UrW�r:;llIndir.g

LANGUAGE AND PROBLEMS OF KNOWLEDGE:

BRIDGING NATURAL AND ARTIFICIAL INTELLIGENCE

RosliOmar

Jabalan Kejuruteraan EJektrik Universiti Malaya

Introduction

In society today, computers are very much part of our lives. They seem to be capable of doing so many "clever" things that we some­

times think they are intelligent. In certain popular books, authors talk about computers which will soon be more intelligent than man. They are actually dumb machines. Any perceived intelligence is due to their programmer A programmer in programming a computer to do a cer­

tain task will lay out all the steps necessary to perform the task. These sleps constitute an algorithm. All the possible situations that can occur will be thought out by the programmer so that when the computer is faced with any one of them, the necessary step to take has already been specified.

This method of specifying a task is possible where there are clearly defined goals, constraints, and where all possible situations can be known in advance. But in the real world, the world we live in, it is not possible to determine beforehand all possibilities that can occur Let us assume that we wish to programme a robot, with a computer as its brain, to go to market. A tree may fall on its path; the road may be closed and a detour provided, a robber may snatch away its basket; the market it usually goes to may be closed. The possiblities are endless. It is just not possible to think ahead about all the foreseeable circum­

stances. Thus the robot must have intelligence to overcome these unthought-of situations. This is because the essence of intelligence according to Winograd (1987: 98), is the ability "to act appropriately when there is no simple pre-definition of the problem or the space of states in which to search for a solution".

The discipline which is involved in trying to make computers more intelligent is called Artificial Intelligence, popularly known as AI.

According to Margaret Boden of Sussex University, "the least tenden­

tious definition is Marvin Minsky's, 'Artificial intelligence is the science of making machines do things that would require inteligence if done by men'" (Boden, 1987 4). Some aspects of intelligence studied by AI are vision (how we understand what we see), reasoning, knowl­

edge representation (how knowledge is stored and retrieved) and natu­

ral language processing and understanding. The term "natural" lan­

guage is used here to distinguish it from the formal languages of computer science such as Fortran or Cobol. Al is multi-disciplinary is character and involves researches from computer sc.ience, matbema-

tir.:,., lIml Ingie. p"y(;hology and philosophy {)f mind, neum�cienc(; and hnllubt.ic� amung others. l!ndcrl>landing intelligence is ^fl(l simple mat·

tc^r Henc;,:, the m;eu fur coop(;ralion from various dicip!ine�. Since human intelligence i� the best intelligence (best in the manner dellncd

<lbc)Yc for intcllit;cnce). AI rescarcher� �eek: 10 um.ler.<i.land this and to simulllte it in computers.

There arc two approaches to Al ac.:ording to Edward Feigenhaum (19R5) of Stanford IJnive'�lfY Thc firM approach is Ihrough the �tudy of the human mind. how human intelligence is 3(;hiewd. From the insight gamed, It h ho�d lhat a model of intelligence might he simulaled in the compulcr

Thi� approach relies hcavily on the ps)'chology anti phllosuphy of minu and geoerally conMIIUles the c(lgnlfl\l� SCIt'I1C� approoch 10 AI The se<:ond is what Feigcnhaum calls the tnginuring approach. Here AI lne� to produce programs that can solve prohil!ms intclligeutly. It does not mailer if the way tbt':� programs solve (hem is nOI Ihe way we would solve them. In thc�e two approache� we can see (hat the computer is jU�1 a tool to implement our ideas.

What are the reas()n.� fur making cumputers understand language?

Por (me. it would be ca.�ler In communicate with them using natural language ratller than formal languages. This would bring about eco­

unffilC benefit� ^-a mlljor morivati(ul in it.o;ell" The <'lbililY to communi­

calC with <:omputers u!>ing language would eliminate une sense of the

"alienatil)n 01 man fwm technology". cspedally high technology which one tines nllt under�tand. Another reason o f greater interest is that.

language is at the root of intelligence. Whereas olher attributes of intelligence. �ueh as vi.�i()n, arc f!re�ent in other creatuTes. the ability ttl speak and understand language i.� the sole prcrogaUvc of human inteJ1\gence. the !lbility of certain primale� 10 understand a very lim­

ited numher of wnnls n(ltwith�tanding

It i� nur ahillty 10 u.�e language tbal has given us our pre�enl civilisatinn. Acwrding to King (1990). thc genl!tic differences between man 11l1U the (.:himpanzee, the I\(.:arest Ii) man i n terms of intdligeuce, i � ks.' than ()n(.: percent But the difference in tcrms o f the civilisation produced by the two is ^Sil vast. Tbis is due t o our language capability The abIlity to pa�� knowledge fwm one persnn 10 another anti from (ltle generation to the next through written texts enormously speeds up clvili�;\lion [I] It would he I1\h:resting to understand this unique Jan·

guage feature Ilf ours by �imulaling it in the computer Of course this j� (lllly onc way of trying tn um!er.o;tand the language phenomenon.

Another way originates Irom the humanistic school. "understanding hnmanity i.\ hest done hy �Iudying humans" This is, of course, a very valid approach The ta.�lc of under.,tanding the language phenomenol1 is srI very dllTicu!l thal !'Ill)' help from any quarter ought 10 be welcomed.

I_,� IIftd Pn>NUIU of K"",,�<:<1� "

LanJluage is at the roOI or intclligt.nce. But to undetstnnd language.

lOtclligence is required. Is thi� a chIcken IIntl egg :situation? As will be elaboratetl, intelligence requires knowledge: oC the worltl we live in Since: language cannot be separated rrom il1lclligence, some general II!peclS or intelligence will have 10 be tliscussed in un.kr 10 underSUllltI aur language c:lpabililY

Natuflll Language J·roces.. ^.. lng.nd

UrW�r:;llIndir.g

In the computer. the lirst stage 10 under�landing language Shuts With Slfoctural analysb. The slfUClure or II sentence is produced through

!he pars 109 process, which result5 In a parse tree. Parsing follows the

!rammar or ^It given language in building the parse tree. Grammar can he coosidered II.'; knowledge ahout the language structure Ihnt needs 10 be known berore any :Htcmpl LS ffilU.Ic 10 UI1(Jersllmtl a particular Inn·

guagc. Thus in English, n sentence S is made or a "noun phrase" NP anti It Mverb pbrasc" VI'

The NP and the VP can be broken tlown rurther inlU their respective components. This clln be wnllcn a�.

GRAMMAR

S -> NI} VI'

NP -> proper-noun

NP ^-> pronoun

NP ^-> detenniner NP2

NP ^-> NP'1

NP2 ^-> noun

NI-'2 -> adjective NP2

NP2 -> NP2 PP

rr -> preposition NP

VI' -> verb

VI' -> verb NP

VI' -> VP PP

t!etenniner adJecllve rrepositioll

(Grammar adapted from (Winograd.

I!lS3])

DICfIONARV

Bill. Mary She, he

tose�

gave, �aw

the that red, beauliful to. with

Tbe diclJonary of Ihe compute, stores aU the wOrd); needed :IS Ib

vocabulary Wurds not in the tliclionary. needltu tu say, will not be untlerstuod by the computer For c.lI3ffiple. parsing thc sentence

Mile gave Lbc beautiful red roses to Mary" will result in the parse

Jwmal B";"""Modm

tree .�hown in figure I.

s NP

/�

/\

pronoun

/

He

I

IIcrb _NP

gave determiner

/\

NP2 the adjt:Clive

/\ /

^P2

^\

beautiful adjectille NP2 reu NP2

/\

PP

noun

/\

''''''''''/",

^roses

N?

'" N?"

noun

Mary

Figure I. The pilfse tree for the nntenee ^. ... "He gave the heautiful red .. o�e� to l\111ry"

Bul syntuclic analysis by i!!iClf is 001 enough 10 determine the

\nCowing of a sentence. Consider tbe sentence: I read an article about Ihe Vie/n.am War in the n.ewJpuper

Here, syntaC!lC analysis ha� two chuices:

I) ··lhe Vietnam War in the newspaper" - war occur� in the new�pa­

""

/..nHxw"X" and 1'",hI�rru; '" ���d� "

2] the prepositional phrase "in lh� newspaper" is m!lditier for "read"

I.e. "rc^ad ill the newspaper" [f'iJarniak. 19R51

(lnly scmamics can detCfmine Ih;ll {he �ec()nd nll::anm� is the correct I'n� One Or Ihe classic AI progrit1lIS to demullstratc langllage U^ll⁽lcr­

$l:lllllillg tblOugh th� use of semantics (and lliso that Illngllage under­

$l:lll.ling requires knowledge of Ihe topic heing di�cu�scd. [Ipan from

�r:lIIl1^Jlar knowlt:dge) is Winngrau's SHRDLU, a fragment or whl�h is ,hOII'll in tigur� 2 lWin{1grad, 1972]

Figure 1

J. Pick up ^i! big red olock (Winograd's input)

OK. (Co^mputer rcspunse)

The srsl�m ,IIlSWCJS "OK" when it carnes QUI a comm",lt!.

In order to pick: up th^e red hll'x:k, it hlld to clear it off by finding a ."pace lor the green om: and moviⁿg Ihe green (lne away The ahove figure sh()ws the initial

�t;lle 1)1' the seen\!, (Winograd's ellmment.�) 2 Gra�p the pyramid.

I DQ�'T UNDEKSTANI) WlHCIi PYRAMID YOU MEAN There arc three pyralllid� in tlie ^scene^. so the pWt!ram UOC$ nothing ^.

. t lind a hlock whiCh ^IS taller then the one you are holdiu1; alld 1'111 II into the h�Jx..

The �)'�tem ha� t(J figure out what is meant by "(Jlle"

and ';;1." as well all u!,iug deduction l!) decide which

block it is holding

BY "IT:' J ASStJ�1E Yon MEAN THE BI.OCK WlITCH IS TAI.LER

THAN THE ONE I AM HOLDIl'\G PruJled our til teer us informed. OK.

[lut el'en scm\'nti(:s will he lIu�ure ^III thi\ .�itua\ion;

""I rcau an article aNlut the ^WM III the new�papcr

lIere. ^was 1n.1)' ^meall a Wi\( I'Ictwecn UIC cditHr and lhc rcpllrteI\, in whcih case rhe Wolf i� happeniⁿg ill Iht: ncw�p:lJ",cr

..

Thu�. aeUla! world knowledge i� requIred in order to understand the sentence. It i� the problem of integrating world knowledge in under­

standing language that is the main hurdle for computers in undcr�tand­

ing 11'.ullullge. Why thi� is so will be discussed below L ^.. ngulI/,oe Underslanrlingand Wurkl Kl'tf)wlOOge

AI's apPw8ch to language untlcrslllOtling and the cognitive science appruitch in general is largely based on [he reductionist approach. 'fhe reductjnni�l lhctum �ays that "the meaning of the whole is the sum of the meaning of its parts" Thu� to unde(ijtand a phenomenon, the pbl:onmcnon i$ broken into its conSli!uent parts. The total understand·

ing of all the parts will be the under�tanding of the whole phenom­

enon. The basic assumption in the rcducttollist approach is that there 1�

no cross-Jntcraction between the parts. Thi� approach has been sue­

ees�rully lip plied in the phy�ic;!1 science� where the cross-interaction is minim,,1 (But note its failure in quantum mecb.:mics - one of the most, if not the most, successful tlranch of modcm pbysics) Reductiollism's beilci In Wi dicl Um springs from its helief thaI the world i� ubjcclive:

!lHU knuwkdge is L'Ompo�ed of context-free data, Lc. there is only one way Ilf Interpreting a phenomenon. In the physical world, this aSl<UlJ'Ip­

tion i.� perhaN more valit! than in the social world, the world of language ant! human interaction

1n language understanding, the reductionist dictum means that the meamng of a sentence is merely the sum of the meaning u f all the w{lrd.� in the scntence. Just the oPPllsitc is the holistic approach which says "'1 he whnJt is greater than the Sllm of its parts", i.e. meaning (lr a SCll1\:nl'e 1.� mllfe than jU"t the meaning uf ils parIs. To show thill a lIcnll'''I:C is more than jusl Ihe sum of 11� pans, consider this fragmelll 1)1 �on\'crsation helween two children (Paper! and Minsky's example in Drc),lu)', 1(85):

.Il/flr/: '"Thill isn't ^u Vt'ry good ball )"Qu hallt. GIVi' it 10 � and I'll Ril't yo//.

111)" lollll>op .. ^,

Tu understand this, a jOl of world knowledge and concepb are re{luired:

Tllne, �ra(;e. wort!s, thoLlghb. talking (explaining, ordering, per­

lIulldln1:, pretending). social relations (giVing, buying, bargaining. beg­

glllg. slealing). playing (real <lnll unreal, pretending), eating (bow does nnc �'om[larc the value� (If fut>d with the values of toys?). owning (bdnnjls 10. master 01), livmg (girl, awake, rlay�), m!ention (want, pllll, gllaJ), cmtllion (moods. llispOSilionll). Those items in parentbe�l�

art l'nlleep!S in their own right and need to be further elaborated. The

lIop:mJIS41! )0 S�;'t;)OJtJ

:'Iql pa[JlI:;l 'Io:!

IIJIjM . (4 �IJOp sl 1J0!WIU3\JJIbJ

;JUp:lIMilU'!

SUIl:'lUi SIIO!J!!" JU :III[

J:lmdUlOl ;)ql Ihl!q JSp�I"'UU� JU

:'11:11 uuoJ III'" pUll :'I:U)-J1:1ltlO:l

.(1'3"!HllaJ P:llu)s:'IJd;u

�! :lSpalt.1llu,

:Iq.L '::11:1 li:llnJ 'saWllJ) ')(JOIIII'311

;)!IUI!UJ;I� ':11801 ICWJOJ :SUIIOIUI

�nmJl!,\

alII :u:nu, LJ:llnl1lUOJ ;1111

U! palUa�;wJ;u

;)W;11I1\OU)( ptJONi

"'1111 snJ.\;\J(] SI JU .(1:l1(J;I(I 'fiL61) '��()I (88(11

IJ:I\.Inl1 llall! S!1I1 �!

;)1111:) )0 JOft:w

V '(HY61 '"mU:l(]) 3:>ua81IPIU!

01 �ql Jno Il;1P! IS!"IIIlISO:'l 33p<lIIl\()U'l 1I!'1" II! I';)lu'3S:;lJd;u U!llJq alii ;')�!J �;)",3

1l11Ni IlIlmOSIIO:' sl l>!II.L

(\861 '1j1!WS) JOIlIiUlIJO:!

.(q aql u(}!llllnulU�w

;')Sp�III\OU'l )0 Ilns:).!

;nu ;'IQ pallla:'lp 01

lil a:m�81IPIUI ;llU!�

tpJUi

·IU IV II;){lW

Ull:llI ;)ql )0 S! IIUlIIIIU:;l��ld;u 1t:

'33paINiOU)I

·�3lUJIII�

3UIlIUUJ;)WI II!

:'I3p'3INiOU)(

[lpOfll 1/1111\

IlIU:llU! UI q!'fAjIlUIl

J1IIIIIW:n

puc J!I:'IJIU,(S '3lljIJU

II!II\ OI S!II,L .I:'1lndulO:)

'3111

" III

1 3 11 I! 11 U

"' 1

apllUl PUI

POIIII:ls:;lJd;u

;)11 ,,,nw ',)Ju;,3m:'IlI11 llU: I!

'IIIa3 JOj pUP.

;,lh10811111 3U!Plllllf

'Japun II!

,(Jussa:>:,m :lru IlIJOII\ aql JO :tSpa[lI\uu't 10 s;1d};1

illO(j ::I) II!S

(�861 'pcJ8nUlA\) slhl!1fI

op NiOIl JO OJ :'ISpal""ou'!

::'Iip:lINinU1(

:lAli

'lIJUp:'lp 1 0 �:xJ};1

�nO!Je" }O SS;')UIl:lIUIOIJ-J;)IU! aql

uo ilSpOIIII\OU'I ',,11\011

ilU!l\'\(lU:l[" S!

ilS[l:'lllllOLl)(

I'lJnpJ;)(lJu sll;u�111I\

":)I�

Jn01 S!

011\1 [lUU

IIMI

':'IJJqds

�I PPONi 11

a,,1 1l)'111

\pm: sa PIJtlM ;)IP

III f):'IU}

'lIIlqll ;)SI);)[MnU�

•

"IUlIl 3U!NiOU!1 .Ill 1t!,,11\

Il'U!NiOU1(,. 'Io:!

;13p;)IJII

;M!IUJUIZ);)(I OW[

;)jfl�

'1II\OlJ!1 IIIJnP:'IXlJU

PUl:

:'''IIIIJUj:I:'IP lill

1l;)!J!�SIlP '3Ij UC:>

JSflJ1NiOIl)j

-'!�u, ';1[II!'Io:\mlw!

IOU J! ''In:'JUllp ,(ja" e

Il!II\ 1I0!ll!I:lJdJ:lIUI dq j3;!lJo:)

;)1/1 puU 111 SUIAJI

'SUtI!l1!nI!q ..

n"!JlIA aliI U;);)MI;')q :)

11I111I�JaJJ!fl IOIIUII:) tp!IINi

JalOlIUJOJ

Il JO:.! 'P!nu 8U!'!U!Jp

};UII Iluam un:)

'J�lI'fJl)r 11

I!

lOa 01 -s:lln,lalnm

Ull:1UJ II!!:) O'U J�1If1l\ ';'IOIJ1

J;)IUNi '(1'311111)'Io:qc

SIU�lw:)q;'l )111 1.1:1:'1'1 01

�'lIt"

oqNi 1"!III;)tp O.L 'SUOlffil:I.JW;))UI u

III:1J:;lJJ!fl :xlll;),;;) IIlUUIl:t ;).'U;)IU�

�[�

'W!S S!lIl U;'i";)

10(1 'IlJ-UN.J01IqS!llJl�

SI J[:;l�'I!

OI:)U�IU;'l� Jill pun SUIUC�W

a"'p3:'1fqll Ull ';1"'111

SW';I:'IS 01

;'i::)U;)JlIa� Sllli

�llJOM U!

alii JO :'lUll lP\13

:1I!1I1

:luI s,p11J80U!M UOlll'lillf (L861) Jluwu;: S! :I[dwux:I :lA!j:'):lJJ3 /.JOI" ^{'Nioqs U!}

01 :1"!IUI';lJ II

'JaqlIlJ wq '.\p"'[I:I:'If411

IUt" 'S!

UU!IIlIOSI U

! pas};[lllle

<Xl IJ;xl� JO SI;)t:J:lIU! Bu[",,, qllNi SJ:I(jltl aSunSUU1 lI:'Jql SU1UlI:'JUi ^10IlUl!.)

,(J;'I,,;'J ';I;)1I

lI;)lIUl :lGJno:ti!p 1U

;)8\1n8UIII :lUll

...

PI.OM Il!pn�

:lql 11

1 II Jaql0 01 111111 'SllpUIIIOJ;)!W ;)41 'S! 1/:>I1OJUUIl IS!:I0!I:')n!l';lJ

lIu!

'q:)1l1J1l 11;141

!lllll uOPlllos1 :lA!llllaJ

Il!

.. ,(elu '

.. JO PPOII\OJJ!w

;)AlXjJ!

alii

':JluUIIIJ::l JtI.:I "PpolIIU!:)lw H q:!ca

lhu/.pIlIS)O t[:'lROJudll

';I�I jO IlllIIL l:m

.(1I1l,:');)US:,) S!

;)11 'IU!If! 'os

s;)OP IOU snJ,(':JJ(I ".,(:;lI:J)(J3Q

[ll :lUO :1811:[

,(IOlJ;JUJ '1011 I:

l! sl

�U!ql ,(:'ISU!�

lJiXllld 48nc,1I[1[1I pm:

'SSOllpU:;l Stu3;)�

I IS!

[Brnok�,

19861 The abstraction process abstracts only those facts that are con,�ldered to be re\evant and pertinent

10

the task at hand. Ab­

slraoing only facts thaI are relevant is necessary bccau� world knowl­

edge is s o vast and probably limitless. Furthennorc, in the beginning,

AI thought

Ibis was cDough ror

understanding. Bul 10 Brooks, an imporlant essence

of

intelligence

is

tbe

ability

t

o

detennine from the myriad

racts of a given situation, wbictl ones

^are

relevant.

Ir h

u

ma

n

s

do

the abstraction for computers, then the latter will never be intelli­

gent

since

our ability to select relevant factS is very much associated with Icarning, which is anmher aspect of intelligence. Thus, to be truly intelligent,

10

learn what is relevanll!.nd what is not, computers need to do it� own abstraction.

Humans

also engage in abstraction.

But

we seloct relevant facts of a particular silUation from a large knowledge base formed from a whole lifetime or living experience in the world (Dreyfus 1979)

From thb

lifeti

me

experience. some of the knowledge forms tacit knowledge

(Wiugenstein, 1953:

Polanyi

1967),

knowledge

tbat

is

^so deep that

we cannllt e\'en fannulate il into woftls. As Polanyi

puts

i t

"We

can know

more

than we can tell" This lifetime experience forms ollr broad knowledge base, descrihed as an outer·horizon by the philusopher Husserl. a pte-understanding hy Winograd, from whieh a background for understanding language is

made

possible. Just as explicit knowl­

edge is necessary to understand language, so is tacit knowledge.

Tacit knowledge at the syntactic level, is shown by the fact that we cannot completely explain or provide rules on how we structure a sentence, eitber for understanding a given

sentence

or ullering. As to the

erfect

of tacit knowledge on ordinary day-to-day situations, con­

sider

the

classic

example:

the concept

of "bactlelor"

(adapted from Dieyfus, 1985).

Say iliat a

computer

is

provided

with

the definition

"an adult human

male

wtlo has never

married" ^-

which seems reasonable enough. But in our everyday use of the word, the above definition is clearly not enough. Note this converl\B.tion:

HO$t to computer: "I'm having a patty next weekend. Do you know any nice bachelors I could invite."

Suppose the compllter's answer is:

I) Arthur.

(He has lived witb Alice for several years now and has a Child.) 2) Charlie.

(Charlie is 17, lives al home and is still at school.)

'PI10""' JlIl ul :I·\H

:lit\.

1l?1I1 p�J <l1I1 JO :mU!h hq

S S ;:O

U

� ;J

}nO q8nuJIJJ sP�J

:'J�J41 �JJlnb:n�

:lM '(j)u:llf!((JJU!

SJ�lnUUnY.'l ;))jCUI OISU!hJI

U:all ....

(PM) li C

.{J�Ss":'J;m AIJ)nlosqll s�'lIn

[Ill W;)1j1 Inoqe III :IIU!tll

loti Up

J ill

leql

� nO h ! o q OS Wql :w:

Sl:>IlJ 'J::))ntlwo."

Jl[1 II! JSJIJI U�

Ill<lSJJd 0) ;}h\

�le

",,('IJ[

'�lJ p

<l lls nd Inq IOU

pal1nu ;Ii.{

U1:J 2!UPIS 'awn auo IU sa.)!:!d OMl

?q 1t .) ue 8U1111(1O

'JUI!}

10 ,e,\J<llU!

PHoS JOJ Il J'\!l slcwllle

:8U!!tl.OW1J

JlIl Jpnpu!

:'!lip<lIMOIl)j lpns JO l :'!

U W1 t n:

l

�quuq :nu:;,u;J;J

p:Jpnpu! UJ

Jq 01 '

! sno h u q UUI ',IlJ!SUJSUOUlUl\).)

UUI S!

Illljl J!P;I ...

UU}j 51 SPl.L '�1J!1

-uJP�·;lJd S ! P p OM Jql U! MOlD(

0\ p;:>u Jill WI!III .IU )SOW

Iuq) hl?S

([l'� :ISPJI;I\UU� JScq lll�Ulnq 3u!uu�dS " o u :)S u s n s IIIUU":t .. J81�I ^'1!lInf» :0661

;JSUJS UOUlWO;,) JSJI!!

e�

I P

! q 0) u XUlhn

<lJe

OqM IClla, pUl:

cqn9

(00£

:��6I) ,.J8pJ,.'tI.{lUl[

IIlUOrt!sodOJU Jq 00 UB:! Jl;"lql :lSp<l\

...

OU:II llJIl) Inoql!M -

aS p:l ... mq l JI I!

I I

jO U1nn�.rlsqos Il

ut) 'JpnpuOJ

U�J a M 'SISJ1 J?rP'dIMOIl'l

\IlU0!1ISOOI.l1J ..

:JilpJIII'IOlDI IPI)) pUll

;l?rPJIII'IUIl'l [110tlnS!1lhHd SIlE�

"II 1"111M '?f!lm�U1q

UJ:Jh'lI;)(j dlq�l/')nel;lJ ;llJ)

�CltjS!lqBISJ h[l:)IIl;}�nS

UJSSJUUl:!q(l[

'SUOlll�IlI!�

Ilt?J:JJJ!P

� U!

SJ OJS O

p l

JO E

;Y.)u:J1UJS ld e '

<l;}

u J O

'[HOM v.

PUl?ISJJIHln \l

0) sn

? S

q l C

? U

lel[1 Clh EII JM JIIIII a8palill

01Dj lPCI �111

I) Sl

i.JOI:l1pcq

IOU P.

;:III 1>1 'JIIO flllql

Jill/I.[ E

01 SmdUll S

! pU!:

0111 seq 1 MOil J�I i.5"JMtt.

mOJ

"' .... �II U 0)

:f,p O!

sllJ hq J PJMUlll!

olJld. S!

'!IV

�'I!I :'IUOJUIOS

Iuolle

lllll/t\

i, pJ S M U 01 I!

SS:lJ!.h:;l hIlPTtU);:J

01 pUll - sU0!lRnl!� ?Nl�SUJ

Jql

lie )U U :':

!ll 01 l :1M JJU MOll lua

·�1l0pp.nl1S J

! I 4 S S

1ItI od JO pe'dqe )[lllljl

U) P:J:JU :1M 'unpJl:Jlsqe

J J w lnd o 10.1 !O8 :) 'JEpJIh'lOUlj

l1'el tplq/t\.I0 S!

IJHtI e ':lJU?pJd'll:J OIIU!)JJ!I)O

;seq S ;l J p I OU It\.

:':

;lEJIII 01111 W01) p;lIJ\)'J1>llt

:1lP.

�JI!J

;lSCllIl U p V 'OO!ICnl!S

tpta 01 lfu!pJo:J:J'e

SI:"l�J

\Uu.�;lIClJ 011[1

?U!UU?PIl II' ...

mS!UI:IjJOIul uU!l:)\lJ1sql!

JOO 'su1!mnq JOrl 'un!l!u!PP

aliI

P:lP!hoJd JO) Jq

01 seq u0!1r:J()qt:[� JClquoJ

'SU0!1IWI!1>

CI.�oqU lljl In 'peOl

ul ,.?rU!u1!OIUll!

J '

.�

!E I�J!J 11:111

H1aW:JlllIii q:mJ P I/I a\.! 111 li

() U ls ! h ;m

�";4 V

SUll:lUI lJUrepEO

1I:4M

�! S!lI.I.

,,(.MOU'I 01 lUHM nOh op hlJM"

01

�lJItl.Su\l

jHII U

;llO d

"'ql liU!p!.\{udlnUI!1!M

p"'J:JM�UIl .Jq WUUE:)

"i,JOI:'Jq,wq '\: X

�l .. ld;1JUUJ J41 .. JO q m ?t[, JUI " SV 'peJ8'oU!M (L861) SInd 'I! 1J011�'dnb 'Jt/l.

;lJllldl?J 01

� t/

( n ) u IOU �""! :'J

UU!l!ll!.PP :'J41 Illlll

?JS ue:J :1M .)&'\1' I

£ p

III J (JJUOWWOJ o�J

(",l:nxasoUloll S! Il

H:rJ ('S�",U!snq IR)�'S:n.l[1S

w.� £ qllM puu 'UMO SN

uo 'ifm:h!i ...

S! JH TI ou 1'1:

I<"II )'PS I!nb pm!

S!:'JIO LI

"

AS�lll, some of the prescientific knowledge i� so deep that it forms tacit knowleJge.

One lisped ... hich is �aid

til characterise

human

intelligence is

our imuifion, the

ability

to perceive �omething usually as being "not­

qnile'right" or "I can't (,juile put my finger on il. hut I think this is

right' without being ahle

to

fully explain

why i t i

s

so.

Think

again

of

our

c(lJlcepl

"hac

h

e

l

or" Here I'm nOlltlinking

of

my

s

tical or religious

inluili(JII.�.

It

is my contention (intuition?)

that intu

i

t

ion arises from tacit knowledge. This knowledge is too deep in the region or the suhcnllsciou� fllr

os to he

nble to express it in words. We just know

it.

linw is intuition to work in computers'! For Ihis,

cou'pulers tldlnilcly w

il

l

Dced tacit knowledge.

Dut computer!> by virtue

of

having explicit

knowledge,

through hUrna!1

ah�traclion, will not hc able

to h

av

e

thi�.

Even if it i� possible (never mind the paradox of tacit knowlcdge in

computcrs for

a while,) how

i� thi� [0 he brought

to

the s

u

rf

a

ce? Thi�

b

hecause any question (If

bringmg knowledge 10 LIle surface

for

a c

mnp

ul

er requires that Ib�' kn

�)w

led

gc be e

xp

li

c

il

l

y

known'! How

is it th'lt man

can knuw yet he unable 10 exptess

what he

knows

^or how

is it

Ihnt he knows hut is nol consdou� that he knows,! How ducs the

sllhcon�cious relate

w

[he conscious'! These are �ome of the questions whl"e answus

are pcninelll

if computers arc

to achieve

intelhgem.:e.

Tacit knowledge rcsull.� rrom experience gained hy living in the w(,rl<1. Living in the world enables !el1ming to di�tillguish between w(lf\b and c()neepl� alld htlW they arc used and pmvides opportunities fllr practice in their usc. Practice makes perfect. 'I"his is true not just lor hlOguage skills hUI for practical .�kllls as well. To Dreyfus, (1985) h,ldily ^;,kilh form a pan III our intelligence. And hodily ski1l� are only acqlllrcd hy living and praeti�ing in the world. Cousider bicycle riding :mu 11ft and COlftS. Furthermore^, we cnnnOl expn:ss our skills in words.

Smce complllers do not have t>lldies and do lIot live in (he world, the fact Ihnt Ihc�e �kil1s cannOt be formulated as ^I1bstraetion for the com­

puter, means that they can never he truly intelligent, at least nOt like hUlllan hcing� But having .�nid that, wh<lt alxJUt rohots'! They can have btllhcs /;If ⁽nohillly and other purposes, and sensors 10 inter^act with the eunwnmcnl. Can they -live in the world'·, and as weh, achieve tnlclli·

geu;;e in the �:Ime way that we do? If �O, does this refute Dreyfus?

Illtcrprdalion.. Cuntext and t:xpcctatilln

Since language is nOI objective hut relative to a given situation, i!l1erprcration is required before the correct meaning is arrived at. To arriv!'; at Ihi� interprcta!ion, the c(⁾nte^.l:t of the particular situatiOn has 10 !'Ie determincd. Con.�i(kr �gajll the examples mentioned earlier of "h Ihere water In the fridge,!", the cuncerll of bachelor; alld lile Papert.

Mm,ky example. The tlr�t example uses contexi to determllle meanillg

/...dJlguClge (UI(I PmMl!111S uf Kn(Jwledge 49

on the basis of differenl needs. The second uses context to differentiate between various tacit knowledge sublleties and the, third ^lO differenti­

ate between various motivations. Tbe sarne conversation among adulls may show different motivation and thus different meaning

The sentences "to wreck a nice beach" and "to recognise speech"

(from Winograd 1987) wben spoken, sound very similar It is the context of tbe situation in wbicb the sentence is uttered that will determine the meaning. For example. if the topic of conversation is aboUl pollution and oil spills, tbeu the former inlerprelation is logi­

cally consistent. However, if il is about language and. speech under­

standing, then the laller interpretalion is more likely

Language understanding is nOl only relative 10 different situations but also relative to ditTerenl cultural praclices that is. different social situations. HislOry, geography, religion, children's stories, folklore, literature and technological level are some of the factors that will need to be taken into accounl. For example, ¹⁰ undersland modern Hebrew fluently, il is necessary 10 understand the Bible in Hebrew since many words used bave meanings related 10 biblical connotations (Hofstadler 1980: 377) Furthermore, to Hofstadter (the author of the classic Code/, Escher, Bach) language is relative not only 10 culture but also lO subculture. To those in farm areas, the difference bel ween a pickup and a truck is more pronounced than LO those living in cities. Here, a knowledge of the cullUral and social contexl is necessary for interpre­

tation. Language practice within a cui lUre forms a world view, a background from whicb we interprel language meaning. Computers wbich are not grounded in the world and thus not immersed in culture will not be able 10 distinguish subtle differences in meaning

Ultimately. we understand language because we are human. We know the meaning of hunger because we bave a body which gets hungry And so il is for olher bodily altribules. We may nicely put into a thousand words the concepl of bunger bUI these words are jusl emply symbols 10 the compuler Intrinsic underSlanding will forever elude it.

For Weizenbaum,(l984) to be human is to live in a society with its associaled values, objecti ves and illleresls. Wbat does il mean for computers to bave values, objectives and interests? Speaking of whal it is 10 be human, Weizcnbaum wriles: "His life is full of risks, but risks he has the courage to accepl. because, like Lhe explorer, he learns to Irust his own capacities to endure, to overcome. What could it mean 10 spe�lk of risk, courage. trust, endurance, and overcoming when one speaks of machines?" (1984 280)

Even LO understanu a .simple concept such as a cbair, onc needs knowledge of what it is to be human It presupposes certain facts about Ihe human body (fatigue, comfort) and a nelwork of other culturally determined equipment (Iables, floors) and skills (writing, reading)

l'If'!:""", ,,,,,I p"""�",, .1 Kn" .. ��.J/l<' "

011 Ihe ha�is of differenl n�cds The ^{� ..} ;cnnd u�cs UlntC:o:t to dIITcrenli:m:

OClwo.:ell varin^\l� lacit tllllwlcd��' suhllclic� and the, third III lIillcrellll­

ale 11c\wcen l'ari(Jll� moti\'ariOn�, The �amc cOllversarion a'lu)u!, adults ttI�y �how different n1oli\'a[ioll and thn� diffcrC!l1 meaning

Th� �C"lcll('es "10 "'rerk a nke beacb" and ",() rccoglll:>C speech"

(from Winogr:t(l 1987) when ,pokell, H)\tlld n:ry �imilar 11 i� Il\e wnl�XI of tlit' �iluati(⁾n in which the ^�cnt^cnl'e i� uttered Ihat will dttl'�I1H"e the ^{me ••} mng_ For example, If Ihe lopi(' of COIl\'o::r�a(ion I�

alxlUt pollulion ;JIlIJ uil �pllh, then the former inlnpru^:t1ion j� logl' c!llly \.'on,istcn1. Howc\'cr. if It is ahout I;tngua!l;e ;tnd specdl untler·

\tandl1lg, thcn Ihe latter Hilerprcw\iull IS m(lr!;: ^Jitel�'

Latlgu.lgc: IJn(\cnlandlnf I� nOI only rclltli\'e I(l lllflc:rCIII 'llualion�

hut al�(J rdal1\'c 1!l lIilferent cultural pra�tkes thai i�, differellt �odl1l

�LlUluion� I1istory, gC(lgraphy, religion, .,:hildrcu'.� �I(lri>:�, flliidurc:, IitCflllure and Icchnoln�kal h'\'cI :trc !'omc of 11lc f�clor� lhflt ^.... i1 1 need

lil he takcll into lKl'OUllt. r(lr cltampk, to un(ler.qantl modern Hehrcw tlu�ully, it i� lIeccssarr w undcnt:lmJ Ihe Bi^ble ill Hebrl:w �in("c tTl!\ny w(ff[h ".,ed ha\'(' meamng!' rel:ucd to hiblical conn\llati(ln� (Hllhmdter 19�O: 377) Furthcrmore, ((J lI()f�I:Lllll:r (the �\lthor uflhc cla", ... it,; Go(ld f.'cher, Bach) l;tIlguage IS rdal¹v^c not only to C^Ul^{l l}lf^C bm abn to

\uhculturc_ To those in farm :lJell�, the differen.,:e betwecn ^" pICkufJ altll ,L \ruck IS Inllfe pronouliceu than 10 thosc Iilllll� III citlc�. Here, ¹¹ kllOwlcdge of the l'llllm'Jl and �{}o.;lal l'OIHext is lLt!CC��lIry f\)l' intcq)I'C' lation. Langu;lge pnKllCc witlull ^{a c}ulture form .... a world \'icw, a baci::gn)und from whIch we Interpret lallguagc I!leaning. ('omplller�

whkh �lre not grounded ^III the; world �nd Ihus nnt hnmcrsed in nlhure

"ill nm be able to di!.ting^llish suhtle tlillcrcnce .• in m�aning.

lJh"nalely. we undCfl-tr>ml J:mguage bcc"u�e we are human Wo:

binw lh� meaning of llutlg^�r hc("au�e we h�wc a ho⁽ly wtlkh liCH hU1I!;ly And Sll it is for ()Ihcr hodil� attribUles_ We may !lied)' put illlo

� tholls;md words tbe CIluccpt Cli hunger hUI thc�e "'nrtl� arc jU�1 empty j\LlIbIlJ� \(l Ihe compll1er lntrin�k undcr�lall"ing will fore\'er dude It l'nr Wdzenoaum,(19H4) to hc hUtllan is l!) live ill a wdety \vith It�

:l.��{lci;lh�d lIalue�, o hjcCtivc .^... and inh.:resis. What dOt� ;t mean for wmputers tn have ^... "Iut" Ohjt:CIIH�� and mlue�h'! Spe,lklllg o. wh!u 11 i' to he human, Wei^�enllaum writes: "His life I� full of nsks, hUI ri.\b tic ha\ the <:oufa�(! II^I acc;,:pl. hecause, like tlie explorer, hc learn, tIllru,Q his own <:^apa^citi^e.� 10 cndure, to overcome. Wbal could n JIIe:'"

l\I '[1l:ak of risk, C(lur"l;t!, trll�I, euduranCIl, and {lllt!fcoming wh^en ^ono.:

}rx:a\;� of T!lachi1J(;�'!" (]!)!i4 2XO),

!::.\·cn 10 tlnller�land <I �ilnJJIe con.:ept !(uch ^a" a chair. nne needs

�now1c(lgc of what it i� 10 he human It f!resupf!0se� ^cerlain laCh lIbl'Ul lh� h\111l�n hotly (tariguc_, C(lm!"(lt'tl lind l network of otht!r cuJlllr�JJy dct<'r11lined equipment (tahlc�, noms) and �kills (wrilillJ;, readill)ll

"

There are all surts of chairs (arm-chairs, dentist's chairs, beanbags).

Chairs would not be equipment (or us if ollr knees bend bnckwards, or if we have no tnbles as is the elise in traditionni Japan. Undr:rsHlnding chain; also includes social 5kill.\ such as being able 10 sil appropriately (�edalcJy, seductively, oaturall)'. casually, pmvocatively) al dinners, ilHervicws, cOlleens. in Jiving rooms, courts and bars. A I'unctiona!

uescription as "�omething onc can sit on" treated in a context-free manner will nol even distinguish conventionlll chairs from snddles, thrones from toilet seats (Dre)'ru�, 1985. 83). Thus, when we .�peak a word, a whole world of relaletJ concepts is behind !.hal word. 10 under�tanding then, "what is unspOKen is as much a pari of the mean­

ing as what is spoken" (Winuj!rad, 1987).

Another aspect of human intelligence is the importance of expeeta­

ti()n in understanding langulige. '1'0 Husserl, inlelligence is not based on passively receiving conteu,free facts into an already Siored data hUI ^r.llher, it is a conlext-delermined, goal- directed se:trch of antici­

pated facts (Dreyfus. 19K5). Con�ider again !be sentences k[O w^reek a nice heach" and "to rec()gni�e speech". Contexl will detenninc the toph; of conversation, and expecrntion will ensure that the interpreted meaning is comiRtent with the context. In a crow\led alld noisy situa­

tion, one may !'Ie ncar to tWO persons in a conversation. It may seem

�trange that one c^annol make _head or tail of wbat tbe conversation is all about, at lea�l in the beginning, whereas tbe two particip.ants i>eem to carry on the conversation may be easily enough. This is because, initially the context is not clear After a few related words are heard, the context can be determined and expectatiun can then function - and Ihe cnnversalion may be ea,�i1y followed.

t::llpecl:tlion is nol only necessary in language understanding but al�o in another a�pect of intelligence. that of vision under�tandiug:

how to make sense of what we see. In an airport scene, we would ellpect to sec aircraft, haugan and possibly helicopters. Looking at x­

my pictures, the layman caunot M:e the difference between those of a disc:l."cd perljon and a hc:tllhy person. To the expert, knowledge bas rrovided expeCllllion on whal llsllecls 10 look out for (Cbalmer�, 1982).

r�)r a person wbo ba� been !'Ilind since birtb, gainiug vision after su many years of adjusting 10 a dark world, will be quite disorienlating.

There is no knowledge of what to expect. To Chalmers, even the physical !>Cience�, which some 01 iL� practitioners proudly proclaim is compmed of conlext-free knowledge, cannol escape from context­

dependent data. Observations in the physical sciences are guided by theory which colollr� anti provide� a background of expectation for interpreting the observed data.

The fact that computers arc not grounded in the world would make it difricult for them to determine context. For humans, becau� we live in the world, we are always in (.'OOIexl or always-already-in-a-situatioll

U"'l""l� Md I',obltm., of K"o..,{t'dl� "

l1.S Dreylus (1985) puts il. This b a necessary OUleomc \11 Ih,t1 hnllg.

Living in the world enables us to develop our liletime kmlwlellge ha.�

and tacit k.nowledge, Out outlook. background, pre-underManding, (InTer­

bOIi1on, worlu-view, for undcrst<1nding and interpreling language mC:Ulint,

�nd Cor intelligence in genera!. Living in the world pr^Gvide� upp0rluni­

tics fur the learning and practicc uf language and bodily skills amI contc1>HJctennined e:o..pectations. Living in a society with its �hart:d ,ultural practices give� U� our values, intere�ts and obje" ivcs th;\I define what it is to be human, something that no mere knowlcd�t:

representation a.� object iv!,)^. context-free data in empty (rcpre.�t:J)t .. ti{ll1) symbols will ever capture. The involvement of living is llilt tlu:re tll provide causal, deep iUlentional 1>emantics to the symh()I.�.

ArIOIlog)'. Md31,hnr and IIlhtl" hit.;

Not only is word meaning deeply fOOled in tacit knowledge^. HI turthcr Compllnnd the problem for computt:rs. word meaning changc�

with the us^c of analogy and metaphor "To sec analogically IS to �cc (llle thing in another, not in thc �t:n�e that one mistakes one for the other, hut that she [31 conceiveS uC the one in terms of the other"

(Boden, 134). To Chomsky "language is a hahit system, a Sy�tt:lll or dispositions to behavinur, acquircd Thruugh lraiuing and c{)lldithming.

,\11), irmovalive aspech of this hchaviour is due to 'anal^{( )}gy'" (iIJM9' P7). According to Boden, there are TWO import�IILt creau"e uscs of analogy The first is the use or Ihe famIliar Iramc to prompt uHluLry at dc\'Cloping the nO\'el (rllme ill an economical way r-or (lI.'nnple, the Ba� laws were upiaincd In lerm� of Dilliard halls. Secondly, creative u�c (If" analogy enahles (lnt: nm merely to gather ^Ill:W factual knowl­

edge about the nuvel phen(llnenort, hut correlatively to understand and explain it by relating it to the concepts aln^:ady lIt:eessil;lle in the lamiliar frame (Boden, 326). Arthur Koe�t1er talk.s of ··bi�llCi:tllon".

th� merging of two different ml.Ltrict:5 ur puints of view in ero.:ating new in�ights, whether in science, humour or the arts. For cl(amp)e, in

j!').-t,

Freud describt:d the Chrit>lma� ^.,e^:t^._on a� "the alcnholidays" ^. from The 111.'0 m^atric es of alcohol and holidlt), (Marlin. 1975).

AU(lthcr impt.lrtant facet of human language and cullure I� that 01 poeLIy Martin thinks that mefaphor i, the hasls of poetry (1975 201)·

210) The use of a word in a (hfferelll contell.t, a� in the ea!'t of mtlurh,lf. brings with it certain ^,onnotations of that won..! to con·

�ci(ILlsncs.� - or at lea�t to the (nnge of conscionsness. For c.lnrnple. the U.IC of tht metaphor 'a ro.�e in hloom' to dqcribc the qualities of a jt)vcd onc. Furthermore, Manin, in defen�e of poetry against charges (of �ome logicians, fOT namplc^, Ayer in hi� Langua,�tj, Trulll Imd Lngic) that it is impreci�c, argues that the use of logic can at !lest. Onl)' modrl reaJit)·, whercas poetry ICJiVt:S and evokes image^. n f fClllity

Ctmsilier Melville's Mohy DIck: "And heaved Rnd heavell, still unre�tingly heuved the black sea, a$ if ib \'a�t tide� were a conscience" Tn Murtill. (1975'2) "If Ihere is anYlhmg c!-.pecially valuable about po­

etry, it i� n vallie that beiou£� to the real world and i� expressed in the sp!.:eeh of the unregener,tle human animal. poetry's means are Iinguis' tic alld semalllic, and Its �ubJccHn:\IIer i$ experience" C^(lmplete un·

dentanoing (If languuge must t11erefore include thl� very important :I�peel of human culture.

To un(kn;Lal\d language Ih!.:" computers cannot jusl interprel words ritnall) when they arc use\l analogically or metaphorically Examples include dead mctaphors �u(.;h as "I �ee yOllr point of view", to use M:'lrll1l'S c]I(ample. The use of "see" hcre i� aecepled pmctice anti thu�

the metaphor is -tleao". th:u is II()! seen as novel or slmnge, and nOI even nllllced thai il is not liSCO literally Dead met.arhors and idioms may be colleetell and represented in the compmer Bnt words are always u�ed aJlillogieally and mCI:lphoric�lIy III new �ituatjon�. Com.

pUlers mU�1 be able to dctermine new meaning� using somc theory on how analogy and metaphor work, otherwisc humans will always need 10 update new word meaning:. for Ihem.

How lines one chome the matrices Koe�tler lalks about? Since analogy ttlll.l metaphor work around (.;enain relevant And common fae·

tors ocllveen IWO 1Jlatrkc�, how arc the relevant and common factors Ilelami!)ed? How tloe� the merging, the I:>isocialion work? More needs to he .�tudico bel'ore ruks on how they work can be formulated. Out the pwble!lls of tacit knowledge and living in the world IIgain �urfaee. II may 1101 he [Kl^s�ible to determin^e e�aclly how we do an�logies and metaphor are produced. Thj� call nnly be dune through the eltperienee of living

Wunb can al$O change I.heir meaning through lime as people use them ^III !lew ways. fnr cltl\!llpJe. the word 'nicc' originally, in the Ihirteenth century, meant fooli�h. It has been variously Ilscd (0 mean iglloranl. SIlly and wanton How il ch:mgcs ! 11 would nnt do tn describe a respeclflhlc lady ^:l� mce when the WI/f(1 could mean wan Ion. A compult:r which ha� no facility 10 learn and make these ehange� will be limited ill ils u�cfulnc.�s ir nm outright dangcruus 10 use.

New words are eOlillnually l>eillg coined, either fonnulated hy lin·

1l1J1�1s, experts through COlisellsm or through being accepted by the populace through COll.�lallt U!>llge. Cumputen; WIll need 1(l have a mechanism to incorporate such word� and their meaning::..

Sp<."t'ch Act�, S^ea^rl^II' and ollll'r hit�

Ap:UI (rom sentences Ihal have :t truth value mere are omers mal are meant for ael/ons to t,lke place. An example of tnt titllt type

"Stepllt;u Hawking i.� the grc:tle.�t particle physicist today " An exam·

Language and Problenu of Know/edge 53

pIc of the second is "Remove the above false sentence." The second Iype of sentences was studied by AUSIin and his student Juhn Searle of Berkely 'Such sentences are called "speech acts". There are three types of speech acts. The firsl is locurionary, Ihe act of building a senlence which obeys the syntactic anu semanlic rules, which can be said IU be the aim of context-free-grammar The second is illocutionary, which is Ihe purpose of the sentence, and thirdly perlocutionary, requiring an action to be taken.

The sentence "Would I abandon you?" is structurally interrogalive, but might have the illocutionary force uf a statemenl, and ille perloculionary force of reassuring (Ritchie, 1988). To interpret the various aclS re­

quires the context of utterance with its associated prohlems already mentioned.

Some other issues of language that need 10 be deaH wilh by the computer are "conversational implicarure" and the "cooperative princi­

ple" of Grice and the "relevance Iheory" of Sperber and Wilson. How do we effectively represent these in the computer? How about Ihe tone, limbre and body language of speech? It is said that only ten percent of meaning is derived from the literal meaning The rest comes from the other three factors. Consider the classic example, "I hale you" CODl­

ing from a wife to her husband it may mean the end of their roau;

coming from a mistress to her man, it may mean the very opposite. It all depends on the lone, limbre and body language. Can a compulcr ever hope to understand all these subtleties, all this deviousness of human language?

Searle and th� Chinese Room

Searle in his celebrated article "Minds, Brains and Program" (1980, 1990), argues Ihrough his thought experiment "The Chinese Room"

that computers by these very nature cannot achieve human intelligence and understanding. His argument. runs as follows: Suppose someone or something, let it be Searle, were 10 be in a room, opa4ue 10 the outsiue worlu. Suppose that he were given all the rules of grammar of a language which he uoes not understand, in this case, Chinese, because Searle dues not understand Chinese, and also a basketful of Chinese symbols. (Of course, Ihe rules of grammar anu how to use the symbols will be in a known language, say, English.) The Chinese symbols will be identified by the rules according to their shapes anu no unucrslanu­

ing of the symbols is required. Imagine that people outside the room, who understand Chinese, send in statements in Chinese symbols into Ihe room. By manipulating the symbols, Searle can senu out the cor­

rect response. To the outside world. Searle (or whatever is inside the room) understands Chinese.

Juma/ BaluL", Modm

But he i.� just manipulating symbols without any under�tandin8.

This is just like the computer' the "rules" are "programs", "people" are programmers and Searle the computer Searle is trying to show that the Turing leSI for understanding is nOI enough. The Turing test of under­

standing is passed if, to people Outside the room. the response is indistinguishable, whether it is from a human or otherWise. The test is namcd arter Alan Turing. one of the falbers of AI (Turing, 1950). The Turing test is essentially a behaviouristic test for intelligence. which to Searle is not enough since there is no real unuerstanding by computers.

Searlc states this as "computer programs are formal (syntactic). Human minds have mental coments (semantic)" Thus. according to Searle, computer�. by their very nature, inherently cannot achieve human level undcrstanding

His argumenlS, published in 1980. have for a decade gen�rated a heated debate, especially within the AI community, the community uescribed as the "artificial intelligentsia" by Weizenbaum.

Conelus!u"

It doe� scem that the computer has a real problem on its band, so 10 speak, in trying to understand !:mguage. To us. langnage is such a nalural thing lhat we do not attribute any intelHgcnce to it. We marvel al robou wbich can do many mechanical feats, wheLber in fiction or in rea lily, but we lIlink nolhing of the abiljty of HAL the robot (from tbe 2001 Space Odyssey fam!;:) to conver�e with his human counterparts.

But AI re�earehers, trying \("I make computers literate, are involved in a never-ending regf�!<o)Oion ill undcutanding the pbilo:.ophicaJ. psycho­

logical and biological foundations of intelligence, which, in turn, are the foundations of langnage understanding. In trying \("I nnrJcrstand language, one is led to an understanding of intelligence. Which, in IUrn, leads to an understanding of knowledgc. ils reprc!lentation. tacil lmowledge (how to represent tacit knowledge - is that not a contradiction in terms?), lind a whole philosophical debate between objectivity aou relalivily!l;ubjeetivity (of knowledge. of lIle world); between holisDi and reductionism, and question� on the nature of consciousness anu the !lalUre of being Unul these and other deep issues are bettcr undcrswod. until we unucrstand our tJwn intclligence better. computer intelligence anu language understanuing will still be at a very rudi­

mentary stage. Some have compared computer intelligence to tbat of In�1Io Slill. this might he ^:w in.�ult to in�ect.s. whieh al least can negotiatc ,.nU survive in (heir environment.

It al�o seems that (he areas which AI finds so difficult to make compmerll uodeutand occur in (lur everyuay Jrnowledge, our common

�ense ahility 10 gel altlng well in the world. Remember our markel­

going rnhot? Apart from language. this inclnues vi�ion nnder�taoding,

Uonlf"" le atld P",Nwu �f K,,,,,,,'k'/re ^�5

how we understand what we see. Again, JIISI ^as in language under�

Uanding, it IS preci�cly world knowledge that h3� to bi.: incorporaled in lfllcrpreting the image seen. Th:>.! is the rout proh\em . AI has been relatively succc�sful In lurmalismg tilt: intelligence ot experts in the so-called knowlcdge·ba�ed systell\\, ur expert systems.. how they go alJout djagno�illg, suh'ill!', problems. This i� so because Iheir knowl·

edge is highly formali�ahle dill: to the pled,.ion of the knowledge. It is nlll so of our everyday knowledge. This is precisely the prublem. huw to reprc:.ent imprcci� knowledge. myriad i^nter-relat^ed facts, facL ..

whkh we may not even know exist in (lur mintl/brain, tadt knowkdge.

Thu., there is a paradox of intelli�ence hClween men and computers.

COUlputer� find it ca.�y 10 achieve expert iJHclJigcnce but do pClOIly ⁰¹¹ mundane, e�eryday, wnullon sense int^clligence^. following Boden, (I<)R7:47�) Mmsky's defintion of 1\1. ⁱⁿ the ligbt of the�e problems, hl'� to be mO(!ified: It i� "the study of how \0 huild and program machines thal can do the snrt of things which human mind� can do" The study of A!.

far trom dehmnamsins man. as �omc hal'e ^complaUled. ha� In fact, produced a profound re�pect for the human mind/brain.

The pmhlems mentjoned ahove, tacit and prescicnlific knowledge, ab�lfaction, hnli�tic and relativistic intcrrrclation, booily skill�. COIl­

tc�t, expeet<lli(lII. ete ^{. •}Me laken care of by us hceau�e we li^ve ^HI Ibe WllTld. CllmputCrs, in order 10 �Ivercome the infinite regress of knowl­

edge repre�entation, and those other prohlems, need 10 ·'live in the .... orld ._{. They} need 10 have seu.�ors and a hody ^to intcraCl with lhe enviwnment. They need to have an allwmatic capabitity to acquire their llwn knowkdge. A major re,�earch area in A I i� knowledge acqui.

sition, which is, therefore, prior to kllowlt:dge reprc�enLati(}n. 0111)' in this way will there be a chancc til ha�'e the large knowledge bll.�e of hletlme experience, rilld the ptl!'Slbihty of tadt knowlellge. Fmm bere, abstraction proper can he attempted. Knowlcdge acquisition anti ah­

straction require learning of wh;1l is deemed as us�ful and rdevant.

Machine learning is another major rel>earcb area. Living in the world .... 'iIl provide the always-a1ready.in-a-�itualton. the context. necessary for expectation and, C(ln�equently, imcrprCliltion of sentences.

Ncvccth<:le�s, will computer understanding be the �aluc as ours'!

May he nOI. Computer.� may hrlve a hody ^• bUI it is not a body that feelS hunger, that feels pain [41 Can our values. objectives anti inter­

ests be internalised hy the computer'! Internillisation Seems to TC4uire consciouslles�. or mOTe precisely, sclf-con�ciou�ness of olle sclf-oll what It is 10 he human^. 10 be a pan of �ociety. Agalll. self-con�ciou�­

ne�s .�eems to he the crucial fac((If behind the Chinese Room argument against computer illfclligcllce. It pr()vidcs the causal power, semanLic�.

and the intemiollS (beliefs, de�ire.� etc) 10 otherwise empl�' �ylllh()ls 01

representation (5]

"

Still, computer inlclli�en(;c-w ithout deep undcrSlanding, j� useful as lun� as what is m:mife�ted ^to the out�iue world, i� no different from our own manifested intc!liYl:ucc. Anyhow, it would uOl do to have rnachillcs which are conscious. A wll!)ie new problem area of defiDing a legal ptr�on. uf law lmd re.�p<)n�ibility. questiuns of ethics, of a human-machine fc!ationship, will haye to be thought ouL But, having

�aid Ihal, without consciousness. how different will computer ,ntelli­

gelKe he? As T Edchon of Gc�)rsetown University puis it, �Can a system be llltclligent if it never give� a damn?"

Notes

[11 Bul this is al�() the civilisation that produces the IIirnshimas and Ihe Nagasakis.

[21 P.1 Hayes (l'JR5) thinks thaI the numl'ler of concepts required 10 represent eoU\mon-sen�e, every oay knowledge, I� in the order of iO,OOO 10 IOU,OOO. A^ccording In Minsky, "a lIlachine will quite cri!ieally Iteell to acquire 011 the order of a huudred thousand elelnent� or knowledge in order to behave with reasonahle sen�i.

bili!)' in n[[linary �ituatHms. A million, if properly organi�ed, should be enough for a very grelu inlelligenr.:e." (Dennett: 45) Dennell thillk� our knowledge IS Ycry much larger, but othcr J'lieee� ot knowk<lge coulll he gCTlerated by mind/brah} inferences from the ru:1I1i knuwledg^e body

13] TInden u;;c� 'she" in place uf Ihe usual 'he" to �how the �te!cutypi­

cal l'e rre�entation Ill" '!le' a� repl'e�ellting both sexes -but even she woultl nol Iry usillJ; 'woman' in place of 'mao' LO rcpresem' 'lilankind{�) To �ee the [m ^... erful effect-largely ullconscious-of thi� :-tercotYPlflg, see Hofstadler (1 9lS7), and Smullyan (I9RlS) [4J Sec Dennett's chapter " Why yOIl can', makc computers tllat feel

pain"

[51 Paul Churehlalill (I9!\M), ocsn"s psyChologist and 1.R, Lueas (1961), an Oxfmu philosopher, however think lhat sclf-consciou.�­

ness ^1S jusl :1 mailer Ilf Ihe i.:1.mplclity of Ibe brain Beyond a cerl:l.in level uf complexity, �elf-nlllsciou�nes� aut\lmatii.:ally ap­

pcar�

nod�ll, �., (1').r.:7) Al"lljicUI/ lllldll)jente and Natural Mlln, London: MIT Prt'�,

Bro(lk�, R.A., (I\j!!fl) "Acl\ievill,g Arillial Intelligence Through Bmidinll Rob(lt�", MIT AI Memo t109

Ch;llmcj")', A F (19"'2) \\'11111 ^IS HilS Tiling Cillfel/ Sci�n.ce?, Sl Lucia:

1)11I\�I"�ily of Quc:e!I.�lalld Prc,s.

Lallguage alld Problmu of Knowledge 57

Chamiak, E. and D. McDermon (1985) Introduction to Artificial Intelb- gence, Reading, Ma: Addison-Wesley.

Churchland, P.M., (1988) Matter and ConSCIOusness, London: Bradford Books.

Dennen, D.C., (1988) Br(//nstorms, London: Bradford Book,.

Dreyfus, H.L., (1979) What Compl/ters Can't Do: Critique of Artificial Reason.

New York: Harper & Row (2nd. Edition).

Dreyfus, H.L., (1985) "From Micro-Worlds to Knowledge Represcmation:

AI at an Impasse" in R.1. Brachman and H.L. Levesque (eds.),

Readings III Knowledge Representation.

Los Altos. Morgan Kaufman.

Dreyfus, H.L., (1988) "The Socratic and Platonic basis of Cognitivism", in

AI &: Society. The Journal of HUllum and Machllle Intelligence,

vol. 2 nn.

2. Springer International.

Feigenbaum, E., (1985) in E.A. Torrera (ed.) (1985)Computers, Next-Genera­

t/On New York: Spectrum Series. IEEE Press.

Guha, R.V., and D.B Lenat (1990) "Cyc: A Mid-Terrn Report" in FAI Magazine, Fall., vol. II No. 3

Hayes. P.J (1985) "The Second Naive Physics Manifesto" in R.J Brachman and H. J Levesque (eds),

Reading in Knowledge Represenll/tion.

Los Altos: Morgan Kaufman.

Hofstadter. D.R. (1980) Godel. Escher Bach. An Eternal Golden Braid, Hannondsworth: Penguin Books.

Hofstadter. D.R., (1987) Metamflgical ThellUls. London: Penguin Books.

Jobannessen. K.S., (1988) "Rule Following and Tacit Knowledge" in AI &:

SOCiety, The Journal of HUIIUln and Machine intelligence

vol

²

no.

4, Dec.,

Springer Internationall

King, M., (1990) "Genes of War" by D Noonan in

Discover.

Oct.

Lucan. l.R., (1961) Minds, Machines

and

Godel:,

Pllliosophy 36

Also in Andersons "Minds and Machines" "

Martin, G.D., (1975) Language. Truth and Poetry. Edinburgh: The Univer­

sity Press.

Polanyi, M., (1967) The Tacit Dimension. New York: Doubleday, Anchor Ritchie, G., (1988) "Natural Language Processing", MSc Notes. Edinburgh

University

Searle,

J.,

(1980) "Minds, Brains and Program", in

The Belll/vlOural lind Brain Science

3 Sept.

Searle, l., (1990) "Is the Brain's Mind a Computer Program", SCIentifiC American, Jan. Vol 262 No. I New York.

Smith. B.C., (1985) "Prologue to 'Renection and Semantics in a Procedural Language" in R.1 Brachman and H. J Levesque (eds). Read,ng.l· tn Knowledge RepresentatIOn, Los Altos: Morgan Kaufman.

Smullyan, R., (1988)

Wliat IS tlie Name of TillS Book,

Hannondworth: Pen­

guin Books.

Turing, A.M., (1950) "Computing Machinery and Inteligence", in

Mind.

vol.

59, Reprinted in A. Anuerson. (ed.)(l964)

Minds and Macliines,

Englewoods Cliffs, NJ Prentice-Hall.

58 Jumal Bahasa Moden

Weizenbaum, J., (1984) Computer Power and HuTrUlll Reason, Harmon<lworth:

Penguin Books.

Winograd, T., (1972) Understanding Natural Language, New York: Aca­

demic Press.

Winograd, T., (1983) Language as a Cognitive Process, Reading, Ma: Addison Wesley

Winograd, T., (1985) "Frame Representations and the DeclarativelProcedural Controversy" in RJ. Brachman and H. J Levesque (eds), Readings in Know/edge Representation, Los Altos: Morgan Kaufman.

Winograd, T., and F. Flores (1987) Understanding Computers and Cognition Reading & New York: Addison-Wesley.

Witlgenstein, L., (1953) Philosophical Investigations, Oxford: Blackwell.