretation: Inferences are sought to

arguments.. Among o t h e r t h i n g s , t h i s o p e r a t i o n r e c o v e r s o m i t t e d

m a t e r i a l , f o r example when t h e whole i s used t o r e f e r to t h e

p a r t . In

"T p o i n t s to a b i n a r y t r e e w

t h e p r e d i c a t e " p o i n t " r e q u i r e s i t s second argument t o be a

node. The knowledge s t o r e d i n the d a t a base a b o u t b i n a r y

trees i s searched f o r a dominant node, t h e r o o t node i s found,

and t h e s e n t e n c e becomes

'IT points t o the r o o t node of a b i n a r y tree. I t

-

-

2 . I n t e r s e n t e n c e r e l a t i o n s a r e determined by matching

s u c c e s s i v e s e n t e n c e s a g a i n s t a small number of p a t t e r n s .

These p a t t e r n s a r e s t a t e d i n terms of i n f e r e n r e s t o b e drawn from t h e c u r r e n t sentence and a previous s e n t e n c e . A n

o r d e r e d h e u r i s t i c s e a r c h of t h e d a t a base i s used t o l o c a t e

t h e desired i n f e r e n c e s . I n t e r s e n t e n c e r e l a t i o n s a r e e s p e c i a l l y

important i n a l g o r i t h m d e s c r i p t i o n s , f o r they determine the

flow of c o n t r o l i n t h e program. The most common and e a s i l y

handled patterns in t h e s e texts a r e Temporal Succession,

cause, and Enablement. They a r e t r a n s l a t e d i n t o s u c c e s s i v e

l i n e s of code.

But a l s o important i s t h e Contrast p a t t e r n : "The p r e d i -

c a t e s of s e n t e n c e s S1 and S 2 a r e t h e same. One p a i r of c o r -

responding arguments a r e c o n t r a d i c t o r y . The o t h e r p a i r a r e

different: but s i m i l a r . ' Consider

"If VAL(M) VAL(N), s e t N e q u a l t o L I N K ( N ) .

1 4 t h ACL Meeting

I t i s intended t h a t t h e s e t r a n s l a t e not i n t o successive i n - s t r u c t i o n s , but i n t o a c a s e statement. This i s keyed by t h e

Contrast p a t t e r n . (The top-lev-el predicate i s "imply". )

S i m i l a r l y , consider the sentences

(1) "Decrease N by 1,. and if i t i s zero, r e s e t i t t o MAX. t I (2) "Decrease N by 1, but i f i t i s z e r o , r e s e t i t t o MAX."

I n (1) the t e s t f o r zero comes a f t e r decreasing N , i n ( 2 ) before. This i s because i n ( 2 ) , "but" invokes the Contrast

p a t t e r n , f o r c i l g us t o recover t h e implicit " I f N is not zero"

before "Decrease". ( 2 ) is t r a n s l a t e 3 i n t o a c a s e statement.

3. Knitting: Redundancies are recognized and merged This f r e q u e n t l y r e s o l v e s anaphora a s a byproduct.

4 . Resolving anaphora : Antecedents not Found by k n i t t i n g

are searched for by t r y i n g t o maximize t h e redundancy of t h e t e x t . I n

"Decrease N by J , and i f it i s z e r o , r e s e t i t t o MAX." ltN" is chosen over "J" a s t h e a n t e c e d e p t o f "it1' s i n c e i t i s N whose value i s subject- t o change.

1 1

5 . Search f o r P r i m i t i v e s : Certain predicates a r e p r i -

mitive'' tb t h i s a p p l i c a t i o n , i . e the t a s k component has

s u b s t i t u t i o n r u l e s f o r them. Where p o s s i b l e , a decomposition of t h e t e x t i n terms of t h e s e p r i m i t i v e s i s looked f o r . For

i n s t a n c e , a t r e e i s , i n p a r t , a s e t of nodes, a node i s a s e t of f i e l d s , and a f i e l d may b e a number. "Set1' and "number"

are p r i m i t i v e s . To decrease N by 1 is t o cause N t o be equal t o one l e s s than b e f o r e , and "cause t o be e q u a l t o " is a

Semantic Analysis augments and interrelates the texk

What i s produced i s then given t o t h e t a s k component which

per

forms

two Punc

t i o n s :

1.

The temporal succession r e l a t i o n s d i s c o v e r e d by o p e r a t i o n 2 may form onrly a partial ordering on actions. A ltnear ordering is imposed.

2 . Sobstitution _{r u l e s a r e a p p l i e d} t o primitives t o

t r a n s l a t e t h e semantic representatidn i n t o code. A sebt

translates i n t o an array declaration, a s e t of s e t s i n t o a two-dimensional a r r a y . "Carse t o be equal to" t r a n s l a t e s

i n t o an assignment s t a t e m e n t .

Bibliography

Hobbs, Jerry R . A g e n e r a l system f o r semantic analxsis

of English and i t s u s e in drawing maps Prom d i r e c t i o n s .

AJCL M i c r o f i c h e 3 2 , 1 9 7 5 .

Hobbs,.J. R . , and Grishman, Ralph. The automatic t r a n s -

formational analysis of English sentences: an implemen- tation. International Journal of Computer Mathema-ics,

t o a p p e a r .

Knuth, Donald. The art o f computer programming,. 1. Addison-Wesley, Reading, Mass., 1968.

Sager, Naomi, and Grishman, Ralph. The restriction

language f o r computer grammars o f natural l a n g u a g e .

14th ACL Meeting 13

GENERATING NATURAL, LANGUAGE EXPLANATIONS OF COMPUTER PROGRAMS

GEORGE E. H E I D O R N I B M RESEARCH

.

An automatic programming system that converses with a u s e r

in a natural language must be able to generate explanations of

portions of the program being produced. In the system that wc

are developing for producing customized business application

programs, these explanations.must be given in application terms

rather than program terms.

For

example. the system might sey , "The quantity shipped is the lesser of the quantity o r d e r e d and

the quantity available," rather t h a n "QS=MIN(QO,QA)".

One could certainly imagine generating such sentences by

having an inventory of sentence patterns and name phrases and

using a fill-in-the-blanks approach. However, in addition to

not being very interesting, such an approach would certainly

not provide the same flexibility of expression that a more lin-

guistically motivated approach would. Also, it seems reasonable

to do this generation b y utilizing information that,. the system

already has for understanding the user's English utterances.

In our system, information is stored in the form of a se-

mantic network with three p a r t s .

The

program part contains

information about the programing language and about a specific

program; the application part contains information about business

concepts; and the language part contains information about English words. Each of these parts has its own internal structure, and

within the application part the node for the concept

QUANSHIPD

has a S U P e r s e t arc to QUANT and a KIND a r c t o 1 . Hctwccn

parts of the network, QUANSHIPD is linked as t h e M O ( R e l i l t e d

A p p l i c a t i o n Object) arc of the program node

QS,

the language

node QUANTIT (the stem of "quantity") is linkcd as the RLO

(Related Language Object) of

QUANT,

and t h e 1angua);c node SHIP

i s the RLO of SHP. I n g e n e r a l , t h e links bet~,*\.ecn cliljnccnt

p a r t s a r e many-to-many.

The linguistic processing in our system is specified by

augmented phrase structure rules, uniformly for the three

levels of processing: semantic, syntactic, and morphological.

Given a p o i n t e r to a node in the program part, a l ~ n g with some

information about what aspect of that node i s to b e explained,

these rules can g e n e r a t e an appropriate English statement by

utilizing the ~tructure of the network. Because these s t n t e -

ments are constructed from very small p i e c e s ( i . e . word stems),

considerable flexibility of expression is possible, including

taking into account dialogue context.

This paper will describe in some detail the overall

method of generation, with special emphasis on complex definite

14th

ACL

Meeting

THE SEMANTIC INTERPRETATION OF MASS NOUN EXPRESSIONS

I N THE P H L I Q A 1 SYSTEM

A large number of the most commonly used,English nouns

belongs to the category of mass nouns.

In

recent years it

has become increasingly clear that the semantic analysis of

expressions containing mass nouns encounters great difficulties.

because of the uncertain logical status of mass noun referents.

This uncertainty is reflected in the appearance of a variety

of vague terms, such as "substance", "quantity", "matter",

"bits", "pieces1', "stuff", etc. in recent literature on the

subject (see Pelletier 1975)

.

It

is indeed fair to say that,

"the logic of mass expressions

. . .

at present is grdssly

underdeveloped" (McCawley 197 5)

Work in computational linguistics so far has not contri-

buted to a solution or even a better understanding of the

problems that mass nouns pose. In language understanding sys-

tems that have been developed in the past few years these

problems have been evaded, either by restricting the admitted

use of mass nouns trivially, as in the SHRDLU dialogue system

(Winograd 1972), or by explicitly excluding their use altoge-

ther, as in the LSNLIS question answering system (Woods 1972)

.

The recently implemented question answering system

PHLIQA 1 (~andsbergen 1976, Medema et al. 1 9 7 5 , Scha 1976)

respect. This system

performs

the semantic analysis a f corn-

plex mass noun expressions, including quantifying phrases, adjectives, amount expressions, etc

.

The analysis is based

on a theory, i n which a ceritral role is played by a l o g i c a l

formalism,

developed for the semantic representation

of

mass expressions. These r e p r e s e n t a t i o n s were r e q u i r e d t o meet t h e

following conditions:

1. They should be in agreement with the ontological commit- ments of mass noun uses. In our view. the t y p i ~ a l use of a mass 'noun indicates that the noun referent is viewed as a

continuum; i . e . no articulation of the r e f e r e n t into individual

elements

is

taken into consideration--which is of course n o t to

say that the r e f e r e n t actually is a continuum.

2. They should

bring

the logical properties of mass expression

to the light, e.g. rendering analytic sentences tautologically true.

The

paper develops this semantic theory of mass noun expres-

sions, and describes its incorporation in the semantic analysis

performed by t h e PBLIQA 1 sys tern.

Bibliography

Landsbergen, S. P. J . Syntax and formal semantics of English. in PHLIQA 1. Preprints, COLING 76.

14th

ACL Meeting

Medema, f . , W.

J. Bronnenberg, H.

C. Bunt, S.

P .

J. Landsbergen,

R.

J.

fl. Scha,

W.

J.

Schoenmakers, and

E.

P . C. van U t t e r e n .

PHLIQA

1:

:multi~evel

semantics

i n question answering.

AJCL Microfiche

32, 1 9 7 5 .

P e l l e t i e r , F . J . , e d . On mass terms. Synthese 31 ( 3 / 4 ) , . 1 9 7 5 . Scha, R .

J.

H . Semantic t y p e s i n PHLIQA 1. Preprints,

COLING

7 6 *

Winagrad,

T.

Understanding natural language. Edinburgh

University Press,

1972.

Woods,

W.

A , , R.

M.

Kapl-an, and B. Nash-Webber. The lunar sciences natural language informatian system. Final R e p q r t ,

E M P I R I C A L STUDIES OF NOUN MEANING

A c e n t r a l problem for computational models of understand-

ing is to formalize the meanings of words which are manipulated

by the model. This paper reports on research directed towards

the formalization of the

meanings of

nouns. The g o a l of t h e

research is to d e e c r i b e nouns sufficiently t o enable the per-

formance of a particular task of understanding: the appropriate

selection of verb senses for multiply-sensed verbs.

The hypothesis being tested is that noun meaning can be d e s c r i b e d by primitive "units of meaning, " here called features

.

The concepts of semantic features and componential analysis are

.

familiar; the contributions of this work are the development of

empirical methods of analysis and the generation of a general

set af features.

The effect of noun meaning on the sense of a verb can b e

clearly seen by examining simple sentences with identical syn-

tactic structure and no extra-sentential context. Tb study

noun meaning in a regular, non-subjective fashion, a large number of simple sentences of the form noun-verb-noun have

been generated. In groups of the sentences, the subject noun

and verb have been fixed, while the object noun is varied.

The variance in the object noun causes the verb t o have diffe-

rent readings, or different mappings of the lexical verb onto

14th ACL Meeting 19

From t h a t group of nouns, i n the object position; which

cause t h e selection of the same sense for the verb in question,

the common t h r e a d of meaning is extracted and posited as a

semantic feature descriptive of that group of nouns. R e p e t i -

t i o n of this process for various verbs and nouns yields a s e t

of semantic features. These features are c o n s i d e r e d as f o r - malizing noun meaning at least f o r purposes of verb sense

selection.

The major problems in the above approach are 1) the i n e v i -

table intrusion of subj e c t i v i ty i n t ~ the feature extraction

task and 2) the l a c k of generality of the f e a t u r e s e t beyond t h e particular group of nouns and v e r b s s t u d i e d . Several

methods have been used t o c o h t r 0 1 s u b j e c t i v i t y : r e p e t i t i o n of

the process using the same verbs with d i f f e r e n t nouns and vice-

versa, and t h e g e n e r a t i o n of features from t h e r e g u l a r study

of other tasks of understanding which involve nouns--noun-

adjunct:noun pairs ( e . g . s c h o o l house) and noun a n a l o g i e s . To

ensure as much generality a s p o s s i b l e , t h e words c o n s i d e r e d i n

t h e s t u d y have been chosen from the s e t of (statistically) most

common English words and d o n o t comprise any unified semantic

f i e l d .

Results have been i n t e r e s t i n g and e n c o u r a g i n g . In a d d i - t i o n t o t h e commonly used features ("human, 11 I t animate, 1 1

" o b j e c t " ) , features like "emotion, " "authority, " "concept , 11

and "avocation" have been g e n e r a t e d . F u r t h e r work will look

t h e u s e

of empirical

methods t o study

verbs,

and

the

imp-lica-

tions of features

f o r

semantic

f i e l d s ,

meaning

extension,

and

14th

ACL

Meeting.

PANEL; EVALUATION OF NATURAL LANGUAGE SYSTEMS

JOYCE 0 . FRIEDMAN, CHAIR

U N I V E R S I T Y O F MICHIGAN

WILLIAM A . MARTIN, f6.I.T.; WILLIAM H. PAXTON, STANFORD RESEARCH

INSTITUTE; STANLEY PETRICK, I B M RESEARCH^ NAQMI SAGER,, NEW

YORK U ~ I V E R S I T Y ; ERIC VAN U T T E R E N , PHILIPS RESEARCH LABORATORIES;

Diverse natural language systems share

a

common problem

of evaluation. This panel will address the topic of evaluation,

emphasizing the evaluation of performance:

The

panelists will

consider such questions a s : What measures are used to evaluate

a system? How sensitive is performance to vocabulary size and scope of syntax? What alternatives can be studied analytically

or experimentally?

Performance can be considered absolutely (how e f f e c t i v e is

a system,

with

respect to resources and goals?), relative to

problem

variation (hvw would it r e s p ~ n d to changes in, e . g . ,

vocabulary size)

.

relative to system variation, and relative to other systems. In each case both the results of evaluation of

particular sys terns and procedures

of

evaluation are of interest

.

The panel will consist of experts, all of whom have written

large systems. These systems include speech systems and text

systems, business-oriented question-answering systems and

artificial intelligence systems, and systems with contrasting

METHODS F O R MODELING DIALOGUE

WILLIAM C. MANN, JAMES A . LEVIN: AND JAMES A . M O O R E

RESEARCH METHODS F O R M O D E L I N G D I A L O G U E

When p e o p l e communicate w i t h machines, t h e y do so by spew

cializing

and extending

their

existing ability to communicate.

To design systems which interact effectively with humans, we

need to

better understand

how people communicate with each other.

In

particular,

we

need to view human communication as a

problem-solving activity, in which the people so engaged are

using language as a method to achieve certain of their goals.

The

existing research into natural language has not focused

on this aspect, so from our point of

view the

work so

far

has

been fragmentary and hard to use, particularly in terms

of

h e l p i n g to enhance man-machine corn-nication.

Constructing a useful model of human communication is an

extremely complex

task.

We view t h e controlling of this com-

plexity (without sacrificing utility) as the central problem

in designing

our

research approach.

In

response to this

chal-

lenge, our research methodology contains some innovations,

specifically intended to maintain control on the complexity,

while

retaining

the

usefulness of

our

results

for

designers

1 4 t h ACL Meeting

S p e c i f i c a l l y , t h e s e i n n o v a t i o n s i n c l u d e

--Case a n a l y s i s , r a t h e r t h a n f u n c t i o n a l d e s i g n .

--Independent judgment of t h e adequacy-of t h e model.

- - R e s u l t s i h t h e form of i n d i v i d u a l a l g o r i t h m s , n o t

f u l l s y s t e m s .

--Algorithms which a r e t r a n s f e r a b l e into n o n r e s e a r c h

s y s terns,

We d e s c r i b e a r e s e a r c h methodology, and t h e r e s u l t i n g

Dialogue Modeling System which i n c o r p o r a t e s these i n n o v a t i o n s .

The methodology a n a l y z e s n a t u r a l l y - o c c u r r i n g d i a l o g u e s r a t h e r

t h a n prepared examples. For each d i a l o g u e w e choose t o ana-

l y z e , a s e p a r a t e model i s c o n s t r u c t e d . Each model i s judged

a g a i n s t a n n o t a t i o n s of t h e d i a l o g u e prepared by a t r a i n e d ob- s e r v e r who u s e s a p r e d e f i n e d set of c a t e g o r i e s of phenomena.

Although t h e p r o c e s s e s employed i n each model a r e t o b e

a s g e n e r a l as w e f i n d p r a c t i c a l , t h e y may a l s o b e a s ad-hoc a s n e c e s s a r y t o make t h e model work. A f t e r s e v e r a l models have

been produced i n t h i s f a s h i o n , we e x p e c t t o f i n d t h a t , a l t h o u g h

many p r o c e s s e s have been t o o i d i o s y n c r a t i c t o be used i n more t h a n one model, many others have seen r e p e a t e d application and t h u s have demonstrated t h e i r more g e n e r a l u t i l i t y . These pro-

c e s s e s are regarded a s the p r i n c i p a l r e s u l t s , s i n c e t h e y have

been v e r i f i e d a g a i n s t a d i v e r s i t y of o b s e r v e r s ' a n n o t a t i o n s .

Subsequent s p e a k e r s w i l l p r o v i d e d e t a i l s on t h e memory

organization of t h e model, the p r o c e s s o r s which o p e r a t e on t h e s e

memories, and t h e i s s u e s r a i s e d by t h e i n c l u s i o n of t h e o b s e r v e r

KNOWLEDGE STRUCTURES F O R M O D E L I N G D I A L O G U E

This paper

describes the know7 edge structures developed

f o r our Dialogue Modeling System. Permanent knowledge is

stored as Concepts in e semantic n e ~ w o r k Long

Term

Memory

(LTM)

, using a predicate/parametcr representational format Concepts

t h a t a r e currently salient t o t h e Processors of t h e System a r e

represented by tokens of t h o s e concepts, c a l l e d A c t i v a t i o n s ,

i n t h e Workspace (WS). A c t i v a t i o n s are transient entities,

w i t h new ones f r e q u e n t l y appearing and e x i s t i n g ones changing o r d i s a p p e a r i n g , The WS a t any moment forms the c u r r e n t con-

~ e x t , which d r i v e s c u r r e n t p r o c e s s i n g . A l l Processors inter-

communicate through the WS, modifying the set of Activations

and b e i n g influenced by them. The WS simplifies the integra- tion of multiple sources of i n f o r m a t i o n about any given e n t i t y

Two k l n d s of h i g h e r l e v e l knowledge s t r u c t u r e s are used

t o model d i a l o g u e s . There a r e Execution Scenes, r e p r e s e n t i n g o r g a n i z e d c l u s t e r s of a c t i o n s and/or objects being d i s c u s s e d

in t h e dialogue'. These are s i m i l a r t o the o t h e r c u r r e n t l y

proposed h i g h e r level units, like "scripts" or "frames".

There

a r e a l s o higher l e v e l u n i t s called Dialogue-games, spe-

cifications concerning t h e topic of discussion, t h e partici-

p a n t s , the g o a l s t h e y a r e trying to achieve with a given type

14th ACL Meeting

I n our a n a l y s i s of n a t u r a l l y occurring d i a l o g u e s , we have*

found many d i f f e r e n t k i n d s of Dialogue-games. including Helping C

Smalltalk, Gripe, P o l i t e - c o n v e r s a t i o n , Order, Information-

seeking, and Information-probing. An a n a l y s i s of f o u r t e e n

helping dialogues uncovered a s u r p r i s i n g amount of consistency

i n the s t r u c t u r e of t h e s e i n t e r a c t i o n s , which i s captured i n

t h e r e p r e s e n t a t i o n of t h e Helping Dialogue-game.

PROCESWRS F O R M O D E L I N G D I A L O G U E

This paper provides d e t a i l s of the major Processors i n c o r -

porated i n t o our Dialogue Modeling System. These Processors

a r e independent and asynchronous, monitoring the a c t i v i t y i n

the Workspace, and modifying i t as a p p r o p r i a t e .

PROTEUS PROCESSOR

--

Every a c t i v a t i o n i n the Workspace

has a "level of a c t i v a t i o n " (a number) which corresponds t o

i t s degree of s a l i e n c e t o t h e model's c u r r e n t activity. Each

of t h e s e a c t i v a t i o n s imparts some of i t s s a l i e n c e t o its i m e -

d i a t e neighbors (including those i n Long T e r m Memory). S i m i -

l a r l y , the a c t i v a t i o n l e v e l of each node i s augmented by t h a t

imparted t o i t by a l l of i t s neighbors. Whenever a concept i n

LTM accumulates a c t i v a t i o n above a certain t h r e s h o l d , an acti-

v a t i o n of t h a t concept i s c r e a t e d i n the Workspace, Thus, if

enough concepts, closely r e l a t e d t o X a r e active, X w i l l also

become a c t i v e . Proteus i s the Processor which a t t e n d s t o a l l

MATCH PROCESSOR

--

This i d e n t i f i e s concepts i n LTM that

are congruent to existing.activations in _the WS The Dialogue

Modelling S y s t e m contains a number of e q u i v a l e n c e - l i k e rela-

t i o n s . which Match uses to identify a concept i n LTM a s r e p r e -

senting t h e same, t h i n g as an a c t i v a t i o n of some s e e m i n g l y -

d i f f e r e n t concept Once such an e q u i v a l e n t concept i s f o u n d , i t i s a c t i v a t e d . Match thus p r o v i d e s the same sort of a t t e n t i o n

directi.cn that Proteus d o e s , except t h a t Proteus i s d r i v e n by

the e x p l i c i t c o n n e c t i v i t y of t h e memory, w h i l e Match responds

t o more i n d i r e c t l y - r e p r e s e n t e d s i m i l a r i t i e s .

DEDUCE PROCESSOR

--

T h i s attempts t o apply a r u l e , whenever

ehat r u l e B a s become- a c t i v e . Rules are concepts of the f o r m .

"(Conditior)) -> (Action) "

.

They become a c t i v e by v i r t u e of t h e a c t i v i t y of the concept which i s t h e i r c o n d i t i o n h a l f . Deduce

*I

senses t h e a c t i v i t y of a r u l e and attempts t o a p p l y i t by a c t i -

vating t h e concept f o r t h e a c t i o n . Whateve'r correspondences

were evolved i n t h e course of c r e a t i n g t h e a c t i v a t i o n of t h e

condition (left) half of the rule are c a r r i e d over into t h e

a c t i v a t i o n of t h e a c t i o n (right) h a l f . The combination of

Match and Deduce g i v e s us a l l the c a p a b i l i t y o f a production

system, where t h e system i s represented by a l l t h e Rules i n LTM.

DIALOGUE-GAMES PROCESSOR

--

Dialogue-games a r e a theore-

tical construct w e have developed t o represent c e r t a i n c o m u n i -

cative f o m s used by people t o i n t e r a c t i n achieving g o a l s

each person has. Each Dialogue-game has a s e t of parameters,

14th

ACL

Meeting

specifications on

what can

serve as a

value

f o r itself. W e

have found that these dialogue forms are identified i n conver-

sation by utterances that attempt to establish these parameters.

Once a Dialogue-game has been activated as possibly the cornmu-

nication form being proposed for a dialogue, the Dialogue-game

Processor operates

on

i t to verify that the parameters are

properly specified, and then to establish the subgoals that

are specified in LTM as the components of the particular

Dialogue-game

.

PRONOUN PROCESSORS

--

The Dialogue Model System contains

a set of Pronoun processors, including an I-Processor, a You-

Processor, and an It-Processor. Each of these is invoked

whenever the associated surface word appears in an input

utterance, and operates to i d e n t i f y some preexisting activation

that

can

be seen as r e f e r r i n g t a the same o b j e c t .

OBSERVATION METHODS FOR MODELING D I A L O G U E

Most models of natural language are constructed using only

the designer's own linguistic intuitions. and'evaluated by

others using t h e i r own intuitions. This paper describes obser-

vation methods which have been used as a guide for the genera-

tion of process models of dialogue phenomena and as a basis f o r

e m p i r i c a l evaluation of these models.

The methods have concentrated on a number of specific

dialogue phenomena. Observers w e r e given the annotation in-

annotations of :

Requests

-

Annotations of requests and the various kinds

of responses to these requcsts by t h e e t h e r p n r t i c i -

pant in a dialogue were useful for evaluating various

control issues.

Repeated r e f e r e n c e

-

Annotations of repeated r e f e r e n c e

were used to evaluate models of anaphoric r e f e r e n c e

(cases in which two sets o f words in a dialogue

describe the same thing).

Topic

-

Observers' anriotations of the topic units in

naturally-occurring dialogues provide a basis for

specifying high level multi-utterance knowledge

structures.

Expressions of comprehension

-

Observers annotated t h e expressions of comprehension ( o r l a c k thereof)

by

each

participant of the other's previous utterances. These

annotations a r e being used t o evaluate our model of

the stopping rule for comprehension, which determines

when the model has successfully comprehended an utte-

rance.

Similar expressions

-

Observers judged possible paraphrases

of the utterances of dialogues for similarity in meaning

in isolation and also in the context of the dialogue.

This

is serving as useful d a t a for determining the

14th ACL

Meeting

Correction actions

-

Observers' annotations of utterances

t h a t corrected some

information conveyed previously

i n t h e d i a l o g u e allow us to evaluate our representa-

tion of the comprehension of the previous utterances.

While

these phenomena don't span a l l possible i s s u e s of

i n t e r e s t in natural language comprehension, t h i s same g e n e r a l

methodology can be ext-ended to other 1anguage.phenomena. It

can provide a valuable guide for building n a t u r a l language

A PROCESS MODEL FOR SPEECH ACT UNDERSTANDING

In a canversation, hearers must make judgments about the

i n t e n t i o n of a s p e a k e r ' s utterance. A restaurant waiter r e a - lizes t h a t when a p a t r o n s a y s , "Waiter, t h i s soup i s c o l d , " he is not merely t h e r e b y making an a s s e r t i o n , b u t is a l s o making

a c o m p l a i n t , and perhaps requesting r e c ~ i f i c a t i o n .

Utterances can be understood t o hav'e more than one i n t e n -

t i o n . Two t y p e s of k n t e n t i o n s are d e s c r i b e d i n t h i s p a p e r , b o t h of which depend l a r g e l y on c o n t e x t u a l f a c t o r s for t h e i r r e c o g n i t i o n o r understanding i n a d i a l o g u e s i t u a t i o n . Those

of t h e f i r s t t y p e ( t y p e A) are broadly a p p l i c a b l e t o many

d i s c o u r s e s i t u a t i o n s and a r e understood p r i m a r i l y on t h e b a s i s

of t h e syntactic s t r u c t u r e o f utterances and certain f a c t s

a b o u t t h e s t a t u s r e l a t i o n s h i p s that hold between t h e speaker

and t h e hearer. Type A i n t e n t i o n s are very similar to the

Speech Acts of Searle o r of Gordon and Lakoff. Speech a c t s

of t h e second t y p e (type B) a r e much more dependent on t h e nature of t h e p a r t i c u l a r contextual situation. They. a r e

recognized o r understood on t h e b a s i s of shared cultural norms concerning the p a r t i c u l a r kind of c o n v e r s a t i o n the p a r t i c i p a n t s

a r e engaged i n . T y p e B speech a c t s a r e much l e s s g e n e r a l in

n a t u r e t h a n type A . I f a salesman says, "This l i t t l e two-tone

model has been p r i c e d t o s e l l , " then t h e customer should recog-

nize that the type A intention is to make an assertion or a

14th ACL Meeting

A prdcessing model for speech act recognition or under-

standing is presented, in wMch procedures c a l l e d spe-ech a c t

schemata are activated. In the example with the salesman, the customer should have experienced the a c t i v a t i o n of b o t h an

assertion schema and a sales-offer schema The schemata are

formulated i n SOL, the h i g h l e v e l language of the LNR memory model

orma man,

Rumelhart, and LNR), which permits t h e modeling

of conceptual e n t l t i e s which have b o t h s t r u c t u r a l and proce-

dural p r o p e r t i e s . I n a d d i t i o n , the t y p e B speech a c t schemata

are subprocedures of l a r g e r i n f o r m a t i o n o r g a n i z e r s which a r e

sensitive t o t h e p a r t i c u l a r c o n t e x t . These l a r g e r d e v i c e s

are called scripts and are r e l a t e d t o the computational e n t i -

t i e s of t h a t name developed by Schank and Abelson. In the

example with t h e salesman, t h e s a l e s - o f f e r speech a c t i s p a r t

of the "Commercial Transaction" s c r i p t

.

A detailed model for one such script, an "Airline Reser- v a t i o n Phone Call" s c r i p t i s presented, t o g e t h e r w i t h an account

of the i n t e r a c t i o n s between s c r i p t s , t h e two t y p e s of speech

NATURAL LANGUAGE UNDERSTANDING BY C O G N I T I V E NETWORKS

A model of human cognition which stores and utilizes l i n -

guis tic knowledge in the same structures as 'general

'

or 'world

'

knowledge. supports an efficient and tntuitively satisfying

technique for the understanding of natural language. Cognitive

comprehension and syntactic passing can proceed in parallel,

cooperatively, using the same procedures, without the loss of

the syntax/semantics disrinc~ion

.

In this paper, a system w h L ~ h uses such a representation

is described. The representation is a cognitive network, a

directed graph model of human cognition. The grammatical

knowledge it contains is s h o w to be similar in structure to

an augmented recursive transition network, but the understanding

process is not ubilt around an

ATN

parser. Iflstead

a

"syntac-

tically aided comprehension" procedure is used; cogn-itive

processes, taking advantage of the uniform representation of

knowledge, utilize linguistic knowledge as is required< for

tasks such as the disambiguation of conceptual relationships,

and the location, in the network, of referents designated by

phrases instead of names. Altheugh the system has the syntactic

knowledge needed to produce a correct par* for an utterance, it does not always have to apply it.

A few brief comparisons with other natural language

14th ACL Meeting

HOW TO REPRESENT AND USE KNOWLEDGE ABOUT C A U S A L I T Y

CHARLES J . RIEGER

U N I V E R S I T Y O F MARYLAND

A s y s t e m for representing and using knowledge of cause

and effect, called Cormnonsense Algorithms (CSA), will be de-

scribed. The CSA project, ongoing for about a year now, has

as i t s g o a l the unification of current ideas a b o u t problem

solving w i t h current ideas about conteatual language compre-

hension. To this end, the CSA project currently consists of

e k r e e gart's, all built on top of the declarative CSA repre-

sentation formalism: (1) a plan synthesizer, (2) a sentence-

in-context interpreter (designed as the 2-sentence case of

an eventual n-sentence story comprehender), and (3) a mecha-

nism descriptllon "laboratoryr', who~se purpose is to provide a

framework &or describing the "causal topology" of man-made

devices and mechanisms. The

LISP

sys tern -which implements t h e

CSA theory incorporates processes which are demon-like and

processes which are more planful, and the CSA theory identi-

fies how and why the planner and the population of demons

interact. The presentation will cover the organization and

access techniques for large numbers of CSA causal patterns in

these two- modes (planful and demoiic) , drawing on scenarios

which illustrate the level of sophistication of the current

implementation. One scenario will be taken from the children's

s t o r y , The Magic Grinder (a Walt Disney book-of-the-month

comprehension part of t h e p r o j e c t .

It

is believed t h a t

any

language

comprehender must

possess

CSA-like

knowledge, and

t h a t

any

t h e o r y of _language

comprehension must

d e a l w i t h

24th ACL Meeting

A H E U R I S T I C SEARCH APPROACH TO DISCDURSE ANALYSIS

JERRY R. HOBBS

CITY COLLEGE OF CUNY

This paper describes a computational approach t o t h e

problems of discourse s t r u c t u r e of real-world English para-

graphs, u s i n g the techniques G £ h e u r i s t i c search. Procedures

for detecting intersentence relations a r e being developed

w i t h i n the framework of a system f o r semantic analysis o f , English texts (1). This system t a k e s a s input the t e x t i n a

p r e d i c a t e c a l c u l u s - l i k e n o t a t i o n produced by a s y n t a c t i c

f r o n t end ( 3 , 2 ) . Various s e m a ~ t i c o p e r a t i o n s a r e a p p l i e d :

1) Inferences are sought t o s a t i s f y the demants made by

p r e d i c a t e s on t h e n a t u r e of their arguments ( p r e d i c a t e i n t e r -

p r e t a t i o n ) .

2) Redundancies a r e recognized and merged ( k n i t t i n g ) .

3 ) Anaphora a r e resolved.

The operations work by accessing a d a t a base of world know-

ledge i n f e r e n c e s , which a r e drawn s e l e c t i v e l y i n response t o

the o p e r a t i o n s , and which carry a measure of their salience

that varies with t h e c o n t e x t .

I n t e r s e n t e n c e r e l a t i o n s are determined by matching the

current sentence and the previous text a g a i n s t a s m a l l number

of p a t t e r n s . These patterns are stated i n terms of i n f e r e n c e s

t o be drawn from the c u r r e n t sentence and an " e l i g i b l e " pre-

vious sentence, and t h e modification t o be performed on t h e

kinds--coordinating and subordinating. One coordinating

relation is

Temporal

Succession : "The c u r r e n t sentence

asserts a change whose initial s t a t e i s implied by an e l i g i b l e

previous sentence." Other coordinating relations are Cause,

Contrast, Parallel, and Paraphrase. Two subordinating rela-

tions are

1)

Example: "The predicate and arguments of the asser-

tlon of the current: sentence stand in a subset or element-of

relation to those of the assertion of ah eligible previous

sentence.

"

(See below.)

2) "Relative clause in disguise": "The current sentence

provides further information about an entity fairly deeply

embedded in an eligible previous sentence.

"

(E.g. the second

and third sentences of this abstract.) When this pattern is

matched, the modiffcation made to the text is equivalent to.

turning the current sentence into a relative clause.

As analysis proceeds, a tree-like structure is constructed

for the paragraph, with subordinating relations building the

tree downward and coordinating links building it to the right.

A previous sentence is "eligible" if it is on the right fron-

tier of this tree. For example, in a text S1 S 2 Sg, if S 2 is

a disguised relative clause on S1, then Sg may be coordinated with either S1 or S 2 , but if S g succeeds S1 temporally, then

Sg may Be coordinated only with S p .

The heart of the operation is an ordered heuristic

14th ACL Meeting

desired inferences are kept on a goal list, and have strengths

associated with them. The s t r e n g t h of a goal depends on the

type of text and the patterns found previously in

the

text.

Also, the presence of conjunctions and some other elements

advance certain patterns--'>.cndu promotes Temporal Succession

and a repetition of the previo~sly recognized pattern, a dash

and "l.e." promote Paraphrase, and the a r t i c L e "this" fre-

quently signals a disguised relative clause. The evaluation

function which orders the heuristic search is based on the

srength of the goal, the length of the chain of inference, and

the current saliencerof inferences in the chain.

When a partial match is found, the difference, or the

remainder of the goal, is placed high on a goal list for

subsequent processing. Consider the text

"Republicans. are discouraged about their prospects

The party chairman is convinced that many GOP congress-

aen will lose their bid for reelection. ? 1

Suppose for simplicity "be discouraged" decomposes into

"believe something bad will happen" We have a partial

match with the Example pattern since the secdnd sentence

asserts that a particular Republican believes something.

Hence a principal goal for our processing of the "that"

clause is to show that what is believed is than an event bad for a Republ-ican will occur. This is shown by accessing the

fact about a political party that one of its purpdses is to

Goals generated by a p a r t i a l match a r e also passed on

to subsequent sentences. I n a Newsweek paragraph analyzed, t h e f i r s t sentence S1 asserts a change. The next three

sentences S 2 S g S4 assert stages along the course of t h i s

change. S2 is matched w i t h the i n i t i a l s t a t e of the change

in S1, thus g e n e r a t i n g as a goal the matching of S g and S4

w i t h succeeding s t a t e s . Finally, the structure

"s2

--

then S3 then S4" is taken as an Example of S1.

This work has significance f o r f u t u r e attempts t o sum-

marize paragraphs a u t o m a t i c a l l y . For, i n s t a n c e , i n a t e x t

, t i e d together by the Example pattern, t h e sentence "exampled" must be a major p a r t of the summary.

Bibliography

1. Hobbs, J e r r y R . A general system for semantic ana-

l y s i s of E n g l i s h and i t s use in drawing maps from d i r e c t i o n s .

AJCL

-

M i c r o f i c h e 3 2 , 1975.

2. Hobbs, J . R . and Grishman, Ralph. The automatic

bransf drmational a n a l y s i s o f . English sentences : An implemen-

t a t i o n . I n t e r n a t i o n a l Journal of Computer Mathematics, t o appear

3 . Sager, Naomi, and Grishman, Ralph. The r e s t r i c t i o n language f o r computer grammars of n a t u r a l language. CACM

American Journal of

Computational Linguistics

~ i c r o f i c h e 51 : 3 9 .

ACM

CWPUTER SCI

ENCE

CONFERENCE

E M P L O Y M E N T

R E G I S T E R

The

Fifth

Annual R e g i s t e r w i l l p r o v i d e books of l i s t i h g s of

a p p l i c a n t s and p o s i t i o n s during the Conference

.

Applicant listings fnclude education, p u b l i c a t i o n s , e x p e r i e n c e ,

interests, r e f e r e n c e s , p o s i t i o n , and salary d e s i r e d .

Employer listlngs include p o s i t i o n available, s t a r t i n g date,

salary, and benefits; e d u c a t i o n , experience, and s p e c i a l i z a t i o n

requirements.

Address :

Deadline: January

7,

1977

Fee : $20 f o r employers

$ 5 f o r a p p l i c a n t s , plus $5 f o r anonymous listing

Free to students

A form on which t o submit i n f o m a t i b n w i l l be supplied by

Orrin E. Taulbee

Computer Science Employment Register

Department of Computer Science

University of Pittsburgh

A D V I S O R Y

P A N E L F O R

L I N G U I S T I C S

WILLIAM 0. DINGWALL University of Haxyland

V I C T O R I A

FROMKIN University of California, Los Angeles

IVES GODDARD Harvard Univer s i ty

ROGER SHUY Georgetown University and CAL

CARLOTA S M I T H University of Texas

ARNOLD

ZWICKY Ohio State University

The

Panel

is expected to meet three times annually, in fall, winter, and spring. All proposals are to be reviewed at Panel

meetings; proposals received too late for the spring meeting

w i l l b e h e l d until fall, p o s s i b l y extending the customary s i x -

month interval between submission and decision.

American Journal

of

Computational Linguistics

Microfiche 51 : 41

T H I R D EUROPEAN CONGRESS ON

I N F O R M A T I O N

S Y S T E U S

OVERCOMING

THE

LANGUAGE

BARRIER

EURONET, supplying scientjfic, technical, and economic ipforma-

tion from diverse origins, will begin to operate in 1977. The

Congress deals with the language problems raised by the network.

1. Teaching and utilization of languages in the

European community.

2. Semantics, terminology, and lexicography, including

equipment fo't: rapid access to terminological resources.

3. Linguistics, grammar, and syntax, including computer-

aidedvcontrol and representation of syntactical

structures

4. Human translation and interpretation, translation

aids, and camputet-aided or semi-automatic translation.

5. Multilinguai thesauri and information retrieval systems.

6. Automatic translation.

A closing panel will discuss the evolution of multilingual s y s -

tems from the viewpoint of policy makers and users.

Information: Loll Rolling, Information Management (XIII-B),

Commission of the European Communities, European Center,

T R A N S A C T I O N S

O N

D A T A

B A S E

S Y S T E M S

E d i t o r : D A V I D K. H S A I O

Ohio S t a t e University

Associate editors: STUART

E.

M A O N I C K

Massachusetts Institute of Technology

C H R I S T I N E A. MONTGOMERY

Operating Sys terns, Inc

.

H O W A R D L. M O R G A N

University of ?Pennsylvania

E D G A R S I B L E Y

University of Maryland

Publisher : A S S O C I A T I O N F O R C O M P U T I N G M A C H I N E R Y P . 0. Box 12105, Church S t r e e t Station New York, New York 10249

Subscriptions : $40 for noq~llembers o f ACM

$15 f o r members

In the f i r s t issue: Selected papers from the International

Conference on Very Large Data Bases, Framingham, Massachusetts,