Speech Generation from Semantic Nets

American Journal

of

Computational Linguistics

~ i c r o f i c h e 33 : 6 3

A r t i f i c i a l I n t e l l i g e n c e C e n t e r S t a n f o r d R e s e a r c h I n s t i t u t e Men10 P a r k , C a l i f o r n i a 9 4 0 2 5

ABSTRACT

8atur.l languag8 o u t p u t can b* generatrd fram remantic nets by ptoc*rsing ternplats8 asroeiated with concept# in the n e t r A # q t a t v e r b teaplater i s belng derived from a Study o f t h e

surfrc9 syntax o f ran@ 3000 Englirh V e r b 8 8 f h e actlve forms of

t h e verb$ have b e e n t Z ~ r r i f l s d rtcorb&n$ t o s u b j e c t r o b j e e t C m 1 ~ and compl~aent(r1) there Syntactic P a t t e r n s , augmented w i t h case n a m s , ara used as a grammat t o Cantrol th4 generation O t t e x t , Thlr text

in

turn

i s p a s r a d through a speech syntheri8 program

and output by 4 V O T R A X speech rynth4o&zax, ThLr analyrar r h o u l d

ultimately benefit s y s t e m s a t t s a p t l n g t o understand E r l ~ l i ~ h i n p u t

by p r a v i d i n g surface _{e t r u c t u r s} t o d e e p c u l i t r u e t u r d maps using the rare trmpirter r r e m p 1 Q y t d b y t h e generator.

T h f r

reararch

w4r rupportrd by t h e Detenra Advaneed Reuearch Proleeta Agancy a f th* Dcprrtmmnt @ t D ~ f e n r e and m a n i t 6 r r d by t h e

TNTRQDUCTXON

It

c o m p u t e r 6 r t r t o

canmonicrtr

e f t r e t l v c l y w i t h p e o p l e ,

t h e y nust spark,

or

a t

i r a r t

w r i t e , t h e

urcr*r

nrtur61

Language,

The b u l k o t the work

in

c o ~ p y t a t l o n r l

L l n g u i l t i c r h a s b e e n

d r v o t r d t o computer

undcrstandlnq

o f

n a t u t a l

language i n p u t # b u t

r e l a t i v e l y l i t t i e

r t f o < r t

ha8 been e % p e n d @ d i n d e v e l o p i n $

natural

Language o u t p u t . M o r t Enplirh o u t p u t

systems

h r v e been along t h e

l i n e

o f n f i l l

In

t h e blanku w i t h Perhap& soma semantic

cenrtrrint6 i m p o r a d !

t h e r a

have bean few attempts a t l a n g u a g e

generation

from what

one

could

crLl

wsamrntic

n e t R

s t r u c t u r e s

[8g@aon8 and

8locua,

1972; Sloeuar 1973; Goldman, 1 9 7 4 1 ,

PeFh&p@

generation

i s c a n r l d e r e d a much

a a s i s r

p r o b l e m , The

ruccrrr o f

understanding

e f f o r t s

i r

generally b c l i s v t d t o d r p r n d

on rome warkrble

theory o f

Rdlrcaurrr

6rganlzatlonc

which w o u l d

rccbunt

gar affect8 o f

context

and

w o u l d

ahow how i n i p h o r i c

~ ~ p r e n 8 l ~ n s

(pronoun#

and

noun

phrases)

are

r e s o l v e d

and how

#rnt*nCeS

r r o

order84

t n

t h e b U t P U t r A s t t h r p p d n o t

there

m@chanlr@r

are

p r ~ L # * l Y

those

t h a t

r

w r r 8 p o n r e

g ~ n e r r t o r ~

must

Incorporate

i t i t 1 s t o

appear

Lntrlligent. Tha lrtudy o f

qrnrtrtion

r i l l p l a y

an

important

r o t s

in

r o l v i n g t h e p r o b l e m of

undrrttandino

i f i t

ern

draonrtrrte

& mapping t ~ o m d e r p

ramantic

rtrUctura6

t o r u t f a c e

r t r t n g r ,

b e t

ua

b r i e f l y

outlina

g o m e

relevant

preccrrrs

i n

t h e #peach

understanding

system

b r i n g d r Y c L 0 ~ 1 d by 8RI

and

SDC (Walker a t

a1.t 1 9 1 5 ~

and

Rite.,

1915).

The

urar

inltlatrs

rrrrton

by

( i n p u t and o u t p u t )

is

manlter@d b y a Wdircaurae m a d u l a 8 ( D a u t a c b ,

11975)

to

_maintain

r n

accurate ca~ntrerskition41 c o n t e x t , An

e x e e u t t v c

eserdinata#

v a r i a u r K n a r l a d g e

s a u x c o s

4 c W r t i ~ r

p r o s o d i c , syntactic, r ~ m r n t i e , pzagmrtlc, and d i ~ e e u r r c t o

B u n d e r o t r n b w

c u c c e r s i v t utt@rane@rr

The analyzed U t t a F l n g e i Q then Pas8rB l a ZRa

"resPandarw

PBItll

another

eamponrnt

o f the d i l e a u r t c module. The r e s p o n d e r may

c a l l the

pucrtlonransrcrrr

i t t h e l n p u t

i r

a

q u e s t i o n !

i t may

c a l l

4 d a t a bead u p d a t e

Program

I t ths i n p u t i s

r

otatcmant

o f

fact1

or

i t may d r c l d r an r o m c o t h e r r p p r o p r i a t c r e p l y . Tha

content

ab t h e r@DpDnD@ I s

p r r o r a

t o t h e g s n a r a t a r ~ p e r h a p a # i $ h

S o m e

tndicrtian

of hsa i t is t o b e t a r m u l a t @ d s TRa r e p l y may b s

a r t s r c a t y p s d _rsrpsnss ( w y @ 8 B p

" n o M ,

"1 @e@C)I

noun

p h r a r s

(nodel,

r

o@nt+nes

( v e r b n o d e ) r O r r e ~ ~ n t u 1 1 1 y ~ a paragraph,

The q e n ~ P a Q ~ r o u t p u t s s t e t t o t y p e d r a s p a n r e g Pmmed$atslyt if!

t h e

rcrpanrr

i r

nor@

c o m p l i c a t e d [ a

"nounB

nab@, * v e r b R nads, ar

rvcnturlly r

n r t w o r k ) ,

r

mbrr d e t a l l c d p r o p t a m i r ra g ulr c d. T h i s

p r o g r a m

nbLl

dctrrmina exactly how t h e r e s p o n g e i s t s b s

b o r m u l r t a d m a & g

N B I

61 O H t e q U I R C Q 8 f St! i t m a y bb ~ d ~ ~ l r 6 d

t o Chooag V e r b @ and

noun@

w i t h w h i c h t o @ X g s @ @ @ t h e d e e p C l S c nae

c t r u c t u r r r ,

r r w a l l r8 a ryntaetlc from. f o r t h e genaration. The

g ~ n a r a t o r

P ~ O ~ U C I I tha

r r r p o n r r

in

w f r x t w

form1

t h i r

in

t u r n

i r

p & g s a d t o a B ~ O I C ~ t y n t h r l l r program

t a r

trrnrtarmatlon

a n d

c u t p u t by

r

comn@rCial

VOTRAX r p r a c h r y n t h e ~ i z a r . C u r r ~ n t l y

no

IsntQner

intenstion

sr

r t r ~ a s

c o n % s W i n g i~ b e i n g p @ r t ~ r m e d ,

f u r t h e r r e f e r e n c e t o t h e s y n f h c r l g r t r p r i l l b e m e d e n

CQNSTRAXNTS Or? R E S P O N D I N G

T h e r e a r e s e v e r a l c e n s L d c r 8 t i a n r i n v o l v e d in r e s p o n d i n g

a p p r o p r i a t e b y t o an utterance. F i r s t , t h e r c a r e *conver:ctlanal

P o ~ t U l & t e # " (Gordon a n d Lakoff, 1 9 7 5 1 8 h a r s d b y t h e u 6 e r s a £ e

Lanuurprt

t h e r e p o t t u 2 a t e r s e r v e t o c a n s t r a i n t h e _{c o n t e n t} a n d

f o r m

o f

c a ~ ~ u n i c a t i o n r

from t h e speaker t o t h e h e e r a r e For i n s t a n c e , t h e r p e r k e r s h o u l d n o t t a l l t h e h e a r e r r e ~ s t h i n g t h e hearer all@&dy knows, l c r t h e b e b o r e 6 1 y e t t h e s p e a k e r c a n n o t t e l l t h e h c a r e r s a m e t h i n g _{t h e} h e a r e r k n o w s a b ~ o l u t e l y n a t h l n q

about, or t h e hearer r l l l n o t c o m p r e h e n d . T h e ggenkee c h o u l d

r e l a t e k n t

n c w a

i n h i s marsaga t o t h e p r t o r k n o w l e d g s o f t h e

h e a r e r ; t h i s r e q u i r e s t h e r p e r k c r t e h a v e ti m o d e l of! t h e h a a r c f

There h e u r i s t i c 6 m u r t o p e r a t e in c a n j u n c t b o n w ~ t h c w ~ ~ c ~ ~ n c ~

p r o d u c e r w

t o

constrain

what m a y b e o u t p u t b y a q ~ e n t c n c e n

g e n e r a t o r e

We

a r e

only b e g i n n i n g t o u n d e r s t a n G h o w t o l n c o r p a r r t t t h e r e p a r t u i a t e r in a l a n g u a g c g r o c e 6 ~ b n g k y g t c m ,

Then t h e r e

Is

t h e matter o f c o n 8 t r u c t l n g the b a t i c s e n t e n c e

N o r m a l E n g l i s h r y n t a x r c q u i r r r a t l t a a t a n t v e r b I n t h e s a n t c n c a f

c h o o r i n g a mrln v e r b constrains the r u r f d c t R t r u c t u r e , F o r

inltrncer i n t h r r b l c n c r of campounds a n y v e r b s o t h e r t h a n t h e

main v w b w i l l have t o a p p e a r i n a n o t h e r Fornr nominal,

i n f i n i t i v e ,

g e r u n d r p ~ r t L c I p L @ ~

or

s u b o r d i n a t e C f a u c e , H o w d a e c

t h e r e l e v a n t f n f a r m a t l o n c e n t a i n t d I n & g e m e n t i c n e t I n d i c a t e t h a

r p p r o p r l a t e farm? The t r a d i t l o n d l a n s w e r

I s

" b y m e a n n o f t h t

lexicon

and r d v s n c e a methodology

f a r

r a p r e l c n t l n g a m a p tram

d e e p c a s e

s t r u c t u r a

t a

surface

s t r u c t u r e ,

Thlr prgrr ~ O C U L I I an a p h i l o s o p h y o t s L n g l a m 6 c n f e n c r

farmattinqr

cmeaaing

a

main

vatbr choosing

t n a

g r a s s

s t r u c t u r e

s t

t h e o u t p u t scntencer and deciding how t o generate s p g r o p r i a t a

noun

p h r a r a s ,

Qur

exarnpl~a will

anB1oY

l i m P l i f i a 4

semantic

n e t

r t r u e t u r r r , remcrhat

like

t h o s e

i n t h e a c t u a l 5 R I ~ a a r t i t i a n a d

B l a r n t P ~

n e t a

8 y s t a m

lHandrlxr

1 9 7 5 1 , MMes

In

t h e

net

may

r a p r e r a n t

p h y l i c r l objects, rcllttonshipr, evantr, s b t r , r u l a s ,

a r

u t t s r e n c e s r a0

i n

t h e Q x e m p l a b a l a w , D i r e c t a d l a b @ P l e d areg

conncct n s d a a and r s p r a s a n t c e r t a i n w p r i m i % i r a n tlmc-invariant

r a l a t t s n r h t p a ,

In

t h e n @ t

L r r q m ~ n t

above, t h e U.8. an4 t h r U , K . r r a elcmrntl ( a )

in

OWNLng r l f u r t i o n l l n v o l v l n p as O B J l c t r p a r t i c u l a r r u b m a r i n e r )

reeh

r u b m a r l n t 11 an rlcmsnt o f s a m e c l a s s o t s u b m & r 1 n t s r

r n d

t e ~ p l a f e s f a t Engllah rtntcner8, Bc c h o a ~ e a s i m p l e verb f o r dernmstrrtfin a = OWN, W a

not@

t h a t our v e r b h $ @ v & t & 1

VynanymrRr W A V E , POSSESS, and BELONG C10L S l m c c l f h of t h c s c

v e r b 8 (including OWN1 ha8 o t h e r Ssnsr meanings, r e D a r l t a n o d e

tense

t h e y have

in

Comment this _node w i l l b e & h a m ~ r o t o t y p Z c r l E

OWHI

in

t h s t

It

w 1 1 2

incarpareta

t h e

u m s r n i n ~ R

of t h e a l t u a t l s n

and

Ln

t h a t all

inaerncrs

a f owning r i t u a t i a n s w i l l b 4 r e r a t ~ d P Q

P O S S E S S t HIVEI

BELONG] and

trmplatrr, Wsts t h a t one t e m p l a t e

r i l l

h a t r u f f t e a Ear a l l dour v e r b s ;

For

inatanesr t h e s u b j e c t 0 %

t h e subject i r t h e FXPeriencQsr

EXP

own&

OBJ 1 0BJ i s

owned

b y EXP

EXP p o r o a s s ~ a OBJ OBJ 1 s P a b l @ l l d d b y E X B EXP h r r aBJ OBJ G d r o n v t o EXP

tOWK

(EXP

V&ct

OBJ)

(OBJ V p r r B Y E X D ] )

[PO3SES8 (EX$ Yact OBJ) (063 Ypar B Y E X P I ]

f H A V E (EXP Vact OBJ)] IBEbON6 (084 V a c t T O EXP)]

t % d W r BELONG)

and

an a g ~ o e 1 a t e d t e m p l a t e (OBJ VaeE T O EXPI, and

But we have 4 p r o b l e m !

t h e r e

i s no indication o f how t h o EXP and OBd Qrgumcntt &re t o b e generated, NP w i l l not always

r o f t i c r r note f o r inrtanec t h a t t h e p r e d i c a t e argument o f

in %?shn hapad t o go h o m e n must b a an Infinltlva P h P 4 8 e ( r a t h e r than the QetUfid p h r r e e t h a t NP m i g h t p r o d u c a l , Even a c u r s o r y

8 t U d Y o f a few h u n d r e d

verbs

in

t h @ language s h a w l that t h e y have very defintts Cand r e g u l a r 1 C o n s t t a i n t &

on

t h e syntactic form 0 %

tnrrr

canststucntr, T h a a s c o n s t r a i n t 8 appear ts b c m a t t e r s

% o r

t h e

lexicon

r r t h a r than t h e grammar, we a s s o c i a t e

net) r a t h e r t h a n imwemsnt them

v i a

QP4MmBF TUI~II and w g cxpllcltly

incorporate

t h s conrtltutnt t y p e s

in

t h e t a m p l a t c s t

COWM

((NP

EXP) V a ~ t (NP OBJ)) (CNP OBJ)

Ypar

BY (NP EXP))I

(POSSES6 ((NP

EXP)

V a c t

(NP

OBJ)) ((NP O B J ) Vpe8 BY (NP EXP))]

[HAVE

((HP

EXPI

Vact (NP OBJ))]

[BELONG ((NP OBJ] Yaet TO (NP

EXP))I

A a c t

o f

p a t t e r n s liLe t h t r a i s a s s o c i a t e d * I t h r v s r y " P I O ~ ~ Y P . v e r b n nods

in

t h e k n s w Z ~ d g @ b r a @ * $t w o u l d s e a m that

m l l r a need i s an P n t n r g s e b r t k a b g i v ~ n any w ~ e ~ b i n s t a n c e R

node fn t h e knowb@dg@ b ~ g e r l 0 b k o up t h e p s t t b r n s P Q T t h a t t y p e of! n o d h C ~ O Q G Q ~ Q V Q t b r 1 e O r r a a p d n d i n g tsmglate

f o r

t h e v e r b @

and t h e n P r o e a e d r t s * e v r l u a t a n f h a p a t t e r n o

v e r b

[ O W N , S - 3 8 , D W N I sms b Q l ~ n $

tamp [(MP OBJ) Vact TQ (NP EXpll

(NP

Odd) 913 t h e S ~ a w o L f Y I C ~ -13 B c l o n q s

T O --3 t o

B u t

us

t t l i l

r u n

i n t o

t r o u b l e

w i t h

o u r

r i m p l c cchema,

Consldsr

t h e

rent@ncc,

*John b u r n e d the

t a a r f

b l a c k , "

\

ACT

\ ~ B J

colorl

By u g i n g t h e simple p a t t e r n U N P AGT) Vact (HP O B J I ) we c o u l d e a s i l y generate t h e

* L n c o r r r c t w

sentence,

rJahn burned t h e b l a c k

tmr;trW

since

(NP QBJ) might include the

color

o f t h e t e 8 l t t 8 Me

need a

p a t t e r n

m o r e

llke

((NP

AG,T) Vact [NP

O W )

( n o d R ~ S I I I

Zn

w h i c h t h e RESult of t h e a c t i o n w f l k b c d i r e c t l y r e l a t e d t o t h e

verb. Ha*rvcrr t h l r 1%

n o t

q u i t e enough a = rt l a a r t , n o t W t t h o u t

a

very

c0mpfic.tad Lntcrpretcr -*

because

t h e i n t e r p r e t e r

m u s t

RnoW t h a t (NF OBJ)

cannot

i n e l u d e h a v r a b P s RE8 ergurnawl

[ b l a c k ) .

Thus#

by

convention,

nay

indicate

an

e x t r a

argument

t o bs Passed t o r

eon6tttuant

q a n s r d o r

(such as t h e

Ounetisn

NP)

t e denote the

item(@)

not

t e

appear

in

t h e

resultant

canrtiturnta

( I N P ACT) Vact (HP OBJ RES) (Nod RES))

*he

pattern

(NP

OBJ RES)

mrrnr

Vgrnaratr

an NP u s i n g t h e O B J s c t

of

t h e VQrbr b u t do

n ~ t

i n e l u d e

t h a RESult of t h s v e r b in t h e

p a t t a r n o ( L e e r a

p a t t e r n

copy f a r

e v e r y

p o r r / b l r *mmlrrLngfl

constituent),

This

l e v e l

o f detail w o u l d b e

unrdasonable

i d f @ w o t h e r verbs c o u l d we t h i s template; however,

there

a r e rnt?t@ khan

rstrtlvaly

few

tern~lrtarr

baCh s h w ~ d b y

several

t a n g BB h u n d r e d @

There?

rrr

other

r o u r e q r

o t potantla1 p a t t e r n

p r o l t f a r a t i ~ n ~

an i m p a t t a n t

one

being t h e c a m b i n d t o r i a l arrangamants o f CQea

arguments

o f t i m e ,

manner,

and a t h a r s u c h ~ B ~ W b i ~ l l r &it well aa

other ( p o t 8 i b l Y

non*advrrbiall

ease arguments such as

sourer,

$ 0 4 1 ,

inttrumrnt,

e t c . Some a t $ha@@ a r g u m a ~ t d a r e r a t h @ r

censtralnsd

I n

t h e i r p a a i t i o n a

in

t h e

rantencar

b u t o t h e r s may

"esterday t h e s h i p

satled

f r o m t h e l i g h t h o u s e t o t h e d o c k , "

"The

# h i p

s a i l r d

f r o m

t h s

lighthau8a t o t h e dock yesterday." wYast@rday t h e s h i p r a i l e d t o t h e dock from t h e l l g h t h o u r ~ . ~

g t i s a f c a u a r a

unrsrranabl@

t a t r y t o

maintain

a l l t h e ~ a s 8 i b X e

p r f t r r n r ! Lnefcad r s

leave

Lnrsrtlon o f f h r t e r d v c r b i a f

rrguncntr

t o r ring11 heuristic

routine

( d c r e r i b c d below), There are

rrvcral

jurtitlcations f o r t h j l , amanp them8 t i ) t h e p a r t i c u l a r

farm 09 t h e

verb

cannot

b e grneobatsd u n t i l t h e s u b j e c t b b j e c t ( a 5 )

@ r ~ r t @ L

p o s s i b l e

P L & C I ~ ~

and ( 3 1 t h e t a

r t a

some

heuristic

one

may

question

whether passive t e n p l a t a o a h o u l d b e a t o r e 4 1

c@tt@inlyr they _{c o u l d} b e d e r i v e d r On t h e o t h e r hand, n e g l e c t i n g

t~

rtare

thorn woWd

farce

ur t o I n d % b a t @ wAth

each

Verb ( 8 4 n 0 3 ) ~

whether Lt can

(or,

ronrtimar, m u s t ) be p a r r i v i z c d , I n d i c a t i n g

" t r . n ~ l t l v @ V i r naf enough

since

there are tranritlvc vrrbr ( i . a , , vcrbr t h a t taka

an

object) t h a t

cannot

b c p s s s l v i z a d .

8incs we hava t o r t o r a t h e information a n y w a y p we Can r a v e dame

There

arr

rtvsrrl

_rraronr

f o r g c n r r a f i n q t h e v e r b a f t e r t h e

mafor

@rgumentr. F t r ~ t t h e l u b j r e t a u a t b e p a n e r r f a d ro t h a t t h e

v e r b

can

b r

mrdc

t o a p r r r

i n

nuabnr. S e c o n d , c e r t a i n rare t r n r r r

8tr

t r u e o f

v r r b - p r r t i c l r conblnitionc rkiir n o t cL t h e i c o l a t r d v e r b ,

Btncr,

in addition, prrticl@c

must a p p e a r r f t t r o b j e c t 8

t h a t F S h o r t (like

p r a n o w n r l

b u t b s f ~ r s a b j a e t e t h a t a r e l o n g

(like

noun

phraras), t h e p a r t i c l e nurt b e p o s i t i o n a d a f t e r t h e

o b j e c t j S

grncrrtcd.

F f n a L 1 ~ 1

i n o c r t i a n

a f soma a d v a r b i r b r

t r . ~ ,

annot"

rrqulrrr

an r u x l l i r r y v e r b I- t h u g v e r b g r n a r a t l o n

aurt

fotlor r d v r r b i r l grncratlon,

VERB PATTERNS

T h i s #tudy s t a r t e d with t h e 25 V e r b p a t t c r n r " p r c r c n t c d b y

Hornby ( 1 9 5 4 ) . T h e ~ e

Ln

t u r n

crwr f r o m a d l c t g o n a r y b y Hornby c t

el.,

( 1 9 4 8 ) . V c t M in t h e dict$onary _&re c Z & i r l t i r d & c ~ o r d i n g t a t h e i r g r o s s

syntactic

p r t t r r n ~ a f l u b j r c t p a b j r c t ( s 1 , end

c o m ~ ~ r a e n t C s ) ~

most

o t t h e ~ a f t q r n r r r a r u b - d l v g d a d e Thc rutkorr

cl@La

t h a t

there

p a t f r r n r WXXNnt f o r a l l

conttruetlanr

i n v o l v i n g

& I & t h e

verbs

I n t h e L r 4 i e t i a n a r y d n d r b y N t t n e f o n ,

I n

t h e

IangUagr, I tl&SrifiertLon i s n o t Z m m c d i & t e l Y u s e f u l t o

C ~ t h p U t d h t i ~ n a l l i n g u l r t r r l n c a i t d o a s not a d d r a g s u n d e r l y i n g

~ e m a a t i c s ~

H a v a t k h e l e ~ ~ ~ i t $ r c l a r r t h a t i t can ssrve ae t h s

b r r l r

f o r

r

&artvatIan

o l

U n d @ r l Y l n g c a r r r t r u c t u h a r

prrtlculatly, &r a bar16 f o r Vplnrrrtion t c n g l & t r o , *

There pattern8

&re

b @ l n g c a n v a r t a d i n t a t a m p l e t @ @ much llkt

tho#* d e r i v e d

a r r l f e t r

t h a t n r l y r i s i s B a i n g p e r f a r m e d w i t h

a p p e a r ) , There t e m p l a t e a s e r v e ro t h o major portion o f a m o d u l a r *gan@ratLon g r a m m a r r * w i t h t h e

r@aarlnder

in

t h r farm o f h s u t i a ; t i c

tunctianr

t o r c o n s t r u c t i n g syntactic constituentr,

N O U N PHRASES

What t o I n c l u B b

Ln

a

noun

phrase should b e another

mrttar

f a r

t h e d i r c a u r t e module t o j u d g e , There are

no

w ~ 1 l ~ E a r m ~ l a C a d

f u l r l

eccountlng

f a r

anaphorl

In

EngliShj i n d e a d , there a r c

t r w

w s f l e e 8 t r b l i e h a d parameters a t h e r than t h a t %ha hearer m u s t b a

a b l e t o r ~ a a l v a t h o (pto)nouns t o t h e i r r e f o r a n t s , T h e &D@ak@a?p

S h o u l d t m p l o y anaphoxa

In

order t o a v o l a t @ p e t i t l o n r b u t o n l y P t h i s B0deX of! t h e hearer i n d i c r t o ~ t h a t t h e ~ ~ Q P C P can t d s o l v d t h e

a m b i g u i t y , Thcrc

arc

same lawcpowrr ptaneminalization _{r u l e r} t h a t could be d i r e c t l y i n c a r p 0 r a t s d in a generator r a f g @ x $ v % z a l P a n r

f o r

s ~ 1 1 p 1 0 . N c V ~ P ~ ~ B ~ ~ S S ~ f t

IS

i m p a s t a n t L O

P ~ B Z I Z B

that When

r

generator

i r

unaware

o f t h e

e a n v c r t a f l ~ n a l

context, it l h a u l d

not indepsndantLy d e c i d e haw t o ganaratr

noun

p h p a 1 4 8 1 I t CQR

only d e c i d e when t o d6

r o e

T h i s r b t u a t l o n har not been u n i v + r o a l L y r @ c a q n f i e d r b u t I t 1 s becoming hnercarlngZY c l e a r

t h r t a Q Z ~ c s u r t e m s d u h

m u s t

b e c a n r u b t s d d u r i n g t h e ~ s n a r r t l a n p h r r s . The d i r e o u r r e modal. w i l l not know ahead of timr what NPI

are

% O b e p r ~ d u e @ d u n ~ @ s g $t p4rforrn\@ many o f ths g e m @ ~ P o r a t i a n a t h a t t h e paneratar w o u l d d o anyr4y. Y a t t h e cantcxt-8cnritivo d e c i r i o n s t r r t ~ g ~ may h . v e t o r e s o r b ts r u s h m a a r u t b r a r d l a e w b $ q u a t i n g t h e p r o p o r r b o u t p u t u o L w g khs model of &he hrsrer

in

order t a d ~ t e r m b n r what rnaphdra if4

t e s e a v a b l a ,

t %

I r

f o r rany

rrrsonr

i t nurt b e p a r t o f t h e d l r c o u r r r m o d u l c ,

T h e r e f a t e t h r g e n r r r t o r S h o u l d p a 6 S a n y R n c u n U c o n s t i t u c n t

t o t h @ dJscourre m o d u l e ( p l r h s p t r i t h i t s r r c o m m c n d a t i o n a b o u t

how t o P I g d U c e the c o n s t i t u e n t , ; t h n o d u l e ~ m u s t d c t c r m i n r i f a

p r o n o u n

or

b & g a

n o u n

i t ambiguour t o t h e h e a r e r , a n d , i f s o , w h a t

t o add t o t h e n o u n ln order t o makt t h e _{d e s i r e d r e f e r e n t e l e a r ,}

Ln

the

current

SRI s y s t e m ,

n o u n

p r t t c r n r [ S l o c u m , t a r p P c a r ) a r a

U S I ~ t o

control

noun

phrase

grnrrrtian,

Much l i k e v e r b p a t t e r n s ,

noun

p r t t c r n r order t h e ~ o n ~ t l f u @ n t 8 i n t h e p h r a r r a n d i n d i c a t e

how

each

eon8tLturnf Is t o b e generated b y naming a f u n c t i c n t o

b e Called w i t h t h e

nQtwerK

~ a n s t f t u e n t f

((DET]

[Ad3 QUA&) ( A d j SIZE) ( A d j SHRFE) ( A d 1 C O L O R ) (N)]

Batttrhr like t h i r

e r a

d i r t r i b u t r d a b o u t t h e network h i @ r @ r e h y j

in

t h e f u t u r e @ t h @ Q L ~ c ~ u r r c aadulc r i l l dccLde t o r e a c h p q t t r r n

constituent

whather f t i 8 t o appear i n t h e p h r r c s ,

MEURZSTXC RULES

R ~ r n b y d+rcrlb@r t h r e e b s r l c p o r l t l a n r f a r a d v c r b r in t h e

clrussr

"rentff

pa#b$tionp R ~ t d n pssltlonr and @ a n d u ~ e r i t l o n ~

Front

poritgen r d v r r b r

occur

b a t a r e

t h e s u b j r c t t w V e r t a ~ d a y h @

want

hornat from these ha t o o k a t a x i , * The i n t o r r o q l t i w a r d v a r b r

(a,g,

now,

when?

r r r

t y p i c a l l y c o n s t r d k n ~ d t o

front

P o o i t i o n l

o t h e r 8 miy r p p r r t t h r r r t a t p u r p o r r r o f r m p h a r l a

cr

e a n t r a l t .

M i d p e r l e i o n

r d v a r b a

occur

v i r h t h e v e r b ( a t r i n g l t Sf t h e r e

ar*

medal

s r

ru#ilLrrv

v a r h r s

t h e dbvrrls

occurs

a f t o r t h e f i r s t

one,

Oth@twi8r the r d ~ ~ r b

w ~ L L

appear b t k ~ r @ t h e V e r b l cxcapt

t h a r a r l @ o h @ if8 t y p i ~ l l l y buw8] *he I t 1 t 1 1 I waiting,@

End p o s i t i o n a d v e r b s

occur

r t t r r

t h e v e r b and

a f t e r

any

d b t e e t or indirect object pteacnt, While t a l a t i v ~ l y f e n

clrurss

have mare t h r h

one

a d V Q r b

i n

tront parttian

or

mara

t h r n

sns

In

@ i d

porltianr

i t i r

common

t a t r r v c r r l .dV@rbS t o appear

in

and

~ a r l t l o n

in

t h a

rams

c l a u s a l * t h a V p l a y t h e

P i a n a

PeerlY

t a g c t h r r w ,

h d v a r b l r l r

o f

t i m e

(rnrrrring

t h e 4uartiont w r h @ n ? w ) usually

d e e u t

in a n d

_{p o a l t i o n ,}

b u t may appear

In grsnt

p o s i t i o n

_{f a r}

aaphrrtr

or

contrast.

k d v e r b i a l r o f frequency (anrwsrlng t k a

quettbon, wRaw ~ f t a n ? ~ l

_can

b e s p l i t

i n Q

t w o g r o u p s , The f i r s t

group is comgo@@d ~f ringla-nerd a d v e r b s t h a t t y a l c a l l y s e e u i fn

& i d porition b u t a I r o may b r

in

i n d p a t L t i a n j t h @ s a c a n d i r eosparsd a $ m u l t i p l s = w o r d phrrrrr t h a t dgp&ar Pn end p a r i t i o n

e r ~

&err L r r 9 U r n t l Y r

in

t r e n t

Position. A d v e r b s of duration [ V l f ~ r l

hew

lonqtw)

u d u ~ l l y

have end

parltlan,

wbth %song p6sBtPon

f b r

empha8fs

or

csntxartr Adverb8 o f p l a c e and d i r a e t l e n narrnall~

hivQ and

porition.

A d v c r b r o f 4 r g r c a and manner h a v e m i d o r and

~ O S I ~ L Q R I

d e p b n d i n g

on

t h r a d v a r b ,

Along W i t h r u a h r u l e r c o n ~ ~ r n l n g t h e B a r L t i o n r O L Vsriout t y p e s o f a d v r r b r ,

t h e r e

murt b r a mcchanirm t o 0rd.f t h e r d v c r b r t h a t arr L a

s ~ c u r

In

t h a "maQn p e t i t L o n , T h r r r

r r a

e o m e

h @ u a l r t i c l ~ among rdverbi.11 o f

t i m e

Car

p l a ~ 1 1 t h e s m a l l a t

unit

81 u r u r l l y pL4ccb

t l r l t ,

u n l r r l i t

t r

44604 4 8 &R s f t a r t h a u p h t l

usually precede t h o r e

of

f r e q u e n c y , w h i c h i n

t u r n

p r a e o d e t h o o ~ a t time,

T h e r e r u l e 8 r e i a p l r m o n t s d in th@ raRr

r a u t L n ~

M a t produce@ t h e v e r b 1 when @ t e m p l a t e i s f i r a t f n e a r p r c k e d =- N W h

eo B 8 e q u e n c c o f

function

e r l l r t h e R V a ~ t w o r "Vpao k a y a a r c

lonored.

once

t h e l U b j C C t , ~ b j C C t ( S 1 a n d c o m p l t m c n t ( a ) l n d i c r t r d

by t h e

t c n p l r t r

rrr

O a n r r r f r d r t h i s

* c l e a n

u p m

r o u t i n e

$ 6 c a l l e d . Et employg the h a u x i t t i e k d ~ r c r f b e d a b a v a L a a d d t h e a d v e r b i a l C ~ r t & t f t U @ n t & and V h F b r t h e n concatsnatss t h e c a n s t b t u t n t s t a

produce

r

complete

@IruseI

DXSCUSSTON

f n

theory, t h e s e t o f p o r s i b l c E n p L l r h r a n t c n c r s i a

i n f fnZtsr Tha o b v l a o r q u s a t l b n m a n

~ t @ @ @ r

O l f

sne

t r i e s t o

account

ror t h e m with t a r n p l i t r r r

won

t t h e r e b c an i n f i n i t e

n u m b e r eL

t a a p l a t t ~ ? ~ The s i m p l e antxcr i r r "Nor t o r romc of t h e

Sam@

r e l l ~ n l

t h a t

allow

a f i n i t e grammar t o g c n r r a t a an i n t l n i t e

n u m b e r

of

itr2ngr.R One can p r e d u c a r a n t r n c e r o f a r b i t r a r y L e n g t h

by ( 1 ) ~ T ' b l t t ' a t ~ a n b @ d d l n g r and ( 2 1 a r b i t r a r y c o n j u n c t t a n @ Caw@

B o t o not do s o b y includdng a r b i t r a r y n u m b e r s sf d i s t i n c t c a r @

erpuacntl. Evrn r a t h e numbcr o t b a s i c p a t t e r n s c o u l d b e t n l ~ a m e l Y I a f gc

,

E v i d r n c a , horevrrr i l t c t h e c o n t r a r y ! t h e

l V e n t U I I nUnbqr o f [email protected]@# w o u l d appear t o b r r e v e r a l t i m e 8 t h e

number O I p l t t e r n r ~ a w i n g t o t h r S U b l t i t U t i o n o f p a r t t c u l a r

p r a ~ a r i t l o n e

Ear

n p r ~ p w

In

t h e r ~ ~ t i c t l c attaro or, a n d t h a

arrlgnaent

o t dltfetent c r l e nancr t o a p a r t i c u l a r canstitusnt

Dcotsch,

_Barbara

G, Establishing

Context

in

Task-Oriented

b l a l d g ~ .

Prrrcnted

a t

t h e

Thirteenth

Annual

Meeting o t t h e

Atsooiatian

f a t

Computational

Llngutrflcr,

Barton,

M a ~ r a c h u ~ d t t ~ ,

S O

October

* 1

Navarnber

1 9 7 5 ,

OoLdmrnr Neil

H e

Computcr

G ~ n e r a t i o n o f

Natural L@nguaga

tram

a

D I I ~

Conceptual

I r r e .

A 1

M e m o

2 4 7 , A r t i t i ~ i a l

I n t e l l i g e n ~ ~

Laboratory, 8trnford Univcr8ityt S t r r n f

ord,

C q i f o r n i b 1974,

Gordon

D I v i d t

and

Lakaff, George.

Convearettonal

Portulater.

Syntax s n d

Senanticst

V o l u m e

3 1 S p e e c h A c t s , E d i t e d by Peter Cole and J e r r y

L. Morganr

A c a d r m i a

P r s a ~ ,

N e w Y g r k r

1975 rn

H@ndtixt

Gary G. Expanding thr U t i l i t y o f

Bemantie

Nafworkr f hrough

Pact

i t

toning,

Advance

_Pav@r

_r

o f t h e

F o u r t h

Internationcl

J o h t

Contcrrncs

on A r t i f i c i a l

Z n t b l l l g e n e e , T b i i i o l r C s o r ~ i a ,

U6SRl _3.8 G e p t a m b c r 1 9 7 5 , I I S a I Z l r

Hornby, 8 t r Gatrcby, V e t a n d W~lK@flb@Xdr H Q The Advmced

Laatnet@@

Dictionary

o f

C u r r a n t

Engllrh. O x f o r d

P r @ r r ~

London@ 1 9 4 8 ,

Rita.#

H, Barry,

Autoastic S p e e c h

U n d e r r f a n d i n ~

SYrtamr.

Proessdlngr,

Eleventh

Annual

IEEE

Camputst S a e l e t y

Contcranea,

Harhlngton,

D.

C e r 9 - i 1 S ~ p t r m b e r 1 9 7 5 ,

Siamanrr R o b t r t & I and S ~ O C U ~ ~

Jonathan.

Gaa@rartinq

English

Dlrcourcr

f r o m

Senrntlc

N I E W O ~ K C . CommUnicstlonr ~t the

ACH, 1 9 9 2 , 1 s t 8 9 1 1 9 0 9 n

S J o c u l ,

Jonathan.

Q u a r t i o n

Anarrring v i a Csnanlcrl V e r b 8

a n d Bsrnrnt

l c

Madcllat t C ~ n s r a t i n g %ngllrh

t r o w

t h e W O ~ Q ~ ,

Technical

Report H t = i 3 t

Department

a & Csapartrr bcbtawcar r

Univeoritp

a t T a ~ a r ~ Austin@ Ttx.8,

January

1 9 1 3 .

$locum,

Jonathan,

Verb Patterns and

Noun

P a t t a r n s

I n

E n g l i ~ h t A C l & r

Anrlyrir.

A r t l f i ~ L a l

Lntrlliprncs Center,

SRX4