swartout-gil-kaw95.ps

%!PS-Adobe-3.0
%%Title: (Microsoft Word - swartout-gil\311)
%%Creator: (Microsoft Word: LaserWriter 8 8.2.2)
%%CreationDate: (4:01 PM Thursday, April 20, 1995)
%%For: (swartout)
%%Pages: 19
%%DocumentFonts: Times-Bold Times-Italic Times-Roman Courier Courier-Oblique Courier-Bold Helvetica Helvetica-Bold Helvetica-Oblique
%%DocumentNeededFonts: Times-Bold Times-Italic Times-Roman Courier Courier-Oblique Courier-Bold Helvetica Helvetica-Bold Helvetica-Oblique
%%DocumentSuppliedFonts:
%%DocumentData: Clean7Bit
%%PageOrder: Ascend
%%Orientation: Portrait
%%DocumentMedia: Default 612 792 0 () ()
%ADO_ImageableArea: 12 12 600 780
%%EndComments
userdict begin/dscInfo 5 dict dup begin
/Title(Microsoft Word - swartout-gil\311)def
/Creator(Microsoft Word: LaserWriter 8 8.2.2)def
/CreationDate(4:01 PM Thursday, April 20, 1995)def
/For(swartout)def
/Pages 1 def
end def end
/md 184 dict def md begin/currentpacking where {pop /sc_oldpacking currentpacking def true setpacking}if
%%BeginFile: adobe_psp_basic
%%Copyright: Copyright 1990-1993 Adobe Systems Incorporated. All Rights Reserved.
/bd{bind def}bind def
/xdf{exch def}bd
/xs{exch store}bd
/ld{load def}bd
/Z{0 def}bd
/T/true
/F/false
/:L/lineto
/lw/setlinewidth
/:M/moveto
/rl/rlineto
/rm/rmoveto
/:C/curveto
/:T/translate
/:K/closepath
/:mf/makefont
/gS/gsave
/gR/grestore
/np/newpath
14{ld}repeat
/$m matrix def
/av 81 def
/por true def
/normland false def
/psb-nosave{}bd
/pse-nosave{}bd
/us Z
/psb{/us save store}bd
/pse{us restore}bd
/level2
/languagelevel where
{
pop languagelevel 2 ge
}{
false
}ifelse
def
/featurecleanup
{
stopped
cleartomark
countdictstack exch sub dup 0 gt
{
{end}repeat
}{
pop
}ifelse
}bd
/noload Z
/startnoload
{
{/noload save store}if
}bd
/endnoload
{
{noload restore}if
}bd
level2 startnoload
/setjob
{
statusdict/jobname 3 -1 roll put
}bd
/setcopies
{
userdict/#copies 3 -1 roll put
}bd
level2 endnoload level2 not startnoload
/setjob
{
1 dict begin/JobName xdf currentdict end setuserparams
}bd
/setcopies
{
1 dict begin/NumCopies xdf currentdict end setpagedevice
}bd
level2 not endnoload
/pm Z
/mT Z
/sD Z
/realshowpage Z
/initializepage
{
/pm save store mT concat
}bd
/endp
{
pm restore showpage
}def
/$c/DeviceRGB def
/rectclip where
{
pop/rC/rectclip ld
}{
/rC
{
np 4 2 roll
:M
1 index 0 rl
0 exch rl
neg 0 rl
:K
clip np
}bd
}ifelse
/rectfill where
{
pop/rF/rectfill ld
}{
/rF
{
gS
np
4 2 roll
:M
1 index 0 rl
0 exch rl
neg 0 rl
fill
gR
}bd
}ifelse
/rectstroke where
{
pop/rS/rectstroke ld
}{
/rS
{
gS
np
4 2 roll
:M
1 index 0 rl
0 exch rl
neg 0 rl
:K
stroke
gR
}bd
}ifelse
%%EndFile
%%BeginFile: adobe_psp_colorspace_level1
%%Copyright: Copyright 1991-1993 Adobe Systems Incorporated. All Rights Reserved.
/G/setgray ld
/:F/setrgbcolor ld
%%EndFile
level2 startnoload
%%BeginFile: adobe_psp_patterns_level1
%%Copyright: Copyright 1991-1993 Adobe Systems Incorporated. All Rights Reserved.
/patfreq Z
/patangle Z
/bk Z
/fg Z
/docolorscreen Z
/graystring Z
/pattransf{}def
/initQDpatterns
{
/patfreq 9.375 store
/patangle
1 0 $m defaultmatrix dtransform
exch atan
por not
{90 add}if
normland{180 add}if
store
:a
}def
/docolorscreen
/setcolorscreen where
{
pop/currentcolorscreen where
{
pop/setcmykcolor where
{
pop true
}{
false
}ifelse
}{
false
}ifelse
}{
false
}ifelse
def
/setgraypattern
{
/graystring xs
patfreq
patangle
{
1 add
4 mul
cvi
graystring
exch get
exch
1 add 4 mul
cvi
7 sub
bitshift
1 and
}setscreen
64 div setgray
}bd
/:b
{
/pattransf load settransfer
pop pop pop
setgraypattern
}bd
docolorscreen startnoload
/screensave 5 array def
/:a{currentgray currentscreen currenttransfer screensave astore pop}bd
/:e{screensave aload pop settransfer setscreen setgray}bd
/:d
{
pop pop pop
/pattransf load settransfer
setgraypattern 8{pop}repeat
}bd
/:c
/:d ld
docolorscreen endnoload docolorscreen not startnoload
/screensave 20 array def
/:a{currentcmykcolor currentcolorscreen currentcolortransfer screensave astore pop}bd
/:e{screensave aload pop setcolortransfer setcolorscreen setcmykcolor}bd
/rstring Z
/grstring Z
/blstring Z
/convroll{64 div 4 -1 roll}bd
/setcolorpattern
{
/graystring xs
/blstring xs
/grstring xs
/rstring xs
patfreq
patangle
{
1 add 4 mul cvi rstring
exch get exch 1 add 4 mul
cvi 7 sub bitshift 1 and
}
patfreq
patangle
{
1 add 4 mul cvi grstring
exch get exch 1 add 4 mul
cvi 7 sub bitshift 1 and
}
patfreq
patangle
{
1 add 4 mul cvi blstring
exch get exch 1 add 4 mul
cvi 7 sub bitshift 1 and
}
patfreq
patangle
{
1 add 4 mul cvi graystring
exch get exch 1 add 4 mul
cvi 7 sub bitshift 1 and
}
setcolorscreen
convroll convroll convroll convroll
setcmykcolor
}bd
/:d
{
pop pop pop
/pattransf load settransfer
pop pop setcolorpattern
}bd
/:c
/:d ld
docolorscreen not endnoload
%%EndFile
level2  endnoload level2 not startnoload
%%BeginFile: adobe_psp_patterns_level2
%%Copyright: Copyright 1990-1993 Adobe Systems Incorporated. All Rights Reserved.
/pmtx Z
/BGnd Z
/FGnd Z
/PaintData Z
/PatternMtx Z
/PatHeight Z
/PatWidth Z
/$d Z
/savecolor 4 array def
/savecolorspace Z
/:a{
mark 0 0 0 currentcolor savecolor astore pop cleartomark
/savecolorspace currentcolorspace store
}bd
/:e{
savecolorspace setcolorspace
mark savecolor aload pop setcolor cleartomark
}bd
/initQDpatterns
{
gS
initmatrix
mT dup 4 get exch 5 get :T
1 0 dtransform round exch round exch idtransform
dup mul exch dup mul exch add sqrt
0 1 dtransform round exch round exch idtransform
dup mul exch dup mul exch add sqrt
neg
scale
0
por not{90 add}if
normland{180 add}if
rotate
matrix currentmatrix
gR
/pmtx xs
:a
}bd
/:t
{
14 dict begin
/BGnd xdf
/FGnd xdf
/PaintData xdf
/PatternType 1 def
/PaintType 1 def
/BBox[0 0 1 1]def
/TilingType 1 def
/XStep 1 def
/YStep 1 def
/PatternMtx[24 0 0 24 0 0]def
/PaintProc
BGnd null ne
{
{
begin
BGnd aload pop :F
0 0 1 1 rF
FGnd aload pop :F
24 24 true PatternMtx PaintData imagemask
end
}
}{
{
begin
FGnd aload pop :F
24 24 true PatternMtx PaintData imagemask
end
}
}ifelse
def
currentdict
PatternMtx
end
gS $c setcolorspace pmtx setmatrix makepattern gR
}bd
/:u
{
14 dict begin
/$d 8 dict def
/PatternType 1 def
/PaintType 1 def
/BBox[0 0 1 1]def
/TilingType 1 def
/XStep 1 def
/YStep 1 def
/PaintData xdf
/PatHeight xdf
/PatWidth xdf
/PatternMtx[PatWidth 0 0 PatHeight 0 0]def
$d begin
/ImageType 1 def
/MultipleDataSource false def
/Height PatHeight def
/Width PatWidth def
/Decode[0 1 0 1 0 1]def
/ImageMatrix PatternMtx def
/DataSource PaintData def
/BitsPerComponent 8 def
end
/PaintProc
{
begin
$d image
end
}def
currentdict
PatternMtx
end
gS $c setcolorspace pmtx setmatrix makepattern gR
}bd
/bk[1 1 1]def
/fg[0 0 0]def
/:b{
:t
setpattern
pop pop
}bd
/:d{
:t
setpattern
10{pop}repeat
}bd
/:c{
:u
setpattern
10{pop}repeat
}bd
%%EndFile
level2 not endnoload
%%BeginFile: adobe_psp_uniform_graphics
%%Copyright: Copyright 1990-1993 Adobe Systems Incorporated. All Rights Reserved.
/@a
{
np :M 0 rl :L 0 exch rl 0 rl :L fill
}bd
/@b
{
np :M 0 rl 0 exch rl :L 0 rl 0 exch rl fill
}bd
/arct where
{
pop
}{
/arct
{
arcto pop pop pop pop
}bd
}ifelse
/x1 Z
/x2 Z
/y1 Z
/y2 Z
/rad Z
/@q
{
/rad xs
/y2 xs
/x2 xs
/y1 xs
/x1 xs
np
x2 x1 add 2 div y1 :M
x2 y1 x2 y2 rad arct
x2 y2 x1 y2 rad arct
x1 y2 x1 y1 rad arct
x1 y1 x2 y1 rad arct
fill
}bd
/@s
{
/rad xs
/y2 xs
/x2 xs
/y1 xs
/x1 xs
np
x2 x1 add 2 div y1 :M
x2 y1 x2 y2 rad arct
x2 y2 x1 y2 rad arct
x1 y2 x1 y1 rad arct
x1 y1 x2 y1 rad arct
:K
stroke
}bd
/@i
{
np 0 360 arc fill
}bd
/@j
{
gS
np
:T
scale
0 0 .5 0 360 arc
fill
gR
}bd
/@e
{
np
0 360 arc
:K
stroke
}bd
/@f
{
np
$m currentmatrix
pop
:T
scale
0 0 .5 0 360 arc
:K
$m setmatrix
stroke
}bd
/@k
{
gS
np
:T
0 0 :M
0 0 5 2 roll
arc fill
gR
}bd
/@l
{
gS
np
:T
0 0 :M
scale
0 0 .5 5 -2 roll arc
fill
gR
}bd
/@m
{
np
arc
stroke
}bd
/@n
{
np
$m currentmatrix
pop
:T
scale
0 0 .5 5 -2 roll arc
$m setmatrix
stroke
}bd
%%EndFile
%%BeginFile: adobe_psp_basic_text
%%Copyright: Copyright 1990-1993 Adobe Systems Incorporated. All Rights Reserved.
/S/show ld
/A{
0.0 exch ashow
}bd
/R{
0.0 exch 32 exch widthshow
}bd
/W{
0.0 3 1 roll widthshow
}bd
/J{
0.0 32 4 2 roll 0.0 exch awidthshow
}bd
/V{
0.0 4 1 roll 0.0 exch awidthshow
}bd
/fcflg true def
/fc{
fcflg{
vmstatus exch sub 50000 lt{
(%%[ Warning: Running out of memory ]%%\r)print flush/fcflg false store
}if pop
}if
}bd
/$f[1 0 0 -1 0 0]def
/:ff{$f :mf}bd
/MacEncoding StandardEncoding 256 array copy def
MacEncoding 39/quotesingle put
MacEncoding 96/grave put
/Adieresis/Aring/Ccedilla/Eacute/Ntilde/Odieresis/Udieresis/aacute
/agrave/acircumflex/adieresis/atilde/aring/ccedilla/eacute/egrave
/ecircumflex/edieresis/iacute/igrave/icircumflex/idieresis/ntilde/oacute
/ograve/ocircumflex/odieresis/otilde/uacute/ugrave/ucircumflex/udieresis
/dagger/degree/cent/sterling/section/bullet/paragraph/germandbls
/registered/copyright/trademark/acute/dieresis/notequal/AE/Oslash
/infinity/plusminus/lessequal/greaterequal/yen/mu/partialdiff/summation
/product/pi/integral/ordfeminine/ordmasculine/Omega/ae/oslash
/questiondown/exclamdown/logicalnot/radical/florin/approxequal/Delta/guillemotleft
/guillemotright/ellipsis/space/Agrave/Atilde/Otilde/OE/oe
/endash/emdash/quotedblleft/quotedblright/quoteleft/quoteright/divide/lozenge
/ydieresis/Ydieresis/fraction/currency/guilsinglleft/guilsinglright/fi/fl
/daggerdbl/periodcentered/quotesinglbase/quotedblbase/perthousand
/Acircumflex/Ecircumflex/Aacute/Edieresis/Egrave/Iacute/Icircumflex/Idieresis/Igrave
/Oacute/Ocircumflex/apple/Ograve/Uacute/Ucircumflex/Ugrave/dotlessi/circumflex/tilde
/macron/breve/dotaccent/ring/cedilla/hungarumlaut/ogonek/caron
MacEncoding 128 128 getinterval astore pop
level2 startnoload
/copyfontdict
{
findfont dup length dict
begin
{
1 index/FID ne{def}{pop pop}ifelse
}forall
}bd
level2 endnoload level2 not startnoload
/copyfontdict
{
findfont dup length dict
copy
begin
}bd
level2 not endnoload
md/fontname known not{
/fontname/customfont def
}if
/Encoding Z
/:mre
{
copyfontdict
/Encoding MacEncoding def
fontname currentdict
end
definefont :ff def
}bd
/:bsr
{
copyfontdict
/Encoding Encoding 256 array copy def
Encoding dup
}bd
/pd{put dup}bd
/:esr
{
pop pop
fontname currentdict
end
definefont :ff def
}bd
/scf
{
scalefont def
}bd
/scf-non
{
$m scale :mf setfont
}bd
/ps Z
/fz{/ps xs}bd
/sf/setfont ld
/cF/currentfont ld
/mbf
{
/makeblendedfont where
{
pop
makeblendedfont
/ABlend exch definefont
}{
pop
}ifelse
def
}def
%%EndFile
/currentpacking where {pop sc_oldpacking setpacking}if end
%%EndProlog
%%BeginSetup
md begin
countdictstack[{
%%BeginFeature: *ManualFeed False
1 dict dup /ManualFeed false put setpagedevice
%%EndFeature
}featurecleanup
countdictstack[{
%%BeginFeature: *PageSize Letter

    1 dict
    dup /Policies 2 dict dup /PageSize 2 put dup /MediaType 0 put put
	setpagedevice
	2 dict
    dup /PageSize [612 792] put
    dup /ImagingBBox null put
    setpagedevice
%%EndFeature
}featurecleanup
(swartout)setjob
/mT[1 0 0 -1 12 780]def
initQDpatterns
/sD 16 dict def
300 level2{1 dict dup/WaitTimeout 4 -1 roll put setuserparams}{statusdict/waittimeout 3 -1 roll put}ifelse
%%IncludeFont: Times-Bold
%%IncludeFont: Times-Italic
%%IncludeFont: Times-Roman
%%IncludeFont: Courier
%%IncludeFont: Courier-Oblique
%%IncludeFont: Courier-Bold
%%IncludeFont: Helvetica
%%IncludeFont: Helvetica-Bold
%%IncludeFont: Helvetica-Oblique
/f0_1/Times-Bold
:mre
/f0_16 f0_1 16 scf
/f0_14 f0_1 14 scf
/f0_12 f0_1 12 scf
/f0_10 f0_1 10 scf
/f1_1/Times-Italic
:mre
/f1_12 f1_1 12 scf
/f2_1/Times-Roman
:mre
/f2_80 f2_1 80 scf
/f2_12 f2_1 12 scf
/f2_10 f2_1 10 scf
/f2_8 f2_1 8 scf
/f3_1/Courier
:mre
/f3_72 f3_1 72 scf
/f3_56 f3_1 56 scf
/f3_40 f3_1 40 scf
/f3_12 f3_1 12 scf
/f4_1/Courier-Oblique
:mre
/f4_12 f4_1 12 scf
/f5_1/Courier-Bold
:mre
/f5_96 f5_1 96 scf
/f5_72 f5_1 72 scf
/f6_1/Helvetica
:mre
/f6_128 f6_1 128 scf
/f6_112 f6_1 112 scf
/f6_96 f6_1 96 scf
/f7_1/Helvetica-Bold
:mre
/f7_128 f7_1 128 scf
/f7_112 f7_1 112 scf
/f7_96 f7_1 96 scf
/f8_1/Helvetica-Oblique
:mre
/f8_112 f8_1 112 scf
/Courier findfont[10 0 0 -10 0 0]:mf setfont
%%EndSetup
%%Page: 1 1
%%BeginPageSetup
initializepage
(swartout; page: 1 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
103 96 :M
f0_16 sf
(E)S
114 96 :M
f0_14 sf
(XPECT)S
f0_16 sf
(: Explicit Representations for Flexible Acquisition)S
269 121 :M
f0_12 sf
(Bill Swartout)S
272 133 :M
(Yolanda Gil)S
212 157 :M
(USC/Information Sciences Institute)S
249 169 :M
(4676 Admiralty Way)S
233 181 :M
(Marina del Rey, CA  90292)S
255 205 :M
(tel: \(310\) 822-1511)S
212 217 :M
(email: )S
247 217 :M
f1_12 sf
(swartout@isi.edu, gil@isi.edu)S
96 257 :M
f0_12 sf
.073 .007(Abstract: )J
149 257 :M
f2_12 sf
.082 .008(To create more powerful knowledge acquisition systems, we not only need)J
96 269 :M
1.409 .141(better acquisition tools, but we need to change the architecture of the knowledge)J
96 281 :M
2.281 .228(based systems we create so that their structure will provide better support for)J
96 293 :M
.616 .062(acquisition.  Current acquisition tools permit users to modify factual knowledge but)J
96 305 :M
1.635 .164(they provide limited support for modifying problem solving knowledge.  In this)J
96 317 :M
1.326 .133(paper, we argue that this limitation \(and others\) stem from the use of incomplete)J
96 329 :M
4.055 .406(models of problem solving knowledge and inflexible specification of the)J
96 341 :M
.267 .027(interdependencies between problem solving and factual knowledge.  We describe the)J
96 353 :M
(E)S
f2_10 sf
(XPECT)S
135 353 :M
f2_12 sf
2.607 .261( architecture which addresses these problems by providing an explicit)J
96 365 :M
1.123 .112(representation for problem solving knowledge and intent. Using this more explicit)J
96 377 :M
1.15 .115(representation, E)J
f2_10 sf
.302(XPECT)A
213 377 :M
f2_12 sf
2.27 .227( can automatically derive the interdependencies between)J
96 389 :M
.838 .084(problem solving and factual knowledge.  By deriving these interdependencies )J
487 389 :M
f1_12 sf
(from)S
96 401 :M
.103 .01(the structure of the knowledge-based system itself )J
f2_12 sf
(E)S
f2_10 sf
.041(XPECT)A
378 401 :M
f2_12 sf
.126 .013( supports more flexible and)J
96 413 :M
(powerful knowledge acquisition.)S
60 449 :M
f0_12 sf
(1. I)S
77 449 :M
f0_10 sf
(NTRODUCTION)S
60 466 :M
f2_12 sf
(E)S
f2_10 sf
(XPECT)S
99 466 :M
f2_12 sf
.391 .039( is a framework for knowledge based systems that we are developing to support knowledge)J
60 478 :M
.363 .036(acquisition and explanation.  A central idea behind )J
312 478 :M
(E)S
f2_10 sf
(XPECT)S
351 478 :M
f2_12 sf
.386 .039( has been the notion that more powerful)J
60 490 :M
1.263 .126(acquisition and explanation tools can be constructed if acquisition and explanation concerns are)J
60 502 :M
.273 .027(reflected in the structure of the knowledge based systems we create.  That is, rather than developing)J
60 514 :M
.399 .04(tools that operate on conventional knowledge based systems, we first modify the architecture of the)J
60 526 :M
.303 .03(target knowledge based systems so that they will be structured in a way that provides better support)J
60 538 :M
1.792 .179(for knowledge acquisition and explanation.  It is then possible to build tools that exploit this)J
60 550 :M
(additional information to provide enhanced capabilities.)S
60 567 :M
.905 .091(In prior work on the EES framework [Neches et al, 1985; Swartout et al., 1991] we explored the)J
60 579 :M
1.929 .193(architectural modifications that are needed to support explanation.  )J
415 579 :M
(E)S
f2_10 sf
(XPECT)S
454 579 :M
f2_12 sf
2.258 .226( extends the EES)J
60 591 :M
.102 .01(framework to support knowledge acquisition.  In this paper, we discuss the architectural features that)J
60 603 :M
.441 .044(support knowledge acquisition.  It is interesting to note that some of the features that are important)J
60 615 :M
(for explanation also provide leverage for knowledge acquisition.)S
60 632 :M
1.585 .158(There are several knowledge acquisition capabilities that we seek to support with the E)J
f2_10 sf
.652(XPECT)A
60 644 :M
f2_12 sf
(architecture:)S
60 660 :M
(1)S
66 660 :M
(.)S
69 660 :M
3 .3( )J
78 660 :M
f1_12 sf
2.212 .221(Users should be able to modify both factual knowledge and problem solving knowledge.)J
78 672 :M
f2_12 sf
1.374 .137(Although many acquisition systems provide good support for modifying factual knowledge,)J
50 10 512 42 rC
63 23 :M
f3_12 sf
-.202(From the Proceedings of the Ninth Knowledge Acquisition for)A
60 11 1 1 rF
60 11 1 1 rF
61 11 490 1 rF
551 11 1 1 rF
551 11 1 1 rF
60 12 1 13 rF
551 12 1 13 rF
552 12 1 13 rF
63 35 :M
-.201(Knowledge-Based Systems Workshop \(KAW\32595\) Banff, Canada, February 26-)A
60 25 1 12 rF
551 25 1 12 rF
552 25 1 12 rF
63 47 :M
-.215(March 3, 1995)A
60 50 1 1 rF
60 50 1 1 rF
61 50 490 1 rF
61 51 490 1 rF
552 50 1 2 rF
551 51 2 1 rF
551 50 1 1 rF
551 50 1 1 rF
60 37 1 13 rF
551 37 1 13 rF
552 37 1 13 rF
endp
%%Page: 2 2
%%BeginPageSetup
initializepage
(swartout; page: 2 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(2)S
gR
gS 0 0 588 768 rC
78 69 :M
f2_12 sf
.324 .032(support for modifying problem solving knowledge is more limited.  In early acquisition systems)J
78 81 :M
.289 .029(\(such as MORE [Eshelman 1988], SALT [Marcus and McDermott 1989], and ROGET [Bennett)J
78 93 :M
.087 .009(1985]\), a single problem solving strategy \(e.g. heuristic classification\) was built into the tool.  As)J
78 105 :M
.448 .045(a result, when one selected a tool, one also determined the problem-solving strategy.  However,)J
78 117 :M
.772 .077(many realistically-sized knowledge-based systems use several problem solving strategies, so a)J
78 129 :M
.57 .057(tool that only supports a single strategy won\325t be much help.  Additionally, if the user changes)J
78 141 :M
.222 .022(his mind about which problem-solving strategy is appropriate, he may also have to change tools,)J
78 153 :M
1.151 .115(potentially losing a lot of work in re-configuring the domain knowledge.  Recent knowledge)J
78 165 :M
2.03 .203(acquisition work [Klinker et al. 1991; Musen and Tu 1993] has partially addressed these)J
78 177 :M
1.162 .116(problems by creating tools that can use multiple problem-solving methods.  In PROT\203G\203 II)J
78 189 :M
.392 .039([Musen and Tu, 1993], problem solving methods are composed of pre-encoded building blocks.)J
78 201 :M
.793 .079(PROT\203G\203 II permits modifications to problem solving by substituting one building block for)J
78 213 :M
1.286 .129(another, however, users cannot modify the building blocks themselves.  By constraining the)J
78 225 :M
2.04 .204(user\325s options to just those that have been pre-encoded, he may not be able to make the)J
78 237 :M
(modification he wants.)S
60 253 :M
(2)S
66 253 :M
(.)S
69 253 :M
6 .6( )J
78 253 :M
f1_12 sf
.343 .034(Internal models \(based on the knowledge based system being modified\) should guide knowledge)J
78 265 :M
1.788 .179(acquisition rather than external ones \(based on the acquisition tool\). )J
f2_12 sf
2.24 .224( Current acquisition)J
78 277 :M
1.945 .195(systems cannot allow much modification to problem solving knowledge because they use)J
78 289 :M
1.166 .117(external models of problem solving that are built into the acquisition tool. By understanding)J
78 301 :M
1.477 .148(what factual knowledge is needed to support problem solving the acquisition tool can form)J
78 313 :M
.947 .095(expectations about what additional knowledge is required when the user adds or modifies the)J
78 325 :M
.162 .016(system\325s knowledge base.  In early acquisition tools, these expectations were built, by hand, into)J
78 337 :M
-.002(the tool itself.  In more recent work, such as PROT\203G\203 II, the interdependencies between factual)A
78 349 :M
1.019 .102(knowledge and problem solving building blocks are represented, but they are still entered by)J
78 361 :M
1.505 .15(hand and interdependencies between factual knowledge and what happens inside a building)J
78 373 :M
(block are not modeled.)S
60 389 :M
14 1.4( )J
78 389 :M
.8 .08(These limitations could be overcome and a wider range of modifications and problem solving)J
78 401 :M
1.234 .123(methods could be supported if we could use an )J
323 401 :M
f1_12 sf
1.028 .103(internal )J
366 401 :M
f2_12 sf
1.145 .115(model for acquisition, that is, if the)J
78 413 :M
.691 .069(expectations for acquisition could be derived from the system itself.  Furthermore, because the)J
78 425 :M
.966 .097(interdependencies would be derived from the system rather than entered by hand, they would)J
78 437 :M
.298 .03(change as the system changed, and would be more likely to be correct and consistent.  Thus, we)J
78 449 :M
.97 .097(need architectures that allow us to derive the interdependencies between problem solving and)J
78 461 :M
.155 .016(factual knowledge from the knowledge structures themselves  Such architectures will allow us to)J
78 473 :M
.907 .091(build knowledge acquisition tools that permit direct modification of problem solving methods)J
78 485 :M
(and use their understanding of the program to guide the user.)S
60 501 :M
(3)S
66 501 :M
(.)S
69 501 :M
6 .6( )J
78 501 :M
f1_12 sf
.522 .052(KA tools should support modifications at a semantic \(or knowledge\) level rather than  just at a)J
78 513 :M
.108 .011(syntactic level.  )J
156 513 :M
f2_12 sf
.108 .011(KA tools should help ensure that the knowledge a user adds makes sense, that is,)J
78 525 :M
2.03 .203(that it is coherent and consistent with the rest of the knowledge base\321not just that it is)J
78 537 :M
1.761 .176(syntactically correct.  In addition, we want to facilitate the addition of new knowledge by)J
78 549 :M
.779 .078(reducing the distance between what the user understands and way the system represents them.)J
78 561 :M
.507 .051(The system must be able to represent and manipulate conceptual entities that are meaningful to)J
78 573 :M
.356 .036(users and that are used to describe the domain.  To further facilitate acquisition, KA tool should)J
78 585 :M
.951 .095(allow users to make modifications that are local to these meaningful conceptual entities,  and)J
78 597 :M
.461 .046(guide them in resolving the global implications of these changes.  In addition, if the user enters)J
78 609 :M
.387 .039(knowledge using somewhat different terminology, the acquisition tools should be able to weave)J
78 621 :M
1.43 .143(the new knowledge into the existing system based on the meaning of the terms.  Providing)J
78 633 :M
.499 .05(assistance at the conceptual level and allowing greater flexibility in the use of terminology will)J
78 645 :M
(free users to focus on what matters: getting the knowledge right.)S
60 662 :M
1.906 .191(We have introduced several features into the architecture of )J
377 662 :M
(E)S
f2_10 sf
(XPECT)S
416 662 :M
f2_12 sf
1.955 .195( that directly support the)J
60 674 :M
.306 .031(acquisition goals we outlined above.  We have used the )J
334 674 :M
(E)S
f2_10 sf
(XPECT)S
373 674 :M
f2_12 sf
.262 .026( framework to construct knowledge)J
endp
%%Page: 3 3
%%BeginPageSetup
initializepage
(swartout; page: 3 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(3)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
1.697 .17(based systems in several domains.  Our main application is for evaluating alternative military)J
60 81 :M
.54 .054(transportation plans for moving personnel and materiel from bases to crisis situations.  The system)J
60 93 :M
1.635 .164(we have been developing takes a high-level description of a possible plan, or course-of-action)J
60 105 :M
1.804 .18(\(COA\),  and produces an evaluation of the plan from a logistics perspective.  The evaluation)J
60 117 :M
.238 .024(indicates, for example, how many logistic personnel will be required to execute it, and how long the)J
60 129 :M
.212 .021(plan will take \(the closure date\).  Most of the examples in this paper are based on this domain, some)J
60 141 :M
(are from another domain concerned with the diagnosis of faults with local area networks.)S
60 158 :M
1.281 .128(The next section presents a high level overview of the )J
342 158 :M
(E)S
f2_10 sf
(XPECT)S
381 158 :M
f2_12 sf
1.203 .12( architecture, and is followed by)J
60 170 :M
.521 .052(detailed discussions of its features.  Section 3 is concerned with how we represent various kinds of)J
60 182 :M
.316 .032(knowledge in E)J
f2_10 sf
.113(XPECT)A
168 182 :M
f2_12 sf
.492 .049(.  Section 3.1 deals with the need to separate different kinds of knowledge to)J
60 194 :M
.105 .01(make it easier to understand and modify.  Section 3.2 describes how we represent the intent behind a)J
60 206 :M
.081 .008(problem solving method using a structured representation for goals and method capabilities.  Section)J
60 218 :M
.537 .054(4 then describes how we bring knowledge back together again to actually solve problems.  Section)J
60 230 :M
.282 .028(4.1 details how \322loose coupling\323 between knowledge resources can facilitate knowledge acquisition)J
60 242 :M
.926 .093(and knowledge re-use, and describes how we achieve loose coupling through the use of semantic)J
60 254 :M
1.604 .16(match and reformulation.  Section 4.2 describes how we use partial evaluation to  capture the)J
60 266 :M
2.342 .234(interdependencies between problem solving and factual knowledge that are needed to guide)J
60 278 :M
(knowledge acquisition.)S
60 314 :M
f0_12 sf
(2. E)S
f0_10 sf
(XPECT )S
f0_12 sf
(A)S
125 314 :M
f0_10 sf
(RCHITECTURE )S
202 314 :M
f0_12 sf
(O)S
f0_10 sf
(VERVIEW)S
60 331 :M
f2_12 sf
1.082 .108(A diagram of the overall E)J
f2_10 sf
.507(XPECT)A
230 331 :M
f2_12 sf
1.36 .136( architecture appears in Figure 1.  As we describe in the next)J
60 343 :M
.162 .016(section, E)J
f2_10 sf
.054(XPECT)A
139 343 :M
f2_12 sf
.306 .031( provides explicit and separate representations for different kinds of knowledge that)J
60 355 :M
.876 .088(go into a knowledge based system.  E)J
f2_10 sf
.407(XPECT)A
283 355 :M
f2_12 sf
.851 .085( distinguishes domain facts, domain terminology and)J
60 367 :M
1.252 .125(problem solving knowledge.  These different sources of knowledge are integrated together by a)J
60 379 :M
1.208 .121(program instantiator to produce a domain-specific knowledge based system.  While the domain-)J
60 391 :M
.512 .051(specific system is being created, the program instantiator also creates a design history.  The design)J
60 403 :M
.402 .04(history records the interdependencies among the different kinds of knowledge, such as what factual)J
60 415 :M
.847 .085(information is used by the problem solving methods.  This information is used by the knowledge)J
60 427 :M
.254 .025(acquisition routines to form the expectations that guide the knowledge acquisition process \(see [Gil,)J
60 439 :M
1.118 .112(1994; Gil and Paris, 1994] for a description of the knowledge acquisition tools\).  An interpreter)J
60 451 :M
1.58 .158(executes the knowledge based system, and the user interacts with it through a user interaction)J
60 463 :M
(manager.)S
60 499 :M
f0_12 sf
(3. R)S
81 499 :M
f0_10 sf
(EPRESENTING )S
f0_12 sf
(K)S
f0_10 sf
(NOWLEDGE IN )S
f0_12 sf
(E)S
f0_10 sf
(XPECT)S
60 516 :M
f2_12 sf
.037 .004(In order to provide the capabilities that we discussed in the introduction, E)J
f2_10 sf
.016(XPECT)A
450 516 :M
f2_12 sf
.033 .003( represents different)J
60 528 :M
1.345 .135(types of knowledge separately and explicitly.  After describing the kinds of knowledge that are)J
60 540 :M
1.717 .172(represented in )J
137 540 :M
(E)S
f2_10 sf
(XPECT)S
176 540 :M
f2_12 sf
1.834 .183( we present in detail our approach to representing goals.  This explicit)J
60 552 :M
(representation of goals provides )S
217 552 :M
(E)S
f2_10 sf
(XPECT)S
256 552 :M
f2_12 sf
( with a better understanding of the problem solving process.)S
60 576 :M
f0_12 sf
(3.1 Separation of Knowledge)S
60 593 :M
f2_12 sf
.257 .026(It is well established that a major source of difficulties in understanding, modifying and augmenting)J
60 605 :M
3.118 .312(first generation knowledge based systems stemmed from the use of low-level knowledge)J
60 617 :M
1.21 .121(representations that failed to distinguish different kinds of knowledge \(see [Chandrasekaran and)J
60 629 :M
1.564 .156(Mittal, 1982; Clancey, 1983b; Swartout, 1981; Swartout, 1983]\).  In a first generation system,)J
60 641 :M
.068 .007(domain facts, problem solving knowledge, and terminological definitions were all expressed in rules.)J
60 653 :M
.914 .091(A single rule might mix together clauses concerned with the user interface, the system\325s problem)J
60 665 :M
2.255 .226(solving strategy and internal record-keeping. Because none of these different concerns were)J
60 677 :M
1.092 .109(distinguished, it was often difficult to understand exactly what the rule was supposed to do, and)J
endp
%%Page: 4 4
%%BeginPageSetup
initializepage
(swartout; page: 4 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(4)S
gR
gS 0 0 588 768 rC
60 382 :M
f2_12 sf
.129 .013(when modifying a rule, it was difficult to see what the effect of the modification might be. Although)J
60 394 :M
1.907 .191(early critiques of these representations focused on their failure to provide good explanations,)J
60 406 :M
.592 .059(explanation and knowledge acquisition can be viewed as inverse processes, and these architectural)J
60 418 :M
(flaws create problems for both.)S
60 435 :M
1.255 .125(A number of second generation expert system frameworks that support better explanations have)J
60 447 :M
.047 .005(emerged in recent years \(see [Chandrasekaran, 1986; Clancey, 1983a; Hasling et al., 1984; Neches et)J
60 459 :M
.639 .064(al, 1985; Swartout, 1983; Swartout et al., 1991; Wielinga and Breuker, 1986]\).  A common theme)J
60 471 :M
.153 .015(among these frameworks is that they encourage a more abstract representation of domain knowledge)J
60 483 :M
1.718 .172(and problem solving knowledge that makes distinctions between different kinds of knowledge)J
60 495 :M
(explicit.)S
60 512 :M
2.03 .203(By moving toward architectures that allow a system builder to distinguish different kinds of)J
60 524 :M
1.992 .199(knowledge and represent them separately and more abstractly, second generation frameworks)J
60 536 :M
.59 .059(increased the modularity of an expert system.  It has been shown that the use of these higher-level)J
60 548 :M
.439 .044(representations helps make second generation systems easier to understand.  These frameworks can)J
60 560 :M
1.639 .164(also make it easier to extend and maintain an expert system, and enhance the possibilities for)J
60 572 :M
(knowledge re-use across systems [Bachant and Soloway, 1989].)S
60 589 :M
1.53 .153(The )J
84 589 :M
(X)S
93 589 :M
f2_10 sf
.356(PLAIN)A
f2_12 sf
1.147 .115( [Swartout, 1983] and EES [Neches et al, 1985; Swartout et al., 1991] frameworks,)J
60 601 :M
.392 .039(which were precursors to E)J
f2_10 sf
.156(XPECT)A
226 601 :M
f2_12 sf
.481 .048(, were among the first to explicitly distinguish the different kinds)J
60 613 :M
1.197 .12(of knowledge in a knowledge based system.  In )J
307 613 :M
(E)S
f2_10 sf
(XPECT)S
346 613 :M
f2_12 sf
1.077 .108(, we distinguish three different kinds of)J
60 625 :M
(knowledge: )S
119 625 :M
f1_12 sf
(domain facts)S
181 625 :M
f2_12 sf
(, )S
f1_12 sf
(domain terminology)S
284 625 :M
f2_12 sf
(, and )S
f1_12 sf
(problem solving knowledge)S
442 625 :M
f2_12 sf
(.)S
60 642 :M
1.115 .112(Domain facts)J
f1_12 sf
.114 .011( )J
130 642 :M
f2_12 sf
.83 .083(are the relevant facts about a system\325s domain.  For example, the domain facts in a)J
60 654 :M
.252 .025(system for transportation planning might include the fact that the naval port of Los Angeles is Long)J
60 666 :M
(Beach and the fact that the maximum depth of Long Beach berths is 50 feet.)S
.38 .56 .992 :F
60 35 475 312 rC
221 154 51 103 rF
1 G
217 149 51 104 rF
0 G
217.5 149.5 50 103 rS
gS
.09 .09 scale
2468.441 1925.238 :M
f6_112 sf
(Domain-)S
gR
gS
.09 .09 scale
2490.679 2058.781 :M
f6_112 sf
(Specific)S
gR
gS
.09 .09 scale
2568.513 2192.323 :M
f6_112 sf
(KBS)S
gR
gS
.09 .09 scale
2512.918 2459.408 :M
f6_112 sf
(Design)S
gR
gS
.09 .09 scale
2512.918 2592.951 :M
.38 .56 .992 :F
f6_112 sf
0 G
(History)S
gR
.38 .56 .992 :F
77 40 483 105 8 @q
.548 .956 .916 :F
72 35 479 101 8 @q
0 G
72.5 35.5 478.5 100.5 7.5 @s
gS
.09 .09 scale
1423.245 534.17 :M
1 G
f7_128 sf
0 G
(Expectation-based Tools for Knowledge Acquisition)S
gR
1 G
238 55 312 86 4 @q
0 G
238.5 55.5 311.5 85.5 3.5 @s
gS
.09 .09 scale
2802.014 756.741 :M
f6_96 sf
(expectation)S
gR
gS
.09 .09 scale
2902.086 868.026 :M
1 G
f6_96 sf
0 G
(builder)S
gR
1 G
352 55 422 86 4 @q
0 G
352.5 55.5 421.5 85.5 3.5 @s
gS
.09 .09 scale
4136.307 745.613 :M
f6_96 sf
(agenda)S
gR
gS
.09 .09 scale
4102.949 856.898 :M
1 G
f6_96 sf
0 G
(manager)S
gR
1 G
137 55 203 86 4 @q
0 G
137.5 55.5 202.5 85.5 3.5 @s
gS
.09 .09 scale
1812.414 756.741 :M
f6_96 sf
(KB)S
gR
gS
.09 .09 scale
1690.104 868.026 :M
.38 .56 .992 :F
f6_96 sf
0 G
(modifiers)S
gR
.38 .56 .992 :F
412 142 464 266 5 @q
1 G
407 137 460 261 5 @q
0 G
407.5 137.5 459.5 260.5 4.5 @s
gS
.09 .09 scale
4692.262 1914.11 :M
f6_112 sf
(User)S
gR
gS
.09 .09 scale
4547.714 2047.652 :M
f6_112 sf
(Interaction)S
gR
gS
.09 .09 scale
4569.952 2181.195 :M
f6_112 sf
(Manager/)S
gR
gS
.09 .09 scale
4581.071 2314.737 :M
.38 .56 .992 :F
f6_112 sf
0 G
(Explainer)S
gR
.38 .56 .992 :F
65 126 79 139 rF
1 G
61 122 79 138 rF
0 G
61.5 122.5 78 137 rS
gS
.09 .09 scale
722.742 1302.04 :M
1 G
f7_96 sf
0 G
(Knowledge Bases)S
gR
1 G
64 126 70 22 rF
0 G
64.5 126.5 69 21 rS
gS
.09 .09 scale
833.933 1546.868 :M
1 G
f6_96 sf
0 G
(Terminology)S
gR
1 G
65 194 69 57 rF
0 G
65.5 194.5 68 56 rS
gS
.09 .09 scale
922.886 2325.866 :M
f6_96 sf
(Problem)S
gR
gS
.09 .09 scale
945.124 2437.151 :M
f6_96 sf
(Solving)S
gR
gS
.09 .09 scale
856.171 2548.437 :M
f6_96 sf
(Knowledge)S
gR
gS
.09 .09 scale
856.171 2670.851 :M
.38 .56 .992 :F
f6_96 sf
0 G
(\(strategies\))S
gR
.38 .56 .992 :F
134 281 209 329 6 @q
.888 .744 1 :F
130 277 204 325 6 @q
0 G
130.5 277.5 203.5 324.5 5.5 @s
gS
.09 .09 scale
1623.389 3305.178 :M
f7_112 sf
(Program)S
gR
gS
.09 .09 scale
1545.555 3438.72 :M
f7_112 sf
(Instantiator)S
gR
-1 -1 105 299 1 1 104 260 @b
60 35 59 312 rC
104 299 -1 1 127 298 1 104 298 @a
gR
gS 60 35 475 312 rC
113 296 -1 1 127 298 1 113 295 @a
-1 -1 114 303 1 1 113 295 @b
-1 -1 114 303 1 1 126 298 @b
np 126 298 :M
113 295 :L
113 302 :L
126 298 :L
eofill
60 193 153 154 rC
-1 -1 170 277 1 1 215 186 @b
gR
gS 60 35 475 312 rC
-1 -1 208 197 1 1 216 186 @b
207 197 -1 1 215 200 1 207 196 @a
-1 -1 215 201 1 1 216 186 @b
np 217 186 :M
207 196 :L
214 200 :L
217 186 :L
eofill
.38 .56 .992 :F
312 178 381 209 4 @q
.992 .752 .896 :F
307 173 377 204 4 @q
0 G
307.5 173.5 376.5 203.5 3.5 @s
gS
.09 .09 scale
3535.875 2125.552 :M
.38 .56 .992 :F
f6_112 sf
0 G
(Interpreter)S
gR
.38 .56 .992 :F
320 122 53 35 rF
1 G
316 117 52 36 rF
0 G
316.5 117.5 51 35 rS
gS
.09 .09 scale
3546.994 1457.839 :M
f6_112 sf
(Execution)S
gR
gS
.09 .09 scale
3647.066 1591.382 :M
f6_112 sf
(Trace)S
gR
60 158 475 189 rC
-1 -1 342 174 1 1 341 152 @b
gR
gS 60 35 475 312 rC
-1 -1 339 166 1 1 342 152 @b
338 166 -1 1 346 165 1 338 165 @a
342 153 -1 1 346 165 1 342 152 @a
np 342 152 :M
338 165 :L
346 165 :L
342 152 :L
eofill
60 107 475 240 rC
-1 -1 342 118 1 1 341 100 @b
gR
gS 60 35 475 312 rC
-1 -1 339 114 1 1 342 100 @b
338 114 -1 1 346 113 1 338 113 @a
342 101 -1 1 346 113 1 342 100 @a
np 342 100 :M
338 113 :L
346 113 :L
342 100 :L
eofill
60 35 241 312 rC
268 182 -1 1 308 181 1 268 181 @a
gR
gS 60 35 475 312 rC
294 179 -1 1 308 182 1 294 178 @a
-1 -1 295 187 1 1 294 178 @b
-1 -1 295 187 1 1 307 182 @b
np 307 182 :M
294 178 :L
294 186 :L
307 182 :L
eofill
60 107 475 17 rC
-1 -1 437 131 1 1 436 100 @b
gR
gS 60 35 475 312 rC
-1 -1 434 114 1 1 437 100 @b
433 114 -1 1 441 113 1 433 113 @a
437 101 -1 1 441 113 1 437 100 @a
np 437 100 :M
433 113 :L
441 113 :L
437 100 :L
eofill
-1 -1 438 131 1 1 440 117 @b
433 118 -1 1 441 117 1 433 117 @a
433 118 -1 1 438 130 1 433 117 @a
np 437 130 :M
441 117 :L
433 117 :L
437 130 :L
eofill
60 35 340 312 rC
368 143 -1 1 408 142 1 368 142 @a
gR
gS 60 35 475 312 rC
394 140 -1 1 407 143 1 394 139 @a
-1 -1 395 148 1 1 394 139 @b
-1 -1 395 148 1 1 406 143 @b
np 407 143 :M
394 139 :L
394 147 :L
407 143 :L
eofill
60 35 341 312 rC
269 234 -1 1 408 233 1 269 233 @a
gR
gS 60 35 475 312 rC
394 231 -1 1 408 234 1 394 230 @a
-1 -1 395 238 1 1 394 230 @b
-1 -1 395 238 1 1 407 234 @b
np 407 234 :M
394 230 :L
394 238 :L
407 234 :L
eofill
60 35 340 312 rC
376 191 -1 1 408 190 1 376 190 @a
gR
gS 60 35 475 312 rC
394 188 -1 1 407 191 1 394 187 @a
-1 -1 395 195 1 1 394 187 @b
-1 -1 395 195 1 1 406 191 @b
np 407 191 :M
394 187 :L
394 194 :L
407 191 :L
eofill
60 35 339 312 rC
341 251 -1 1 407 250 1 341 250 @a
gR
gS 60 35 475 312 rC
393 248 -1 1 407 250 1 393 247 @a
-1 -1 394 255 1 1 393 247 @b
-1 -1 394 255 1 1 406 250 @b
np 406 250 :M
393 247 :L
393 254 :L
406 250 :L
eofill
-1 -1 342 347 1 1 341 250 @b
78 346 -1 1 343 345 1 78 345 @a
-1 -1 79 347 1 1 78 260 @b
60 107 475 240 rC
-1 -1 290 225 1 1 289 101 @b
gR
gS 60 35 475 312 rC
-1 -1 287 115 1 1 290 101 @b
286 115 -1 1 295 114 1 286 114 @a
290 102 -1 1 295 114 1 290 101 @a
np 290 101 :M
286 114 :L
294 114 :L
290 101 :L
eofill
60 35 223 312 rC
268 159 -1 1 291 158 1 268 158 @a
gR
gS 60 35 475 312 rC
277 156 -1 1 291 158 1 277 155 @a
-1 -1 278 163 1 1 277 155 @b
-1 -1 278 163 1 1 290 158 @b
np 290 158 :M
277 155 :L
277 162 :L
290 158 :L
eofill
268 223 -1 1 291 222 1 268 222 @a
60 107 475 240 rC
-1 -1 169 222 1 1 168 100 @b
gR
gS 60 35 475 312 rC
-1 -1 166 114 1 1 168 100 @b
165 114 -1 1 173 113 1 165 113 @a
168 101 -1 1 173 113 1 168 100 @a
np 168 100 :M
165 113 :L
172 113 :L
168 100 :L
eofill
146 35 389 312 rC
140 221 -1 1 169 220 1 140 220 @a
gR
gS 60 35 475 312 rC
140 222 -1 1 153 225 1 140 221 @a
-1 -1 153 226 1 1 152 217 @b
-1 -1 141 222 1 1 152 217 @b
np 140 221 :M
153 225 :L
153 217 :L
140 221 :L
eofill
147 35 388 312 rC
140 138 -1 1 169 137 1 140 137 @a
gR
gS 60 35 475 312 rC
140 138 -1 1 154 141 1 140 137 @a
-1 -1 154 142 1 1 153 134 @b
-1 -1 141 138 1 1 153 134 @b
np 140 137 :M
153 141 :L
153 134 :L
140 137 :L
eofill
np 516 256 :M
516 257 :L
515 259 :L
515 259 :L
515 260 :L
515 261 :L
515 262 :L
515 262 :L
516 263 :L
517 263 :L
518 263 :L
519 263 :L
520 262 :L
521 261 :L
522 261 :L
523 260 :L
523 259 :L
523 258 :L
523 257 :L
523 257 :L
523 256 :L
523 254 :L
516 256 :L
eofill
.498 G
np 518 261 :M
518 261 :L
518 262 :L
518 262 :L
519 262 :L
520 261 :L
521 261 :L
522 260 :L
522 259 :L
522 260 :L
521 260 :L
520 260 :L
520 260 :L
519 260 :L
518 260 :L
517 260 :L
516 260 :L
516 260 :L
517 260 :L
518 260 :L
519 260 :L
519 260 :L
519 260 :L
519 260 :L
518 260 :L
517 260 :L
516 260 :L
515 261 :L
516 260 :L
517 260 :L
518 260 :L
519 260 :L
519 261 :L
519 261 :L
518 261 :L
518 261 :L
eofill
.247 G
-1 -1 519 258 1 1 519 255 @b
518 258 -1 1 520 258 1 518 257 @a
0 G
np 491 255 :M
490 255 :L
489 257 :L
489 258 :L
488 259 :L
488 259 :L
487 260 :L
487 261 :L
488 261 :L
490 261 :L
491 261 :L
493 260 :L
494 260 :L
495 259 :L
496 258 :L
496 258 :L
497 257 :L
497 257 :L
498 257 :L
499 256 :L
500 256 :L
500 253 :L
491 255 :L
eofill
.498 G
np 489 259 :M
489 259 :L
489 260 :L
489 260 :L
489 260 :L
491 260 :L
492 260 :L
492 259 :L
493 259 :L
494 258 :L
493 258 :L
492 257 :L
491 257 :L
490 257 :L
490 257 :L
491 257 :L
492 258 :L
492 258 :L
491 258 :L
491 258 :L
490 258 :L
489 258 :L
491 258 :L
491 258 :L
491 259 :L
490 259 :L
490 259 :L
489 259 :L
eofill
.373 G
np 505 161 :M
501 164 :L
500 165 :L
495 166 :L
493 177 :L
491 188 :L
492 190 :L
499 192 :L
497 209 :L
498 209 :L
493 234 :L
488 255 :L
492 256 :L
500 254 :L
504 229 :L
510 212 :L
511 229 :L
513 256 :L
519 257 :L
525 255 :L
521 229 :L
521 227 :L
522 209 :L
521 205 :L
523 203 :L
524 197 :L
525 194 :L
527 190 :L
530 183 :L
525 168 :L
524 164 :L
521 164 :L
518 163 :L
515 163 :L
513 162 :L
513 162 :L
512 163 :L
510 165 :L
509 165 :L
508 165 :L
506 164 :L
505 163 :L
505 161 :L
eofill
.247 G
np 498 170 :M
498 173 :L
499 176 :L
499 179 :L
499 180 :L
499 181 :L
500 182 :L
500 183 :L
500 184 :L
501 185 :L
497 184 :L
497 182 :L
498 181 :L
498 180 :L
498 180 :L
497 178 :L
498 175 :L
498 174 :L
498 170 :L
eofill
np 522 173 :M
521 177 :L
520 180 :L
520 182 :L
520 184 :L
520 186 :L
521 188 :L
521 191 :L
521 192 :L
521 188 :L
521 187 :L
522 186 :L
522 185 :L
523 185 :L
523 184 :L
522 184 :L
523 184 :L
524 183 :L
525 183 :L
526 183 :L
525 182 :L
524 182 :L
525 182 :L
525 181 :L
524 180 :L
523 180 :L
522 180 :L
522 179 :L
522 177 :L
522 173 :L
eofill
np 495 169 :M
492 187 :L
494 188 :L
493 189 :L
502 191 :L
503 192 :L
504 192 :L
507 188 :L
507 191 :L
510 189 :L
515 192 :L
513 197 :L
511 198 :L
517 208 :L
521 205 :L
522 209 :L
519 211 :L
515 216 :L
515 221 :L
515 224 :L
516 227 :L
516 231 :L
518 230 :L
516 233 :L
517 233 :L
516 235 :L
519 257 :L
513 256 :L
511 221 :L
510 212 :L
507 221 :L
504 230 :L
500 254 :L
492 256 :L
488 255 :L
493 231 :L
496 220 :L
498 210 :L
509 211 :L
510 199 :L
510 199 :L
508 203 :L
506 198 :L
504 199 :L
504 198 :L
504 200 :L
502 200 :L
501 201 :L
501 201 :L
502 204 :L
500 204 :L
500 204 :L
500 207 :L
499 210 :L
497 209 :L
499 192 :L
492 190 :L
491 188 :L
495 169 :L
eofill
np 519 194 :M
520 197 :L
521 199 :L
522 199 :L
523 199 :L
522 198 :L
520 196 :L
519 194 :L
eofill
np 519 201 :M
521 201 :L
522 199 :L
522 203 :L
520 203 :L
519 201 :L
eofill
np 529 184 :M
528 185 :L
527 186 :L
526 187 :L
524 187 :L
526 188 :L
523 188 :L
525 190 :L
521 190 :L
523 193 :L
521 193 :L
523 195 :L
523 198 :L
529 184 :L
eofill
1 .624 .498 :F
np 504 158 :M
505 159 :L
505 160 :L
505 161 :L
506 162 :L
507 163 :L
508 163 :L
509 163 :L
510 163 :L
511 162 :L
512 161 :L
513 160 :L
513 160 :L
514 159 :L
514 158 :L
514 157 :L
513 156 :L
513 155 :L
513 153 :L
512 151 :L
512 150 :L
512 151 :L
511 152 :L
509 153 :L
508 153 :L
507 153 :L
508 152 :L
506 152 :L
505 153 :L
504 153 :L
504 157 :L
504 157 :L
504 158 :L
504 158 :L
eofill
np 506 161 :M
505 162 :L
505 163 :L
507 164 :L
508 165 :L
509 165 :L
509 166 :L
511 165 :L
513 164 :L
514 163 :L
514 159 :L
506 161 :L
eofill
1 .498 .247 :F
np 511 152 :M
511 152 :L
510 153 :L
510 153 :L
509 154 :L
509 154 :L
508 154 :L
507 153 :L
507 153 :L
506 154 :L
506 153 :L
505 153 :L
505 154 :L
505 154 :L
505 155 :L
505 155 :L
505 155 :L
505 155 :L
506 155 :L
506 155 :L
507 155 :L
507 155 :L
508 155 :L
508 156 :L
507 158 :L
507 158 :L
508 158 :L
509 158 :L
509 158 :L
508 159 :L
508 159 :L
508 159 :L
510 160 :L
510 160 :L
510 160 :L
507 160 :L
508 160 :L
508 160 :L
509 160 :L
509 161 :L
508 161 :L
508 161 :L
508 162 :L
508 162 :L
508 161 :L
507 161 :L
507 160 :L
507 159 :L
508 159 :L
507 159 :L
507 158 :L
507 157 :L
507 156 :L
506 156 :L
506 156 :L
505 156 :L
505 157 :L
505 157 :L
504 158 :L
505 158 :L
505 159 :L
505 158 :L
505 159 :L
505 160 :L
505 160 :L
506 161 :L
506 161 :L
507 162 :L
508 162 :L
509 162 :L
509 162 :L
509 162 :L
509 163 :L
510 163 :L
510 163 :L
512 162 :L
513 161 :L
513 160 :L
513 161 :L
512 162 :L
511 165 :L
509 166 :L
508 165 :L
507 165 :L
505 163 :L
505 162 :L
505 161 :L
505 160 :L
504 158 :L
504 157 :L
504 153 :L
506 152 :L
509 152 :L
511 152 :L
eofill
np 513 153 :M
512 152 :L
512 153 :L
512 154 :L
512 155 :L
511 155 :L
512 155 :L
512 156 :L
512 156 :L
512 157 :L
512 157 :L
512 158 :L
512 158 :L
512 158 :L
512 158 :L
513 159 :L
512 159 :L
512 160 :L
512 161 :L
513 159 :L
514 159 :L
514 158 :L
513 158 :L
513 157 :L
513 156 :L
513 156 :L
513 155 :L
513 153 :L
eofill
np 509 155 :M
509 156 :L
509 156 :L
510 156 :L
511 156 :L
511 156 :L
512 156 :L
511 155 :L
511 155 :L
510 155 :L
510 155 :L
509 155 :L
eofill
1 G
np 511 156 :M
511 155 :L
510 155 :L
510 155 :L
509 156 :L
509 156 :L
510 156 :L
511 156 :L
511 156 :L
eofill
0 G
510.5 155.5 0.5 @i
1 G
np 505 156 :M
506 155 :L
506 155 :L
507 155 :L
507 156 :L
507 156 :L
506 156 :L
506 156 :L
505 156 :L
eofill
0 G
506.5 155.5 0.5 @i
np 509 154 :M
509 154 :L
510 154 :L
510 154 :L
511 154 :L
511 154 :L
512 154 :L
512 155 :L
512 155 :L
512 155 :L
511 155 :L
511 154 :L
510 155 :L
510 155 :L
509 155 :L
509 155 :L
509 155 :L
509 154 :L
eofill
np 508 155 :M
507 155 :L
507 154 :L
506 154 :L
506 154 :L
506 154 :L
505 155 :L
505 155 :L
505 155 :L
505 155 :L
505 155 :L
506 155 :L
507 155 :L
507 155 :L
507 155 :L
508 155 :L
508 155 :L
508 155 :L
eofill
np 504 158 :M
504 158 :L
503 157 :L
503 157 :L
503 156 :L
503 156 :L
503 154 :L
503 153 :L
503 152 :L
503 151 :L
504 150 :L
505 149 :L
506 148 :L
507 148 :L
508 147 :L
510 147 :L
511 147 :L
511 147 :L
512 148 :L
513 148 :L
513 148 :L
514 149 :L
514 150 :L
515 150 :L
515 151 :L
515 151 :L
515 152 :L
515 153 :L
515 154 :L
516 154 :L
516 154 :L
516 155 :L
516 155 :L
515 156 :L
515 157 :L
515 157 :L
514 158 :L
514 157 :L
514 156 :L
513 156 :L
513 157 :L
513 156 :L
513 155 :L
513 153 :L
512 152 :L
513 151 :L
512 151 :L
512 151 :L
511 152 :L
511 152 :L
510 153 :L
510 153 :L
509 153 :L
508 153 :L
508 153 :L
507 153 :L
508 153 :L
508 152 :L
508 152 :L
507 152 :L
506 153 :L
505 153 :L
505 153 :L
505 153 :L
504 153 :L
504 156 :L
504 157 :L
504 157 :L
504 158 :L
504 158 :L
504 158 :L
eofill
.247 G
np 504 150 :M
504 151 :L
505 150 :L
505 150 :L
506 151 :L
505 151 :L
506 151 :L
507 151 :L
508 151 :L
509 151 :L
508 151 :L
510 150 :L
510 151 :L
511 150 :L
512 150 :L
513 149 :L
512 150 :L
513 150 :L
513 152 :L
514 152 :L
514 153 :L
515 152 :L
515 153 :L
516 154 :L
515 155 :L
515 156 :L
515 157 :L
514 158 :L
514 158 :L
514 157 :L
513 156 :L
513 156 :L
513 156 :L
513 155 :L
513 154 :L
513 153 :L
512 152 :L
513 151 :L
512 150 :L
512 151 :L
511 152 :L
510 153 :L
509 153 :L
508 153 :L
507 153 :L
508 153 :L
508 152 :L
506 152 :L
505 153 :L
504 153 :L
504 154 :L
504 156 :L
504 157 :L
504 158 :L
504 158 :L
503 157 :L
503 156 :L
503 156 :L
503 153 :L
503 152 :L
503 151 :L
504 150 :L
eofill
.624 G
np 505 165 :M
505 168 :L
505 169 :L
506 171 :L
510 176 :L
513 170 :L
514 168 :L
514 165 :L
514 164 :L
515 168 :L
515 172 :L
514 177 :L
511 183 :L
510 186 :L
508 181 :L
506 177 :L
505 174 :L
505 172 :L
504 170 :L
504 168 :L
505 165 :L
eofill
.874 G
np 505 162 :M
507 164 :L
507 164 :L
508 165 :L
509 165 :L
506 169 :L
506 170 :L
505 169 :L
505 168 :L
505 167 :L
505 166 :L
505 165 :L
505 162 :L
eofill
np 514 162 :M
512 163 :L
511 164 :L
510 165 :L
509 165 :L
512 169 :L
513 170 :L
514 169 :L
514 168 :L
514 167 :L
514 166 :L
515 164 :L
514 162 :L
eofill
0 0 .502 :F
np 507 168 :M
508 169 :L
507 173 :L
510 176 :L
512 172 :L
512 171 :L
511 169 :L
511 167 :L
509 165 :L
507 168 :L
eofill
.498 G
np 506 161 :M
504 163 :L
502 164 :L
501 165 :L
500 167 :L
500 167 :L
502 168 :L
501 169 :L
501 173 :L
503 176 :L
503 177 :L
504 179 :L
508 185 :L
511 185 :L
507 177 :L
505 171 :L
505 169 :L
505 166 :L
505 165 :L
505 165 :L
505 164 :L
505 162 :L
506 161 :L
eofill
np 513 161 :M
516 163 :L
517 163 :L
518 164 :L
518 165 :L
519 166 :L
519 167 :L
517 168 :L
519 168 :L
518 174 :L
517 178 :L
512 186 :L
510 185 :L
511 182 :L
513 174 :L
515 169 :L
515 168 :L
514 166 :L
514 164 :L
513 163 :L
513 162 :L
513 161 :L
eofill
1 G
np 512 181 :M
511 182 :L
510 184 :L
509 189 :L
508 193 :L
515 194 :L
518 186 :L
520 185 :L
512 181 :L
eofill
.874 G
np 509 189 :M
513 189 :L
516 187 :L
518 186 :L
520 185 :L
512 181 :L
511 182 :L
510 184 :L
509 189 :L
eofill
0 G
np 504 185 :M
505 185 :L
506 186 :L
505 189 :L
505 190 :L
505 191 :L
504 191 :L
503 191 :L
503 190 :L
502 190 :L
502 189 :L
502 188 :L
502 188 :L
502 187 :L
503 186 :L
504 185 :L
eofill
1 .624 .498 :F
np 504 186 :M
505 186 :L
505 186 :L
506 185 :L
508 184 :L
509 183 :L
509 183 :L
510 183 :L
510 184 :L
513 185 :L
513 185 :L
513 186 :L
513 186 :L
513 186 :L
513 187 :L
512 187 :L
512 187 :L
512 188 :L
511 188 :L
511 189 :L
510 189 :L
509 189 :L
508 189 :L
507 190 :L
506 190 :L
505 190 :L
503 189 :L
503 188 :L
503 187 :L
504 186 :L
eofill
1 .498 .247 :F
np 502 189 :M
504 189 :L
505 188 :L
505 188 :L
506 187 :L
506 186 :L
508 184 :L
508 185 :L
509 185 :L
509 185 :L
508 185 :L
508 186 :L
508 187 :L
508 187 :L
508 187 :L
508 188 :L
508 188 :L
508 189 :L
507 189 :L
506 190 :L
506 190 :L
505 190 :L
504 190 :L
503 189 :L
502 189 :L
eofill
1 G
np 504 185 :M
503 186 :L
503 186 :L
502 186 :L
502 187 :L
502 187 :L
502 188 :L
502 188 :L
502 189 :L
502 189 :L
502 190 :L
502 191 :L
503 191 :L
504 191 :L
504 191 :L
503 190 :L
503 189 :L
503 188 :L
503 187 :L
504 187 :L
505 186 :L
506 185 :L
504 185 :L
eofill
.498 .373 .247 :F
np 509 187 :M
510 187 :L
511 188 :L
511 188 :L
512 188 :L
511 188 :L
511 189 :L
510 188 :L
510 187 :L
509 187 :L
eofill
np 509 186 :M
512 187 :L
512 187 :L
512 187 :L
509 186 :L
eofill
np 510 185 :M
513 186 :L
512 186 :L
510 185 :L
eofill
.247 G
2 1 510 177.5 @j
510 179 1 @i
510 181 1 @i
.874 G
2 1 510 177.5 @j
510 179 1 @i
510 181 1 @i
0 G
464 35 19 312 rC
457 193 -1 1 490 192 1 457 192 @a
gR
gS 60 35 475 312 rC
457 194 -1 1 471 196 1 457 193 @a
-1 -1 471 197 1 1 470 189 @b
-1 -1 458 194 1 1 470 189 @b
np 457 193 :M
470 197 :L
470 189 :L
457 193 :L
eofill
477 190 -1 1 490 193 1 477 189 @a
-1 -1 478 197 1 1 477 189 @b
-1 -1 478 197 1 1 489 193 @b
np 489 193 :M
477 189 :L
477 197 :L
489 193 :L
eofill
gS
.09 .09 scale
5348.289 3093.735 :M
f7_112 sf
(User/)S
gR
gS
.09 .09 scale
5092.55 3227.278 :M
f7_112 sf
(Domain Expert)S
gR
:a
32<CCCCCCCCCCCCCCCC><CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC>fg bk
:b
217 205 -1 1 268 204 1 217 204 @a
:e
1 G
68 156 66 27 rF
0 G
68.5 156.5 65 26 rS
gS
.09 .09 scale
945.124 1847.338 :M
f6_96 sf
(Domain)S
gR
gS
.09 .09 scale
1000.719 1958.624 :M
f6_96 sf
(Facts)S
gR
gR
gS 51 31 493 333 rC
223 361 :M
f2_12 sf
(Figure 1. E)S
f2_10 sf
(XPECT)S
308 361 :M
f2_12 sf
( Architecture)S
endp
%%Page: 5 5
%%BeginPageSetup
initializepage
(swartout; page: 5 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(5)S
gR
gS 0 0 588 768 rC
60 329 :M
f2_12 sf
.224 .022(The domain terminology \(or )J
f1_12 sf
.065(ontology)A
243 329 :M
f2_12 sf
.3 .03(\) provides the conceptual structures that are used to describe a)J
60 341 :M
.604 .06(domain.  In a transportation planning domain, the domain terminology would include concepts for)J
60 353 :M
.07 .007(various kinds of ports, such as airports and seaports, and concepts for describing the various kinds of)J
60 365 :M
(movements)S
115 362 :M
f2_8 sf
(1)S
119 365 :M
f2_12 sf
.917 .092( and materiel to be moved, among other things.  Concepts can be defined in terms of)J
60 377 :M
.8 .08(other concepts.  For example, an airlift-movement is a kind of movement whose destination is an)J
60 389 :M
.368 .037(airport.  Domain terminology and domain facts may seem to be quite similar, but they are different.)J
60 401 :M
1.904 .19(Domain terminology provides a set of terms, or concepts, that can be used to describe some)J
60 413 :M
.335 .034(situation.  The existence of a concept does )J
f1_12 sf
.103(not)A
f2_12 sf
.383 .038( imply that the situation it describes actually exists\321)J
60 425 :M
-.001(concepts are just descriptions that may or may not apply in any given situation.  Domain facts, on the)A
60 437 :M
(other hand, use domain terminology to describe the state of the world at some point in time.)S
60 454 :M
.917 .092(To represent both domain facts and terminology, )J
308 454 :M
(E)S
f2_10 sf
(XPECT)S
347 454 :M
f2_12 sf
.968 .097( uses Loom [MacGregor 1988, 1990], a)J
60 466 :M
1.973 .197(knowledge representation system in the KL-ONE family.  Loom provides a descriptive logic)J
60 478 :M
1.337 .134(representation language, and provides a classifier for inference.  Facts are represented as Loom)J
60 490 :M
f1_12 sf
(instances)S
105 490 :M
f2_12 sf
.398 .04(, while terminology is represented using Loom )J
f1_12 sf
.114(concepts.)A
382 490 :M
f2_12 sf
.515 .051(  Both instances and concepts are)J
60 502 :M
1.674 .167(structured, frame-based representations with slots that indicate relations in which the object is)J
60 514 :M
(involved.  Instances denote objects that actually exist in the domain, while concepts are descriptions.)S
60 531 :M
1.17 .117(Like other description logic representations, Loom provides a tool, called the )J
455 531 :M
f1_12 sf
(classifier)S
499 531 :M
f2_12 sf
1.378 .138(, that has)J
60 543 :M
.923 .092(unique advantages for knowledge acquisition.  Given a set of defined concepts, the classifier can)J
60 555 :M
.192 .019(automatically organize them into an AKO \(or subsumption\) hierarchy by analyzing their definitions.)J
60 567 :M
.533 .053(For example, if we defined an express-movement as a movement whose duration is less than a day)J
60 579 :M
.286 .029(and whose destination is an airport, the classifier would figure out that an express-movement is also)J
60 591 :M
1.4 .14(a kind of airlift-movement, given the definition of airlift movement above \(see Figure 2\).  The)J
60 603 :M
.032 .003(classifier helps support the acquisition of new concepts, because it ensures that the concept hierarchy)J
60 615 :M
2.227 .223(is maintained consistently and it can find subsumption relations that a user might overlook.)J
60 627 :M
.193 .019(Furthermore, if a new concept is added to the knowledge base and the classifier places it in a bizarre)J
60 639 :M
1.832 .183(place, that usually indicates a modeling error of some sort.  For knowledge acquisition these)J
60 656 :M
(                                                )S
60 653.48 -.48 .48 204.48 653 .48 60 653 @a
60 665 :M
f2_8 sf
(1)S
64 668 :M
f2_10 sf
.43 .043(In transportation planning, a \322movement\323 specifies what is to be moved from an origin to a destination at a particular)J
60 679 :M
(time using some set of vehicles.)S
145 97 393 167 rC
141 91 -1 1 158 125 1 141 90 @a
143 90 -1 1 160 123 1 143 89 @a
gR
gS 60 60 478 204 rC
141 90 -1 1 145 105 1 141 89 @a
-1 -1 145 106 1 1 153 101 @b
141 90 -1 1 154 101 1 141 89 @a
np 141 89 :M
144 105 :L
153 101 :L
141 89 :L
eofill
1 G
105 28 180.5 130 @j
0 G
104 27 180.5 130 @f
gS
.095 .095 scale
1432.021 1397.961 :M
1 G
f6_128 sf
0 G
(Airlift Movement)S
gR
1 G
83 28 123.5 76 @j
0 G
82 27 123.5 76 @f
gS
.095 .095 scale
989.779 819.857 :M
1 G
f6_128 sf
0 G
(Movement)S
gR
1 G
73 23 246.5 187.5 @j
0 G
72 22 246.5 187.5 @f
gS
.095 .095 scale
2400.74 1997.087 :M
f6_128 sf
(airport)S
gR
60 60 163 110 rC
201 145 -1 1 228 175 1 201 144 @a
gR
gS 60 60 478 204 rC
223 166 -1 1 229 176 1 223 165 @a
-1 -1 219 170 1 1 223 165 @b
218 170 -1 1 229 176 1 218 169 @a
np 228 176 :M
223 165 :L
218 169 :L
228 176 :L
eofill
gS
.095 .095 scale
2253.326 1650.225 :M
1 G
f8_112 sf
0 G
(destination)S
gR
1 G
97 37 108.5 185.5 @j
0 G
96 36 108.5 185.5 @f
gS
.095 .095 scale
884.483 1902.489 :M
f6_128 sf
(Express )S
gR
gS
.095 .095 scale
831.835 2049.642 :M
f6_128 sf
(Movement)S
gR
60 60 142 204 rC
156 186 -1 1 208 186 1 156 185 @a
gR
gS 60 60 478 204 rC
196 184 -1 1 208 186 1 196 183 @a
-1 -1 197 190 1 1 196 183 @b
-1 -1 197 190 1 1 207 186 @b
np 207 186 :M
196 183 :L
196 189 :L
207 186 :L
eofill
gS
.095 .095 scale
1632.082 1881.467 :M
1 G
f8_112 sf
0 G
(destination)S
gR
1 G
82 27 164 250.5 @j
0 G
81 26 164 250.5 @f
gS
.095 .095 scale
1505.728 2659.28 :M
f6_128 sf
(< 1 day)S
gR
60 60 478 174 rC
124 205 -1 1 140 240 1 124 204 @a
gR
gS 60 60 478 204 rC
138 229 -1 1 141 240 1 138 228 @a
-1 -1 133 232 1 1 138 228 @b
132 232 -1 1 141 240 1 132 231 @a
np 140 240 :M
138 228 :L
132 231 :L
140 240 :L
eofill
gS
.095 .095 scale
1442.55 2322.928 :M
f8_112 sf
(duration)S
gR
60 98 478 166 rC
-1 -1 93 168 1 1 109 90 @b
-1 -1 95 168 1 1 111 90 @b
gR
gS 60 60 478 204 rC
-1 -1 103 106 1 1 110 90 @b
102 106 -1 1 112 107 1 102 105 @a
110 91 -1 1 112 107 1 110 90 @a
np 111 90 :M
102 105 :L
112 107 :L
111 90 :L
eofill
400 96 138 168 rC
396 90 -1 1 413 124 1 396 89 @a
397 89 -1 1 414 122 1 397 88 @a
gR
gS 60 60 478 204 rC
396 89 -1 1 400 104 1 396 88 @a
-1 -1 400 105 1 1 407 100 @b
396 89 -1 1 408 100 1 396 88 @a
np 396 88 :M
399 104 :L
407 100 :L
396 88 :L
eofill
1 G
106 28 435 129 @j
0 G
105 27 435 129 @f
gS
.095 .095 scale
4117.059 1387.45 :M
1 G
f6_128 sf
0 G
(Airlift Movement)S
gR
1 G
82 29 378 74.5 @j
0 G
81 28 378 74.5 @f
gS
.095 .095 scale
3674.817 809.346 :M
1 G
f6_128 sf
0 G
(Movement)S
gR
1 G
74 23 501 186.5 @j
0 G
73 22 501 186.5 @f
gS
.095 .095 scale
5085.779 1986.576 :M
f6_128 sf
(airport)S
gR
60 60 418 109 rC
455 144 -1 1 483 174 1 455 143 @a
gR
gS 60 60 478 204 rC
478 165 -1 1 483 174 1 478 164 @a
-1 -1 474 169 1 1 478 164 @b
473 169 -1 1 483 174 1 473 168 @a
np 483 175 :M
478 164 :L
473 168 :L
483 175 :L
eofill
gS
.095 .095 scale
4927.835 1639.714 :M
1 G
f8_112 sf
0 G
(destination)S
gR
1 G
96 38 363 184 @j
0 G
95 37 363 184 @f
gS
.095 .095 scale
3558.992 1891.978 :M
f6_128 sf
(Express )S
gR
gS
.095 .095 scale
3516.874 2039.131 :M
f6_128 sf
(Movement)S
gR
60 60 396 204 rC
410 185 -1 1 463 185 1 410 184 @a
gR
gS 60 60 478 204 rC
451 183 -1 1 463 185 1 451 182 @a
-1 -1 452 189 1 1 451 182 @b
-1 -1 452 189 1 1 462 185 @b
np 462 185 :M
451 182 :L
451 188 :L
462 185 :L
eofill
gS
.095 .095 scale
4317.121 1870.956 :M
1 G
f8_112 sf
0 G
(destination)S
gR
1 G
82 28 419 249 @j
0 G
81 27 419 249 @f
gS
.095 .095 scale
4190.766 2648.769 :M
f6_128 sf
(< 1 day)S
gR
60 60 478 173 rC
379 204 -1 1 395 239 1 379 203 @a
gR
gS 60 60 478 204 rC
393 228 -1 1 395 238 1 393 227 @a
-1 -1 388 230 1 1 393 227 @b
387 230 -1 1 395 238 1 387 229 @a
np 394 239 :M
393 227 :L
387 229 :L
394 239 :L
eofill
gS
.095 .095 scale
4127.589 2312.417 :M
f8_112 sf
(duration)S
gR
.75 G
60 97 478 167 rC
-1 -1 348 166 1 1 364 89 @b
-1 -1 349 166 1 1 365 89 @b
gR
.75 G
gS 60 60 478 204 rC
-1 -1 358 104 1 1 365 89 @b
357 104 -1 1 367 106 1 357 103 @a
365 90 -1 1 367 106 1 365 89 @a
np 365 89 :M
357 104 :L
366 106 :L
365 89 :L
eofill
0 G
340 121 -2 2 374 141 2 340 119 @a
-2 -2 339 143 2 2 367 117 @b
60 148 332 116 rC
-1 -1 381 164 1 1 396 140 @b
-1 -1 382 166 1 1 397 142 @b
gR
0 G
gS 60 60 478 204 rC
-1 -1 386 151 1 1 398 140 @b
385 151 -1 1 393 156 1 385 150 @a
-1 -1 393 157 1 1 398 140 @b
np 398 140 :M
385 151 :L
392 156 :L
398 140 :L
eofill
gR
gS 51 60 504 221 rC
115 278 :M
f0_12 sf
(Before Classification)S
374 278 :M
(After Classification)S
gR
gS 157 298 291 17 rC
166 312 :M
f2_12 sf
(Figure 2.  Loom Classifier Organizes Concept Hierarchy)S
endp
%%Page: 6 6
%%BeginPageSetup
initializepage
(swartout; page: 6 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(6)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.784 .078(capabilities are important, because they mean that a user can add knowledge to a knowledge base)J
60 81 :M
.944 .094(and maintain its consistency without needing to be aware of all the other concepts in the system.)J
60 93 :M
.055 .005(Concepts from the domain terminology are used as the building blocks for )J
421 93 :M
(E)S
f2_10 sf
(XPECT)S
460 93 :M
f2_12 sf
.056 .006(\325s two other kinds)J
60 105 :M
(of knowledge: domain facts and problem solving knowledge.)S
60 122 :M
2.461 .246(As Doyle and Patil [Doyle and Patil, 1991] have pointed out, the advantages of automatic)J
60 134 :M
1.807 .181(classification are vitiated somewhat if primitive concepts are used.  Primitive concepts lack a)J
60 146 :M
.766 .077(definition, and the knowledge base builder explicitly indicates where they should be placed in the)J
60 158 :M
.114 .011(concept hierarchy, instead of having the concept\325s place inferred by the classifier reasoning about its)J
60 170 :M
1.25 .125(definition.  Primitive concepts are used when the model builder either cannot or chooses not to)J
60 182 :M
.045 .005(provide an explicit concept definition.  Doyle and Patil\325s critique of NIKL \(an early description logic)J
60 194 :M
.293 .029(representation\) stated that in realistically sized applications, most of the concepts are primitive.  We)J
60 206 :M
.65 .065(have not found this to be the case in our work, perhaps because Loom provides a more expressive)J
60 218 :M
.743 .074(language for defining concepts, making it easier to define concepts.  Regardless, it is important to)J
60 230 :M
3.296 .33(point out that the most important capabilities of our architecture, such as capturing the)J
60 242 :M
.258 .026(interdependencies between domain and problem solving knowledge, can be provided whether or not)J
60 254 :M
.071 .007(primitive concepts are used.  The use of primitive concepts reduces our framework\325s ability to verify)J
60 266 :M
(the consistency of a knowledge base, and to automatically find new uses for knowledge.)S
60 283 :M
1.253 .125(Loom benefits us by providing a declarative foundation for representing concepts, relations and)J
60 295 :M
.865 .087(facts. However, to be able to analyze the interdependencies between the conceptual structure of a)J
60 307 :M
.994 .099(system and its problem solving methods we need a highly declarative representation for problem)J
60 319 :M
.284 .028(solving knowledge as well.  In the next section, we describe how we have built on Loom to create a)J
60 331 :M
1.931 .193(declarative representation for goals and methods that captures their intent and promotes their)J
60 343 :M
(analysis.)S
60 367 :M
f0_12 sf
(3.2 Capturing Intent with Structured Goal Representation)S
60 384 :M
f2_12 sf
.03 .003(In the introduction, two of our goals were to allow users to be able to make modifications to problem)J
60 396 :M
.52 .052(solving as well as factual knowledge, and to make modifications locally, so that a problem solving)J
60 408 :M
1.603 .16(method could be modified without having to understand how other methods in the system are)J
60 420 :M
.749 .075(implemented. Problem solving knowledge in )J
286 420 :M
(E)S
f2_10 sf
(XPECT)S
325 420 :M
f2_12 sf
.675 .068( is represented as strategies \(called )J
f1_12 sf
.261(methods)A
542 420 :M
f2_12 sf
<29>S
60 432 :M
.541 .054(for achieving particular goals.  Each method has a )J
311 432 :M
f1_12 sf
.452 .045(capability description  )J
f2_12 sf
.483 .048(associated with it, which)J
60 444 :M
1.543 .154(states what the method can do \(e.g. \322diagnose a component\323\), and a )J
417 444 :M
f1_12 sf
1.708 .171(method body  )J
492 444 :M
f2_12 sf
1.649 .165(which is a)J
60 456 :M
.116 .012(sequence of steps that attempt to achieve the advertised capability of the method.  The steps can post)J
60 468 :M
2.058 .206(further subgoals for the system to achieve.  The language used in the method bodies allows)J
60 480 :M
(conditional expressions and iteration.)S
60 497 :M
1.038 .104(One of E)J
f2_10 sf
.531(XPECT)A
140 497 :M
f2_12 sf
1.405 .14(\325s architectural features that helps users make modifications to problem solving)J
60 509 :M
.808 .081(knowledge is a rich representation for goals and method capabilities that provides a more explicit)J
60 521 :M
.553 .055(representation of intent than is available in most systems.  This representation both makes it easier)J
60 533 :M
.047 .005(for people to understand what a method is supposed to be doing, and it makes it easier for )J
495 533 :M
(E)S
f2_10 sf
(XPECT)S
534 533 :M
f2_12 sf
.058 .006( to)J
60 545 :M
(analyze a system\325s structure and hence provide guidance to a user in making modifications.)S
60 562 :M
.711 .071(In most programming languages, function names have no real semantics associated with them.  A)J
60 574 :M
.466 .047(good programmer may indicate what a function is supposed to do by giving it an appropriate name)J
60 587 :M
.215 .021(\(e.g. )J
f3_12 sf
.102(clear-screen)A
f2_12 sf
.217 .022(\) but as far as the system is concerned, the name is just a string of characters.)J
60 599 :M
1.061 .106(Furthermore, the relationship between the function name \(what is to be done\) and the functional)J
60 611 :M
.612 .061(parameters \(the things that will be involved in doing it\) is completely implicit.  This is one reason)J
60 623 :M
1.246 .125(why conventional programs are so difficult to modify: before one can make a modification to a)J
60 635 :M
-.001(function, one needs to understand not only the function itself, but also how the function is called and,)A
60 647 :M
2.204 .22(in turn, how the functions it calls work.  Only then will the information that is needed for)J
60 659 :M
.263 .026(modification but not represented be available.  AI systems such as planners  provide a more explicit)J
60 671 :M
.242 .024(representation of what is to be done.  Usually the goal is represented as some sort of state expressed)J
endp
%%Page: 7 7
%%BeginPageSetup
initializepage
(swartout; page: 7 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(7)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.134 .013(in some form of predicate logic.  A problem with this representation that affects both acquisition and)J
60 81 :M
.354 .035(explanation is that it seems to be a bit removed from the way people think and talk about what they)J
60 93 :M
.752 .075(are doing, since protocols of people solving problems show that they use verb clauses to describe)J
60 105 :M
(what they are doing rather than state descriptions \(see for example, [Anzai and Simon 1979]\).)S
60 122 :M
3.531 .353(Thus, to address these difficulties and decrease the distance between )J
452 122 :M
(E)S
f2_10 sf
(XPECT)S
491 122 :M
f2_12 sf
<D5>S
495 122 :M
f2_10 sf
.896 .09(S )J
f2_12 sf
.699(internal)A
60 134 :M
1.53 .153(representation and how people talk about what they do, )J
351 134 :M
(E)S
f2_10 sf
(XPECT)S
390 134 :M
f2_12 sf
1.53 .153(\325s representation is based on a)J
60 146 :M
.055 .006(vocabulary of )J
130 146 :M
f1_12 sf
(verbs.)S
159 146 :M
f2_12 sf
.034 .003(  In E)J
f2_10 sf
.037(XPECT)A
217 146 :M
f2_12 sf
.058 .006(, we represent both goals \(what is to be done\) and method capability)J
60 158 :M
1.212 .121(descriptions \(what a method can do\) using a verb clause representation based on case grammar.)J
60 170 :M
1.344 .134(Each verb clause consists of a verb and a set of slots \(or \322cases\323\) that are the parameters.  For)J
60 182 :M
-.002(example, the goal of evaluating a particular COA is represented as a verb clause where the main verb)A
60 194 :M
(is \322evaluate\323 and its \(direct\) object slot is filled by the COA instance \(e.g., \322coa-3\323\) as follows:)S
96 219 :M
f3_12 sf
-.204(\(evaluate \(obj \(instance coa-3\)\)\))A
60 247 :M
f2_12 sf
.434 .043(The capability descriptions associated with methods are represented similarly, except that they may)J
60 259 :M
1.163 .116(contain variables.  For example, the following capability description would be associated with a)J
60 271 :M
(method that has the capability to evaluate a course of action for force deployment:)S
96 296 :M
f3_12 sf
-.202(\(evaluate \(obj \(?c is \(instance-of deployment-coa\)\)\)\))A
60 323 :M
f2_12 sf
(where:)S
96 348 :M
f3_12 sf
-.207(deployment-coa is a coa)A
204 360 :M
-.207(that has force-movements)A
60 387 :M
f2_12 sf
( )S
63 387 :M
2.321 .232(For matching, the goals and capability descriptions are translated into corresponding Loom)J
60 400 :M
1.22 .122(concepts.  The goal above would match this capability description if )J
413 400 :M
f3_12 sf
.278(coa-3)A
f2_12 sf
.77 .077( had movements of)J
60 413 :M
(forces, since )S
123 413 :M
f3_12 sf
-.055(coa-3 )A
f2_12 sf
-.039(would then be a kind of)A
279 413 :M
f3_12 sf
-.211( deployment-coa.)A
60 430 :M
f2_12 sf
.252 .025(This approach gives us a rich representation for what is intended by a goal and what the capabilities)J
60 442 :M
2.146 .215(of a method are.  Unlike the state-based representation, goals and capabilities can be easily)J
60 454 :M
.298 .03(paraphrased into natural language.  Furthermore, goals can be expressed at an abstract level without)J
60 466 :M
(forcing over-commitment.)S
60 483 :M
.789 .079(Figure 3 shows a simple method from our system for evaluating transportation plans.  It finds the)J
60 495 :M
.487 .049(unloading time of a movement by calculating the unloading time of the set of available vehicles of)J
60 507 :M
1.726 .173(the COA movement.  There are two things to notice about this method.  First, it is relatively)J
60 519 :M
.45 .045(straightforward to paraphrase the representations of the capability description and the method body)J
60 531 :M
.067 .007(into understandable natural language, because their structure)J
352 528 :M
f2_8 sf
(2)S
356 531 :M
f2_12 sf
.081 .008( mirrors natural language.  Second, this)J
60 543 :M
.555 .055(method illustrates the use of two general kinds of parameters that can be passed in )J
471 543 :M
(E)S
f2_10 sf
(XPECT)S
510 543 :M
f2_12 sf
.663 .066(.  They)J
60 555 :M
.54 .054(distinguish between the kind of data that will be provided to the method and the kind of task to be)J
60 568 :M
.103 .01(accomplished by the method.  They are indicated by )J
f3_12 sf
.041(instance-of)A
f2_12 sf
.038 .004( or )J
f3_12 sf
.041(specialization-of)A
f2_12 sf
.059 .006( in)J
60 580 :M
1.464 .146(the capability description.  These keywords indicate how the matcher should match these slots)J
60 592 :M
(against the corresponding slots in goals.)S
60 634 :M
(                                                )S
60 631.48 -.48 .48 204.48 631 .48 60 631 @a
60 643 :M
f2_8 sf
(2)S
64 646 :M
f2_10 sf
.272 .027(One weakness of our current representation for goals and capabilities is that some slots that are attached directly to the)J
60 657 :M
.027 .003(main verb should really be embedded within other parameters.  Figure 3 illustrates that problem, since the \322of\323 slot in the)J
60 668 :M
1.041 .104(capability description of the method should be attached to unloading-time rather than directly to the verb.  We are)J
60 679 :M
(currently developing a more structured representation to support this embedding.)S
endp
%%Page: 8 8
%%BeginPageSetup
initializepage
(swartout; page: 8 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(8)S
gR
gS 0 0 588 768 rC
60 223 :M
f3_12 sf
.667(Instance-of)A
f2_12 sf
1.188 .119( indicates that the slot is a )J
286 223 :M
f1_12 sf
3.192 .319(data parameter)J
f2_12 sf
1.617 .162( and will match an instance of the)J
60 236 :M
2.075 .208(indicated type.  Thus, )J
180 236 :M
f3_12 sf
3.075 .307(\(instance-of movement\))J
f2_12 sf
1.088 .109( will match a movement instance or an)J
60 248 :M
.322 .032(instance that is more specialized.  \(Recall that in Loom, instances are representations of objects that)J
60 260 :M
.883 .088(actually exist, whereas concepts are just descriptions.\)  Instance-of slots work much like function)J
60 272 :M
2.615 .262(parameters in conventional programming languages: they supply the data that the function)J
60 284 :M
.836 .084(manipulates.  However, in E)J
f2_10 sf
.324(XPECT)A
235 284 :M
f2_12 sf
1.143 .114(, slots on goals do more than just provide data; they can also)J
60 297 :M
.283 .028(further specify the task to be done.  )J
f1_12 sf
.623 .062(Task parameters)J
f2_12 sf
.281 .028( are indicated by )J
401 297 :M
f3_12 sf
-.14(specialization-of)A
f2_12 sf
-.097( slots)A
60 309 :M
2.868 .287(and match against )J
164 309 :M
f1_12 sf
3.515 .352(concepts )J
f2_12 sf
2.672 .267(that appear in corresponding slots in the goal.  For example,)J
60 322 :M
f3_12 sf
-.03(\(specialization-of unloading-time\) )A
311 322 :M
f2_12 sf
1.971 .197(will match )J
371 322 :M
f3_12 sf
(unloading-time )S
479 322 :M
f2_12 sf
2.111 .211(or any of its)J
60 334 :M
(specializations.  The capability description of the method in Figure 3 will match a goal such as:)S
96 352 :M
f3_12 sf
-.203(\(find \(obj \(specialization-of unloading-time\)\))A
60 369 :M
-.203(           \(of \(instance movement-23\)\)\))A
60 400 :M
f2_12 sf
(or it will match the goal:)S
96 417 :M
f3_12 sf
-.202(\(find \(obj \(specialization-of emergency-unloading-time\)\))A
60 433 :M
-.203(           \(of \(instance movement-23\)\)\))A
60 464 :M
f2_12 sf
.074 .007(Notice the \322obj\323 slot in both cases does not )J
f1_12 sf
.025(supply)A
f2_12 sf
.061 .006( data but instead it )J
f1_12 sf
.022(specifies)A
f2_12 sf
.083 .008( the sort of information)J
60 477 :M
.895 .09(that is supposed to be found by the method.  )J
288 477 :M
f3_12 sf
.124(Specialization-of)A
f2_12 sf
.256 .026( slots allow us to re-use the)J
60 489 :M
.044 .004(same method in several different contexts, and they are one of the ways we achieve \322loose-coupling\323)J
60 502 :M
.109 .011(in E)J
f2_10 sf
.062(XPECT)A
111 502 :M
f2_12 sf
.274 .027(, which we will discuss in detail in Section 4.2.  In the example in Figure 3, )J
483 502 :M
f3_12 sf
-.097(?t )A
505 502 :M
f2_12 sf
.235 .024(is bound)J
60 515 :M
.603 .06(to the concept in the goal that matches )J
f3_12 sf
2.201 .22(\(specialization-of unloading-time\))J
509 515 :M
f2_12 sf
1.899 .19(. )J
f3_12 sf
2.733 .273( )J
527 515 :M
f2_12 sf
(The)S
60 528 :M
.992 .099(variable )J
104 528 :M
f3_12 sf
.557(?t)A
f2_12 sf
1.165 .116( is then used in the method body to pass the goal context on to subgoals so that the)J
60 541 :M
1.312 .131(method body will compute for example the )J
f3_12 sf
.499(emergency-unloading-time)A
f2_12 sf
1.113 .111( rather than the)J
60 554 :M
f3_12 sf
.164(unloading-time)A
f2_12 sf
.378 .038( if emergency time was specified in the original goal.  Like other variables )J
f3_12 sf
.328(?t)A
60 566 :M
f2_12 sf
.383 .038(can be used in conditional expressions so that methods can be written that branch depending on the)J
60 578 :M
(nature of the task specified.)S
60 595 :M
.471 .047(In most systems, the only parameters passed are data objects that will be used during the execution)J
60 607 :M
.184 .018(of the method.  Distinguishing task parameters adds an additional dimension for method abstraction,)J
60 619 :M
.36 .036(so that methods can be abstracted not only in terms of the data that they will operate on, but also in)J
60 631 :M
1.344 .134(terms of the )J
126 631 :M
f1_12 sf
(classes)S
160 631 :M
f2_12 sf
1.266 .127( of tasks they can address.  This allows us to create a common method that)J
60 643 :M
.648 .065(abstracts both the instances it operates on and the classes of tasks it can be used for and provide a)J
60 655 :M
1.366 .137(single, common skeletal representation that captures both.)J
355 655 :M
f0_12 sf
2.536 .254(  )J
367 655 :M
f2_12 sf
1.624 .162(This helps both method re-use and)J
60 667 :M
(knowledge acquisition by providing an additional way of abstracting methods.)S
51 60 510 141 rC
60 60 1 1 rF
60 60 1 1 rF
61 60 490 1 rF
551 60 1 1 rF
551 60 1 1 rF
60 61 1 13 rF
551 61 1 13 rF
63 84 :M
f3_12 sf
-.297(   )A
84 84 :M
f4_12 sf
-.198(capability:)A
f3_12 sf
-.238( \(find)A
60 74 1 12 rF
551 74 1 12 rF
135 96 :M
-.202(      \(obj \(?t is \(specialization-of unloading-time\)\)\))A
60 86 1 12 rF
551 86 1 12 rF
135 108 :M
-.203(      \(of \(?m is \(instance-of movement\)\)\)\))A
60 98 1 12 rF
551 98 1 12 rF
60 110 1 12 rF
551 110 1 12 rF
63 132 :M
-.297(   )A
84 132 :M
f4_12 sf
-.198(result-type:)A
f3_12 sf
-.208( \(inst-of time-value\))A
60 122 1 12 rF
551 122 1 12 rF
63 144 :M
-.297(   )A
84 144 :M
f4_12 sf
-.198(method:)A
f3_12 sf
-.218( \(calculate)A
60 134 1 12 rF
551 134 1 12 rF
171 156 :M
-.227(\(obj ?t\))A
60 146 1 12 rF
551 146 1 12 rF
171 168 :M
-.205(\(of \(available-vehicles ?m\)\)\))A
60 158 1 12 rF
551 158 1 12 rF
60 183 1 1 rF
60 183 1 1 rF
61 183 490 1 rF
551 183 1 1 rF
551 183 1 1 rF
60 170 1 13 rF
551 170 1 13 rF
146 198 :M
f2_12 sf
(Figure 3. A method to calculate the unloading time of a movement)S
endp
%%Page: 9 9
%%BeginPageSetup
initializepage
(swartout; page: 9 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
300 732 6 12 rC
300 741 :M
f2_12 sf
(9)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.196 .02(The key features of our goal representation are: 1\) it is structured, 2\) the structure is based on a verb)J
60 81 :M
2.644 .264(clause representation, and 3\) in addition to data parameters, task parameters are explicitly)J
60 93 :M
1.547 .155(distinguished.  Our representation captures more of the intent behind a goal than conventional)J
60 105 :M
.394 .039(representations.  Thus it provides more of the information that acquisition tools need to analyze the)J
60 117 :M
.313 .031(knowledge base and capture interdependencies. Our representation also structures goals and method)J
60 129 :M
.053 .005(capabilities in a way that is quite close to the way people express them in natural language, making it)J
60 141 :M
(easier to build tools that explain methods and help users modify them.)S
60 177 :M
f0_12 sf
(4. B)S
f0_10 sf
(RINGING )S
f0_12 sf
(K)S
f0_10 sf
(NOWLEDGE )S
199 177 :M
f0_12 sf
(B)S
f0_10 sf
(ACK )S
232 177 :M
f0_12 sf
(T)S
f0_10 sf
(OGETHER)S
291 177 :M
f0_12 sf
(: L)S
306 177 :M
f0_10 sf
(OOSE )S
f0_12 sf
(C)S
345 177 :M
f0_10 sf
(OUPLING IN  )S
f0_12 sf
(E)S
f0_10 sf
(XPECT)S
60 194 :M
f2_12 sf
3.391 .339(We have argued that by separating different kinds of knowledge and providing explicit)J
60 206 :M
.984 .098(representations for them knowledge based systems created within the )J
412 206 :M
(E)S
f2_10 sf
(XPECT)S
451 206 :M
f2_12 sf
1.12 .112( framework can be)J
60 218 :M
.886 .089(easier to understand, and because the separation provides increased modularity, they are easier to)J
60 230 :M
.491 .049(augment and modify.  In this section, we describe how the program instantiator works to match up)J
60 242 :M
.481 .048(and integrate different knowledge resources.  We refer to the matching process we use as )J
f1_12 sf
.163(semantic)A
60 254 :M
.308(match)A
f2_12 sf
.902 .09( because resources are matched up based on their )J
f1_12 sf
.31(meaning)A
f2_12 sf
.796 .08( as opposed to their syntax.  We)J
60 266 :M
1.486 .149(argue that this loosens the coupling between resources, which can have distinct advantages for)J
60 278 :M
(knowledge re-use and acquisition.)S
60 302 :M
f0_12 sf
(4.1 Resolving Goals: Semantic Matching And Goal Reformulation)S
60 331 :M
f2_12 sf
.815 .082(In many systems, matching of goals and methods is done on a fairly syntactic basis.  Lexemes in)J
60 343 :M
.913 .091(goals must match those in methods, and variables are matched by position.  In )J
458 343 :M
(E)S
f2_10 sf
(XPECT)S
497 343 :M
f2_12 sf
1.011 .101(, we have)J
60 355 :M
.255 .026(tried to move toward a matching process that is based on the semantics of the goals rather than their)J
60 367 :M
.246 .025(syntax, and one in which reformulation can be used to achieve a match when a more direct match is)J
60 379 :M
2.669 .267(not possible.  In )J
155 379 :M
(E)S
f2_10 sf
(XPECT)S
194 379 :M
f2_12 sf
2.319 .232(, the ability to provide looser coupling between goals and method)J
60 391 :M
1.104 .11(capabilities depends on the way that they are represented.  As we described in Section 3.2, both)J
60 403 :M
.04 .004(goals and method capabilities are represented as verb clauses.  A main verb states what is to be done,)J
60 415 :M
1.185 .119(using a number of slots that act as \322cases\323 \(as in case grammar\).  For matching, both goals and)J
60 427 :M
(method capabilities are translated into Loom concepts that mirror their structure.)S
60 455 :M
f0_12 sf
.433 .043(Semantic Match)J
145 455 :M
f2_12 sf
.691 .069( One of the mechanisms )J
269 455 :M
(E)S
f2_10 sf
(XPECT)S
308 455 :M
f2_12 sf
.627 .063( provides to achieve looser coupling is based on)J
60 467 :M
.507 .051(Loom\325s classifier \(one of the reasoners that Loom provides\).  Given an existing hierarchy of Loom)J
60 479 :M
.824 .082(concepts \(and their definitions\) organized according to the subsumption relations \(AKO\) between)J
60 491 :M
.638 .064(them, the classifier is capable of figuring out where in the hierarchy a new concept belongs, based)J
60 503 :M
.654 .065(solely on the definitions of the concepts.   To find possible methods for accomplishing a goal, the)J
60 515 :M
.345 .035(Loom classifier is used to find those methods whose capability descriptions subsume the goal.  The)J
60 527 :M
.685 .069(classifier provides a form of semantic match, because match is based on the meaning of concepts,)J
60 539 :M
.356 .036(not on their syntactic form.  Section 3.2 described how goals are represented as Loom concepts and)J
60 551 :M
(showed some simple examples of how the semantic matcher works.)S
60 568 :M
( )S
63 568 :M
.076 .008(The semantic matcher finds methods whose capabilities subsume the posted goal not only when it is)J
60 580 :M
.8 .08(given the class name but also when it is given a description of the type of object that needs to be)J
60 592 :M
.271 .027(matched.  An example of this kind of matching occurs in our network diagnosis domain.  )J
498 592 :M
(E)S
f2_10 sf
(XPECT)S
537 592 :M
f2_12 sf
<D573>S
60 604 :M
(terminological knowledge contains the definition:)S
96 629 :M
f3_12 sf
-.206(lanbridge-23 is a COMPONENT)A
240 641 :M
-.205(that is CONNECTED-TO 2 networks)A
96 665 :M
-.204(CONNECTED-COMPONENT is a COMPONENT)A
240 677 :M
-.204(that is CONNECTED-TO some NETWORK)A
endp
%%Page: 10 10
%%BeginPageSetup
initializepage
(swartout; page: 10 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(10)S
gR
gS 0 0 588 768 rC
60 81 :M
f2_12 sf
(When )S
92 81 :M
(E)S
f2_10 sf
(XPECT)S
131 81 :M
f2_12 sf
( is reasoning with its problem solving knowledge about how to achieve the goal:)S
96 106 :M
f3_12 sf
-.203(\(diagnose \(obj \(instance lanbridge-23\)\)\))A
60 129 :M
f2_12 sf
(It will be able to match that goal with a method with the capability description:)S
96 154 :M
f3_12 sf
-.202(\(diagnose \(obj \(?c is \(instance-of CONNECTED-COMPONENT\)\)\)\))A
60 178 :M
f2_12 sf
2.245 .225(because the classifier can figure out that )J
281 178 :M
f3_12 sf
.675(lanbridge-23)A
f2_12 sf
1.452 .145( is a kind of component which is)J
60 191 :M
3.54 .354(connected to a network, based on the definitions of )J
358 191 :M
f3_12 sf
.267(lanbridge-23)A
f2_12 sf
.345 .035( and )J
f3_12 sf
.296(CONNECTED-)A
60 204 :M
.453 .045(COMPONENT )J
133 204 :M
f2_12 sf
1.777 .178(above.  This kind of subsumption matching allows )J
402 204 :M
(E)S
f2_10 sf
(XPECT)S
441 204 :M
f2_12 sf
1.928 .193( to reason about the)J
60 216 :M
(semantics of a goal in terms of the )S
228 216 :M
f1_12 sf
(meaning)S
f2_12 sf
( of its parameters.)S
60 245 :M
f0_12 sf
1.717 .172(Reformulations )J
f2_12 sf
1.087 .109(The second mechanism that )J
291 245 :M
(E)S
f2_10 sf
(XPECT)S
330 245 :M
f2_12 sf
1.433 .143( provides for looser coupling of goals and)J
60 257 :M
.847 .085(methods is reformulation.  Usually if a method cannot be found to achieve a goal using semantic)J
60 269 :M
1.109 .111(match, the system will attempt to reformulate the goal and then look for methods to achieve the)J
60 281 :M
.347 .035(resulting goal\(s\).  Goal reformulation involves decomposing a goal and then assembling a new goal)J
60 293 :M
.148 .015(\(or goals\) by transforming pieces of the original goal based on their meaning.  To be able to perform)J
60 305 :M
.493 .049(goal reformulation, one needs an explicit, decomposable representation for the goal, definitions for)J
60 317 :M
(the terms the goal is constructed from, and domain facts to drive the reformulation process.)S
60 334 :M
2.927 .293(In )J
77 334 :M
(E)S
f2_10 sf
(XPECT)S
116 334 :M
f2_12 sf
2.035 .203(, we provide two general types of reformulations, )J
f1_12 sf
2.923 .292(conjunctive )J
f2_12 sf
1.724 .172(and )J
469 334 :M
f1_12 sf
(disjunctive.)S
524 334 :M
f2_12 sf
3.193 .319(  A)J
60 346 :M
.803 .08(conjunctive reformulation involves transforming some goal into a set)J
404 346 :M
f1_12 sf
1.475 .147( )J
409 346 :M
f2_12 sf
.95 .095(of goals, where )J
f1_12 sf
1.001 .1(each of the)J
60 358 :M
.313(goals)A
f2_12 sf
.917 .092(  in the set must be performed to achieve the intent of the initial goal.  Thus, a conjunctive)J
60 370 :M
.921 .092(reformulation is a form of divide-and-conquer: it splits a problem up into subpieces that together)J
60 382 :M
.2 .02(achieve the original goal.  As in divide-and-conquer, the system must find a way of recombining the)J
60 394 :M
1.942 .194(results of each of the subproblems back into an appropriate result for the original goal for a)J
60 406 :M
.999 .1(conjunctive reformulation to be successful.  A disjunctive reformulation may also reformulate an)J
60 418 :M
.213 .021(initial goal into several goals, but  at runtime, )J
284 418 :M
f1_12 sf
.22 .022(only one of the goals)J
386 418 :M
f2_12 sf
.226 .023(  needs to be executed to achieve)J
60 430 :M
(the intent of the original goal.)S
60 447 :M
(E)S
f2_10 sf
(XPECT)S
99 447 :M
f2_12 sf
1.357 .136( provides three types of conjunctive reformulations: covering, individualization, and set)J
60 459 :M
(reformulations.)S
60 476 :M
1.276 .128(A )J
f1_12 sf
2.172 .217(covering  )J
128 476 :M
f2_12 sf
1.582 .158(reformulation occurs when a goal can be transformed into several new goals that)J
60 488 :M
.411 .041(together \322cover\323 the intent of the initial goal.  Suppose that the following goal is posted to estimate)J
60 500 :M
(how many support personnel are required for a COA:)S
96 525 :M
f3_12 sf
-.202(\(estimate \(obj \(specification-of support-personnel\))A
168 537 :M
-.207(\(for \(instance-of coa\)\)\))A
60 553 :M
f2_12 sf
-.004(Using subsumption matching, the system would try to find methods for achieving this goal.  Suppose)A
60 565 :M
.836 .084(none were found, because the system had no general method for estimating the support personnel)J
60 577 :M
.703 .07(needed for a COA.  Suppose, however, that the system did have methods for estimating particular)J
60 589 :M
1.234 .123(types of support personnel needed for a COA.  How could these methods be found?  When the)J
60 601 :M
.417 .042(system failed initially to find a method, it would then try to reformulate the goal into new goals.  If)J
60 613 :M
(the domain model contained the fact:)S
96 638 :M
f3_12 sf
-.204(support-personnel is partitioned by)A
132 650 :M
-.202(airport-support-personnel and seaport-support-personnel)A
endp
%%Page: 11 11
%%BeginPageSetup
initializepage
(swartout; page: 11 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(11)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
(the system could reformulate the original goal into two new goals:)S
96 94 :M
f3_12 sf
-.202(\(estimate \(obj \(specification-of airport-support-personnel\))A
168 106 :M
-.207(\(for \(instance-of coa\)\)\))A
60 117 :M
f2_12 sf
(and)S
96 130 :M
f3_12 sf
-.202(\(estimate \(obj \(specification-of seaport-support-personnel\))A
168 142 :M
-.207(\(for \(instance-of coa\)\)\))A
60 170 :M
f2_12 sf
.831 .083(The system would then be able to find the two methods for estimating particular types of support)J
60 182 :M
1.017 .102(personnel needed for a COA.   For conjunctive reformulations, part of the reformulation process)J
60 194 :M
.843 .084(involves finding a function for re-combining the results of each of the reformulations to form the)J
60 206 :M
.133 .013(result for the original goal.  The appropriate function for combining results is determined by the type)J
60 218 :M
(of the goal.  In this example, the system adds the estimates together.)S
60 235 :M
1.175 .117(This sort of reformulation process reduces the need for different parts of the knowledge base to)J
60 247 :M
1.338 .134(match up exactly, which enhances the possibilities for knowledge re-use across systems.  Also,)J
60 259 :M
.456 .046(because the system explicitly reasons through the reformulation process, more of the design can be)J
60 271 :M
.959 .096(captured to support knowledge acquisition.  In this case, if the user added a new type of support)J
60 284 :M
1.049 .105(personnel to the knowledge base, for example )J
f3_12 sf
.389(unloading-support-personnel)A
f2_12 sf
.826 .083(, then the)J
60 296 :M
.25 .025(system would use its record of this reformulation to determine that it would be necessary to perform)J
60 308 :M
.091 .009(the reformulation over again to capture the new type of support personnel.  )J
425 308 :M
(E)S
f2_10 sf
(XPECT)S
464 308 :M
f2_12 sf
.092 .009( could detect that)J
60 320 :M
(the user needs to provide a method for estimating the law enforcement personnel needed for a COA.)S
60 337 :M
f1_12 sf
.21 .021(Individualization  )J
f2_12 sf
.216 .022(reformulations are similar to covering reformulations, except that they decompose)J
60 349 :M
.357 .036(a goal over a set of objects into a set of goals over individual objects \(i.e., instances\).  For example,)J
60 361 :M
(given the goal of calculating the employment personnel of the force modules in a COA:)S
96 386 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of employment-personnel\)\))A
132 398 :M
( )S
168 398 :M
-.206( \(of \(force-modules coa-2\)\)\))A
60 421 :M
f2_12 sf
(and the domain fact that:)S
96 446 :M
f3_12 sf
-.203(force-modules of coa-2 are the instances:)A
132 458 :M
-.206(\0003rd-ACR 57th-IMF CVN71-ACN)A
67 470 :M
-.264(    )A
60 481 :M
f2_12 sf
(the system could transform the original goal into three goals:)S
96 506 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of employment-personnel\)\))A
132 518 :M
( )S
168 518 :M
-.207( \(of \(instance 3rd-ACR\)\)\))A
96 542 :M
-.202(\(calculate \(obj \(specification-of employment-personnel\)\))A
132 554 :M
( )S
168 554 :M
-.206( \(of \(instance 57th-IMF\)\)\))A
96 578 :M
-.202(\(calculate \(obj \(specification-of employment-personnel\)\))A
132 590 :M
( )S
168 590 :M
-.206( \(of \(instance CVN71-ACN\)\)\))A
60 618 :M
f2_12 sf
(where each of these goals corresponds to one of the instances force modules of coa-2.)S
60 652 :M
.492 .049(Individualization reformulations are used when the set of objects denoted by a concept is known at)J
60 664 :M
1.22 .122(program instantiation time.  Sometimes, however, it may be known that a set of objects will be)J
60 676 :M
.019 .002(passed to a method, but the elements of that set may not be known at program instantiation time, that)J
endp
%%Page: 12 12
%%BeginPageSetup
initializepage
(swartout; page: 12 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(12)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.55 .055(is, they may only be known later, at runtime.  To deal with that situation, E)J
f2_10 sf
.275(XPECT)A
465 69 :M
f2_12 sf
.591 .059( provides a third)J
60 81 :M
-.004(kind of conjunctive reformulation, the )A
246 81 :M
f1_12 sf
(set)S
f2_12 sf
-.005( reformulation.  When no method is found to achieve a goal)A
60 93 :M
-.005(that has a set of objects in its parameters, )A
260 93 :M
(E)S
f2_10 sf
(XPECT)S
299 93 :M
f2_12 sf
-.004( tries to solve the goal, at runtime, for each element)A
60 105 :M
.831 .083(of the set in turn.  For example, suppose that the following goal is posted to calculate the closure)J
60 117 :M
(date)S
80 114 :M
f2_8 sf
(3)S
84 117 :M
f2_12 sf
( for several movements:)S
96 147 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of closure-date\)\))A
168 159 :M
-.204( \(of \(set-of \(instance-of movement\)\)\)\))A
60 182 :M
f2_12 sf
.246 .025(and there are no methods that operate on a set of movements.  E)J
f2_10 sf
.123(XPECT)A
404 182 :M
f2_12 sf
.255 .026( reformulates this goal over a)J
60 194 :M
(set into a goal over an individual movement:)S
96 219 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of closure-date\)\))A
168 231 :M
-.205( \(of \(instance-of movement\)\)\))A
60 254 :M
f2_12 sf
.737 .074(The matcher will return the method for calculating the closure date of a movement.  At execution)J
60 266 :M
-.002(time, the system will loop over each of the movements in the set and calculate one by one the closure)A
60 278 :M
(date of the movements that are included in the set.)S
60 302 :M
.155 .016(Finally, E)J
f2_10 sf
.052(XPECT)A
139 302 :M
f2_12 sf
.362 .036( provides the )J
206 302 :M
f1_12 sf
.399 .04(input  )J
238 302 :M
f2_12 sf
.296 .03(reformulation, which is a form of )J
f1_12 sf
.3 .03(disjunctive  )J
f2_12 sf
.38 .038(reformulation.  It)J
60 314 :M
.344 .034(occurs when no method can be found to handle one of the inputs to a goal, but several methods can)J
60 326 :M
1.661 .166(be found that together will cover the range of possible inputs that will occur at runtime.  For)J
60 338 :M
(example, given the goal:)S
96 363 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of closure-date\)\))A
168 375 :M
-.205( \(of \(instance-of movement\)\)\))A
60 398 :M
f2_12 sf
.195 .02(suppose that the method library contained no method for calculating the closure date of a movement)J
60 410 :M
.339 .034(in general, but there were methods for calculating the closure date of particular types of movements)J
60 422 :M
(and there was a domain fact that told the system that:)S
96 447 :M
f3_12 sf
-.203(movement is partitioned by airlift-movement and)A
276 459 :M
-.21(  sealift-movement)A
60 482 :M
f2_12 sf
(then the system could create the goals:)S
96 507 :M
f3_12 sf
-.202(\(calculate \(obj \(specification-of closure-date\)\))A
168 519 :M
-.204( \(of \(instance-of airlift-movement\)\)\))A
96 543 :M
-.202(\(calculate \(obj \(specification-of closure-date\)\))A
168 555 :M
-.204( \(of \(instance-of sealift-movement\)\)\))A
60 578 :M
f2_12 sf
1.062 .106(the system would expand methods for each of these goals, and then create dispatching code that)J
60 590 :M
.802 .08(would select which method to use at runtime, based on the type of the instance of movement that)J
60 602 :M
.933 .093(was actually passed in.  Note that unlike a covering reformulation, only one of the branches of a)J
60 614 :M
(disjunctive reformulation needs to be executed.)S
60 631 :M
.125 .013(In summary, the loose coupling that )J
238 631 :M
(E)S
f2_10 sf
(XPECT)S
277 631 :M
f2_12 sf
.113 .011( provides through semantic match and reformulations is)J
60 643 :M
.516 .052(crucial to our approach to knowledge acquisition.  First, by moving away from syntactic matching,)J
60 655 :M
.054 .005(users can add knowledge to a system without being as concerned with issues of form.  This opens up)J
60 667 :M
(                                                )S
60 664.48 -.48 .48 204.48 664 .48 60 664 @a
60 676 :M
f2_8 sf
(3)S
64 679 :M
f2_10 sf
(The closure date is the date when all the material to be shipped has arrived at the destination.)S
endp
%%Page: 13 13
%%BeginPageSetup
initializepage
(swartout; page: 13 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(13)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.343 .034(the possibility for greater knowledge re-use and eases collaborative work on knowledge bases.  The)J
60 81 :M
1.01 .101(second benefit is that by having the program instantiator reason extensively about the process of)J
60 93 :M
1.972 .197(matching up goals and methods, more of the design of the knowledge based system and the)J
60 105 :M
2.936 .294(interdependencies between parts of the system can be captured \(and hence, used to form)J
60 117 :M
.388 .039(expectations for knowledge acquisition\).   In reformulating goals, the program instantiator develops)J
60 129 :M
.992 .099(a rationale for how a high-level goal can be achieved in terms of lower level goals.  This sort of)J
60 141 :M
1.101 .11(processing is often exactly the sort of reasoning that one wants to explain and use as a basis for)J
60 153 :M
(knowledge acquisition.)S
60 177 :M
f0_12 sf
(4.2 The Program Instantiator: Capturing Interdependencies)S
60 194 :M
f2_12 sf
1.944 .194(The )J
85 194 :M
(E)S
f2_10 sf
(XPECT)S
124 194 :M
f2_12 sf
1.501 .15( program instantiator works in a refinement driven fashion.  Initially, the program)J
60 206 :M
.219 .022(instantiator starts with a high level goal that specifies what the expert system is supposed to do.  For)J
60 218 :M
.329 .033(example, to create an expert system for evaluating a particular COA, the system would be given the)J
60 230 :M
(following goal:)S
132 255 :M
f3_12 sf
-.203(\(evaluate \(obj \(instance-of deployment-coa\)\)\))A
60 283 :M
f2_12 sf
.589 .059(This high level goal determines the scope of the knowledge based system that E)J
f2_10 sf
.253(XPECT)A
488 283 :M
f2_12 sf
.637 .064( will create.)J
60 295 :M
.646 .065(The goal above would create a knowledge based system that could evaluate a deployment-COA\321)J
60 307 :M
(but nothing else.  On the other hand, a goal such as:)S
132 332 :M
f3_12 sf
-.204(\(evaluate \(obj \(instance-of coa\)\)\))A
60 364 :M
f2_12 sf
2.058 .206(would create a knowledge based system that could evaluate any COA that was in )J
498 364 :M
(E)S
f2_10 sf
(XPECT)S
537 364 :M
f2_12 sf
<D573>S
60 376 :M
1.066 .107(knowledge base \(assuming appropriate problem solving knowledge was also available\).  Thus, a)J
60 388 :M
.418 .042(single E)J
f2_10 sf
.153(XPECT)A
132 388 :M
f2_12 sf
.873 .087( knowledge base can be used to create a variety of knowledge based systems, each)J
60 400 :M
(scoped to cover a different \(or possibly overlapping\) set of problems.)S
60 417 :M
1.065 .106(References to instances that appear in goals during the program instantiation phase act as \322place)J
60 429 :M
.153 .015(holders\323 for the actual data object that will appear when the program is executed.  The use of a more)J
60 441 :M
.098 .01(general or abstract instance results in the creation of a system that can handle a wide range of inputs,)J
60 453 :M
(i.e., all the instances of that type.)S
60 470 :M
.864 .086(When a goal is posted, the program instantiator searches its problem solving knowledge to find a)J
60 482 :M
.927 .093(method whose capability description matches the goal.  This matching of goals and methods is a)J
60 494 :M
.526 .053(critical step in the program instantiator\325s reasoning and was the main topic of the previous section.)J
60 506 :M
.527 .053(How it is done, and the representations that are used, have a direct effect on how maintainable and)J
60 518 :M
(reusable the knowledge base will be and )S
257 518 :M
(E)S
f2_10 sf
(XPECT)S
296 518 :M
f2_12 sf
(\325s ability to acquire new knowledge.)S
60 535 :M
1.557 .156(Once a method is selected to achieve the posted goal, the variables in the method\325s capability)J
60 547 :M
.318 .032(description are bound to corresponding instances and concepts in the goal.  The body of the method)J
60 559 :M
.393 .039(is expanded by plugging in the bindings for the variables in the body and then posting its subgoals.)J
60 571 :M
.253 .025(During this process, if any of the slots of an instance are accessed by the method, those accesses are)J
60 583 :M
.941 .094(recorded in the design history.   This record of the interdependencies between factual knowledge)J
60 595 :M
2.069 .207(\(instances\) and problem solving knowledge is later used to form the expectations that guide)J
60 607 :M
1.065 .106(knowledge acquisition.  For example, if the program instantiator notes that it uses a method that)J
60 619 :M
1.2 .12(requires information about the berths of a port, then when a new port is entered, the knowledge)J
60 631 :M
(acquisition routines will know that information about the berths of that new port needs to be added.)S
endp
%%Page: 14 14
%%BeginPageSetup
initializepage
(swartout; page: 14 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(14)S
gR
.852 G
gS 120 381 116 119 rC
-4 -4 113 512 4 4 243 370 @b
gR
gS 43 44 469 633 rC
-1 -1 227 382 1 1 246 370 @b
226 382 -1 1 237 390 1 226 381 @a
-1 -1 237 391 1 1 246 370 @b
np 246 370 :M
226 381 :L
236 390 :L
246 370 :L
eofill
-1 -1 110 512 1 1 129 500 @b
120 492 -1 1 130 500 1 120 491 @a
-1 -1 110 512 1 1 120 491 @b
np 109 511 :M
129 500 :L
120 491 :L
109 511 :L
eofill
0 G
74.5 289.5 317 86 rS
gS
.113 .113 scale
715.999 2650.775 :M
f3_72 sf
(goal: \(calculate \(obj \(spec-of employment-personnel\)\))S
gR
gS
.113 .113 scale
715.999 2730.299 :M
f3_72 sf
(                 \(of \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
715.999 2818.658 :M
f3_72 sf
(achieved-by: \(calculate \(obj \(?p is \(spec-of )S
gR
gS
.113 .113 scale
715.999 2907.017 :M
f3_72 sf
(                                      employment-personnel\)\))S
gR
gS
.113 .113 scale
715.999 2995.376 :M
f3_72 sf
(                 \(of \(?c is \(inst-of coa\)\)\))S
gR
gS
.113 .113 scale
715.999 3074.899 :M
f3_72 sf
(bindings: \(\(?p \(spec-of employment personnel\)\))S
gR
gS
.113 .113 scale
715.999 3163.258 :M
f3_72 sf
(           \(?c \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
715.999 3251.618 :M
f3_72 sf
(method: \(add \(obj \(personnel \()S
gR
gS
.113 .113 scale
2015.404 3251.618 :M
f5_72 sf
(force-modules)S
gR
229 370 -1 1 293 369 1 229 369 @a
gS
.113 .113 scale
2581.132 3251.618 :M
f3_72 sf
( ?c\)\)\))S
gR
-1 -1 262 290 1 1 298 270 @b
-1 -1 323 126 1 1 436 111 @b
1 G
356 109 29 17 rF
gS
.113 .113 scale
3208.736 1051.474 :M
0 G
f3_72 sf
(...)S
gR
0 G
216.5 53.5 264 57 rS
gS
.113 .113 scale
1971.207 565.499 :M
f3_72 sf
(goal: \(evaluate \(obj \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1971.207 653.858 :M
f3_72 sf
(achieved-by: \(evaluate \(obj \(?c is \(inst-of coa\)\)\)\))S
gR
gS
.113 .113 scale
1971.207 742.217 :M
f3_72 sf
(bindings: \(\(?c \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1971.207 830.576 :M
f3_72 sf
(method: \(evaluate \(obj transportation-factors\))S
gR
gS
.113 .113 scale
1971.207 910.1 :M
f3_72 sf
(                  \(of ?c\)\))S
gR
156.5 126.5 255 37 rS
gS
.113 .113 scale
1440.837 1210.521 :M
f3_72 sf
(goal: \(estimate \(obj \(spec-of support-personnel\)\))S
gR
gS
.113 .113 scale
1440.837 1298.88 :M
f3_72 sf
(                \(for \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1440.837 1378.403 :M
f3_72 sf
(achieved-by: covering-reformulation obj)S
gR
137.5 232.5 294 37 rS
gS
.113 .113 scale
1272.887 2147.128 :M
f3_72 sf
(goal: \(estimate \(obj \(spec-of seaport-support-personnel\)\))S
gR
gS
.113 .113 scale
1272.887 2235.487 :M
f3_72 sf
(                \(for \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1272.887 2315.01 :M
f3_72 sf
(achieved-by: \(estimate \(obj ?ssp is \(...)S
gR
367 165 -1 1 402 230 1 367 164 @a
-1 -1 312 176 1 1 366 164 @b
1 G
68 176 295 39 rF
0 G
68.5 176.5 294 38 rS
gS
.113 .113 scale
662.962 1661.152 :M
f3_72 sf
(goal: \(estimate \(obj \(spec-of airport-support-personnel\)\))S
gR
gS
.113 .113 scale
662.962 1740.676 :M
f3_72 sf
(                \(for \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
662.962 1829.035 :M
f3_72 sf
(achieved-by: \(estimate \(obj ?asp is \(...)S
gR
gS
.113 .113 scale
530.37 4674.2 :M
f3_72 sf
(coa)S
gR
gS
.113 .113 scale
565.728 5345.73 :M
f3_72 sf
(deployment-coa)S
gR
gS
.113 .113 scale
760.196 4824.411 :M
f5_72 sf
(movements)S
gR
86 548 -1 1 131 547 1 86 547 @a
gS
.113 .113 scale
662.962 4612.349 :M
f5_72 sf
(force-modules)S
gR
76 524 -1 1 140 523 1 76 523 @a
2 lw
49 493 137 134 rS
192 493 319 134 rS
gS
.113 .113 scale
1962.367 4727.216 :M
f3_56 sf
(capability: \(calculate \(obj \(?d is \(spec-of closure-date\)\)\))S
gR
gS
.113 .113 scale
1962.367 4789.067 :M
f3_56 sf
(                       \(of \(?c is \(inst-of coa\)\)\)\))S
gR
gS
.113 .113 scale
1962.367 4859.755 :M
f3_56 sf
(result-type: \(inst-of time-value\))S
gR
gS
.113 .113 scale
1962.367 4930.442 :M
f3_56 sf
(method: \(calculate \(obj ?d\))S
gR
gS
.113 .113 scale
1962.367 4992.293 :M
f3_56 sf
(                   \(of \(movements ?c\)\)\))S
gR
1 lw
220.5 524.5 448.5 569.5 5.5 @s
1 G
np 479 498 :M
476 501 :L
476 506 :L
479 509 :L
495 509 :L
498 506 :L
498 501 :L
495 498 :L
479 498 :L
eofill
0 G
-1 -1 477 502 1 1 479 498 @b
-1 -1 477 506 1 1 476 501 @b
476 506 -1 1 480 509 1 476 505 @a
479 510 -1 1 495 509 1 479 509 @a
-1 -1 495 510 1 1 498 505 @b
-1 -1 499 506 1 1 498 501 @b
494 499 -1 1 499 501 1 494 498 @a
479 499 -1 1 495 498 1 479 498 @a
21 17 66 529 @f
89 16 101 603.5 @f
17 10 124 532.5 @f
19 12 118 560.5 @f
43 44 66 633 rC
77 528 -1 1 117 532 1 77 527 @a
gR
0 G
gS 43 44 469 633 rC
103 528 -1 1 117 532 1 103 527 @a
-1 -1 103 535 1 1 103 527 @b
-1 -1 103 535 1 1 116 532 @b
np 116 532 :M
103 527 :L
102 534 :L
116 532 :L
eofill
43 44 59 514 rC
71 536 -1 1 109 562 1 71 535 @a
gR
gS 43 44 469 633 rC
100 553 -1 1 110 563 1 100 552 @a
-1 -1 97 559 1 1 100 552 @b
96 559 -1 1 110 563 1 96 558 @a
np 109 563 :M
100 552 :L
96 558 :L
109 563 :L
eofill
43 547 469 130 rC
62 539 -1 1 71 597 1 62 538 @a
64 539 -1 1 73 597 1 64 538 @a
gR
gS 43 44 469 633 rC
-1 -1 62 558 1 1 64 538 @b
-1 -1 62 558 1 1 72 556 @b
64 539 -1 1 73 556 1 64 538 @a
np 64 538 :M
61 557 :L
72 556 :L
64 538 :L
eofill
43 44 428 633 rC
-1 -1 435 523 1 1 480 509 @b
-1 -1 435 525 1 1 480 511 @b
gR
gS 43 44 469 633 rC
461 511 -1 1 481 510 1 461 510 @a
461 511 -1 1 464 521 1 461 510 @a
-1 -1 464 522 1 1 480 510 @b
np 480 510 :M
461 510 :L
463 521 :L
480 510 :L
eofill
43 515 440 162 rC
-1 -1 473 532 1 1 486 506 @b
-1 -1 475 534 1 1 488 508 @b
gR
gS 43 44 469 633 rC
-1 -1 475 520 1 1 488 506 @b
474 520 -1 1 484 525 1 474 519 @a
-1 -1 484 526 1 1 488 506 @b
np 488 506 :M
474 520 :L
484 525 :L
488 506 :L
eofill
gS
.113 .113 scale
857.431 5504.777 :M
f5_72 sf
(TERMINOLOGICAL KB)S
gR
gS
.113 .113 scale
1900.491 5504.777 :M
.852 G
f5_72 sf
0 G
(PROBLEM SOLVING METHODS)S
gR
.852 G
141 44 165 633 rC
-4 -1 131 547 4 1 317 464 @b
gR
.852 G
gS 43 44 469 633 rC
-1 -1 296 469 1 1 317 466 @b
295 469 -1 1 301 480 1 295 468 @a
-1 -1 301 481 1 1 317 466 @b
np 317 466 :M
295 468 :L
300 480 :L
317 466 :L
eofill
-1 -1 131 546 1 1 152 542 @b
147 532 -1 1 153 542 1 147 531 @a
-1 -1 131 546 1 1 147 531 @b
np 130 545 :M
153 542 :L
148 531 :L
130 545 :L
eofill
.568 G
43 386 469 181 rC
131 376 -4 1 220 578 4 131 375 @a
gR
.568 G
gS 43 44 469 633 rC
133 377 -1 1 136 398 1 133 376 @a
-1 -1 136 399 1 1 147 393 @b
133 377 -1 1 148 393 1 133 376 @a
np 133 376 :M
135 398 :L
147 393 :L
133 376 :L
eofill
215 557 -1 1 219 578 1 215 556 @a
-1 -1 204 562 1 1 215 556 @b
203 562 -1 1 219 578 1 203 561 @a
np 218 578 :M
215 556 :L
203 561 :L
218 578 :L
eofill
0 G
2 lw
47 45 463 439 rS
gS
.113 .113 scale
910.468 5062.981 :M
f3_72 sf
(jscp)S
gR
1 lw
19 11 117 585 @f
43 44 59 533 rC
65 538 -1 1 108 583 1 65 537 @a
gR
0 G
gS 43 44 469 633 rC
102 572 -1 1 109 584 1 102 571 @a
-1 -1 97 577 1 1 102 571 @b
96 577 -1 1 109 584 1 96 576 @a
np 108 584 :M
102 571 :L
96 576 :L
108 584 :L
eofill
gS
.113 .113 scale
495.012 3993.835 :M
f5_72 sf
(DOMAIN-SPECIFIC KBS)S
gR
gS
.113 .113 scale
495.012 4082.194 :M
f5_72 sf
(        +)S
gR
gS
.113 .113 scale
495.012 4161.717 :M
f5_72 sf
(DESIGN HISTORY)S
gR
gS
.113 .113 scale
450.814 5628.479 :M
f2_80 sf
2.67 .267(Figure 4.  During problem solving, the program instantiator integrates the different types of knowledge that the diffferent )J
gR
gS
.113 .113 scale
450.814 5725.675 :M
f2_80 sf
5.093 .509(knowledge bases contain and keeps track of how they are used \(use of problem solving methods is in dark grey, use of )J
gR
gS
.113 .113 scale
450.814 5822.87 :M
f2_80 sf
5.589 .559(aspects of domain entities is in light grey\).  These interdependencies, dynamically generated by EXPECT based on the )J
gR
gS
.113 .113 scale
450.814 5920.065 :M
f2_80 sf
(current available knowledge, are the basis for guiding knowledge acquisition.  )S
gR
gS
.113 .113 scale
1697.183 556.663 :M
f5_96 sf
(n1)S
gR
gS
.113 .113 scale
1211.011 1175.177 :M
f5_96 sf
(n2)S
gR
gS
.113 .113 scale
1051.9 2120.62 :M
f5_96 sf
(n4)S
gR
gS
.113 .113 scale
1529.232 3490.187 :M
f5_96 sf
(n6)S
gR
gS
.113 .113 scale
459.654 1634.645 :M
f5_96 sf
(n3)S
gR
gS
.113 .113 scale
512.691 2633.103 :M
1 G
f5_96 sf
0 G
(n5)S
gR
1 G
270 274 26 13 rF
gS
.113 .113 scale
2439.7 2482.893 :M
0 G
f3_40 sf
(...)S
gR
0 G
435 113 -1 1 474 383 1 435 112 @a
1 G
440 226 29 17 rF
gS
.113 .113 scale
3942.414 2085.277 :M
0 G
f3_72 sf
(...)S
gR
gS
.113 .113 scale
1865.133 5177.848 :M
0 G
f3_56 sf
(capab: \(calculate \(obj \(?d is \(spec-of employment-personnel\)\)\))S
gR
gS
.113 .113 scale
1865.133 5239.699 :M
0 G
f3_56 sf
(                      \(of \(?c is \(inst-of coa\)\)\)\))S
gR
gS
.113 .113 scale
1865.133 5310.386 :M
0 G
f3_56 sf
(result-type: \(inst-of person-count\))S
gR
gS
.113 .113 scale
1865.133 5381.074 :M
0 G
f3_56 sf
(method: \(add \(obj \(personnel \(force-modules ?c\)\)\))S
gR
0 G
208.5 579.5 456.5 612.5 3.5 @s
188.5 383.5 309 87 rS
gS
.113 .113 scale
1723.701 3481.351 :M
f3_72 sf
(goal: \(calculate \(obj \(spec-of closure-date\)\))S
gR
gS
.113 .113 scale
1723.701 3569.711 :M
f3_72 sf
(                 \(of \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1723.701 3658.07 :M
f3_72 sf
(achieved-by: \(calculate \(obj \(?d is \(spec-of closure-date\)\)\))S
gR
gS
.113 .113 scale
1723.701 3746.429 :M
f3_72 sf
(                        \(of \(?c is \(inst-of coa\)\)\)\))S
gR
gS
.113 .113 scale
1723.701 3825.952 :M
f3_72 sf
(bindings: \(\(?d is \(spec-of closure-date\)\))S
gR
gS
.113 .113 scale
1723.701 3914.311 :M
f3_72 sf
(           \(?c \(inst-of deployment-coa\)\)\))S
gR
gS
.113 .113 scale
1723.701 4002.671 :M
f3_72 sf
(method: \(calculate \(obj ?d\))S
gR
gS
.113 .113 scale
1723.701 4091.03 :M
f3_72 sf
(                   \(of \()S
gR
gS
.113 .113 scale
2766.761 4091.03 :M
f5_72 sf
(movements)S
gR
313 465 -1 1 358 464 1 313 464 @a
gS
.113 .113 scale
3155.699 4091.03 :M
f3_72 sf
( ?c\)\)\))S
gR
43 550 422 127 rC
-1 -1 447 578 1 1 469 540 @b
-1 -1 449 579 1 1 470 542 @b
gR
gS 43 44 469 633 rC
-1 -1 457 554 1 1 471 540 @b
456 554 -1 1 466 559 1 456 553 @a
-1 -1 466 560 1 1 471 540 @b
np 471 540 :M
456 553 :L
465 559 :L
471 540 :L
eofill
1 G
np 465 533 :M
462 536 :L
462 541 :L
465 544 :L
478 544 :L
481 541 :L
481 536 :L
478 533 :L
465 533 :L
eofill
0 G
-1 -1 463 537 1 1 465 533 @b
-1 -1 463 541 1 1 462 536 @b
462 541 -1 1 466 543 1 462 540 @a
465 544 -1 1 478 543 1 465 543 @a
-1 -1 478 544 1 1 480 540 @b
-1 -1 481 541 1 1 480 536 @b
477 534 -1 1 481 536 1 477 533 @a
465 534 -1 1 478 533 1 465 533 @a
.568 G
43 481 469 31 rC
230 472 -4 1 253 523 4 230 471 @a
gR
.568 G
gS 43 44 469 633 rC
231 472 -1 1 234 493 1 231 471 @a
-1 -1 234 494 1 1 245 489 @b
231 472 -1 1 246 489 1 231 471 @a
np 231 471 :M
233 493 :L
245 489 :L
231 471 :L
eofill
249 502 -1 1 252 523 1 249 501 @a
-1 -1 239 506 1 1 249 501 @b
238 506 -1 1 252 523 1 238 505 @a
np 251 523 :M
249 501 :L
238 505 :L
251 523 :L
eofill
endp
%%Page: 15 15
%%BeginPageSetup
initializepage
(swartout; page: 15 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(15)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.752 .075(A key advantage for knowledge acquisition of )J
294 69 :M
(E)S
f2_10 sf
(XPECT)S
333 69 :M
f2_12 sf
.713 .071(\325s approach is that the program instantiator)J
60 81 :M
.632 .063(explores )J
105 81 :M
f1_12 sf
.875 .088(all )J
122 81 :M
f2_12 sf
.708 .071( the possible execution paths through the knowledge based systems it creates.  It thus)J
60 93 :M
1.692 .169(captures all the interdependencies between factual and problem solving knowledge, something)J
60 105 :M
.173 .017(which is not possible to do by analyzing execution traces, for example, since each analysis will only)J
60 117 :M
(cover one execution path through the system.)S
60 134 :M
.87 .087(Figure 4 shows a partial view of how the program instantiator expands goals.  The top-level goal)J
60 146 :M
.947 .095(given to the system is to evaluate an instance of a deployment course of action.  This is the goal)J
60 158 :M
1.096 .11(posted in node n1.  The matcher finds a method whose capability can achieve this goal, and the)J
60 170 :M
1.77 .177(method's capability, the bindings, and the method's body are recorded in the node.  After the)J
60 182 :M
.394 .039(bindings are substituted in the method body,  the subgoal of evaluating the transportation factors of)J
60 194 :M
(the COA would arise, and successive goal expansions would produce the goals in nodes n2 and n6.)S
60 211 :M
.141 .014(Notice that even though the method used to achieve the goal in n1 can be used for any kind of COA,)J
60 223 :M
.196 .02(the bindings specify that is used for deployment COAs and the system propagates this more specific)J
60 235 :M
1.747 .175(type as it expands the subgoals.  When no method is found to match a goal, )J
466 235 :M
(E)S
f2_10 sf
(XPECT)S
505 235 :M
f2_12 sf
1.825 .182( tries to)J
60 247 :M
1.016 .102(reformulate the goal and try matching again.  For example, the goal in node n2 is achieved by a)J
60 259 :M
(covering reformulation of the object parameter of the goal.)S
60 276 :M
2.268 .227(During the process of expanding the goals, the program instantiator also keeps track of the)J
60 288 :M
1.288 .129(interdependencies between the different components of the knowledge bases.  Let us look more)J
60 300 :M
.883 .088(closely at node n6 in Figure 4.  The goal of calculating the closure date of a deployment COA is)J
60 312 :M
.46 .046(achieved with a method whose body indicates that the system must calculate the closure date of all)J
60 325 :M
1.219 .122(the movements of the COA.  Since )J
f3_12 sf
.53(movements)A
f2_12 sf
.686 .069( is a slot \(or )J
377 325 :M
f1_12 sf
.913 .091(role )J
f2_12 sf
1.276 .128(in Loom terminology\) of the)J
60 338 :M
.377 .038(concept )J
101 338 :M
f3_12 sf
.16(COA)A
f2_12 sf
.379 .038(, the system annotates that COA movements are used by the method in this node.  The)J
60 350 :M
.198 .02(factual domain knowledge is effectively being linked to the problem-solving knowledge that uses it.)J
60 362 :M
.137 .014(This is shown with thick gray lines in the figure.  Furthermore, the bindings in the node indicate that)J
60 374 :M
(movements are used \(and thus needed\) for deployment COAs, but not for other types of COAs.)S
60 410 :M
f0_12 sf
(5. F)S
f0_10 sf
(LEXIBLE )S
126 410 :M
f0_12 sf
(K)S
f0_10 sf
(NOWLEDGE )S
198 410 :M
f0_12 sf
(A)S
207 410 :M
f0_10 sf
(CQUISITION)S
60 427 :M
f2_12 sf
1.365 .137(By representing separately and explicitly knowledge of different types, )J
426 427 :M
(E)S
f2_10 sf
(XPECT)S
465 427 :M
f2_12 sf
1.628 .163( allows users to)J
60 439 :M
.497 .05(make changes to the knowledge bases in terms that are meaningful in the domain.  By deriving the)J
60 451 :M
.761 .076(interdependencies between the different types of knowledge as they are used for problem solving,)J
60 463 :M
(E)S
f2_10 sf
(XPECT)S
99 463 :M
f2_12 sf
1.211 .121( can provide guidance for knowledge acquisition that is dynamically generated from the)J
60 475 :M
1.8 .18(current content of the knowledge bases.  By representing explicitly problem-solving methods,)J
60 487 :M
(E)S
f2_10 sf
(XPECT)S
99 487 :M
f2_12 sf
.809 .081( can reason about their components and support users in modifying any component of the)J
60 499 :M
1.736 .174(methods.  This section describes very briefly how E)J
f2_10 sf
.723(XPECT)A
366 499 :M
f2_12 sf
1.732 .173('s knowledge acquisition tool takes)J
60 511 :M
.605 .06(advantage of the architectural features described in this paper, see [Gil, 1994; Gil and Paris, 1994])J
60 523 :M
(for a more detailed description of the knowledge acquisition tools.)S
60 547 :M
f0_12 sf
(Using Problem-Solving Knowledge to Guide the Acquisition of Factual Domain Knowledge)S
60 564 :M
f2_12 sf
(E)S
f2_10 sf
(XPECT)S
99 564 :M
f2_12 sf
1.868 .187('s knowledge acquisition tool supports users in entering factual domain knowledge by)J
60 576 :M
1.216 .122(automatically generating a dialogue that requires the information needed for problem-solving as)J
60 588 :M
1.359 .136(indicated by the interdependencies captured by the program instantiator.  Let us go back to the)J
60 600 :M
.334 .033(example in Figure 4.  For example, when a user wants to define an instance of a new COA E)J
f2_10 sf
.166(XPECT)A
60 612 :M
f2_12 sf
.445 .045(will examine the interdependencies derived by the Program Instantiator.  )J
420 612 :M
(E)S
f2_10 sf
(XPECT)S
459 612 :M
f2_12 sf
.479 .048( realizes that only)J
60 624 :M
.092 .009(deployment COAs are evaluated \(according to the top-level goal in node n1\),  and that other kinds of)J
60 636 :M
.12 .012(COAs \(such as employment COAs\) are not evaluated.  Based on that information,  )J
463 636 :M
(E)S
f2_10 sf
(XPECT)S
502 636 :M
f2_12 sf
.124 .012( will first)J
60 648 :M
.788 .079(request the user to be more specific about the COA.  To provide maximum support to the user in)J
60 660 :M
1.021 .102(providing this information, )J
199 660 :M
(E)S
f2_10 sf
(XPECT)S
238 660 :M
f2_12 sf
1.212 .121( generates a menu of options that correspond to the different)J
endp
%%Page: 16 16
%%BeginPageSetup
initializepage
(swartout; page: 16 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(16)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.51 .051(kinds of COAs that are known to the system.  The system continues to request the user to be more)J
60 81 :M
(specific for  as long as the subtypes of the currently specified type are used differently by the system.)S
60 98 :M
1.017 .102(Next, )J
91 98 :M
(E)S
f2_10 sf
(XPECT)S
130 98 :M
f2_12 sf
.915 .092( examines the interdependencies to request the data needed about a deployment for)J
60 110 :M
.769 .077(problem solving.  It notices that the roles used in the methods are force-modules and movements.)J
60 122 :M
1.404 .14(According to the knowledge that is currently available to the system, the JSCP \(Joint Strategic)J
60 134 :M
.715 .072(Capabilities Plan\)  information is not used for COA evaluation.  \(JSCP information specifies high)J
60 146 :M
.188 .019(level directives that are irrelevant for COA evaluation.\)  Thus, )J
365 146 :M
(E)S
f2_10 sf
(XPECT)S
404 146 :M
f2_12 sf
.205 .021( requests the user to enter the)J
60 158 :M
1.013 .101(force modules and movements of the COA and makes the JSCP information optional.  Again, to)J
60 170 :M
1.127 .113(support the user as much as possible, )J
253 170 :M
(E)S
f2_10 sf
(XPECT)S
292 170 :M
f2_12 sf
1.132 .113( generates a menu with the force modules and the)J
60 182 :M
(movements that are currently defined in the system as suggestions for possible fillers.)S
60 199 :M
.957 .096(If the user adds a step  to a problem-solving method that uses the JSCP of a COA, E)J
f2_10 sf
.494(XPECT)A
522 199 :M
f2_12 sf
1.078 .108( will)J
60 211 :M
.583 .058(automatically detect this new interdependency and ask the user to provide this information for any)J
60 223 :M
(COAs.)S
60 252 :M
f0_12 sf
(Using Factual Domain Knowledge to Guide the Acquisition of Problem-Solving Knowledge)S
60 269 :M
f2_12 sf
1.163 .116(Suppose now that the domain knowledge changed and a new subclass of support personnel was)J
60 282 :M
-.034(added, e.g., )A
f3_12 sf
-.055(unloading-support-personnel)A
f2_12 sf
-.034(.  )A
319 282 :M
(E)S
f2_10 sf
(XPECT)S
358 282 :M
f2_12 sf
.546 .055( would then detect that to estimate the)J
60 294 :M
2.703 .27(support personnel for a COA in node n2 it now generates three subgoals in the covering)J
60 306 :M
2.077 .208(reformulation instead of two, and as a result it needs to have a problem-solving method for)J
60 319 :M
-.058(estimating the )A
f3_12 sf
-.093(unloading-support-personnel )A
329 319 :M
f2_12 sf
.533 .053(needed for a COA.  )J
430 319 :M
(E)S
f2_10 sf
(XPECT)S
469 319 :M
f2_12 sf
.483 .048( guides the user)J
60 331 :M
.423 .042(in specifying the new method by reusing one of the other two as a rough initial version for the new)J
60 343 :M
(method that the user can correct by changing any of its components.)S
60 360 :M
.44 .044(In addition to adding new problem-solving methods, E)J
f2_10 sf
.162(XPECT)A
360 360 :M
f2_12 sf
.447 .045('s knowledge acquisition tool supports)J
60 372 :M
.018 .002(users in modifying existing methods.  For example, a user may add a new step in a method to use the)J
60 384 :M
.392 .039(priority of a movement in a COA to decide whether or not to use more resources for transportation.)J
60 396 :M
.779 .078(Suppose that the priority of a movement is not defined, and that priorities are defined as relations)J
60 408 :M
.655 .065(that only apply to mission objectives.  Based on this domain knowledge, )J
424 408 :M
(E)S
f2_10 sf
(XPECT)S
463 408 :M
f2_12 sf
.665 .067( notifies the user)J
60 420 :M
.606 .061(that the priority of a movement is not defined and thus cannot be used by a method.  E)J
f2_10 sf
.308(XPECT)A
523 420 :M
f2_12 sf
.689 .069( also)J
60 432 :M
.149 .015(presents to the user with a menu of all the roles that are defined for movements as options to be used)J
60 444 :M
(in the method.)S
60 461 :M
1.717 .172(As always, the knowledge acquisition dialogue is updated if the underlying interdependencies)J
60 473 :M
.146 .015(change.  In this case, if the definition of the role "priority" is changed so that it is defined for a COA)J
60 485 :M
(then the system would allow the user to use the priority of a COA.)S
60 509 :M
f0_12 sf
(Discussion)S
60 533 :M
f2_12 sf
(     )S
75 533 :M
(E)S
f2_10 sf
(XPECT)S
114 533 :M
f2_12 sf
('s knowledge acquisition tool can:)S
96 557 :M
(1\) derive the interdependencies between domain and problem-solving knowledge)S
96 569 :M
(2\) use these interdependencies to guide knowledge acquisition)S
96 581 :M
(3\) provide suggestions to the user about how to resolve an acquisition)S
431 581 :M
( problem based on the )S
96 593 :M
(    nature of the interdependencies)S
96 605 :M
.078 .008(4\) justify, when asked to do so, the reasons for requesting the user's intervention based on the)J
96 617 :M
(    derivation of the interdependencies)S
96 629 :M
(5\) support users in adding new problem solving methods and in changing the internal steps of)S
96 641 :M
(    existing ones because they are explicitly represented and their relation with domain)S
96 653 :M
(    knowledge is automatically derived.)S
endp
%%Page: 17 17
%%BeginPageSetup
initializepage
(swartout; page: 17 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(17)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.537 .054(All these features are important for the flexibility of E)J
f2_10 sf
.241(XPECT)A
360 69 :M
f2_12 sf
.565 .057('s knowledge acquisition tool, and are)J
60 81 :M
(possible because of its architectural features as they are described in this paper.)S
60 98 :M
2.052 .205(Other knowledge acquisition systems, such as BLIP [Morik, 1989] and KREME [Abrett and)J
60 110 :M
.979 .098(Burstein, 1987] have used description logics.  However, their goals have been complementary to)J
60 122 :M
.819 .082(ours.  BLIP has shown how concept definitions can be inferred from domain facts using machine)J
60 134 :M
1.805 .181(learning technology, and KREME provides a variety of tools that primarily support a user in)J
60 146 :M
.054 .005(building a domain model and domain facts.  Our concern, on the other hand, has been to capture how)J
60 158 :M
.079 .008(a domain model is constrained by the way it is used in solving problems, and to use those constraints)J
60 170 :M
(to guide knowledge acquisition.)S
60 187 :M
.167 .017(KSSn [Gaines, 1994] is very close in spirit to )J
283 187 :M
(E)S
f2_10 sf
(XPECT)S
322 187 :M
f2_12 sf
.205 .021(.  Like )J
357 187 :M
(E)S
f2_10 sf
(XPECT)S
396 187 :M
f2_12 sf
.158 .016(, KSSn uses a description logic)J
60 199 :M
1.366 .137(representation to capture domain concepts, facts and relations.  In our view, a major difference)J
60 211 :M
.475 .048(between the two frameworks is the way in which the description logic inference mechanisms \(such)J
60 223 :M
.47 .047(as the classifier\) are used.  In KSSn, the inference mechanisms are used at runtime to help derive a)J
60 235 :M
.108 .011(solution.  )J
109 235 :M
(E)S
f2_10 sf
(XPECT)S
148 235 :M
f2_12 sf
.094 .009( uses the inference mechanisms at a )J
f1_12 sf
.14 .014(meta level)J
f2_12 sf
.104 .01(.  That is, the inference mechanisms)J
60 247 :M
1.125 .113(are used to put together a set of program modules that will solve the problem, but the inference)J
60 259 :M
1.231 .123(mechanisms are not used in the problem solving effort itself.  One of the reasons we chose this)J
60 271 :M
1.187 .119(approach is that we wanted the option of supporting kinds of reasoning \(such as reasoning with)J
60 283 :M
(uncertainty\) that are not yet supported by Loom.)S
60 319 :M
f0_12 sf
(6. SUMMARY)S
60 353 :M
f2_12 sf
.9 .09(Second generation expert system frameworks have advanced over first generation frameworks by)J
60 365 :M
.805 .08(providing ways of representing knowledge that separates different kinds of knowledge and makes)J
60 377 :M
.199 .02(explicit the distinctions among them.  This has resulted in better tools for knowledge acquisition.  In)J
60 389 :M
1.645 .164(this paper, we have argued that additional advances in acquisition can be made by structuring)J
60 401 :M
.168 .017(problem solving knowledge so that intent is more explicitly represented and by the use of automated)J
60 413 :M
.646 .065(tools for recombining the different kinds of knowledge.  We argued that the process of combining)J
60 425 :M
.391 .039(knowledge together to solve problems should be done by the machine so that the interdependencies)J
60 437 :M
.841 .084(between problem solving and factual knowledge that guide knowledge acquisition can be derived)J
60 449 :M
.757 .076(automatically.  We presented a verb clause based representation for goals and method capabilities)J
60 461 :M
.311 .031(that captures the interrelationships between the action to be performed and the parameters involved.)J
60 473 :M
.064 .006(We argued that the architecture should support loose-coupling of different knowledge sources so that)J
60 485 :M
.546 .055(users adding knowledge to a system can focus on the meaning of what they are adding, rather than)J
60 497 :M
1.559 .156(particular terms or syntax.  Finally, we described the major mechanisms that )J
461 497 :M
(E)S
f2_10 sf
(XPECT)S
500 497 :M
f2_12 sf
1.819 .182( uses for)J
60 509 :M
1.308 .131(combining knowledge together, semantic match and reformulation.  These architectural features)J
60 521 :M
.484 .048(make explicit the knowledge  needed to support more flexible and powerful knowledge acquisition)J
60 533 :M
(tools.)S
60 569 :M
f0_12 sf
(ACKNOWLEDGMENTS)S
60 603 :M
f2_12 sf
.506 .051(The authors would like to thank the current and former members of the E)J
f2_10 sf
.229(XPECT)A
454 603 :M
f2_12 sf
.544 .054( and EES projects,)J
60 615 :M
.491 .049(especially Pedro Gonzalez, Bing Leng, Vibhu Mittal, Johanna Moore, Robert Neches, Cecile Paris,)J
60 627 :M
.118 .012(Ramesh Patil, and Marcelo Tallis for their hard work and thoughtful insights that contributed greatly)J
60 639 :M
(to this research.)S
endp
%%Page: 18 18
%%BeginPageSetup
initializepage
(swartout; page: 18 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(18)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
1.549 .155(The work described here has been supported by Rome Laboratory and the Advanced Research)J
60 81 :M
3.067 .307(Projects Agency under contract DABT63-91-C-0025, and under NASA Ames cooperative)J
60 93 :M
(agreement number NCC 2-520.)S
60 129 :M
f0_12 sf
(REFERENCES)S
60 156 :M
f2_12 sf
1.831 .183([Abbrett and Burstein, 1987] G. Abbrett and Mark Burstein. The KREME knowledge editing)J
78 168 :M
(environment.  In International Journal of Man-Machine Studies \(1987\) 27, pp. 103-126.)S
60 183 :M
1.744 .174([Anzai and Simon 1979] Y. Anzai and H. A. Simon.  The Theory of Learning by Doing.  In)J
78 195 :M
(Psychological Review, 86\(2\) 124-140,  1979.)S
60 210 :M
( )S
63 210 :M
2.573 .257([Bachant and Soloway, 1989] J. Bachant and E. Soloway.  The Engineering of XCON in)J
78 222 :M
(Communications  of the ACM, 32\(3\) 311-317,  March 1989.)S
60 237 :M
2.085 .209([Chandrasekaran and  Mittal, 1982] B. Chandrasekaran and S. Mittal. Deep versus compiled)J
78 249 :M
.014 .001(knowledge approaches to diagnostic problem solving. In )J
353 249 :M
f1_12 sf
.013 .001(Proceedings of the National Conference)J
78 261 :M
(on Artificial Intelligence. AAAI, 1992.)S
60 276 :M
f2_12 sf
1.339 .134([Chandrasekaran, 1986] B. Chandrasekaran. Generic tasks in knowledge-based reasoning. )J
f1_12 sf
.469(IEEE)A
78 288 :M
(Expert ,)S
116 288 :M
f2_12 sf
( 1\(3\):23-30, 1986.)S
60 303 :M
1.35 .135([Clancey, 1983a] W.J. Clancey. The advantages of abstract control knowledge in expert system)J
78 315 :M
2.261 .226(design. In )J
134 315 :M
f1_12 sf
1.859 .186(Proceedings of the National Conference on Artificial Intelligence, )J
482 315 :M
f2_12 sf
1.615 .162(pages 74-78,)J
78 327 :M
(Washington, DC, 1983. AAAI.)S
60 351 :M
.372 .037([Doyle and Patil, 1991] J. Doyle and R.S. Patil.  Two theses of knowledge representation: language)J
78 363 :M
2.246 .225(restrictions, taxonomic classification, and the utility of representation services.  )J
503 363 :M
f1_12 sf
(Artificial)S
78 375 :M
(Intelligence )S
138 375 :M
f2_12 sf
(48\(3\) 261-297.)S
60 390 :M
.544 .054([Clancey, 1983b] W.J. Clancey. The epistemology of a rule-based expert system: A framework for)J
78 402 :M
(explanation.  )S
143 402 :M
f1_12 sf
(Artificial Intelligence,)S
249 402 :M
f2_12 sf
( 20\(3\):215-251, 1983.)S
60 417 :M
2.675 .267([Gaines 1994] Gaines, B.R. A situated classification solution of a resource allocation task)J
78 429 :M
.265 .027(represented in a visual language.  )J
244 429 :M
f1_12 sf
.21 .021(International Journal of Human-Computer Studies )J
f2_12 sf
.232 .023(40\(2\) 243\320)J
78 441 :M
(271.)S
60 456 :M
1.719 .172([Gil 1994] Gil, Y. Knowledge Refinement in a Reflective Architecture. In )J
450 456 :M
f1_12 sf
1.537 .154(Proceedings of the)J
78 468 :M
(National Conference on Artificial Intelligence  \(AAAI-94\))S
f2_12 sf
(, 1994.)S
60 483 :M
2.92 .292([Gil and Paris 1994]  Gil, Y., and Paris, C.L.  Towards Method-Independent Knowledge)J
78 495 :M
(Acquisition.  In  )S
159 495 :M
f1_12 sf
(Knowledge Acquisition)S
271 495 :M
f2_12 sf
(, 6 \(2\), pp. 163-178, 1994.)S
60 510 :M
1.133 .113([Hasling et al., 1984] D.W. Hasling, W.J. Clancey, and G. Rennels. Strategic explanations for a)J
78 522 :M
(diagnostic consultation system. )S
f1_12 sf
( International Journal of Man-Machine Studies, )S
465 522 :M
f2_12 sf
(20\(1\), 1984.)S
60 537 :M
2.54 .254([Neches et al., 1985] Robert Neches, William Swartout, and Johanna D. Moore. Enhanced)J
78 549 :M
3.759 .376(Maintenance and Explanation of Expert Systems Through Explicit Models of Their)J
78 561 :M
1.29 .129(Development. )J
f1_12 sf
1.128 .113(IEEE Transactions on Software Engineering,)J
379 561 :M
f2_12 sf
1.009 .101( SE-11\(11\):1337-1351 November)J
78 573 :M
(1985.)S
60 588 :M
.47 .047([MacGregor 1988]  MacGregor, R.  A Deductive Pattern Matcher.  In )J
f1_12 sf
.459 .046( Proceedings of  the Seventh)J
78 600 :M
(National Conference on Artificial Intelligence)S
f2_12 sf
(.  St. Paul, MN, August 1988.)S
60 615 :M
.611 .061([MacGregor 1990]  MacGregor, R.  The Evolving Technology of Classification-Based Knowledge)J
78 627 :M
.388 .039(Representation Systems.  In John Sowa \(ed.\), )J
f1_12 sf
.497 .05(Principles of Semantic Networks: Explorations in)J
78 639 :M
(the Representation of Knowledge)S
238 639 :M
f2_12 sf
(.  Morgan Kaufmann, 1990.)S
60 654 :M
1.385 .138([Morik, 1989] K. Morik. Sloppy Modeling. In )J
f1_12 sf
1.95 .195(Knowledge Representation and Organization in)J
78 666 :M
(Machine Learning.  )S
176 666 :M
f2_12 sf
(Morik, K. \(ed\), Springer-Verlag, 1989.)S
endp
%%Page: 19 19
%%BeginPageSetup
initializepage
(swartout; page: 19 of 19)setjob
%%EndPageSetup
gS 0 0 588 768 rC
297 732 12 12 rC
297 741 :M
f2_12 sf
(19)S
gR
gS 0 0 588 768 rC
60 69 :M
f2_12 sf
.056 .006([Musen and Tu 1993]  Musen, M. A., and Tu, S. W.  Problem-solving models for generation of task-)J
78 81 :M
.556 .056(specific knowledge acquisition tools.  In J. Cuena \(Ed.\), )J
358 81 :M
f1_12 sf
.331 .033(Knowledge-Oriented Software Design)J
f2_12 sf
(,)S
78 93 :M
(Elsevier, Amsterdam, 1993.)S
60 108 :M
( )S
63 108 :M
.974 .097([Swartout et al., )J
148 108 :M
.939 .094(1991] William R. Swartout, Cecile L. Paris, and Johanna D. Moore. Design for)J
78 120 :M
(explainable expert systems. )S
f1_12 sf
(IEEE Expert, )S
280 120 :M
f2_12 sf
(6\(3\):58-64, June 1991)S
60 135 :M
3.343 .334([Swartout, 1981] W. Swartout. Explaining and justifying expert consulting programs. )J
f1_12 sf
1.487(In)A
78 147 :M
(Proceedings of the Seventh International Joint Conference on Artificial Intelligence.)S
484 147 :M
f2_12 sf
( 1981.)S
60 162 :M
.782 .078([Swartout, 1983] W.R. Swartout, Xplain : A system for creating and explaining expert consulting)J
78 174 :M
(systems. )S
122 174 :M
f1_12 sf
(Artificial  Intelligence, )S
234 174 :M
f2_12 sf
( 21\(3\):285-325, September 1983. Also available as ISI/RS-83-4.)S
60 189 :M
.625 .062([Swartout and Moore 1993]  Swartout, W. R., and Moore, J. D.  Explanation in second-generation)J
78 201 :M
1.455 .145(expert systems.  In J.-M. David, J.-P. Krivine, and R. Simmons \(Eds.\), )J
f1_12 sf
2.89 .289(Second Generation)J
78 213 :M
(Expert Systems)S
151 213 :M
f2_12 sf
(, Springer-Verlag, 1993.)S
60 225 :M
.676 .068([Wielinga and Breuker, 1986] B.J.Wielinga and  J.A. Breuker, Models of Expertise, )J
480 225 :M
f1_12 sf
.907 .091( ECAI )J
515 225 :M
f2_12 sf
.756 .076( 1986,)J
78 237 :M
(497-509.)S
endp
%%Trailer
end
%%EOF