connotation as a form of inference - Association for Computational ...

CONNOTATION JB BERTHELIN,

AS

A

University of Essex

FORM

OF

I ~qFERENC

E

(language and l i n g u i s t i c s ) , Wivenhoe Pk, COLCHESTER C043Sq, GREAT BRITAIN,

and INSTITUT DE PROGRAMMATION, Universit~ Paris 6, 75230 PARIS

CEDEX05,

FRANCE.

Abstract Story-processing systems have to deal, or avoid dealing, with INFERENCE CONTROL' ' 6 ' I i , 13.

posed to ("fromage" RC("cheese")) RC("french"),

when designing one such system 2, we were great-

RC("cheese as geological

ly helped by the "(ERC)RC" expression for con-

in the processing of a robot who asked what the

notation I. Our system is specialised in multi -

moon is made of. As Barthes then remarks,

con-

faceted descriptions of characters

notations may be based on a set of i n i t i a l

ex-

ever, in the most d i f f i c u l t

but also s i t u a t i o n - s p e c i f i c

(not, how-

problems of beliefs

ones, as

stuff")

("cheese"

RC("bizarre"),

pressions rather than on a single one, which is

about beliefs 4 and 8): here we present another aspect of the story character processing, name-

expressed by (EiRCl, E2RC2 . . . . . EnRCn) R C This seems to be an essential feature of conno-

ly the recursive EXPLANATION of inconsistencies

t a t i o n s , since i t allows emphasis on 'connoted'

appearing in the description of a character.

contents by means of an accumulation of

signi-

f i e r s . Another feature, elegantly i l l u s t r a t e d We give a very schematic system overview,

then

in 'l'envers des signes ' i 0

by

the

two steps

some details about the CONNOTATIONrules and an

( ( " v o i l e " RC("navire"))RC("po~sie"))RC("rh~to-

example of t h e i r application to a story.

r i q u e " ) , is t h e i r r e c u r s i v i t y . As a beginning we made an attempt to

express

this in AI terms by writing a program, BAQUIL, which finds the connotations with structure(E I

Introduction

RCi, E2RC2)RC in a recursive way. The i n t e r e s t of such connotations can be shown by15:

The allusiveness of human languages, in addition to being quite convenient in social l i f e , fies the use of variable amounts of 1 i g e n c e

.a doctor asks:"how is he feeling ?"

justi-

.the nurse answers:"he is groaning."

i n t e 1 -

where the nurse means, by connotation,that the

in processing a sentencer accor-

ding to the number of reasoning steps

patient is suffering:

leading

("groaning" RC("groaning")

to a position where a s a t i s f a c t o r y reaction be,14 comes possible, l i k e "to redefine 'substance

RC("suffering)),

when reading Spinoza". Let us represent a f i r s t

definitive

step by

be ( " f e e l i n g " R C ( " f e e l i n g " ) ,

"ERC" i . e .

but our understanding is direc-

ted by the previous interrogation,

"an expression is related

so that

the

result we expect from our system w i l l "groaning"RC("groa-

to a content ''1, which Barthes calls the denota-

ning") )RC("suffering").

tion: the second reasoning step w i l l

reached by consulting a semantic network and ob-

be repre-

This result shall

be

sented in the connotation formula: " (ERC) RC".

serving that 'groaning' is not exactly a case of

I t does not determine exactly where the reaso-

'feeling',

ning step leads to: i f we write E's content as

nurse means he is suffering"

C(E), we may have quite general

tation and an inference, and seemingly a

connotations

l i k e ("cheese"RC("cheese"))RC("english")

but an expression of i t . is

Thus both

" the

a connouseful

one: compare with Charniak's 6 "demon-demon

as op-

228

in-

teraction,

'want to buy candies' and 'shake

can run without an elaborate model 5 of persona-

piggy-bank' together t r i g g e r 'need money'; but

lities,

BAQUIL has no such extended world-knowledge.

about foxes and sparrows, which would perhaps

is specialised in DISCOURSE ATTITUDES l i k e rors',

'lies',

'jokes' etc. This also means

It 'er-

fail

it

and so deal with fables and folktales

to have goals 'common to most people', al-

though they are meant as human in a way.

System overview

Before explaining how i t works, we give an idea of BAQUIL's construction ( i . e .

i t s hierarchy).

HAlt

/

14, ~ , .

j,~

,,

,.

,,

".,

/ INPUT

o u'rP~JT

/

¢~/Al~Acr~l~,V'-"

\ ~

~-scmrr'lo~ \

--229--

ou~vr

Schema.tic explanation o f how the system runs

consistency. The metarules are represented in BAQUIL by i n s t r u c t i o n s : Mi in the NOTATION proce-

i) the dictionary input procedure builds, from a

dure, M2 and M3 in the CONNOTATIONprocedure.

f i l e whose structure is shown in the ' sample session ' , a classical semantic network 7 ' 9 " ' '

The l a t t e r also contain the instruction corresponding to the inference rules, while the

com-

parison rules form the comparison procedure.

2) the story s p e c i a l i s t submits sentences to a parser 12 and the resulting case structures

Discussion of the metarules

to the story character s p e c i a l i s t , whose actions include the c o m p a r i s o n s i n f e r e n c e s

and M1expresses the hypothesis that many interes-

detailed here.

ting antinomies can be detected during the 3) output procedures express the inferences English and, on request, detail the

pairwise matching of predicates concerning

in

one story character."

repre-

sentation of each character in the story. M2 aims to expres a more or less syntactic finding about the description of a character (co-presence of two antinomic ' n o t a t i o n s ' )

Metarules of BAQUIL

in terms of the 'notations' ( t h i s page and the following three w i l l what rules are applied by the story specialist:

themselves.

develop M3 means "use the lexical taxonomy when trying

character

to recognize a s i t u a t i o n " ;

this includes METARULES, RULES OF

as a result i t in-

troduce a d i s t i n c t i o n between natural languages,

COMPARISON and INFERENCE RULES.)

for the subcategories of ( f o r instance) "inconMI BAQUIL starts a connotation or inference only when a comparison rule has been applied

are not the same in d i f f e r e n t dic-

tionariesl

Consider French and English, "error"

and "mistake" vs. "erreur" and "m@rise", or

to a pair of predicates which are related to the set of descriptors of one character: the predicates are versions of the semantic case structure in terms of the current

sistency"

worse: the two cases of "to t e l l

a l i e " in Rus-

sian, i . e .

Still

" v r a t ' " vs. " l ' g a t ' " .

perhaps acceptable to

character

i t is

allow for important

pragmatic divergences between languages (and,

description, and are called 'NOTATIONS'

indeed, dialects or sociolects.) Moreover, we did not represent the vocabulary of other lan-

M2 Except when specified otherwise (inference

guages than French and English in our system,

rules R6, R7), the inference is expressed

so we lack precisions about how "whorfian" i t

by a 'notation' whose verb is a subcatego-

would turn out.

ry of either CHANGE or INCONSISTENCY. M3 Those subcategories are examined in the dictionary order and the f i r s t

Although a similar discussion of the comparison

one which

and inference rules would be necessary, we w i l l

permits the application of an inference

simply present them here along with some exam-

rule is selected.

ples. The examples are taken from a set of 20 stories (4 to 200 sentences)which were dealt

To sum up: (M1) comparison and then inference

with by the system at Essex in 1979-1980.

about (M3) subcat(gories of (M2) change or in-

--230--

Compari son rules

Exagpl e s.

Their object is to t e l l whether an inference must be started, or not. C1 the predicates d i f f e r only by a negation in

C1 Confucius is handsome, Confucius is not hand-

one of them ( i . e . same environment, verb etc.

some (in which case 'handsome' need not be

and none was i n f e r r e d . )

a p r i o r i present in the l e x i c o n . ) .

C2 s i m i l a r to C1 but there is a hyponymy be-

C2 Confucius is h o r r i b l e , Confucius is not ugly.

tween the verbs.

C3 lexical exclusion between the verbs of two

C3 Confucius is r i c h , Confucius is broke.

affirmative predicates. C4 transgression of l e x i c a l i n t e r d i c t i o n , or l e x i c a l necessity ignored.

C4 Confucius is human and f l i e s away. (interdiction) Confucius f l i e s away, he does not e x i s t . (necessity)

C5 C1, C2 or C3 applies and the f i r s t

predicate

expresses an inference.

C6 a predicate confirms an inference.

C5 the systeminferred Confucius is l y i n g , the story reveals he is joking ( l i k e C3).

C6 in the s i t u a t i o n above, Lao-tsu says that Confucius is lying BEFORE the story disconfirms i t .

C7 a predicate confirm a discarded inference.

C7 Lao-tsu's remark comes AFTER the revision

( t h i s occurs after a C5 s i t u a t i o n led to

of the inference.

the application of the relevant inference rule. )

The lexicon element has following slots in

The predicate or ' n o t a t i o n ' structure, which

its

permits the comparison, is

structure:

( l i s t of subcategories, are-they-mutually-ex-

( a f f i r m a t . - o r - n e g . , case frame, char. descr.) and the connotation has the same structure aug-

c l u s i v e - o r - n o t , supercategory, property l i s t )

mented by reference to premises ( 2, 3 or a l i s t

and the property l i s t

contain references to

other lexical elements.

of pairs i f an inference has been confirmed.)

--231-

Connotation rules: exam~es

tween levels of discourse independently of which character the inference is about, R3 and

As the rules by themselves do not suggest the

R5 make use of some knowledge associated e i t h e r ,

s i t u a t i o n s which make them useful, l e t ' s

as in the examples, with the name of the charac-

have

t e r , or with previous sentences about i t ,

some examples f i r s t .

e.g.

'Krylov is an author' could be part of the beginning of the story. RI . . . t h e servant said: " t h a t cow is not going to eat you." ( . . . )

The next morning,

she

sees one of them is missing and says: "Oh my God ! the grey cow ate one of these men".

R3 and R5, which connect the inference with a

.INFERENCE ABOUT THE SERVANT'S VIEW OF THE

previously observed d e t a i l , are " c a u s a l i t y "

COW.

rules:

in addition to references to the two

premises whose comparison started i t ,

the in-

ference has one reference to the predicate R2 (same example as

RI)

that j u s t i f i e d verb l i k e

.ACCORDING TO THE SERVANT, THE COW UNDERWENT

the choice of a more precise

'fiction'

or ' j o k e ' .

A CHANGE. A difficulty

(which we p r o v i s i o n n a l l y avoided)

is that some l e x i c a l

R3 A teacher quoting Krylov said that God sent

connections could repre-

sent 'necessary t r u t h s '

a cheese to a raven; a c h i l d objects that

that are j u s t sometimes

there is no God.

true, as "teachers are sometimes i g o r a n t " ; what

.INFERENCE ABOUT AN INCONSISTENCY; USING THE

R5 should do with them is not clear. However, i f

LEXICAL CONNEXION BETWEEN THE TOPIC 'God' AND

one considers f o l k t a l e s and fables, the d i c t i o n 2

'religious

nary connection one uses are almost always

discourse', BAQUIL SELECTS THAT

KIND OF INCONSISTENCY. "foxes are s l y , f u l l R4 the peasant believes that the tree w i l l h i t by other rabbits.. But i t

stop"

be so the question does not arise.

is not.

.INFERENCE: ERROROF THE PEASANT.

A difficulty

R5 (continuation of example f o r R3)

f o r R3 is that a ' t o p i c '

could be

the teacher answers that there is no cheese

connected both to ' r e l i g i o u s discourse' and

either.

' j o k e ' , so in the present state of the system

.INFERENCE ABOUT AN INCONSISTENCY; USING THE

one of the connections would be systematically

LEXICAL CONNECTION BETWEEN 'Krylov' AND 'au-

ignored: but f o r the time being, there are very

t h o r ' , AND ONE FROM 'author' TO ' f i c t i o n ' ,

few l e x i c a l

BAQUIL SELECTS THE LATTER SUBCATEGORYOF

we cannot decide which necessary truths r e a l l y

connections, the reason being that

belong in a lexicon.

INCONSISTENCY.

While the other rules e x p l o i t comparisons be232

-

BAQUIL's connotation rules

'references' are the two i n i t i a l inference;

Rules R6 and R7 are i l l u s t r a t e d by the story of the vegetarian wolf: given a d i c t i o n a r y

'hypotheses come from the l i s t

subcategories of the i n i t i a l

which connects ' w o l f '

premises of the

verb, i . e .

to ' l i e ' ,

of

hence

i n f e r s from the be-

ginning of the story, where the wolf is con-

the

t r a d i c t e d by someone, that i t

d i f f e r e n t cases of 'change' or 'inconsistency' as represented

to 'bad a c t i o n ' ,

the system f i r s t

the s t o r y - t e l l i n g

in the lexicon.

telling

explicitly

lies;

later,

warns the reader about i t , that i t does not. The in-

ference is NOT erased, but earmarked as f a l l a cious; so R7 applies when a character says:

R1 i f both references are about the same character descriptor,

this wolf is l y i n g ; but, i f the story had

the inference is also about

that descriptor.

not denied the p o s s i b i l i t y

( t h i s allows f o r s t o r i e s

of a l i e ,

R6 would

have applied. One can see that R6, R7 and R8

inside the story).

are not quite s a t i s f y i n g .

I suspect them of

being in need of some refinement. The set of situations

R2 i f the i n f e r e n c e ' s verb is not yet specified

or e t c . '

and the inference is about the d e s c r i p t o r of

' c l e v e r character or lucky guess

is not c l e a r l y defined, and I do not

know what to do with the L i a r ' s Paradox.

the second reference, BAQUIL t r i e s the cases of 'change'. R3 i f the verb of a hypothesis is connected to

On the contrary, R3 and R5 give less trouble.

the name of the character in a reference by

This is shown by the 'sample session' next

a 'be about' l i n k ,

page, representing approximately 1.8 second

the hypothesis is selec-

ted.

CPU (using about 25K core) on a PDP-iO.

R4 i f one reference comes from a ' b e l i e f ' the other from the ' s t o r y - t e l l i n g ' ,

and

the verb

of the inference should be ' e r r o r ' .

R5 i f

the verb of a hypothesis is connected to

i t s character's name by a 'tendency' l i n k ,

RULES APPLIED: C4 " l e x i c a l

necessity"

the hypothesis is selected. R3 " t o p i c " R5 "tendency"

R6 i f a character's discourse matches an i n f e rence, BAQUIL checks whether the character is clever or has made a lucky guess etc.

R71if the s t o r y - t e l l i n g

C3 " l e x i c a l exclusion"

contradicts an i n f e r e n -

and~ce, the inference is ' d i s c a r d e d ' ; i f the R8Ydiscarded inference is matched by a discourse, one looks f o r an ' e r r o r '

or ' t r i c k '

etc. -

233--

SamPle session

(STORY INPUT)

(DICTIONARY INPUT)

Food was short. A teacher quoted Krylov.

WORDS

He said that Krylov wrote:

FOOD ISA SUBSTANCE

'God sent a cheese to the raven.'

SHORT ISA STATE

A child said:

TEACHER ISA MAN

'There is no God ! '

QUOTE ISA DISCOURSE

but the teacher replied

AUTHOR ISA MAN

'there is no cheese,

KRYLOV ISA AUTHOR

either . . . '

WRITE ISA DISCOURSE

THE END

SAY ISA DISCOURSE GOD ISA PUTATIVE BEING CHEESE ISA FOOD

(INFERENCES)

RAVEN ISA BIRD CHILD ISA MAN PRODUCE INCONSISTENCY ISA DISCOURSE

(God sent a cheese 1(there is no God

(comment: abbreviated PRODINC)

C4 in description of God and R3/God:

LIE ISA PRODINC

"the child means Krylov has a r e l i g i o u s

MISTAKE ISA PRODINC

discourse".

JOKE ISA PRODINC ILLUSION ISA PRODINC

(Krylov has a r e l i g i o u s discourse

RELIGIOUS DISCOURSE ISA PRODINC

2(the teacher quotes Krylov

FICTION ISA PRODINC

C4 in description of Krylov and default:

BE ABOUT ISA DISCOURSE

"the child means the teacher has an inconsis-

MEAN ISA DISCOURSE

tent discourse".

IGNORANT ISA STATE END WORDS (God sent a cheese 3(there is no cheese C4 in description of cheese and R5/Krylov: "the teacher means Krylov's discourse is f i c LINKS

tion".

USUAL(AUTHOR, FICTION) USUAL(CHILD, IGNORANT) USUAL(IGNORANT, MISTAKE) BE ABOUT(RELIGIOUS DISCOURSE, GOD)

(Krylov has a r e l i g i o u s discourse 4 (Krylov's discourse is f i c t i o n

FORBIDE(RELIGIOUS DISCOURSE, QUOTE)

C3 in description of Krylov and R5/child:

comment: that is to account for the location

"the teacher means the child is mistaken"

being Soviet Russia. NECESSITY(ACTION, EXIST)

END OF INFERENCES

NECESSITY(ACTION, AVAILABLE)

END OF SESSION

END LINKS -234-

Conclusion

of a nationality adjective suggesting a character t r a i t (Cretan for l i a r , e t c . ) , which is

While not achieving much by i t s e l f , Baquil is an important component of the larger system 2

helpful in many cases: when the nationality

currently b u i l t at Paris 6, and could probably

guage (Chines texts about the ancient kingdoms), but also when there are several possible

also be integrated in a large AI MT system as an expert of indirect descriptions of characters, for instance i t could recognize the use

References (1) BARTHES: Elements of semiology, Hill and

adjective is not familiar in the target lan-

interpretations for the nationality adjective in terms of personality ( 'qua' ambiguity).

(i0~ GENETTE: Figures I , Paris, 1966. (11) HAYES: The logic of frames, in METZING, edo

Wang, New York. (2)

BERTHELIN and SABAH: Le traitement des personnages, congr~s AFCET-IRIA, Toulouse

Frames, conceptions and text understanding, De Gruyter, 1980. (12) SABAH: A contribution to story understan-

1979. (3)

BIEN: Toward a multiple environments model

ding, thesis at Universit~ Paris 6, 1978. (13) SCHANKand ABELSON: Scripts, plans, goals

of NL, IJCAI-4. (4) BIEN and WILKS: Speech acts and multiple

and understanding, LEA, 1977. (14) WILKS: Grammar, meaning and the computer

environments, IJCAI-6. (5) CARBONNELL: Computer model of human per-

analysis of language, RKP, 1972 (15) WITTGENSTEIN: Philosophical Investigations,

sonality t r a i t s , IJCAI-6. (6) CHARNIAK: Toward a model of children story understanding, thesis at MIT, 1972.

Blackwell, 1953. IJCAI-4 and IJCAI-6 refer to the Proceedings of the international Joint Conferences on A r t i f i cial Intelligence.

(7) CHOURAQUI: Sur un r~seau s~mantique a c t i f , congr~s AFCET-IRIA, Toulouse 1979. (8) CLARK and MARSHALL: Mutual knowledge and definite reference, mimeo, Stanford Univ.

The story in the sample session is adapted from "The Big Red Joke Book" by Benton and Loomes, Pluto Press, London, 1979.

(9) FAHLMAN: Representing everyday knowledge, MIT, 1977. --235--