Báo cáo khoa học: "TRANSFORMING SYNTACTIC GRAPHS SEMANTIC GRAPHS*" docx

We argue that the resulting ambiguous graph, supported by an exclusion matrix, is a useful data structure for question answering and other semantic processing.. Here, we present meth

Trang 1

T R A N S F O R M I N G S Y N T A C T I C G R A P H S I N T O

H a e - C h a n g R i m

J u n g y u n S e o

R o b e r t F S i m m o n s

D e p a r t m e n t o f C o m p u t e r S c i e n c e s

a n d

A r t i f i c i a l I n t e l l i g e n c e L a b o r a t o r y

T a y l o r H a l l 2 1 2 4 , U n i v e r s i t y o f T e x a s a t A u s t i n ,

A u s t i n , T e x a s 7 8 7 1 2

A B S T R A C T

In this paper, we present a computational

method for transforming a s y n t a c t i c g r a p h ,

which represents all syntactic interpretations of a

sentence, into a s e m a n t i c g r a p h which filters out

certain interpretations, but also incorporates any

remaining ambiguities We argue that the result-

ing ambiguous graph, supported by an exclusion

matrix, is a useful data structure for question an-

swering and other semantic processing Our re-

search is based on the principle that ambiguity is

an inherent aspect of natural language communi-

cation

I N T R O D U C T I O N

In computing meaning representations from

natural language, ambiguities arise at each level

Some word sense ambiguities are resolved by syn-

tax while others depend on the context of dis-

course Sometimes, syntactic ambiguities are re-

solved during semantic processing, but often re-

main even through coherence analysis at the dis-

course level Finally, after syntactic, semantic,

and discourse processing, the resulting meaning

structure may still have multiple interpretations

For example, a news item from Associated Press,

November 22, 1989, quoted a rescued hostage,

"The foreigners were taken to the Estado

Mayor, army headquarters I left that

hotel a b o u t quarter to one, and by the

*This work is sponsored by the Army Research Office

under contract DAAG29-84-K-0060

47

time I got here in my room at quarter to

4 and turned on CNN, I saw myself on

T V getting into the little tank," Blood said

The article was datelined, Albuquerque N.M A first reading suggested that Mr Blood had been flown to Albuquerque, but further thought suggested that "here in my room" probably referred

to some sleeping section in the army headquarters But despite the guess, ambiguity remains

In a previous paper [Seo and Simmons 1989] we argued that a syntactic graph - - the union of all parse trees - - was a superior representation for further semantic processing It is a concise list of syntactically labeled triples, supported by an exclusion m a t r i x to show what pairs of triples are incompatible It is an easily accessible representation that provides succeeding semantic and discourse processes with complete information from the syntactic analysis Here, we present methods for transforming the syntactic graph to a functional graph (one using syntactic functions, SUB-

forming the functional graph to a semantic graph

of case relations

B A C K G R O U N D Most existing semantic processors for natural language systems (NLS) have depended on a strat- egy of selecting a single parse tree from a syntactic analysis component (actual or imagined) If semantic testing failed on that parse, the system would sel~,ct another - - backing up if using a top- down parser, or selecting another interpretation

Trang 2

p p n

0

1 (SNP saw John) 1

2 (VNP saw man) 2

3 ( D E T man a) 3

4 (NPP man on) 4

5 (VPP saw on) 5

6 ( D E T hill the) 6 , 7 ( P P N on hill) 7

S (VPP saw with) S

9 (NPP man with) 9 ,(11) 10 ( N P P hill with) 10

11 ( P P N with telescope) 11

12 ( D E T telescope a) 12

0 1 ~13 4 51617

1

S 9 10 11 12

1 1

Figure 1: Syntactic G r a p h and Exclusion Matrix for "John saw a m a n on the hill with a telescope."

from an all-paths chart Awareness has grown in

recent years t h a t this s t r a t e g y is not the best At-

t e m p t s by Marcus [1980] to use a deterministic

(look-ahead) tactic to ensure a single parse with-

out back-up, fail to account for common, garden-

p a t h sentences In general, top-down parsers with

backup have unpleasant implications for complex-

ity, while efficient all-paths parsers limited to com-

plexity O ( N 3) [Aho and Ullman 1972, Early 1970,

T o m i t a 1985] can find all parse trees in little more

time t h a n a single one If we a d o p t the economical

parsing s t r a t e g y of obtaining an all-paths parse,

the question remains, how best to use the parsing

information for subsequent processing

Approaches by B a r t o n and Berwick [1985] and

Rich e t al [1987] a m o n g others have suggested

what Rich has called a m b i g u i t y p r o c r a s t i n a -

t i o n in which a s y s t e m provides multiple potential

syntactic interpretations and postpones a choice

until a higher level process provides sufficient in-

formation to make a decision Syntactic repre-

sentations in these systems are incomplete and

m a y not always represent possible parses T o m i t a

[1985] suggested using a shared-packed-forest as an

economical m e t h o d to represent all and only the

parses resulting f r o m an all-paths analysis Unfor-

tunately, the resulting tree is difficult for a person

to read, and must be accessed by complex pro-

grams It was in this context t h a t we [Seo and

Simmons 1989] decided t h a t a graph composed of

the union of parse trees from an all-paths parser

would f o r m a superior representation for subse-

quent semantic processing

4 8

S Y N T A C T I C G R A P H S

In the previous p a p e r we argued t h a t the syntactic graph supported by an exclusion m a t r i x would provide all and "only" the information given by a parse forest 1 Let us first review an example of a syntactic g r a p h for the following sentence:

E x l ) J o h n saw a m a n on the hill with a telescope

There are at least five syntactic interpretations for E x l from a phrase structure g r a m m a r

T h e syntactic graph is represented as a set of

d o m i n a t o r - m o d i f i e r triples 2 as shown in the mid- dle of Figure 1 for E x l Each triple consists of a label, a head-word, and a modifier-word

Each triple represents an arc in a syntactic graph in the left of Figure 1 An arc is drawn

f r o m the head-word to the modifier-word T h e label of each triple, SNP, VNP, etc is uniquely determined according to the g r a m m a r rule used

to generate the triple For example, a triple with the label SNP is generated by the g r a m m a r rule,

S N T + N P + V P , V P P is f r o m the rule V P +

V P ÷ P P , and P P N from P P -+ P r e p ÷ N P , etc

We can notice t h a t the ambiguities in the graph are signalled by identical third terms (i.e., the same modifier-words with the same sentence position) in triples because a word cannot modify two different words in one syntactic interpretation In

1 We p r o v e d the "all" b u t h a v e discovered t h a t in c e r t a i n cases to be s h o w n later, the t r a n s f o r m a t i o n to a s e m a n t i c

g r a p h m a y r e s u l t in arcs t h a t do n o t o c c u r in a n y c o m p l e t e analysis

2Actually each w o r d in the triples also i n c l u d e s n o t a t i o n for position, a n d s y n t a c t i c class a n d f e a t u r e s of t h e word

Trang 3

Figure 2: Syntactic G r a p h and Exclusion Matrix for "The monkey lives in tropical jungles near rivers and

streams."

a graph, each node with multiple in-arcs shows an

ambiguous point There is a special arc, called

the r o o t a r e , which points to the head word of

the sentence T h e arc (0) of the syntactic graph in

Figure 1 represents a root arc A root arc contains

information (not shown) a b o u t the modalities of

the sentence such as voice: passive, active, mood:

declarative or wh-question, etc Notice t h a t a sen-

tence m a y have multiple root arcs because of syn-

tactic ambiguities involving the head verb

One interpretation can be obtained f r o m a syn-

tactic graph by picking up a set of triples with no

repeated third terms In this example, since there

are two identical occurrences of on and three of

with, there are 2 3 = 6 possible sentence interpre-

tations in the graph represented above However,

there must be only five interpretations for E x l

T h e reason t h a t we have more interpretations is

t h a t there are triples, called e x c l u s i v e t r i p l e s ,

which cannot co-occur in any syntactic interpre-

tation In this example, the triple ( v p p s a w o n )

and ( n p p m a n w i t h ) cannot co-occur since there

is no such interpretation in this sentence 3 T h a t ' s

why a syntactic graph must m a i n t a i n an e x e l u -

s l o n m a t r i x

An exclusion matrix, ( E m a t r i x ) , is an N • N

m a t r i x where N is the n u m b e r of triples If

E m a t r i x ( i , j ) = 1 then the i-th and j - t h triple

3Once the phrase "on the hill" is attached to saw, "with

a telescope" must be attached to either hill or saw, not

m 0 ~ n

cannot co-occur in any reading T h e exclusion matrix for E x l is shown in the right of Figure 1 In

E x l , the 'triples 5 and 9 cannot co-occur in any interpretation according to the matrix Trivially exclusive triples which share the same third t e r m are also marked in the matrix It is very impor-

tant to m a i n t a i n the E m a t r i x because otherwise

a syntactic graph generates more interpretations

t h a n actually result f r o m the parsing g r a m m a r Syntactic graphs and the exclusion m a t r i x are

c o m p u t e d from the chart (or forest) formed by

an all-paths chart parser G r a m m a r rules for the parse are in augmented phrase structure form, but are written to minimize their deviation from a pure context-free form, and thus, limit b o t h the conceptual and c o m p u t a t i o n a l complexity of the analysis system Details of the graph form, the

g r a m m a r , and the parser are given in (Seo and Simmons 1989)

C O M P U T I N G S E M A N T I C

G R A P H S F R O M S Y N T A C T I C

G R A P H S

49

An i m p o r t a n t test of the utility of syntactic graphs is to d e m o n s t r a t e t h a t they can be used directly to c o m p u t e corresponding semantic graphs

t h a t represent the union of acceptable case analyses Nothing would be gained, however, if we had

to extract one reading at a time f r o m the syntactic graph, t r a n s f o r m it, and so accumulate the union

of case analyses But if we can a p p l y a set of rules

Trang 4

,ubj(~s)

0

1

2

3

9 t01

112

14

15

16

17

50

51

52

53

54

55

0 1 2 3 9 1012141516175051]525354:55

1 1 1 1

1 1 1 i

1 1

1 1 1 1

1 1 1 1 1 1 1 1 1

1 1 1 i 1 1 I 1 1

1 1 1 1 1 1 1 1 1

F i g u r e 3: F u n c t i o n a l G r a p h a n d E x c l u s i o n M a t r i x for " T h e m o n k e y lives in t r o p i c a l j u n g l e s n e a r r i v e r s and

s t r e a m s "

d i r e c t l y to t h e s y n t a c t i c g r a p h , m a p p i n g it i n t o t h e

s e m a n t i c g r a p h , t h e n u s i n g t h e g r a p h c a n r e s u l t in

a s i g n i f i c a n t e c o n o m y o f c o m p u t a t i o n

W e c o m p u t e a s e m a n t i c g r a p h in a t w o - s t e p p r o -

cess F i r s t , we t r a n s f o r m t h e l a b e l e d d e p e n d e n c y

t r i p l e s r e s u l t i n g f r o m t h e p a r s e i n t o f u n c t i o n a l no-

t a t i o n , u s i n g l a b e l s such as subject, object, etc

a n d t r a n s f o r m i n g t o t h e c a n o n i c a l a c t i v e voice

T h i s r e s u l t s in a f u n c t i o n a l g r a p h as s h o w n in

F i g u r e 3 S e c o n d , t h e f u n c t i o n a l g r a p h is t r a n s -

f o r m e d i n t o t h e s e m a n t i c g r a p h o f F i g u r e 5 D u r -

ing t h e s e c o n d t r a n s f o r m a t i o n , f i l t e r i n g rules a r e

a p p l i e d t o r e d u c e t h e p o s s i b l e s y n t a c t i c i n t e r p r e -

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

C O M P U T I N G F U N C T I O N A L G R A P H S

To d e t e r m i n e S U B , O B J a n d I O B J correctly,

t h e p r o c e s s checks t h e t y p e s o f v e r b s in a sentence

a n d its voice, a c t i v e or passive In t h i s process,

a s y n t a c t i c t r i p l e is t r a n s f o r m e d i n t o a f u n c t i o n a l

t r i p l e : for e x a m p l e , ( s n p X Y ) is t r a n s f o r m e d

i n t o ( s u b j X Y ) in a n a c t i v e sentence

However, s o m e t r a n s f o r m a t i o n rules m a p s e v e r a l

s y n t a c t i c t r i p l e s i n t o one f u n c t i o n a l triple F o r

e x a m p l e , in a p a s s i v e s e n t e n c e , if t h r e e triples,

( v o i c e X p a s s i v e ) , ( v p p X b y ) , a n d ( p p n b y

Y ) , a r e in a s y n t a c t i c g r a p h a n d t h e y a r e n o t ex-

clusive w i t h e a c h o t h e r , t h e p r o c e s s p r o d u c e s one

f u n c t i o n a l t r i p l e ( s u b j X Y ) Since p r e p o s i t i o n s

a r e used as f u n c t i o n a l r e l a t i o n n a m e s , two s y n -

t a c t i c t r i p l e s for a p r e p o s i t i o n a l p h r a s e a r e a l s o

r e d u c e d i n t o one f u n c t i o n a l t r i p l e F o r e x a m p l e ,

50

( v p p l i v e s i n ) a n d ( p p n i n j u n g l e s ) a r e t r a n s -

f o r m e d i n t o ( i n l i v e s j u n g l e s ) T h e s e t r a n s f o r -

m a t i o n s a r e r e p r e s e n t e d in P r o l o g rules b a s e d on

g e n e r a l inference f o r m s such as t h e following: ( s t y p e X d e c l a r a t i v e ) & ( v o i c e X p a s s i v e ) & ( v p p X b y ) & ( p p n b y Y ) => ( s u b j e c t X Y) ( v p p X P ) ~ ( p p n P Y) &: n o t ( v o l c e X p a s - sive) => ( P X Y)

W h e n t h e left side o f a rule is s a t i s f i e d b y a set

o f t r i p l e s f r o m t h e g r a p h , t h e e x c l u s i o n m a t r i x is

c o n s u l t e d t o e n s u r e t h a t t h o s e t r i p l e s c a n a l l co-

o c c u r w i t h each o t h e r

T h i s s t e p o f t r a n s f o r m a t i o n is f a i r l y s t r a i g h t -

t o w a r d a n d d o e s n o t r e s o l v e a n y s y n t a c t i c a m b i g u - ities T h e r e f o r e , t h e p r o c e s s m u s t c a r e f u l l y t r a n s -

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

i n t o t h e e x c l u s i o n m a t r i x o f t h e f u n c t i o n a l g r a p h

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

s a m e i n t e r p r e t a t i o n s as t h e s y n t a c t i c g r a p h h a s 4

I n t u i t i v e l y , if a f u n c t i o n a l t r i p l e , s a y F , is p r o -

d u c e d f r o m a s y n t a c t i c t r i p l e , s a y T, t h e n F

m u s t be e x c l u s i v e w i t h a n y f u n c t i o n a l t r i p l e s p r o -

d u c e d f r o m t h e s y n t a c t i c t r i p l e s w h i c h a r e exclusive w i t h T W h e n m o r e t h a n one s y n t a c t i c triple,

s a y T [ s a r e i n v o l v e d in p r o d u c i n g one f u n c t i o n a l

t r i p l e , s a y F1, t h e p r o c e s s m a r k s t h e e x c l u s i o n 4At a late stage in our research we noticed that we could have written our grammar to result directly in syntactic- functional notation; but one consequence would be increasing the complexity of our grammar rules, requiring frequent tests and transformations, thus increasing conceptual and computational complexities

Trang 5

N : the implausible triple which will be removed

The process starts by calling r e m o v e - a l l - D e p e n d e n t - a r c s ( [ N ] )

r e m o v e - a l l - d e p e n d e n t - a r c s ( A r c s - t o - b e - r e m o v e d )

for all Arc in Arcs-to-be-removed do

begin

i] Arc is not removed yet

then

find all arcs pointing to the same node as Arc: call them Alt-arcs find arcs which are exclusive with every arc in Alt-arcs, call them Dependent-arcs remove Arc

remove entry of Arc from the exclusion matrix

r e m o v e - a l l - D e p e n d e n t - a r c s ( D e p e n d e n t - a r c s )

end

Figure 4: A l g o r i t h m for F i n d i n g D e p e n d e n t R e l a t i o n s

m a t r i x so t h a t F1 can be exclusive with all func-

tional triples which are p r o d u c e d f r o m the syntac-

tic triples which are exclusive with a n y of T/~s

T h e s y n t a c t i c g r a p h in Figure 2 has five possible

s y n t a c t i c i n t e r p r e t a t i o n s a n d all a n d only the five

s y n t a c t i c - f u n c t i o n a l i n t e r p r e t a t i o n s m u s t be con-

tained in the t r a n s f o r m e d f u n c t i o n a l g r a p h with

the new exclusion m a t r i x in Figure 3 Notice that,

in the f u n c t i o n a l graph, there is no single, func-

tional triple c o r r e s p o n d i n g to the s y n t a c t i c triples,

(~)-(8), (11) and (13) T h o s e s y n t a c t i c triples are

n o t used in o n e - t o - o n e t r a n s f o r m a t i o n of s y n t a c -

tic triples, b u t are involved in m a n y - t o - o n e trans-

f o r m a t i o n s to p r o d u c e the new f u n c t i o n a l triples,

(50)-(55), in the f u n c t i o n a l graph

C O M P U T I N G S E M A N T I C G R A P H S

O n c e a f u n c t i o n a l g r a p h is p r o d u c e d , it is trans-

f o r m e d into a s e m a n t i c graph T h i s t r a n s f o r m a -

tion consists o f the following two subtasks: given

a f u n c t i o n a l triple (i.e., an are in Figure 3), the

process m u s t be able to (1) check if there is a se-

m a n t i c a l l y m e a n i n g f u l relation for the triple (i.e.,

co-occurrence c o n s t r a i n t s test), (2) if the triple is

s e m a n t i c a l l y implausible, find a n d remove all func-

tional triples which are d e p e n d e n t on t h a t triple

T h e co-occurrence constraints test is a m a t t e r

of deciding w h e t h e r a given functional triple is se-

m a n t i c a l l y plausible or not 5 T h e process uses a

t y p e hierarchy for real world concepts a n d rules

t h a t state possible relations a m o n g them These

relations are in a case n o t a t i o n such as a g t for

agent, a e for affected-entity, etc For example, the

5 Eventually we will incorporate more sophisticated tests

as suggested by Hirst(1987) and others, but our current

emphasis is on the procedures for transforming graphs

51

subject(I) arc between lives and monkey n u m b e r e d

(1) in Figure 3 is s e m a n t i c a l l y plausible since a n -

i m a l can be an agent of live if the a n i m a l is a

subj of the live However, the subject arc between

and and monkey n u m b e r e d (15) in Figure 3 is se-

m a n t i c a l l y implausible, because the relation con- jvp connects and a n d streams, and monkey can not

be a subject of the verb streams In our knowledge base, the legitimate agent of the verb streams is a

f l o w - t h i n g such as a river

W h e n a given arc is d e t e r m i n e d to be semantically plausible, a p r o p e r case relation n a m e is assigned to m a k e an arc in the semantic graph For example, a case relation agt is found in our knowledge base between monkey and lives under the c o n s t r a i n t subject

If a triple is d e t e r m i n e d to be s e m a n t i c a l l y implausible, then the process removes the triple Let us explain the following definition before dis- cussing an interesting consequence

D e f i n i t i o n 1 A triple, say T1, is d e p e n d e n t

o n another triple, say T2, if every interpretation which uses 7"1 always uses T2

Then, when a triple is removed, if there are any triples which are d e p e n d e n t on the removed triple, those triples m u s t also be removed Notice t h a t the d e p e n d e n t o n relation between triples is transitive

Before presenting the a l g o r i t h m to find dependent triples of a triple, we need to discuss the following p r o p e r t y of a functional graph

P r o p e r t y 1 Each semantic interpretation derived from a functional graph must contain every node in each position once and only once

Trang 6

attr(S) ~rles

near(51)

0

1

2

3

9

10

12

50

51

52

53

54

55

Figure 5: Semantic G r a p h and Exclusion M a t r i x for "The monkey lives in tropical jungles near rivers and streams."

Here the position means the position of a word

in a sentence This p r o p e r t y ensures t h a t all words

in a sentence m u s t be used in a semantic interpre-

tation once and only once

T h e next p r o p e r t y follows from P r o p e r t y 1

P r o p e r t y 2 Ira triple is determined to be seman-

tically implausible, there must be at least one triple

which shares the same modifier-word Otherwise,

the sentence is syntactically or semantically ill-

formed

L e m m a 1 Assume that there are n triples, say

7"1 , Tn, sharing a node, say N , as a modifier-

word (i.e third term) in a functional graph I f

there is a triple, say T, which is exclusive with

T 1 , , T/-1, Ti+ l Tn and is not exclusive with

T~, T is dependent on Ti

This l e m m a is true because T cannot co-occur

with any other triples which have the node N as a

modifier-word except T / i n any interpretation By

P r o p e r t y 1, any interpretation which uses T must

use one triple which has N as a modifier-word

Since there is only one triple, 7~ t h a t can co-occur

with T, any interpretations which use T use T/.[3

Using the above lemma, we can find triples

which are dependent on a semantically implausible

triple directly f r o m the functional graph and the

corresponding exclusion matrix An algorithm for

finding a set of dependent relations is presented in

Figure 4

For example, in the functional graph in Fig-

ure 3, since monkey cannot be an a g t of streams,

the triple (15.) is determined to be semantically

52

implausible Since there is only one triple, (1),

which shares the same modifier-word, monkey, the

process finds triples which are exclusive with (1)

Those are triples numbered (14), (15), (16), and (17) Since these triples are dependent on (16),

these triples must also be removed when (16) is re-

moved Similarly, when the process removes (14),

it must find and remove all dependent triples of

(14) In this way, the process cascades the remove

operation by recursively determining the dependent triples of an implausible triple

Notice t h a t when one triple is removed, it removes possibly multiple a m b i g u o u s syntactic

i n t e r p r e t a t i o n s - - t w o interpretations are removed

by removing the triple (16) in this example, but

for the sentence, It is transmitted by eating shell- fish such as oysters living in infected waters, or

by drinking infected water, or by dirt from soiled fingers, 189 out of 378 ambiguous syntactic inter-

pretations are removed when the semantic relation ( r o o d w a t e r d r i n k i n g ) is rejected, e This saves

m a n y operations which must be done in other approaches which check syntactic trees one by one to make a semantic structure T h e resulting semantic graph and its exclusion m a t r i x derived from the functional graph in Figure 3 have three semantic interpretations and are illustrated in Figure 5 This is a reduction from five syntactic interpretations as a result of filtering out the possibility, ( a g t s t r e a m s m o n k e y )

There is one arc in Figure 5, labeled near(51),

t h a t proved to be of considerable interest to us

6 I n "infec'~ed d r i n k i n g w a t e r " , ( r o o d w a t e r d r i n k i n g )

is p l a u s i b l e b u t n o t i n " d r i n k i n g i n f e c t e d w a t e r "

Trang 7

If we a t t e m p t to generate a complete sentence us-

ing t h a t arc, we discover that we can only pro-

duce, "The monkey lives in tropical jungles near

rivers." There is no way that that a generation

with that arc can include "and streams" and no

sentence with "and streams" can use that arc

The arc, near(51), shows a failure in our ability

to rewrite the exclusion matrix correctly when we

removed the interpretation "the monkey lives

and streams." There was a possibility of the sen-

tence, "the monkey lives in jungles, (lives) near

rivers, and (he) streams." The redundant arc was

not dependent on subj(16) (in Figure 3) and thus

remains in the semantic graph T h e immediate

consequence is simply a redundant arc that will

not do harm; the implication is t h a t the exclusion

m a t r i x cannot filter certain arcs that are indirectly

dependent on certain forbidden interpretations

D I S C U S S I O N A N D C O N C L U S I O N

T h e utility of the resultant semantic graph can

be appreciated by close study of Figure 5 The

graph directly answers the following questions,

(assuming they have been parsed into case nota-

tion):

• Where does the monkey live?

1 in tropical jungles near rivers and

streams,

2 near rivers and streams,

3 in tropical jungles near rivers,

4 in tropical jungles

• Does the monkey live in jungles? Yes, by

each other

• Does the monkey live in rivers? No, because

is pointing to jungles not rivers

• Does the monkey live near jungles? No, be-

cause near(50) and conj(12) are exclusive, so

no path from live through near(50) can go

through eonj(12) to reach jungle, and the

other path from live through near(51) goes

Thus, by matching paths from the question

through the graph, and ensuring that no arc in

the answering path is forbidden to co-occur w i t h

any other, questions can be answered directly from

the graph

In conclusion, we have presented a computa-

tional method for directly computing semantic

graphs from syntactic graphs The most crucial

and economical aspect of the computation is the

53

capability of applying tests and transformations directly to the graph rather than applying the rules to one interpretation, then another, and another, etc When a semantic filtering rule rejects one implausible relation, then pruning all dependent relations of that relation directly from the syntactic graph has the effect of excluding sub- stantially many syntactic interpretations from further consideration An algorithm for finding such dependent relations is presented

In t h i s p a p e r , we did not consider the multiple word senses which may cause more semantic ambiguities than we have illustrated Incor- porating and minimizing word sense ambiguities

is part of our continuing research We are also currently investigating how to integrate semantic graphs of previous sentences with the current one,

to maintain a continuous context whose ambiguity is successively reduced by additional incoming sentences

R e f e r e n c e s

[1] Alfred V Aho, and Jeffrey D Ullman, The Theory of Parsing, Translation and Compil-

NJ, 1972

[2] G Edward Barton and Robert C Berwick,

"Parsing with Assertion Sets and Informa- tion Monotonicity," Proceedings of IJCAI-85:

769-771, 1985

[3] Jay Early, "An Efficient Context-free Pars- ing algorithm," Communications of the A CM,

Vol 13, No 2: 94-102, 1970

[4] Graeme Hirst, Semantic Interpretation and

versity Press, Cambridge, 1987

[5] Mitchell P Marcus, A Theory of Syntac-

Press, Cambridge, 1980

[6] Elain Rich, Jim Barnett, Kent Wittenburg and David Wroblewski, "Ambiguity Procras- tination," Proceedings of AAAL87: 571-576,

1987

[7] J u n g y u n Seo and Robert F Simmons, "Syn-

Union of All Ambiguous Parse Trees," Com-

1989

[8] Masaru Tomita, Efficient Parsing for Natu-

Boston, 1985

Tiêu đề	Transforming Syntactic Graphs Semantic Graphs
Tác giả	Hae-Chang Rim, Jungyun Seo, Robert F. Simmons
Trường học	University of Texas at Austin
Chuyên ngành	Computer Sciences and Artificial Intelligence
Thể loại	báo cáo khoa học
Thành phố	Austin

Định dạng
Số trang	7
Dung lượng	330,51 KB