Nominalizations in PUNDIT

(1)

NOMINALIZATIONS IN

P U N D I T

D e b o r a h A. Dahl, M a r t h a S. P a l m e r , Rebecca J. P a s s o n n e a u P a o l i Research C e n t e r

UNISYS Defense Systems 1 Defense Systems, UNISYS

P.O Box 517

Paoli, P A 19301 U S A

A B S T R A C T

This p a p e r describes the t r e a t m e n t of nominalizations in the P U N D I T text processing system. A single semantic definition is used for both nominalizations and the verbs to which they are related, with the same semantic roles, decompositions, and selectional restrictions on the semantic roles. However, because syntactically nominalizations are noun phrases, the processing which produces the semantic r e p r e s e n t a t i o n is different in several respects from t h a t used for clauses. (1) The rules relating the syntactic positions of the constituents to the roles t h a t they can fill are different. (2) The fact t h a t nominailzations are untensed while clauses normally are tensed means t h a t an a l t e r n a t i v e t r e a t m e n t of time is required for nomlnalizations. (3) Because none of the arguments of a nominallzation is syntactically obllga- t o r y , some differences in the control of the filling of roles are required, in p a r t i c u l a r , roles can be filled as p a r t of reference resolution for the nominalization. The differences in processing are cap- t u r e d by allowing the semantic i n t e r p r e t e r to operate in two different modes, one for clauses, and one for nominalizations. Because m a n y nomlnalizations are noun-noun compounds, this

approach also addresses this problem, by suggest- ing a w a y of dealing with one relatively tractable subset of noun-noun compounds.

1Formerly S D C - A Burroughs Company.

1. I n t r o d u c t i o n

In this paper we will discuss the analysis of nominalizations in the P U N D I T text processing system. 2 Syntactically, nomlnalizations are noun

phrases, as in examples (I)-(7).

(1) An inspection of lube oil filter revealed m e t a l particles.

(2) L o u of lube oll preuure occurred during

operation.

(3) S A C received h i f h u e a f e .

(4)

In~eeti#ation revealed adequate lube oil.

(5) Request replacement of SAC..

(6) Erosion of impellor blade tip is evident.

(7) Unit has low output air pressure, resulting

in ale*# gae turbine atarte.

Semantically, however, nominaliTatlons resemble clauses, with a p r e d l c a t e / a r g u m e n t structure like t h a t of the r e l a t e d verb. Our t r e a t m e n t a t t e m p t s to c a p t u r e these resemblances in such a way t h a t very little machinery is needed to analyze nominalizations other t h a n t h a t a l r e a d y in place for other noun phrases and clauses.

T h e r e are two types of differences between the t r e a t m e n t of nomlnalizatlons and t h a t of clauses. T h e r e are those based on linqui~tle

differences, related to (1) the mapping between syntactic arguments and semantic roles, which is

(2)

different in nomlnalisations and clauses, and (2) tense, which nomlnallsations lack. T h e r e are also differences in

control; in p a r t i c u l a r , control of the

filling of semantic roles and control of reference resolution. All of these issues will be discussed in detail below.

2 . C l a u s e a n a l y s i s

T h e semantic processing to be described in this p a p e r is p a r t of the P U N D I T s system for processing n a t u r a l language messages. The PUN- DIT system is a highly m o d u l a r system, w r i t t e n in Prolog, consisting of distinct syntactic, semantic a n d discourse components. ~-lirschman1985], and~-lirschman1986], describe the semantic com-

ponents of

PUNDIT,

while ~)ah11986, Palmer1988, Passonneau1986], describe the semantic and p r a g m a t i c components. The semantic d o m a i n from which these examples are t a k e n is t h a t of reports of failures of the s t a r t i n g air compressors, or s a c ' s , used in s t a r t i n g gas turbines on N a v y ships.

T h e goal of semantic analysis is to produce a r e p r e s e n t a t i o n of the i n f o r m a t i o n conveyed by t h e sentence, b o t h implicit and explicit. This involves 1) m a p p i n g the s y n t a c t i c realization onto an underlying p r e d i c a t e a r g u m e n t r e p r e s e n t a t i o n , e.g., assigning referents of p a r t i c u l a r syntactic consltuents to p r e d i c a t e arguments, and 2) mak- ]Jig implicit a r g u m e n t fillers expllclt. We are using an algorithm for semantic i n t e r p r e t a t i o n based on predicate decomposition t h a t i n t e g r a t e s the p e r f o r m a n c e of these tasks. The i n t e g r a t i o n is driven by the goal of filling in the predicate arguments of the decomposition.~almer1986].

In order to produce a semantic representa- tion of a clause, its verb is first decomposed into a semantic predicate r e p r e s e n t a t i o n a p p r o p r i a t e for the domain. The arguments of the predicates c o n s t i t u t e the SEMANTIC ROLES of the verb, which are slml]ar to cases 4 F o r example,

fall

decomposes into b e c o m e i n o p e r a t l v e , with p a t i e n t as its only semantic role. Semantic roles can be filled either by a s y n t a c t i c c o n s t i t u e n t or by reference

PUBLIC RELEASE, DISTRIBUTION UNLIMITED. s PUNDIT UNDderstands and Integrates Text

4 In this domain the semantic roles include: agent, In- s t i g a t o r , e x p e r i e n c e r , I n s t r u m e n t , t h e m e , I o e a t l o n , actor, p a t i e n t , s o u r c e , r e f e r e n c e _ p t and goal. There are domain specific criteria for selecting a range of semantic roles. The criteria which we have used are described

resolution from default or c o n t e x t u a l information.

We have categorized the semantic roles into three classes, based on how t h e y are filled Seman- tic roles such as t h e m e , a c t o r and p a t i e n t are s y n t a c t i c a l l y OBLIGATORY, and must be filled by surface constituents. Semantic roles are categorized as semantically ESSENTIAL when t h e y must be filled even if t h e r e is no s y n t a c t i c c o n s t i t u e n t avaUahle, s In this case t h e y can be filled p r a g m a t - ically, making use of reference resolution, as explained below. T h e default c a t e g o r i z a t i o n is NON-ESSENTIAL, which does not require t h a t the role be filled. T h e a l g o r i t h m in Figure 1 produces a semantic r e p r e s e n t a t i o n using this information. E a c h step in the a l g o r i t h m will be illustrated at least once in the next section using the following (typical) C A S R E P S t e x t .

~a© failed.

P u m p sheared.

Ineestifatiort reeealed metal eontamlnation

in filter.

2 . 1 . A S i m p l e E x a m p l e

D E C O M P O S E V E R B - T h e first example uses the f a l l decomposition for S a c

failed:

f a l l < -

b e e o m e P ( i n o p e r a t l v e P ( p a t l e n t ( P ) ) ) .

It indicates t h a t the e n t i t y filling the OBLIGA- TORY p a t i e n t role has or will become inopera- tive.

F O R p a t i e n t R O L E -

P R O P O S E S Y N T A C T I C C O N S T I T U E N T F I L L E R - A mapping rule indicates t h a t the syn- t a c t i c subject is a likely filler for a n y p a t i e n t role. T h e mapping rules make use of intuitions a b o u t s y n t a c t i c cues for indicating semantic roles first embodied in the notion of case ~lllmore1968,Palmer1981]. T h e mapping rules can t a k e a d v a n t a g e of general s y n t a c t i c cues like " S U B J E C T goes to P A T I E N T " while still indicating p a r t i c u l a r c o n t e x t sensitivities. (See ~ a l - mer1985] for details.)

in{Paseonneau198611

(3)

CALL R E F E R E N C E R E S O L U T I O N - See is the subject of ma© failed, and is suggested by the mapping rule as a 1Lkely filler of the p a t i e n t role. A t this point the semantic i n t e r p r e t e r asks noun phrase analysis to provide a unique referent for the noun phrase subject. Since no s a c , have been mentioned previously, a new name is created: s a e l .

T E S T S E L E C T I O N R E S T R I C T I O N S - In addition to the mapping rules t h a t are used to associ- ate s y n t a c t i c constituents with semantic roles, there are selection restrictions associated with each semantic role. T h e selection restrictions for

fail test w h e t h e r or not the filler of the p a t i e n t role is a mechanical device. A sac is a mechanical device so the subject of the sentence mac

failed maps s t r a i g h t f o r w a r d l y onto the p a t i e n t

role, e.g.,

b e e o m e P ( i n o p e r a t | v e P ( p a t | e n t ( s a c 1 ) ) ) .

Since there are no o t h e r roles to be filled the algorithm term~-ates successfully a t this point and the remaining steps are not applied. The next example illustrates f u r t h e r steps in the algorithm.

2 . 2 . U n f i l l e d O b l i g a t o r y R o l e s

T h e second u t t e r a n c e in the example, P s m p mheared, illustrates the effect of an unfilled obliga- t o r y role.

D E C O M P O S E V E R B - s h e a r ,

< - e a u s e P ( ! n s t i g a t o r ( I ) , b e e o m e P ( s h e a r e d P

( p a t l e n t ( P ) ) ) )

Sheer is an example of a verb t h a t can be used either transitively or intransitively. In both cases the p a t i e n t role is filled by a mechanical device t h a t becomes sheared. If the verb is used transitively, the i n s t i g a t o r of the shearin¢, also a mechanical device, is mentioned explicitly, as in,

The rotating driee shaft sheared the p s m p . If

the verb is used intransitively, as in the current example, the i n s t i g a t o r is not made explicit; however, the algorithm begins by a t t e m p t i n g to fill it in.

F O R I n s t i g a t o r R O L E - Working from left to right in the verb decomposition, the first role to

and relies heavily on w h a t can be assumed from the context.

be filled is the i n s t i g a t o r role. A mapping rule indicates t h a t the subject of the sentence, psmp,

is a likely filler for this role. Reference resolution returns p u m p 1 as the referent of the noun phrase. Since pump is a mechanical device, the selection restriction test passes.

F O R p a t i e n t R O L E - T h e r e are no syntactic constituents left, so a s y n t a c t i c c o n s t i t u e n t can- not be proposed and tested.

U N F I L L E D O B L I G A T O R Y R O L E S - The p a t l e n t role, a m e m b e r of the set of obligatory roles, is still unfilled. This causes failure, and the binding of p , * r n p l to the i n s t i g a t o r role is undone. The algorithm s t a r t s over again, trying to fill the instigator role.

F O R i n s t i g a t o r R O L E - T h e r e are no other mapping rules for i n s t i g a t o r , and it is non- essential, so Case 4 applies and it is left unfilled, e The algorithm tries again to fill in the p a t i e n t role.

F O R p a t l e n t R O L E - T w o mapping rules can a p p l y to the p a t i e n t role, one of which suggests the subject, in this case, the pump, as a filler. Reference resolution r e t u r n s p u m p 1 again, which passes the selection restriction of being a mechanical device. The final r e p r e s e n t a t i o n is:

e a u s e P ( i n s t l g a t o r ( I ) ,

b e e o m e P ( s h e a r e d P ( p a t l e n t ( p u m p l ) ) ) ) .

The last sentence in the t e x t , "Inveatlga- tion re~ealed metal e o n t a m i n a t l o n ~n filter," is

interesting mainly because of the occurrence of two nomlnallzations which are discussed in detail in a s e p a r a t e section.

2 . 3 . T e m p o r a l A n a l y s i s o f T e n s e d C l a u s e s

The t e m p o r a l component determines w h a t kind of situation a predication denotes and what time it is asserted to hold for ~assonneau1988]. Its input is the semantic decomposition of the verb and its arguments, tense, an indica- tion of whether the verb was in the perfect or progressive, and a list of u n a n a l y z e d constituents which m a y include t e m p o r a l adverbials. It generates three kinds of o u t p u t : an assignment of

(4)

a n a c t u a l t i m e to the p r e d i c a t i o n , if a p p r o p r i a t e ; a r e p r e s e n t a t i o n of the t y p e of s R u a t i o n denoted b y t h e p r e d i c a t i o n as either a s t a t e , a process or a t r a n s i t i o n event; a n d finally, a set of p r e d i c a t e s a b o u t the ordering of the t i m e of the s i t u a t i o n w i t h r e s p e c t to o t h e r times explicitly or implicitly m e n t i o n e d in t h e s a m e sentence. F o r the simple sentence, s a c / ' a i l e d , the input would consist of t h e s e m a n t i c decomposition a n d a p a s t tense m a r k e r :

D e e o m p o s l t l o n s

b e c o m e ( | n o p e r a t i v e ( p a t l e n t

(is s e l l ) ))

3 T e r b f o r m s P a s t

T h e o u t p u t would be a r e p r e s e n t a t i o n of a t r a n s i t i o n a l event, corresponding to the

moment

of

becoming

i n o p e r a t i v e , a n d a resulting s t a t e in which the sac is i n o p e r a t i v e for some

period

i n i t i a t i n g a t t h e

m o m e n t

of t r a n s i t i o n .

8.

N o m l n a l l s a t l o n s

N o m i n a l l z a t i o n s are processed v e r y slml]arly to clauses, b u t w i t h a few crucial d ~ e r e n c e s , b o t h in linguistic i n f o r m a t i o n accessed a n d in the con- t r o l of t h e a l g o r i t h m . T h e first i m p o r t a n t linguistic c h a r a c t e r i s t i c of the n o m ; n a l l z a t i o n a l g o r i t h m is t h a t the s a m e p r e d i c a t e decomposition c a n be used as is used for the r e l a t e d v e r b . Secondly, d ~ e r e n t m a p p i n g rules are required since s y n t a c - tically a n o m i n a l l s a t l o n is a noun phrase. F o r example, where a likely filler for the p a t i e n t of

fail, is

the s y n t a c t i c subject, a llkely filler for the p a t i e n t of failure is a n of pp.

Thirdly, nominal-

isations do not m a k e use of the o b l i g a t o r y classification for s e m a n t i c roles, since noun p h r a s e modifiers are not s y n t a c t i c a l l y o b l i g a t o r y .

In t e r m s of d~rerences in c o n t r o l s t r u c t u r e , because n o m ; n a l l z a t i o n s m a y themselves be a n a - phorlc, there are two s e p a r a t e role-filling stages in the a l g o r i t h m i n s t e a d of j u s t one. T h e first pass is for filling roles which are explicitly given s y n t a c t i - cally; essential roles are left unfilled. If a uomi- n a l i z a t i o n is being used a n a p h o r i c a l l y some of its roles m a y h a v e been specified o r otherwise filled when the e v e n t was first described. T h e a n a - phorlc reference to the event, the nomina]izatlon, would a u t o m a t i c a l l y inherit all of these role

This suggests the hypothesis that OBLIGATORY roles For clause decompositions automatically become BSSeNTL~ roles

for

nominalization decompositions. This hypothesis seems to hold in the current domain; however, it will have to be tested on other domains. We are indebted to James Allen for this observation.

fillers, as a b y - p r o d u c t of reference resolution. A f t e r the first pass, the i n t e r p r e t e r looks for a referent, which, if found, will unify w i t h the noml- n a l i s a t l o n r e p r e s e n t a t i o n , sharing v a r i a b l e bind- ings. This is a m e t h o d of filling unfilled roles p r a g - m a t i c a l l y t h a t is not c u r r e n t l y a v a i l a b l e to clause analysis s. H o w e v e r , the first pass w a s i m p o r t a n t for filling roles w i t h a n y explicit s y n t a c t i c argu- m e n t s of the n o m ; n a l i z a t l o n before a t t e m p t i n g to resolve its reference, since there m a y be more t h a n one e v e n t in the c o n t e x t w h k h n o m i n a l l z a - tion could be specifying. F o r e x a m p l e , failure of

p u m p a n d

failure of sac

c a n only be dis- tinguished b y the filler of the p a t i e n t role. A f t e r reference resolution a second role-filling pass is m a d e , where still unfilled roles m a y be filled p r a g - m a t i c a l l y w i t h d e f a u l t values in the s a m e w a y t h a t unfilled v e r b roles c a n be filled.

S . 1 . T e m p o r a l A n a l y s i s o f N o m l n a l l z a - t l o n s

As w i t h clauses, the t e m p o r a l analysis of norninallsatlons t a k e s place a f t e r the s e m a n t i c analysis. Also as w i t h clauses, one of the inputs to the t e m p o r a l analysis of n o m l n a ] i s a t l o n s is the s e m a n t i c decomposition. T h e c r i t i c a l d ~ e r e n c e b e t w e e n t h e t w o cases is t h a t a n o m ; n a l i s a t i o n does not occur w i t h tense. P U N D I T c o m p e n s a t e s by looking for r e l e v a n t t e m p o r a l i n f o r m a t i o n in the s u p e r o r d i n a t e c o n s t i t u e n t s in which the nomi- n a l i z a t l o n is e m b e d d e d . C u r r e n t l y , P U N D I T processes n o m l n a l i z a t l o n s in three t y p e s of con° texts.

T h e first c o n t e x t for which a n o m l n a l i s a t i o n is t e m p o r a l l y processed is w h e n it occurs as the p r e p o s i t i o n a l object of a t e m p o r a l connective (e.g.,

before, during, after)

a n d the m a t r i x clause denotes an a c t u a l s i t u a t i o n . F o r example,

in the sentence sac lube oil pressure decreased belato 60 pslg after engagement, the temporal

c o m p o n e n t processes the m a i n clause as referring to an a c t u a l e v e n t which h a p p e n e d in the p a s t a n d which resulted in a new situation. W h e n P U N D I T finds the t e m p o r a l a d v e r b i a l p h r a s e

after engagement,

it assumes t h a t the

engage-

m e a t also has a c t u a l t e m p o r a l reference. In such cases, the n o m l n a l i s a t | o n is processed using the

(5)

m e a n i n g of the a d v e r b a n d the tense of the m a i n clause.

T h e second c o n t e x t in which a nominallzation undergoes t e m p o r a l analysis is where it occurs as the a r g u m e n t to a v e r b providing t e m - p o r a l i n f o r m a t i o n a b o u t situations. Such verbs are classified as a s p e c t u a l . O c c u r is such a verb, so a sentence like failure o c c u r r e d would be processed v e r y s~miIarly to a clause with the simple p a s t tense of the r e l a t e d verb, i.e., aomethlng

faile&

A n o t h e r t y p e of v e r b whose n o m i n a l l z a t i o n a r g u m e n t s are t e m p o r a l l y processed is a v e r b which itself denotes a n a c t u a l s i t u a t i o n t h a t is s e m a n t i c a l l y distinct f r o m its a r g u m e n t s . F o r e x a m p l e , the sentence in,aestlgatlon re~ealed metal ¢onfam~natlon i~t oil filter mentions three situations: the s i t u a t i o n denoted b y the m a t r i x v e r b r e v e a l , a n d the two situations denoted b y its a r g u m e n t s , ineemt~gatlon a n d e o n t a m l n a t l o ~ If the s i t u a t i o n denoted by reveal h a s a c t u a l t e m - p o r a l reference, t h e n its a r g u m e n t s are presumed to as well.

8 . 2 . N o m i n a l l s a t l o n M a p p i n g R u l e s

We will Use the previous example, ineestl-

gatlon

revealed

metal eontamlnatlon in filter,

to illustrate the nom~nallsation analysis algo- r i t h m . We will describe the e o n t a m l n a t l o n e x a m p l e first, since all of its roles are filled b y s y n t a c t i c constituents. T h e d o t t e d llne divides the a l g o r i t h m in F i g u r e 2 in the A p p e n d i x into the p a r t s t h a t are the s a m e (above the line), and the p a r t s t h a t differ (below the llne.)

D E C O M P O S E V E R B - Contaminate decomposes into a NON-ESSENTIAL i n s t r u m e n t t h a t c o n t a m - inates an OBLIGATORY l o e a t l o n .

e o n t a m i n a t e < -

e o n t a m i n a t e d P ( i n s t r u m e n t (I), l o e a t l o n ( L ) )

F O R i n s t r u m e n t role - In the example, m e t a l is a noun modifier of contamination, a n d m e t a l l is selected as the filler of the i n s t r u m e n t role.

F O R t h e m e R O L E - T h e t h e m e of a nominaU- nation c a n be s y n t a c t i c a l l y realized by an o f pp or an in pp. T h e role is filled with f l l t e r l , the referent of/~l£er.

A t this point the t e m p o r a l c o m p o n e n t is called for

the n o m l n a l i s a t i o n metal eontamlnatlon in oll f i l t e r w i t h t w o inputs: the decomposition structure a n d the tense of the m a t r i x verb, in this case the simple p a s t . Because this p r e d i c a t e is s t a t i v e , the r e p r e s e n t a t i o n of the e o n t a m l n a - t l o n s i t u a t i o n is a s t a t e p r e d i c a t e with the decomposition a n d a p e r i o d t i m e a r g u m e n t as well as the unique identifier S, (which will be e v e n t u a l l y be i n s t a n t i a t e d b y reference resolution as [ e o n t a m i n a t e l ] ) :

s t a t e ( S ,

e o n t a m l n a t e d P

( i n s t r u m e n t ( m e t a l l ) , ] o e a t l o n ( f i l t e r l ) ) , ( p e r l o d ( S ) )

In this context, the p a s t tense indicates t h a t a t least one m o m e n t within the period of c o n t a m i n a - tion precedes the t i m e a t which the r e p o r t was filed.

CALL R E F E R E N C E R E S O L U T I O N F O R NOlV[I- NALLZATION - T h e r e are no previously mentioned ©ontamlnatlon events, so a new referent, e o n t a m l n a t l o n l is c r e a t e d . T h e r e are no unfilled roles, so the analysis is completed.

8 . 3 . F i l l l n g E s s e n t i a l R o l e s

T h e analysis of the o t h e r n o m i n a l l z a t i o n , in~emtlgatlon, illustrates how essential roles are filled. T h e decomposition of i n v e s t i g a t e h a s two s e m a n t i c roles, a NON-ESSENTIAL a g e n t doing the investigation a n d an OBLIGATORY t h e m e being investigated. 9

i n v e s t i g a t e

< - i n v e s t l g a t e P ( a g e n t ( A ) ~ t h e m e ( T ) )

T h e r e are no s y n t a c t i c constituents, so the m a p - ping stage is skipped, a n d reference resolution is called for the n o m i n a l l z a t l o n . T h e r e are no previously m e n t i o n e d investigative events in this example 10, so a new referent, i n v e s t i g a t | o n l is c r e a t e d . A t this point, a second pass is m a d e to a t t e m p t to fill a n y unfilled roles.

I In other domains, the t h e m e can be essential, as in "I heard a noise. Let's investigate."

(6)

F O R a g e n t R O L E - The role is NON-ESSENTIAL, so Case 4 applies, and it is left unfilled.

F O R t h e m e R O L E - The selection restriction on the t h e m e of an ineestlgation is t h a t it must be a d * m s g e d c o m p o n e n t or a d a u a a g e causing event. All of the events and entities mentioned so far, the , a e a n d the pump, the failsre

of the

sac

and the s h c a r / n g of the pump satisfy this criteria. In this case, the i t e m in focus, the ,hear- ing of the pump, would be selected ~)ah11986].

T h e final decomposition is:

i n v e s t l g a t e P ( a g e n t ( A ) , t h e m e ( s h e a r l ) )

4 . O t h e r C o m p o u n d s

In addition to nom~nalisations, P U N D I T deals with three o t h e r types of noun-noun compounds. One is the c a t e g o r y of nouns with arguments. These include p r e u u r e and temperature,

for example. T h e y are decomposed and have semantic roles like nominalisations; however, their t r e a t m e n t is different from t h a t of nomlualisa- tions in t h a t t h e y do not undergo time analysis, since t h e y do not describe t e m p o r a l situations. As an example, the definition of p r e u u r e , p r e s s u r e P ( t h e m e ( T ) , l o e a t i o n ( L ) ) , specifies t h e m e and l o c a t i o n as roles. T h e analysis of a noun phrase like sa© oil p r e u u r e would fill in the l o e a t l o n with the sac and the t h e m e with the oil, resulting in the final r e p r e s e n t a t i o n , p r e s s u r e P ( t h e m e ( o i l l ) , l o e a t l o n ( s a e l ) ) .

T h e s y n t a c t i c mapping rules for the roles permit the t h e m e to be filled in by either a noun modifier, such as all in this case, or the object of an o / prepositional phrase, as in p r c u u r e o / o i l . Siml-

larly, the mapping rules for the location allow it to be filled in by either a noun modifier or by the object of an in prepositional phrase. Because of this flexibility, the noun phrases, sac all pres- mute, all p r e u u r e in sac, and p r e s s u r e o f oi1 i n s a c , all receive the same analysis.

T h e second class of compounds is t h a t of nouns which do not have semantic roles. F o r these, a set of domain-specific semantic relationships between head nouns and noun modifiers has been developed. These include: a r e a o f o b j e c t , for example, blade tip, m a t e r l a l - f o r m , such as

metal partlclea; and m a t e r | a l - o b j e e t , such as

metal eyllnder. These relationships are assigned by examining the semantic properties of the nouns. The corresponding prepositional phrases, as in tip o/ blade, particle, o/ metal, and

cylinder of metal, have a similar analysis. Finally, m a n y noun-noun compounds are handled as idioms, in cases where there is no rea- son to analyze the semantics of their i n t e r n a l s t r u c t u r e . Idioms in the C A S R E P S domain include

, h i p , f o r c e , gear *hair, and connecting pin.

Our decision to t r e a t these as idioms does not imply t h a t we consider t h e m u n a n a l y z a b l e , or noncompositional, b u t r a t h e r t h a t , in this domain, t h e r e is no need to a n a l y z e t h e m a n y further.

5 . P r e v i o u s C o m p u t a t l o n a l T r e a t m e n t s

Previous c o m p u t a t i o n a l t r e a t m e n t s of nominalizations differ in two ways from the c u r r e n t a p p r o a c h . In the first place, nominallzations have often been t r e a t e d simply as one type of noun- noun compound. This viewpoint is a d o p t e d by ~ i n i n 1 9 8 0 , L e o n a r d 1 9 8 4 , B r a c h m a n ( n u l i ) ] . Cer- t a i n l y m a n y nomlnalizations c o n t a i n nominal premodifiers and hence, s y n t a c t i c a l l y , are noun- noun compounds; however, this a p p r o a c h obscures the generalization t h a t prepositional phrase modifiers in non-compound noun phrases often have the same semantic roles with respect to the h e a d noun as noun modifiers. P U N D I T ' s analysis is aimed a t a uniform t r e a t m e n t o f the semantic s~ml]arlty among expressions like repair of

e n f l n e , enf~ne r e p a i r , and Csomeone) r e p a i r e d e n g l n e r a t h e r t h a n the s y n t a c t i c similarity of

engine repair, sir p r e u u r e , and metal partl-

eles. O f the analyses mentioned above, B r a c h m a n ' s analysis seems to be most similar to ours in t h a t it provides an explicit link from the nominalization to the r e l a t e d verb to relate the roles of the noun to those of the verb. The second way in which our a p p r o a c h differs from previous approaches is t h a t P U N D I T ' s analysis is driven by taking the semantic roles of the predicate and trying to fill t h e m in a n y w a y it can. This means t h a t P U N D I T knows when a role is not explicitly present, and consequently can call on the o t h e r mechanisms which we have described above to fill it in. O t h e r approaches have tended to s t a r t by fitting the explicitly mentioned a r g u m e n t s into the role slots, thus t h e y lack this

flexibility.

6 . L | m l t a t | o n s

(7)

noun phrases like impel[or

blade t~p

erosion

c a n n o t be h a n d l e d b y our s y s t e m in its c u r r e n t s t a t e because impel[or b[a,le tip forms a s e m a n t i c unit a n d should be a n a l y s e d as a a single a r g u m e n t of eroaion. T h e second p r o b l e m k r e l a t e d to the first. T h e s y s t e m does not now keep t r a c k of the relative order of nora|hal modifiers. In this domain, this does not present serious problems, since there are no examples where a different order of modifiers would result in a d ~ e r e n t analysis. Generally, only one order is a c c e p t a b l e , as in mac oil eo~taminatlon, ~o~[

b o t h powerful a n d extenslble, a n d which will provide a n a t u r a l basis for f u r t h e r d e v e l o p m e n t .

A c k n o w l e d g e m e n t s

We would like to t h a n k L y n e t t e H i r s c h m a n a n d Bonnie W e b b e r for their helpful c o m m m e n t s on this p a p e r .

7 . C o n c l u s | o n s

In this p a p e r we h a v e described a t r e a t m e n t of nom~nalisatlons ill which the goal ls to m a x i m - [se the s~m~]arities b e t w e e n the processing of nom- inallsatlons a n d t h a t of the clauses to w h k h t h e y are r e l a t e d . T h e s e m a n t i c s~m~]arltles between nom~nallzatlons a n d clauses are c a p t u r e d by m a k - ing the s e m a n t i c roles, s e m a n t k decompositions, a n d selectional restrictions on the roles the s a m e for n o m l n a ] i s a t i o n s a n d their r e l a t e d verbs. As a result, the s a m e s e m a n t k r e p r e s e n t a t i o n k con- s t r u c t e d for b o t h structures. This s~m;|arity in r e p r e s e n t a t i o n in t u r n anows reference resolution to find referents for n o m ; n a l l s a t i o n s w h k h refer to events previously described in clauses. In addl- tion, it allows the time c o m p o n e n t to i n t e g r a t e t e m p o r a l relationships a m o n g events a n d situations described in clauses with those referred to by non~uaUsations.

On the o t h e r h a n d , where d ~ e r e n c e s b e t w e e n nom~uaUsations a n d clauses h a v e a clear ]ingulstic m o t i v a t i o n , our t r e a t m e n t provides for differences in processing. P U N D I T recognizes t h a t the s e m a n t i c roles of non~na]ised verbs are expressed s y n t a c t i c a l l y as modifiers of nouns r a t h e r t h a n a r g u m e n t s of clauses by h a v i n g a d ~ e r e n t set of s y n t a c t i c m a p p i n g rules. It ls also t r u e in n o m i n a l l s a t l o n s t h a t there are no s y n t a c - ticaUy o b l i g a t o r y a r g u m e n t s , so the analysis of a n o m ; n a l l s a t i o n does not fall when there is an unfilled o b l i g a t o r y role, as is the case with clauses. Finally, the t e m p o r a l analysis c o m p o n e n t is able to t a k e into a c c o u n t the f a c t t h a t nomlnallzatlons are untensed.

(8)

A P P E N D I X

DECOMPOSE VERB;

F O R E A C H S E M A N T I C R O L E

C A S E I: IF T H E R E A R E S Y N T A C T I C C O N S T I T U E N T S - P R O P O S E S Y N T A C T I C C O N S T I T U E N T F I L L E R

C A L L R E F E R E N C E R E S O L U T I O N & TEST SELECTIONAL RESTRICTIONS

C A S E 2: IF R O L E IS O B L I G A T O R Y A N D S Y N T A C T I C A L L Y U N F I L L E D - FAIL

C A S E 3: IF R O L E IS E S S E N T I A L A N D U N F I L L E D -

C A L L R E F E R E N C E R E S O L U T I O N T O H Y P O T H E S I Z E A F I L L E R & T E S T S E L E C T I O N A L R E S T R I C T I O N S

C A S E 4: IF R O L E IS N O N - E S S E N T I A L A N D U N F I L L E D - L E A V E U N F I L L E D

C A L L T E M P O R A L A N A L Y S I S O N D E C O M P O S I T I O N

FIKure 1. C l a u s e AJtalysls AlKorlChm

DECOMPOSE NOMINALIZATION

F O R E A C H S E M A N T I C R O L E :

IF THERE ARE SYNTACTIC CONSTITUENTS -

PROPOSE SYNTACTIC CONSTITUENT FILLER & CALL REFERENCE RESOLUTION

& TEST SELECTIONAL RESTRICTIONS

CALL TEMPORAL ANALYSIS ON DECOMPOSITION

CALL REFERENCE RESOLUTION FOR NOMINALIZATION NOUN PHRASE

FOR EACH SEMANTIC ROLE:

IF ESSENTIAL ROLE AND UNFILLED

CALL REFERENCE RESOLUTION TO HYPOTHESIZE A FILLER TEST SELECTIONAL RESTRICTIONS

ELSE LEAVE UNFILLED

(9)

R E F E R E N C E S

~rachman(nuU)]

Ronald J. Brachman, A Structural Paradigm for Representing Knowledge. In B B N Report No. S605,

Bolt Beranek & Newman, Cambridge, Massachusetts.

~ah11980]

Deborah A. Dahl, Focusing and Refer- ence Resolution in PUNDIT, Presented at AAAI, Philadelphia, PA, 1986.

~ah11987]

Deborah A. DaM, Determ~-ers, Entitles, and Contexts, Presented at TInlap°3, Las Cruces, New Mexico, January 7°9, 1987.

~11more1968]

C. J. F;nmore, The Case for Case. In

Uni, ersal, in Linguimtie Theory, E.

Bach and R. T. Harms (ed.), Holt, Rinehart, and Winston, New York, 1968.

~ininZO80]

Tim Finin,

The Semantic Interpretation of Compound Nominals, PhD Thesis, University of Tlll,ois at Urbana- Champaign, 1980.

[I-Iirschman1985]

L. Hirschman and K. Puder, Restriction Grammar: A Prolog Implementation. In

Lo¢ie Pro¢ramminff and ira Applica- tion,, D~-I.D. Warren and M. VanCaneghem (ed.), 1985.

~'lirschman1986]

L. H~rschman, Conjunction in Meta- Restriction Grammar. d. of Loglc Pro-

grumminq,

1986.

~eonard1984]

Rosemary Leonard, The Interpretation of En¢limh Noun Sequeneem on the Computer. North Holland, Amsterdam, 1984.

[Palmer1981]

Martha S. Palmer, A Case for Rule Driven Semantic Processing. Proc. o/ the 19th A C L Conference, June, 1981.

~almer1985]

Martha S. Palmer, Driving Semantics for a Limited Domain, Ph.D. thesis, University of Edinburgh,

198,5.

~almer1988]

Martha S. Palmer, Deborah A. Dahl, Rebecca J. ~assonneau] Sch~man, Lynette Hirschman, Marcia Linebarger, and John Dowding, Recovering Implicit Information, Presented at the 24th An- nual Meeting of the Association for Computational Linguistics, Columbia University, New York, August 1986.

~assonneau1988] o

Rebecca J. Passonneau,

A

Computa- tional Model of the Semantics of Tense and Aspect, Loglc-Based Systems Technical Memo No. 43, Paoli Research Center, System Development Corporation, November, 1986.

~assonneau198~