G a r y G. Hendrix, C h a i r p e r s o n SRI International Menlo Park, CA 94025
Panelists
Jaime G. Carbonnell, Carnegie Mellon University Jeff Hill, Artificial Intelligence C o r p o r a t i o n Aravind Joshi, University of Pennsylvania Jerry Kaplan, Stanford University
Earl Sacerdoti, Machine Intelligence C o r p o r a t i o n David L. Waltz, University of Illinois
Stan Petrick, IBM C o r p o r a t i o n
1.1 The Interface Problem
A major problem faced by would-be users of com- puter systems is that computers generally make use of special-purpose languages familiar only to those trained in c o m p u t e r science. F o r a large n u m b e r of applications requiring interaction b e t w e e n humans and c o m p u t e r systems, it would be highly desirable for machines to converse in English or o t h e r natural lan- guages familiar to their human users.
Over the last decade, in l a b o r a t o r i e s a r o u n d the world, several c o m p u t e r systems have been developed that s u p p o r t at least e l e m e n t a r y levels of natural- language interaction. A m o n g these are such systems as those '.escribed in the several r e f e r e n c e s at the end of this paper.
1.2 P r o v e n Capabilities
Natural-language (NL) interfaces built so far have primarily addressed the problem of accessing informa- tion stored in conventional data base systems. A m o n g the p r o v e n capabilities exhibited by these systems are those that:
• Provide reasonably good NL access to specific data bases
• Access multiple, r e m o t e data bases. • Answer direct questions.
( " W h a t is Smith's salary?") • C o o r d i n a t e multiple files.
( " W h a t is Smith's l o c a t i o n ? " translates into " W h a t is the location of the d e p a r t m e n t of Smith?.")
• H a n d l e simple uses of pronouns. • Handle many elliptical inputs.
( " W h e r e is John? S a m ? " ) • Do basic r e p o r t generation.
( " B y sex and age, list the salary and title of em- ployees in New Y o r k . " )
• E x t e n d linguistic coverage at run time.
( D e f i n e " J D " as " J e f f e r s o n Davis J o n e s " L e t "Q1 Smith salary" be like " W h a t is the salary of e m p l o y e e Smith? . . . . Q1 JD A G E ? " )
• Analyze NULL answers.
( " H o w many Japanese carriers have inoperative air search r a d a r ? " Response: " T h e r e are no Jap- anese carriers.")
• R e s t a t e in English the system's i n t e r p r e t a t i o n of inputs.
~. C o r r e c t spelling errors.
• E n h a n c e the data in a data base with special-purpose functions.
(E.g., calculate distances b e t w e e n cities.)
1.3 Prominent Potential Applications
A m o n g the many promising prospects for NL inter- faces, in increasing order of perceived difficulty, are interfaces to structured data bases, simulation models (e.g., VISICALC), o p e r a t i n g systems, e x p e r t systems, t r a n s a c t i o n systems (e.g., airline r e s e r v a t i o n s ) , and text data bases.
1.4 Factors for Suitable Applications
F o r NL-interface m e t h o d o l o g y to be suitable for a given application, the construction of such an NL in- terface must be technologically practicable. M o r e o v e r , it should make a positive contribution to the achieve- ment of pragmatic goals in the area of application.
F o r NL-interface m e t h o d o l o g y to be technologically practicable now in a specific area of application:
• T h e r e must be a solid system to i n t e r f a c e with. ( G a r b a g e accessed is garbage retrieved.)
• T h e application domain must be conceptually well- bounded, i.e., there must be a limited n u m b e r of objects in the domain and a limited n u m b e r of rela- tionships among them.
• The domain's objects and relationships must be well- behaved. It is relatively s t r a i g h t f o r w a r d to deal with concrete objects such as ships and employees, but such intangibles as h u m a n goals, beliefs, plans, and wants present very serious problems.
with the fact that " e i t h e r P or Q is t r u e , " without k n o w i n g which is true.
• System users must h a v e r e a s o n a b l e e x p e c t a t i o n s of what the system can do. This is largely a question of the level of a b s t r a c t i o n at which a user wishes to interact with the system. F o r example, a m a r k e t - ing data b a s e m a y easily be built to a n s w e r the specific question " W h a t were our sales in M a y ? " It is far harder to build a system with the a b s t r a c t reasoning needed to handle " W h y are sales slump- ing?", which is p e r f e c t l y natural to ask in exactly the same domain as the previous question.
• Users m u s t be able to t y p e (at least until s p e e c h technology in c o m b i n a t i o n with N L - i n t e r f a c e tech- nology m a k e s it unnecessary).
F o r N L - i n t e r f a c e m e t h o d o l o g y to be useful in a given area, the application must require flexibility and diversity. If the same r e p o r t is to be p r o d u c e d e v e r y month, there is no particular a d v a n t a g e to requesting that it be p r i n t e d b y giving instructions in English. H o w e v e r , if there are hundreds or thousands of possi- ble i n d e p e n d e n t o p e r a t i o n s that a s y s t e m might be called u p o n to p e r f o r m , it m a y t h e n b e c o m e a v e r y difficult task to indicate precisely which o p e r a t i o n is desired. English is much b e t t e r suited for a task of such complexity than m e n u selection systems, and it is easier to learn and r e m e m b e r t h a n a s o p h i s t i c a t e d f o r m a l language.
N a t u r a l language m a y be of value even if there are only a few d o z e n o p e r a t i o n s to be discriminated, p r o - vided the o p e r a t i o n s are not p e r f o r m e d v e r y often. F o r example, a travel r e s e r v a t i o n s y s t e m might have o v e r 100 kinds of o p e r a t i o n s - too m a n y for function keys alone. F o r the p e r s o n w h o uses the m o r e o b s c u r e o p e r a t i o n only occasionally, natural language could be helpful.
The cost of creating an NL interface must be justi- fied either by the s y s t e m ' s volume of usage, or b e c a u s e utilization of the interface expedites access to data or other c o m p u t e r - b a s e d resources w h e n time is a critical factor.
In s u m m a r y , NL-interfaces are suitable for use in applications in which the personal or financial cost of learning a special p u r p o s e f o r m a l language m a y exceed the value of the i n f o r m a t i o n retrieved. This is m o s t likely to occur in situations where the typical user has only o n e (or a f e w ) queries at irregular intervals, needs to use the system only infrequently, is unfamiliar or u n c o m f o r t a b l e with f o r m a l languages, or has only a partial u n d e r s t a n d i n g of the structure or c o n t e n t of the underlying system.
1.5 A d v a n t a g e s of NL I n t e r f a c e s
Among the major advantages of NL interfaces are the following:
• N a t u r a l language is flexible.
• P e o p l e n e e d little training to use it in i n t e r f a c i n g with a c o m p u t e r system, and they have v e r y little difficulty r e m e m b e r i n g it (especially as c o m p a r e d with r e m e m b e r i n g the syntax or r e s e r v e d function terms of f o r m a l languages).
• N a t u r a l language p r o n o u n s , quantification, and c o n - textual c o n v e n t i o n s m a k e it e a s y to p e r f o r m a n u m - ber of o p e r a t i o n s in natural language that are diffi- cult or impossible in o t h e r languages. F o r m a n y applications, the use of natural language is faster t h a n using a m e n u system, c o m p o s i n g f o r m a l quer- ies, or writing c o m p u t e r programs.
• N a t u r a l languages allow f o l l o w - u p questions to build on the linguistic c o n t e x t e s t a b l i s h e d b y p r e v i o u s dialogue.
It is i m p o r t a n t to recognize that natural-language interfaces typically solve two p r o b l e m s simultaneously in p r o v i d i n g users with access to c o m p u t a t i o n a l re- sources.
• T h e y can deal with a natural language articulation of w h a t the user wants.
• T h e y can t r a n s f o r m a description of what the user w a n t s into a c o m p u t e r p r o g r a m that specifics h o w to accomplish it.
T h e aspect of a u t o m a t i c p r o g r a m m i n g p r o v i d e d by the second function is, for m a n y potential applications, at least as i m p o r t a n t and useful as the ability to deal with n a t u r a l - l a n g u a g e syntax.
T h e a u t o m a t i c - p r o g r a m m i n g a s p e c t of m a n y NL-interface s y s t e m s is a k e y b e n e f i t of the interface technology, in that it provides a m e a n s for reducing the high labor cost of using h u m a n s to p r o g r a m c o m - puter algorithms to grapple with the inevitable ad hoc p r o b l e m s that arise in c o n j u n c t i o n with a n y application system.
In general, the p r i m a r y utility of NL interfaces is that they s u p p o r t the user's view of the d o m a i n and the application system. In o t h e r words, t h e y t r a n s f o r m the u s e r ' s c o n c e p t s into t h o s e a c t u a l l y used b y the application s y s t e m - and do so in a m a t t e r of millise- conds. NL syntax provides helpful support, but it is not n e c e s s a r i l y the crucial f e a t u r e t h a t m a k e s these s y s t e m s useful. O t h e r types of interface s y s t e m s can also t r a n s f o r m the user's c o n c e p t u a l i z a t i o n - NL sys- tems do it b y virtue of their essential nature.
1.6 D i s a d v a n t a g e s of NL I n t e r f a c e s
frequently, and, c o n f o r m i n g to user p r e f e r e n c e and convenience, are kept as short as possible. H o w e v e r , one can well imagine wanting to use " p l a i n " English for the less c o m m o n c o m m a n d s or to ask the system for assistance, e.g. " h o w do I change the margins?"
A system that uses natural language does not " w e a r its constraints on its sleeves", i.e., the system's capa- bility is n o t reflected in the input language. This means that users can easily pose questions or give c o m m a n d s that are b e y o n d the ability of the system to interpret. This is in contrast with a menu system, in which the system - always in control of the conversa- tion - constrains the user to select from a limited num- ber of choices. W h e t h e r or not explicit constraints are useful depends largely on the particular application.
P e r h a p s the main d i s a d v a n t a g e of NL systems is that people tend to assume they are " s m a r t . " F o r example, if a system can provide NL access to a data base of information, users will tend to believe it can deduce other facts from that information - facts that, although not explicitly encoded, would be obvious to a n y o n e with c o m m o n sense. More formal systems are not e x p e c t e d to p e r f o r m c o m m o n sense reasoning, because their functionality, and therefore their inher- ent limitations, is readily apparent to the user.
1.7 P r o b l e m s and N e w T e c h n i q u e s
1.7.1 T h r e e Lines of Research and T h r e e Levels of S y s t e m s
There are three major lines of research on natural- language interfaces:
• To make interfaces more easily transportable to new applications.
• To increase the linguistic coverage of systems. • To increase the conceptual coverage of systems. These three lines are so tightly intertwined that re- search focusing on one almost inevitably involves re- search on all three.
The systems already created and the diverse facets of ongoing research can be divided into three levels of complexity. To elucidate the various extensions and new techniques that have been developed in NL inter- faces, let us define and discuss each of these levels in turn.
1.7.2 Level 1 S y s t e m s
Primary C h a r a c t e r i s t i c s
The primary characteristic of a Level 1 system is that the interface per se incorporates only an extreme- ly limited t h e o r y of the d o m a i n of application. 1 In particular, the interface m a y have access to taxonomic information a b o u t the sorts of objects in the applica- tion domain, and m a y have i n f o r m a t i o n a b o u t the names of relationships in which those objects m a y
participate - but it will not have knowledge of specific instances of relationships a m o n g objects. (That is, it m a y k n o w that an e m p l o y e e is a p e r s o n and that ":salary" is the name of an attribute relating employ- ees, but it will n o t explicitly e n c o d e the fact that J o h n ' s salary is $30,000.) As a rule, it will rely on a conventional, external data base as its sole source of such information.
Translation in a Level 1 system is most o f t e n made directly into a data base query. Seldom is there an explicit representation of what the user actually said.
Most NL systems created to date are of the Level 1 category. Moreover, systems at this level are the only ones currently available that are sufficiently fast and robust to be considered for serious applications. In particular, the INTELLECT system (A.I.Corp., 1981), which is typical of a Level 1 configuration, is the only system currently available commercially.
S o m e Level 1 E x t e n s i o n s
T h e p a r a g r a p h s below describe some relatively simple and inexpensive extensions c u r r e n t l y u n d e r d e v e l o p m e n t that should e n h a n c e the utility of Level 1 systems.
Transportability. There is general a g r e e m e n t that the greatest p r o b l e m n o w facing L e v e l 1 systems is h o w to make effective use of existing techniques on new sets of data. W o r k is under w a y on m e t h o d s to ease the transport problem, including work on using interactive dialogues with data base a d m i n i s t r a t o r s (Hendrix and Lewis 1981) for automatic acquisition of the information needed to create new interfaces.
Database Enhancement Tools. As illustrated by the following examples, it is often desirable to extend the conceptual coverage of an interface to include access not only to primary data, but also to functions that can c o m p u t e information derivable f r o m those primary data.
W h e r e is the F o x ?
( C a n be looked up directly in the data base.) H o w far is she from L A ?
(The locations of F o x and L A can be looked up. But the distance b e t w e e n them must be c o m p u t - ed.)
H o w soon could she get there?
(The time needed to travel the distance must be computed, taking into a c c o u n t that ships c a n n o t
follow straight courses if t h e y pass o v e r land masses, ice, or shallows.)
D e v e l o p e r s and users of interfaces n e e d special tools to m a k e it easier to create e n h a n c e m e n t functions and integrate t h e m with the l a n g u a g e - p r o c e s s i n g capability. Database Update. M a n y users would find Level 1 s y s t e m s m o r e s e r v i c e a b l e if t h e y could be e m p l o y e d not only for querying data bases, but also for updating t h e m ( K a p l a n and D a v i d s o n 1981). Admittedly, this can introduce problems. Suppose the user says
CHANGE BOB DAY'S LOCATION TO BLDG. 7. If the data base is c o n s t r u c t e d in such a w a y that an e m p l o y e e is associated with a d e p a r t m e n t , a d e p a r t - m e n t is associated with a location, and an e m p l o y e e ' s location is assumed to be that of his d e p a r t m e n t , then processing the u s e r ' s r e q u e s t will either result in a change in the location of the d e p a r t m e n t (and all its e m p l o y e e s ) or force a r e o r g a n i z a t i o n of the data base.
Multilevel Systems. F o r a n u m b e r of applications, users would like to be able to access data bases b y means of a n y one of several query systems at various levels of relative convenience, precision, and efficien- cy. F o r e x a m p l e , we might imagine a n a t u r a l - language system that:
• A c c e p t s English questions and translates t h e m into predicate calculus.
• Translates predicate calculus into d a t a base queries in a formal language that does not require the user to k n o w the structure of the data base.
• Translates such queries into a f o r m a l language that specifies particular joins b e t w e e n generic files. • T r a n s l a t e s those queries into specific codes t h a t
prescribe the order in which joins are to be m a d e on particular physical files.
Users could interact with the underlying data by ask- ing questions at any one of these levels. This f e a t u r e is available in Waltz 1978 and H e n d r i x et al. 1978.
Context Setting. Users would find it c o n v e n i e n t to restrict the c o n t e x t of evaluation. F o r example, a f t e r saying " C o n s i d e r only US ships," the question " W h a t ships are in the M e d " would retrieve only US ships in the Med. This feature is available in T h o m p s o n and T h o m p s o n 1975.
Graceful Failure. With little effort, the r e s p o n s e of m o s t L e v e l 1 s y s t e m s to failure could be g r e a t l y improved. Ideas for graceful failure include C o d d ' s 1974 notion of r e n d e z v o u s , flexible parsing (as in C a r b o n e l l and H a y e s 1981), and intelligent responses to null answers (as in K a p l a n 1979 and M a y s 1980). W o r k on u n g r a m m a t i c a l and u n p a r s a b l e sentences as in K w a s n y and S o n d h e i m e r 1981 is also relevant here.
Concise Responses. T h e answers p r o v i d e d b y Level 1 systems could be m a d e m o r e intelligent in s o m e spe- cial cases. F o r example, if asked, " W h o has a c o m p a - ny c a r ? " a s m a r t s y s t e m might answer " T h e president
and the V P s , " r a t h e r t h a n p r o d u c e a list of the n a m e s of the p r e s i d e n t a n d v i c e - p r e s i d e n t s . ( A n " o u t s m a r t e d " s y s t e m might a n s w e r " E m p l o y e e s in the ADM Building with ID-NUMBERs less t h a n 1072 whose SPOUSE-MIDDLE-NAME is J a n e . " )
Metaquestions. It has b e c o m e o b v i o u s o v e r the last few years that users of n a t u r a l - l a n g u a g e interfaces to d a t a b a s e s desire far m o r e t h a n m e r e access to the data therein. T h e r e are a n u m b e r of types of " m e t a - q u e s t i o n s " they would like to pose as well. A m o n g t h e m are the following:
W h a t i n f o r m a t i o n is in the db?
W h a t are the allowable values f o r e m p l o y e e j o b titles?
H o w timely are the sales data? H o w were they acquired? H o w reliable is the source? C a n y o u handle relative clauses?
C a n y o u handle a s e n t e n c e t h a t has a relative clause?
W h y might the ship not be r e a d y (causal relation- ships)?
S o m e steps in this direction h a v e b e e n t a k e n in M c K e o w n 1980 and H e n d r i x et al. 1978.
1.7.3 Level 2 S y s t e m s
Primary C h a r a c t e r i s t i c s
L e v e l 2 s y s t e m s m u s t include an explicit t h e o r y of their d o m a i n of application (or be able to acquire one " o n the fly," as in H a a s and H e n d r i x 1980!). T h a t is, t h e y i n c o r p o r a t e internal r e p r e s e n t a t i o n s of s o m e of the objects in the domain, as well as explicit k n o w - ledge a b o u t particular instances of relationships a m o n g those objects. T h e general p a r a d i g m of these s y s t e m s is to:
• Use an explicit t h e o r y of the application d o m a i n to control all processing.
• T r a n s l a t e into an i n t e r m e d i a t e logical f o r m , r a t h e r t h a n into a db query.
• Provide access to multiple k n o w l e d g e sources. • Use deduction techniques to aid translation and fact
retrieval.
• P r o v i d e discourse m o d e l s f o r n o u n - p h r a s e r e s o l u - tion.
• Allow explicit d e s c r i p t i o n s of e v e n t s with c o m p l e x histories.
• F o l l o w the c o u r s e of p r o c e s s e s to d e t e r m i n e the c h a n g i n g physical c o n t e x t against which n o u n -
phrases must be resolved. • Provide for d y n a m i c d a t a bases.
• Use constraints to check the validity of data.
n o u n phrases. L e v e l 2 s y s t e m s m a y also h a v e dis- course models that draw u p o n a source of knowledge as to which objects have b e e n m e n t i o n e d recently or are otherwise in focus (see G r o s z 1977) b e c a u s e of their particular p r o p e r t i e s and the s t r u c t u r e of the discourse.
To c o n t r a s t Levels 1 and 2, consider an NL inter- face to a railroad's i n f o r m a t i o n m a n a g e m e n t system. In a Level 1 system, a question such as " W h e r e are the b o x c a r s ? " will always m e a n " T e l l me the locations of A L L the b o x c a r s in the d a t a b a s e . " In a L e v e l 2 system, if we have just indicated that we w a n t to m a k e up a train f r o m the rolling stock in Yard 3, only b o x - cars in Yard 3 will be retrieved. In a L e v e l 1 system, if we ask " W h a t are the n u m b e r s of the c a r s ? " we will get n u m b e r s for all cars of all types. In a L e v e l 2 system, only the b o x c a r s in Y a r d 3 would be selected.
S y s t e m s of the L e v e l 2 t y p e are b e c o m i n g m o r e c o m m o n in l a b o r a t o r i e s and are likely to provide the basis f o r m o r e sophisticated interface s y s t e m s in the future. Level 2 ideas are also being d e v e l o p e d inde- p e n d e n t l y b y r e s e a r c h e r s c o n c e r n e d with d a t a b a s e s y s t e m s per se (such as W i e d e r h o l d , C h e n and M c L e o d ) .
The I n t e r m e d i a t e - L a n g u a g e Problem
It is generally easier to p r o d u c e a Level 1 s y s t e m that can cope in at least an e l e m e n t a r y m a n n e r with some arbitrary p h e n o m e n o n t h a n it is to p r o d u c e the c o r r e s p o n d i n g Level 2 system. This has led to the belief in some quarters that ad hoc s y s t e m s are actual- ly m o r e flexible than linguistically m o t i v a t e d systems.
This is p r o b a b l y a distortion of the true situation. L e v e l 1 systems translate directly into calls on soft- ware, w h e r e a s Level 2 s y s t e m s force all inputs into a uniform, intermediate logical form. T h e result is that L e v e l 2 s y s t e m s deal with a linguistic p h e n o m e n o n relatively well, once they deal with it at all, but they are forced to c o n f r o n t a general case b e f o r e t h e y c o p e with a n y specific instance. L e v e l 1 s y s t e m s can quick- ly a c c o m m o d a t e simple new e x t e n s i o n s , b u t t e n d to have v e r y u n e v e n linguistic coverage. As m o r e exten- sions are added, an unwieldy " h o u s e of c a r d s " is cre- ated which soon collapses under its o w n weight.
Beyond C o n v e n t i o n a l Data Bases
L e v e l 2 systems are well equipped to m o v e b e y o n d the relatively simple p r o b l e m s of interfacing with c o n - v e n t i o n a l d a t a b a s e s c o m p o s e d of a t o m i c facts. In particular, L e v e l 2 s y s t e m s m a y be easily i n t e r f a c e d with d a t a bases c a p a b l e of storing a n y w e l l - f o r m e d f o r m u l a in f i r s t - o r d e r logic - n a m e l y , with a m u c h richer b o d y of i n f o r m a t i o n t h a n is available in c o n v e n - tional data bases.
We m a y also e n v i s i o n L e v e l 2 s y s t e m s t h a t use m u l t i m e d i a c o m m u n i c a t i o n , e.g., the c o m b i n a t i o n of natural language with graphics and pointing.
Limitations of Level 2 S y s t e m s
A l t h o u g h Level 2 s y s t e m s c o n t a i n explicit theories of o b j e c t s a n d relationships in their a p p l i c a t i o n d o - mains, t h e y do n o t c o n t a i n explicit t h e o r i e s of the knowledge, goals, or plans of external systems, such as external data bases and the user. C o n s e q u e n t l y , L e v e l 2 s y s t e m s are incapable of r e a s o n i n g a b o u t the inten- tion of user inputs or a b o u t h o w to use external data bases. If e x t e r n a l d a t a b a s e s (or o t h e r k n o w l e d g e sources) are used by such systems, the a t t a c h m e n t to t h e m is p r o v i d e d t h r o u g h p r o c e d u r e s whose k n o w l e d g e of the e x t e r n a l s y s t e m s is implicit in the c o m p u t e r codes t h e m s e l v e s and is thus unavailable f o r m e a n i n g - ful e x a m i n a t i o n b y the system.
Basic S t u m b l i n g Blocks
E v e n for Level 2 systems, m a n y f u n d a m e n t a l p r o b - lems r e m a i n u n a n s w e r e d . M u c h of the deficiency cen- ters u p o n the c u r r e n t inability of c o m p u t e r s to r e p r e - sent a n d r e a s o n a b o u t time a n d space, s u b s t a n c e s , collective entities, events, actions, processes, n o n s t a n - d a r d quantifiers, p r o p o s i t i o n a l attitudes, and m o d a l i - ties. T h e s e are t h o r n y p r o b l e m s t h a t p h i l o s o p h e r s , linguists, m a t h e m a t i c i a n s , and c o m p u t e r scientists have b e e n wrestling with for m a n y years. Their solution is not likely to c o m e easily.
1,,7.4 Level 3 S y s t e m s
Principal C h a r a c t e r i s t i c s
L e v e l 3 systems contain explicit theories of e x t e r n a l agents, including i n f o r m a t i o n a b o u t their k n o w l e d g e , goals, and plans. Some possible agents are the user, various d a t a bases, h u m a n experts, and o t h e r s o f t w a r e systems. L e v e l 3 s y s t e m s a l w a y s t r a n s l a t e into an explicit r e p r e s e n t a t i o n of w h a t a user has said; t h a t analysis t h e n b e c o m e s the starting point for r e a s o n i n g about:
D, W h a t the user meant.
D. H o w to use i n t e r n a l a n d e x t e r n a l r e s o u r c e s to ac- quire i n f o r m a t i o n n e e d e d to r e s p o n d to the user. ~, H o w to c o m m u n i c a t e with the u s e r ' s implied goals. Eventually, we m a y see L e v e l 3 interfaces e m e r g i n g as a kind of i n f o r m a t i o n b r o k e r . Such a b r o k e r would:
t, M o d e l multiple e x t e r n a l entities ( b o t h h u m a n a n d mechanical).
=, C o m m u n i c a t e with each in his (or its) o w n language. =, U s e and explicitly u n d e r s t a n d g o a l - d i r e c t e d b e h a v -
ior.
m o r e expensive c o m p u t a t i o n a l l y t h a n the L e v e l 1 and 2 systems n o w available.
1.7.5 T h e N L - I n t e r f a c e P r o b l e m in P e r s p e c t i v e
If we are ultimately to achieve our long-range o b - jective of producing genuinely fluent natural-language interfaces to c o m p u t e r systems, we must recognize and p a y special attention to the f u n d a m e n t a l p r o b l e m s of language understanding. It has recently b e c o m e evi- dent that the use of natural language must be studied as one facet of a general t h e o r y of goal-directed b e - havior. In particular, to use natural language fluently, a s y s t e m m u s t u n d e r s t a n d h o w the c o m m u n i c a t i o n p r o c e s s itself is a f f e c t e d b y l a n g u a g e u s e r s ' goals, plans, and beliefs.
So that the field of C o m p u t a t i o n a l Linguistics m a y b e n e f i t o u r society in b o t h the short a n d the long term, it is i m p o r t a n t to continue w o r k at all three lev-
els of s y s t e m s described above. Because it has only r e c e n t l y b e c o m e technically f e a s i b l e to c o n s i d e r the actual c o n s t r u c t i o n of a L e v e l 3 system, special c o n - sideration should be given to launching a r e s e a r c h p r o g r a m at that level.
1.8 R e c o m m e n d a t i o n s
T h e N a t u r a l - L a n g u a g e I n t e r f a c e P a n e l of the W o r k s h o p on A p p l i e d C o m p u t a t i o n a l Linguistics in P e r s p e c t i v e m a d e the following r e c o m m e n d a t i o n s to the sponsors:
• I d e n t i f y a specific D o D data base a m e n a b l e to L e v e l 1 t e c h n o l o g y and, using p r o v e n techniques, s u p p o r t the construction of an interface to it.
• S u p p o r t A I core research on k n o w l e d g e r e p r e s e n t a - tion, acquisition, and use.
