ࡱ> FLR|T}|-Yu[AJFIFddDuckyPAdobed         !1AQa"q2R#B br3$C%ᒲScs4DdE!1AQaq"2BRbr#3CSc$s% ?Ί)tQK]R袺}ͩ!qA)&yo8WLcQM4N$d[F{(RGTK<-5dZ}ô|"Uq&ARZ`dgg(|<߸ڞc$6FFJ7wڰ-Z~"KK'ypIJ^9덟q+J۷XOi>ǩ/V^cKHYuG u =noւۻ?]S4\?XQs9cBU ֍!t <Өa0R΀.@O7$W3c03w8VH<*s!SM{rw%8ϑ%}>'f?M{3i)%_R:ҴFw 9d9Ȳ:>pz^)㻌t<Bu Kiw')J8K\ `*Ƽyw,? 9rpN [ȵO"RCHBܥm%@^*)մcmbm6?2ε\m?671/miݢ~_b8?~a4Ψ x>Z%ɀ%M:KҐZit-[#q7.= ͹KQ>'QYRɒţaE[ϸ䎸6/-* +xW;*kks| ;D%Iv:ylނ R{kX gpŝl=ˤ"{ߘcZ8ܚңOT+YN:XIhTа} o~ʶقǺKZDqg4!tkob@]YsӿM82Q,U&5JKt*{*wVE.)tQK]WÎ6kuեIS(蔤w$\cKP z֗RdSh,82A qnIH^|[dk|}9yOcs !(`ݘowo,s j~:X:̠hJ %xn'q!猰쳃w^[w?f_]|ꟗg漉zkWu9TQ2[5̸q,L*y#.R 9Obmc+5.<.vڞ6E[45ryזVÍÈI-6WRy+?m&}=f{\Fsgn/8U& R%[/)-R~p vH ^zB.F@G*xÕ-Fcr}Ĝ{ ^ݮc*qԩד!9`< R l%GU6w ȼqP96zzVU~t(ˏ2Md m2<%|n#ԝgU2}:b+3 HSY[H|-)"W(5yN}`9-O&-ȸLxi^ʹ,f1 *RRv{thiZ v{=GnL h]Ol<UM+d4):hc|O}z)l}ԥG ö i}Xrx+'IArkD)i{aAӥn)s\sAvSo&^ Dm[ĊC:[3t{nڴ6>%{gJ,:>A"vJ߁;aؓ^I:5UvZRېthgp(QOFzſķ`m"J7հRKAcuLTJ v E\uk+ FM{=Ʀ_vK'0Fqj\HB.aUr#x 8p?L^U ]R袘dim-I@YA)g Ҿc0?.yc!QUVzhQ{ m~X\M[\'yZ$D5 (K);>o3hOt=Fxow2hO@}gm?`heٙϤ$bqZb8u$foݮC' RV?5K!uV?\u4γO:Ea6u2ٚh[Pڕ)>kXJ@yC%'6q39xcXF: ePnm=n+y24|Jj>T՛ WL3;$'-ZvpdyW?9ur*RJR R 5 P6rZmgXtd"0~ #q{HG{&uC7i#{wV]cu@3q"qގ&32@H ͋1ێךIa)qw-,J}nr5o" z?[.,lkq.Z ;4ƽh.(3g<}N?cPŔkB5Y'C܎ZGݳw};{[:V9 wG3YB/gӜyz|+i`l9RTwkqT%P)ޛNѝ=-ĖH )E xVpwzZbuU:Mܷ-e"÷:}hU$auÞ#bSbW|ypdlldf*mX.jx἗flށǶ“+`K+C!sK>|뛈70YLpk9;`8xl_*?sDJbѷŌEjNP :x[s ^Ӵ&Ȯ!26P L!*MVA3n_tٶ˽4dW[{SsP+ C[o hsshcؠ( \sny9u#6i߈^a9oPbsQoklJ |n)Wclke|R\0h+zfX}EYYysg711<+y$4@ІIRBJHJA'R ȼ-><_#XwH`{-"7dx24H™ߴt 1EK)"$<jOUdK !GuYP{<_%+\D ad}'w]s˛>o]{5|UBzq?wU y?漋w͉Tp9UȘ]5c ,e|5e4jk^cUlhq:7ޙx]ZRClDžJ$ }䀸4Z5;޳Rm -.pz_[G=DE}ߚFr+bT39ȮJb0tY(OuTy=Oyn!q j70Kow yhօ8丒-0rdq{.qQilpzKe7wu'PR}\v3BADTCq\w9>Bִ(pa!q6Z{m<'Ox=9n6m#渌0 ~5| V?1O5^CI95PvL+t" ۉ{j~E뻺g9k8ۄJj\*Ӏtaˆ{@8O{ +WˏIx6i"AKN;dE15 >I JPQuK[-{")p&/ +fWg7UTԸ3D.#]l:`s" XKe|LӘO}5qԘwMHmo!mc7)<-jVRܡVe-=8&ݙ쮢sy=ͨjĪ=9sU3 4#)ף":bTJH<Ϝ{̲iA7`J av2fi<Z4pW0B Md}uckLɰa_S9Եo*vhe 8 `8m'X*>D51lVKS:h)g_s,s!q. 4n S2*6fMm=  RQ2!D?- [)Mr-7*ZJ4D iBը;[Pڿ%[d𞃻v 1673H%/&^iÙ'.YM>d',if3e^&5*ai:)d@{I\‹f3 spС';G̕Kx?WY3]γܞ-)궩{[/ ꂌ!X6SVec^ۉ6ZesqRGMjzO Y#K=GWUi!Y8ulYV`±itɴxi-|)AjQu?zLԡc65{"oUVjjzb~ʈuAWr&??(mP5\Im:ܩ1+t26|(ړj:ioYRGyv@|~O˲/-|MjW*Y8)vv2QRHi%>;_/pzIZ弚6:QC6`ܒӧz{zJ;2%xG44) ]8.`j p_k)R>n@.v7H *ˆMPKbrJ| Ώ3Ř(;Uv--hI_w|WYYy䶦;Ak^r P-g0}v*n9V_\Yzq1߲ǯdk64xz8 I(->ۻsqi86\FpL-ZA=ޝ7[8ٮfY}U{g|W֗ 07 *ڣDjP=.[@,ό"9 8NB\[Hfkwˊ.,~/`RJWn2ٚRH[i *mc@HE|:- GЊӑ<0w E=c$oΜc3,V!b-ck[mr 7@=_>m|_{KQ/rjqԇZ^A&ho~o g2)'g"!Je-9TSi_&scgKNm vO]\|a L=*W5δ?psKx\A('rw mq}R sEM}졊5:ἻϕY^6P@udS»&uQCs4lGNG m;QMyhHRJ͍î瀶--l~)jlzdMllr"@r*,GMc!L$$f_D, u%:,wB;fK UUdۤ:U^l83Q``KF9:-}Ccsʵ~Gָ8ȇ |@N96Ye{29' ZԊ)M&Ӹ2|6 RGZUZqxȅ>tXeIï1+):|TyV!ߔ}byBWk]Di) K=s&XWxkۜC{?e56U rTT?]Qo~oZ}mS˹OŽ9ʫQj1ϱ6ѦEE3BJJ([m ]Ro s)דߨ~MjQk,>Q|*>R_K9QzvfתŖkf^)BRkR N(tHk' ݯ#֮;SƁ3FD jLPT{u 9E)u^2r˥PNuh좫,yXk۵~KJsc4H_e~F7E02Kx3UCoUL}^e+Xel%blv#jR?^HIТ=S>[ o/^di{XQ\~dq!8E<&dV()1$S$:G|x-ӟorkG/Q 1遭454IZPk~oȘi.C+ N*Fђ(6WmI \9sFi[4zN ivHhXCiߤZQ\<A i%8[ϚP@[b@K<@;=&G$l.($N17iܺ-9cL.n:v/?Tj$ POҏُ~%=5V6rEO7fyȞw0N1z8z }jx yUyMNJoBܝuRGOlW7Mu%R#vD &(\!`.֞HSG0?ys'ynf84[A%L4=6Nuߜy K0ማ>{Eu9L}ςIIV&?:_2uW2eb(BԐ+Mw`}uބn~/<p^l;f\$*CqYxl5v_lY}m\Ӫgwt'zW;rWb̨PP!3ޔIBe+n}mg"*GkuO{,׏+2oʺT2:(îh!R*^4B֒U  ETmCЏU+R)K;7vݻNu^{w5I Ӹ)W2C`@/+n`Wβ'K~ n)40JXމ?S^'S<6r?Mn.s$Gu ]V %YD텈l@<ׇ\z_ 6 4u´p"Wdry3=rF{v2֖ڊNt8:Ub2R_2iP+[A[GV)yKL6BRDJz&PBvi̡62iѤ)AZo2(67TʯKI|򄕤ǰ 2[trGCGZ]r,$ O@?vT8NCWg y5-6VV6{zSrft{0S.u3m|I+[KIYUůje22{-)i}Z$#rw/dvMht -phN#4[[+.\;zU`X5WEr6hI[ܤBRBϖ{X-ᳺ1ۖ7I\F$C )dxuH<gkhΛ}6L>d9L]Xy=<v 7$=Q%dF1=(4V5n w'Bqʗ$^H8NT.#i+;Jt*Hצ[/Ӫ"hxBvtǺ]R6x=8p$bNUfa՜O2fCXҿ'jQp4![ ?gLC9{,I+u*/M]ZC[mTOT-=u `zNg0ߑ&ʄ)>J%t܃~jظJrxՓi㶁Ե4v- ^OV=<-o卣MT+u{xƳ[~ݥyr@PP):ZP:tϐOƳ (WhdZX)a X_h릇sqwÃІc n̮*Q=Q*yFT|xeHu= hPT̥HaJֈҊ =&ISEq ʕ۹9xj%ÖCZQTR)ܑ8?UwWUBғ=jwJyPGU­WEVE#A+s:4țb:bvݡ[}V䋊 I<|r)U)C^a VbD.4@J,m6{)ppdiiP W^Sc2/q{|Q\ JuzL|5@iV˦u(bAx-CcL;)vbXH{|!B|g*SV<+)-.=$= 'h)n[E2Ƭr9Rxdn7icͥ2xmKm{Z֢ i(wWr`"ɐE7P2,2 %.$Iԇ4`˝apˋ![@dȋ2RMU{r!MqY$pčT7?H#ZwnQOKyCWlTap}"ejl@T Q9,"}⽺tvmWФ;QYvuvB&Z(Rh)h~#Jf2u7AN"M,\̕%c\R~Շ1_ L^T7\UKXb>/3[0wGCF'7?T8d ^DoPܠ:Něn*|_Ui(HKwuy꼹e"=V-~މDg;$i?CYcn[XͼxnFV>%Hq8uR6Wd1Y5q W+d ;=s#]#QΕ$m? ( m'\(UDE&U;aDq|&nFF.y;kh{w.&kR>ޯEҒ)҆NC0am D Ҵ-2|Ksʸ~ .;7#z5 O-m n,3^/O̭  8qockc,q/@릩hGs6suҎo-cg[k>airm N2@##trvxcy'ѠyҞOX!y ;vgl_N d ˤZAXeۉ>FHmIAZQ4n09~q,Q xDFl;H%Z^s%ZGY}u=Y[=6rvۈdMHBվkfokz:s32WR\BHRwj=ZزqkQ+"㾲V[l5Fxvs<eɬGŸvV*eOPaYJC`Ogiy;GN3|7Q4H!\Wez1L*f0[UƌuL=B >^|?aڸOӄ#%1~8 |ԃI/duQIPZd7ԀTnkÕUqoV)Gwx(Ӭ+#K~$LB/]ry@{rz vO]sq,׈2^3ުux%  JJNJO|:[yeeO$zvH{\qP; M2A{#_Fe^ ɾs$>mgKvMC%D-!GAI41R^@hݒ\|tČ\n񼝂<+:ڒv3: ?Z] Z[``}ږ4cY /4ZfyOخadGAqLҥYO *Cf&4zy(j7UJثF Wf/)G)'Gݏ|ᣮZ5;pAq/dحɮ'>eă%@~ν-?vo%%U)hɨ| T5Cj@XJT>):իqq>NǨQC!ajtO u#Dʐ>! rF<%ge =jcFb9GŸ|cܩKoceu ʲ+-/#}1fe1mJQCpeV]f-k%3Ch JLP\lg@}zu 1oHG"xcz>\zærD_'vUV^khK%)V{n(x,:Ih!Na814`Mq#ՁYi Ib- gmeƥ-Ao/MOǩ{ ! RZ!i g5.d\XI˿%UϪ{BaŻ0֏-Ŏiqi|(A_u ޢeQq_*BF?|ӡ?g<畇څۨ24bѽoZ]C_]0 QC;oV'g@5VPвi!CkԎb7朖AqGbT72)+HƝd p3]qͳ]08WUwnh=Ԍ6\A w ?ƐvӦ&p.4uw7/Kh ;v.)Z t_!s|t^ѕĎ`>E6.d" L\{G=Le[gb7OOB]^ҵF饇# >J"4qIdQZُ@)ic<ՐƑ'8%lּS)^bڻOPБC10Z0!Ml6%J) Evn4KenNtʝʬl`2 T-CGP [ RQܓւ[ a$y{fM8.H87*tAYz<).WeFZc KS{+mZy RpSI9߳D}wOĎ>}ьihM!oO5>ރSP@hiJU\f-8nhv|Y2>BFOu,PHAiٛSC7S+yCRϭS>c971\G$9Mv֞i%[ܤ( BM.A"=W >\xO-l;Pj%#bW`.,v.yl!B$M7) PB{U Xk;ZD, kvLt6zQA^[MpGǴ< U%D wNm#[>+g27Lˌ/$P%R5Oto­;׆kh@A%0׳hsB"K-qH>=eSmJL!A<}5[r1eK ߀ *NkoR1QY!.RA$BA_oR8M͇6Kcfz\z+ӧB_& ;"n]W%Xt3E~"\cUn&VdzpIq_+r[PGk!vLd>8z/X jNӔw ܆v"'ÁcpbWEu6 E HH馚^,W5d2hۼ@M&rs0%|y^7.4QݴJT; #JyYp~ m4W4Sb(>TFP RQR@DOJg1{7|#j{xÎw[Yœ袶 ' cƺ@.W\4G7mִ h=u;c~$l?N.+2#3;@%A*I>73?@F 5$RD&Sｫ N@U-9WǾq:jҐhIۗ":#܈`KmqSc(qnGks<KrR ,9ܤ:t]T.,-i?l>ה娤}D ]E™-'=/ If;ܶ *O@oKrܖ(sp O_GORVL| 87& J^t'LZtԟaxOQ4kv- P->'d<Zm։ mm$uetMP>zEϿu>UN]>w\鋌(̼m>:r@[7 ~D-@x:-wY<)_k,Qn 6ǫiªw<߲[Ds1Ԝ\qQBOUv6Ԗ[khKJRϚN%'Az@ 6 ${jiz[sVb Х 9Mˋ 8%*RpGV-pFw#|6氟W ?PZNIf$}Zϒ匦CtIy +ܤjG\~\?G <v~c_Xgt :Qam_Be6L]f&TiٍAQju+pIP ܍ǧpZ:u?~74lsI6ovi+Q8 ­.Ƭ.; .Ʋ: bM\O.K6*2ƣ$x_i!m-ez?P4ڐϚSI QE<2y9q;Rl*Sc „cik.-vpb5.c=mݟh!!m 4Z64J$ξ꼥E.)ՆeߦUCIT~@"ݧtnwQB16?RDE&+G}#;J&DP;Τ GkzImpG8vF/bc_N?Z7gb1@ʿ,M8N_~ݸ!/ q5o)]]wbKwG G}: ڭ*[.HC.?XCVh0=AX9-pB(/]hgƑ"o\CRbR[%^.YYVdʹ«<ŗܞ$HtĆîʎ p:H#t8MqiV5r@z݂ql,:y2eJ q'=nciGc"7P ,Sx ImJJ$ި;Qjy @8_ZTf/.׼Iݯǿ\5#o<kãLo(,gr%z]u?q^YאVc~rWsie$pyӶCaXE=IessǏ !*4d꭭;@u2\?U AZ;7?)Msx#?ӭx6\S2W]ICosJYaZee$-% BֆhAPISAc9:A)ɤ|66)CjTjؕ>$9 5oTM&May-oV\u^g]6QJףhURPIVz;fk'`(+ck޷lf;H m׶@'NyC]Ai]!t#kN"aFSf9Up`^ǐ(:-^}kjvioI'7.WJ=C`{:QR袗E(Ex'W3;YRʌJh>: RA:ufЕouü?hѨ6DLy;["BT~1gF;@Kg!r94> #4vzκJS>>@mڬ wS M!;]i  :Xy s ?%171qg9ga0 sjжfIjDuŒp<\ae hBTv/9s-4 (n=KVh)7dϴ6ZVB)Т:^Xfm9َmXØ(ɤv-mfik]NE0w/WY.r%EKT*c*B|$}dH8OYvW+!pxv5N;f[Mq"t@j1ԠmI)(wLmxh=A$K3\ڬ$Vz\x֦qr Vr!>1ھ“~~G9,mjߺ#dtaa:y|!T"=d0MuZh?A5-/^-yzHڸGm/:go) ok{3OMd3ejа]\_{z'7R,LWFƦmQ⁍_V-I=SWDd&~g>Ǐg=<؏ ;NR ヺCOw8,0KJ}4RVvyl'*G>VS2oAv6cz]>ҪR袗E(E.)tQK]R袗E(/6>g_:{DQ=~ʎ|߇W}7n5¯Uӿ{~?iO]=??_u3w>w/m?j^'gSjS'OnZCNeKx^ެT(E.)tQK]W3(KkN܂ / 0DTimes New Roman0:A 0DArialNew Roman0:A 0 DImpactew Roman0:A 0"0DWingdingsRoman0:A 0@DArial Unicode MS0:A 0"PDCourier Newe MS0:A 01@ .  @n?" dd@  @@``  P  ,R$R|T}|-Yu[AL 0AA e[|E@ ʚ;~;ʚ;g4dd :A 0p5p%<42d2dL$ 0h0<4ddddL$ 0h0<4BdBdL$ 0h0 80___PPT10 ? %O  =8  ` ` ̙33` 333MMM` ff3333f` f` f` 3>?" ddq |?" ddq  |8" q o E]} n?" dd@   @@``PR  8 q   ` p>o]TT>o]TJJ  V(    TZy<Zy<  E_  T Click to edit Master title style! !4  N Zy<Zy< E4_  RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S  N0Zy<Zy< alh  \*Aq 88q q   NxZy<Zy< a1h   ^*Aq 88q q   NȽZy<Zy< a.6Eh  ^*Aq 88q q H  0I@ ? ̙33 Default Design P8(    Naoao 9"   \*>    Naoao ,N  ^*>    Taoao I|9"ւ   \*>    Taoao ,I|Nւ  ^*>  H  0+ ? 3380___PPT10.!dPeKch gg < 4d(    C B$CDEF[d@A$P @`" B#GF  S BQCDEFd@A , Qx  @`" Hl)mLr h <" `$NDRd g <&m%-j 1 Bd=:H< 4 0(=:H< oKey architectural details$<$ @`  0\(I (RavenClaw: Dialog Management Using Hierarchical Task Decomposition and an Expectation Agenda Dan Bohus Alex Rudnicky School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, 15213D]X66Q(j 4 Bd0HMR 5 s *@MXB 7@ 0DD[ d$XB 8 0DD[ ]#XB 9 0DD[ ]#d$]# E 0L(H#  Abstract RavenClaw is a new dialog management framework developed as the successor to the Agenda architecture used in the CMU Communicator. RavenClaw introduces a clear separation between the specification of task and discourse behaviors, and allows rapid development of dialog management components for spoken dialog systems operating in complex, task-oriented domains. The new system development effort is focused entirely on the specification of the dialog task, while a rich set of domain-independent conversational behaviors are transparently generated by the dialog engine. To date, RavenClaw has been applied to five different domains allowing us to draw some preliminary conclusions as to the generality of the approach. We briefly describe our experience in developing these systems.f < $$$$$ F 0(p@a ; .BXB T@ 0DD[ '+(('XB U 0DD[ '*(*(8p X Hd %)E}'X Y 0 %&(}' Z 6T($&])' :2.B [ 0<( %) I6 ^Overall design RavenClaw is a 2-tier architecture (see below) Dialog Task Specification Layer Captures all the domain-specific dialog (task) logic The system development effort is entirely focused here Domain-independent Dialog Engine Manages dialog by executing the Dialog Task Specification Provides domain-independent conversational strategies? l#p<B$ ,&,!$n$!$ $$6$ @`RB - @ s *DD[8*((8RB / @ s *DD[ =g' =RB 0 s *DD[ =ffhR 2 s *=:9< 3 03(C:< ;Fig6RB 6 @ s *DD[h[hRB 8 s *DD[]S]I]S 9 <'(S}B1I U A Dialog Engine  : <1(pP}BHS MDialog Task Specification 2 F <t*([Ԕ"`-A 4m: C >RoomLine  2 J <)(Ԕ"`Gm"I ;Login2 K B(S(Ԕ"`K!L =Welcome2 L <dW(EԔ"`MmO G AskRegistered2 M Bl[(Ԕ"`M%O =AskName2 N B^(Ԕ"`K$*L ? GreetUser  L O @ c $ԔHiaKL P @ c $ԔcI\ML Q c $ԔI"ML R c $ԔH"'KRB ` @ s *DD['y9'RB a s *DD['xx8j b Bd %+"}'R c s * %K}' d 0d($' :1.BRB f @ s *DD[8x88RB j s *DpM8RRB k @ s *DpMM l <i(LM ORich concept representation; m <@f(5NM Q  Set of confidence / value pairs History of previous values Flags indicating grounding, availability, conveyance status, etc , r u <" `?mLRB w @ s *Dpw?w? x <Dp(q>~? V"Dialog Task Specification (sample)## y 0t(?cL VDEFINE_AGENCY(CLogin, IS_MAIN_TOPIC() DEFINE_SUBAGENTS( SUBAGENT(Welcome, CWelcome) SUBAGENT(AskRegistered, CAskRegistered) SUBAGENT(AskName, CAskName) SUBAGENT(GreetUser, CGreetUser) ) DEFINE_CONCEPTS( STRING_USER_CONCEPT(user_name) BOOL_USER_CONCEPT(registered) ) SUCCEEDS_WHEN(COMPLETED(GreetUser)) PROMPT_ESTABLISH_CONTEXT( establish_context login ) ) DEFINE_INFORM_AGENT(CWelcome, PROMPT( :non-interruptable inform welcome ) ) DEFINE_REQUEST_AGENT(CAskRegistered, REQUEST_CONCEPT(registered) GRAMMAR_MAPPING( [Yes]>true, [No]>false ) ) DEFINE_REQUEST_AGENT(CAskName, PRECONDITION(IS_TRUE(registered)) REQUEST_CONCEPT(user_name) MAX_ATTEMPTS(2) GRAMMAR_MAPPING( [UserName] ) ) ...,  >R RB  s *Dpw?  mLL  c $ԔB4G2  Bx(jJ"`H/=4mI :GetQuery  2  BT(jJ"`J +/=L :DateTime  2  Bt(jJ"`J04=L :Location  2  B(jJ"`J]59=L < Properties  2  B(jJ"`H4 :mI < GetResults  2  Bԕ(jJ"`H:AmI @DiscussResults2  B(Ԕ"`}EAG]G =SuspendL  @ c $jJB14HL  c $jJC=77HL  c $jJB~9E>HR  s *ԔBy:DqEL  c $jJBIM-f0JL  c $jJsI1]2JL  c $jJBI36KR  @ s *jJL*+NR  s *jJL./N2  BР(jJ"`N2}6mO 9Network2  B(jJ"`N 7-;mO ; Projector  2  Bh(jJ"`N;@mO < Whiteboard  L  c $jJL46NL  c $jJCL79NL  c $jJL49->NR  @ s *jJBI:;KR  s *jJCLM-N-NR  s *jJsIE>> LR  s *jJBI@MB LXR  @ 0Zw?5@f2  08c"`F]G  <`(F~G ; user_name f2  08c"`G]H  <(G(H < registered XR  @ 0ZM5DOx  B" `W=W/gRB  @ s *DpqW5]/qW  <,(V\h#oW FExpectation Agenda `  6(E" `1X.Y Nregistered: [No] ! false, [Yes] ! true( Gg GG  6P(" `Y.\ 8zregistered: [No] ! false, [Yes] ! true user_name: [UserName]> Cc C   C   4t  6([" `].d registered: [No] ! false, [Yes] ! true user_name: [UserName] query.date_time: [DateTime] query.location: [Location] query.network: [Network] query.projector: [Projector] query.whiteboard: [Whiteboard] Cc C  A  C          ,3}x  B" `W0=IgRB  @ s *DpqW0=IqW  <(V0f6oW ? User Input:  S  <(%Xb1xD_ wSystem: Are you a registered user? User: Yes, this is John Doe Parse: [Yes](yes / 0.87) [UserName](john doe / 0.46)2x#$$8] "   BCDE4Fp   I z O3~a*XO}: @    "`mU' 7M]   BC7DE(Fp 57 K v;_ @   "`[ $8cx  B" `W \RB  @ s *DpxWxW  <@( V vW SDialog Stack / Agents Execution  x  B" `W \x  B" `W\x  B" `1^ qc~  H8c" `1^qc  6<(["`a[M Q\ <RoomLine    6("`Z[ 9Login  6)E"`Xm]Y ?Welcome  6`)["`a[M Q\ <RoomLine    6)"`Z [ 9Login  6P )["`a[M Q\ <RoomLine    6H)["`!bM c <RoomLine    6l)"`` a 9Login  6)["`!bM c <RoomLine    6)"``a 9Login  6)E"`_m]q` E AskRegistered  6!)"`WsX ;1  6%)"`W sX ;2  6))"`WsX ;3  6H()"`1^ ^ ;4  6/)"`1^^ ;5  3 BCDEFp0p`X@@  "`X_  3 BCDE(Fp 0XHP0X, @   "`a[a  3 BCDE(Fp dpPpx0 @   "`^ icf2  08c"`F.X/G  <5)FN/1G 7queryf2  08c"`F78G  <@9)F8;G 9results  3 B0 C0DEFAp0 0 X  @X@  "`|?-G  3 BCDE(FAp th @0pPX @   "`iG9N^B r 6D8cQ Q p HL=)" `N'Q LJohn Doe / 0.46   s HL3)" `DOQ EJoe Down / 0.33 XB o 0D8cQQRB  s *DԔ(XxYRB  s *DԔY\RB  s *DԔ]dQ e 0^) % [%7  Goals RavenClaw = framework aimed at the rapid development of dialog managers for complex, task-oriented dialog domains Handle a variety of complex domains Easy to develop and maintain systems Developer focuses only on specifying the dialog task Dialog engine handles the rest automatically Architecture supports: Learning (both task and discourse levels) Dynamic generation of dialog tasks Grounding mechanisms zJcb<B$8$9$$$$$%$c$$  $"  $8  $ @`B  s *I@ ?_RJ K O TJ L P VJ M Q XJ N R ]J F _F a F c F e F g i k l m n o p q r s t  ̙33y___PPT10Y+D=' = @B +? ?>0?< Y>(    0`3*H+ K;Conversational behaviors The Dialog Engine automatically provides a basic set of domain-independent conversational behaviors Generic dialog mechanisms Help, Repeat, Suspend, Start over, etc Turn-taking behavior Grounding behaviors Explicit and implicit verifications, disambiguations, context reestablishment, etc')S00$d$$'$)$S  $ @`XR  0Z)*d  <T*J]eZ W 0Խ3 %/ Dialog Task Agents Fundamental Dialog Agents (on leaves) Inform  sends an output Request  requests and listens for information Expect  expects (listens for) information DomainOperation  performs domain operations (i.e. back-end calls, etc) Dialog Agencies (non-terminal nodes) Control the execution of the subsumed agents Agent properties / functionalities: Execute routine Preconditions and triggers Completion criteria (successful / unsuccessful) Effects Hold concepts:&-$q00$%$$$$$'$$$$$8$$$$$$-$$$$  $  $  $  $  $  $  $  $  $   @`d  <M3%<XR  0ZM<->XR  0Z)*aL & "T) #  '"Gr  Bd& "@)Z  s *&@@)  0(3&T) :4.Bd 6@ <eIft 7 c >Asealfabric "fEIid 8@ <i+IiX 9@ 0ZeIIi : <3`f :ELi TSchool of Computer Science, Carnegie Mellon University, 2003, Pittsburgh, PA, 15213.UUXR C 0Zwj D Bd}  $XB @ 0DD[m4d$m4XB  0DD[[4 bp  Hd1 " 4X  01@ 4  6t3m1!4 :3.Bj  Bd + IXB @ 0DD[ j)**XB  0DD[ ))QXB  0DD[ Qp)0*QXR  0Z=%)*'j  Bd( + I-)XR  0Z2)*4j  Bd5 + II5XR  0Z<)*u>j  Bd> + I>j  Bd3>  $c>XR  0ZoQOSj  BdUS  $SXR  0ZmG)*MIj  BdmI+IId  <-J$RB @ s *DD[aQRB  s *DD[yy0j  BdI"]R  s *i]  0pU3q :2.B  03# oo The Dialog Task Specification Generics The Dialog Task Specification = tree of dialog agents, with each agent handling the corresponding part of the dialog task Advantages of hierarchical representation: Dialog task structure naturally lends itself to hierarchical description Ease of maintenance and design; good scalability Implicitly captures context in dialog<B 00$$$$ $E$$ @`j  BdQ))$XR  0ZqQRB @ s *DD[0[K0  01+%b___PPT9bZ  The Dialog Engine Domain-independent component that executes the Dialog Task Specification Dialog flow is generated by alternating Execution Phases and Input Phases Execution Phase The dialog agents in the task tree are executed and generate the system s behavior. Dialog engine uses a stack structure to execute the agents in the task tree: Repeatedly execute agent on top of the stack When agencies execute, they plan one of their subsumed agents for execution (according to preconditions and policies) Completed agents are removed from the stack Request-type fundamental agents can interrupt an Execution Phase and solicit an Input Phase (3-Stage) Input Phase Assemble an Expectation Agenda Expectation Agenda models the system s input expectation at that point in time Bind values from input to concepts Inputs are matched to system expectations Analyze focus shifts Establish if the focus of the conversation should be shifted in light of the recent input & then, continue with another Execution Phase.[,!P$*[/<B0I$($$$ $ 0U$$$($,$0,!$P$$$*$  $[$$.$( @`RB  s *DD[{){)RB @ s *DD[])*RB @ s *DD[UT)*URB  s *DD[Up)p)}dj  BdR*GmUR  s *R'0*mU  0p+R'*U :5.BRB   s *DD[md])*md   0!R*J.d = Conclusions RavenClaw = Dialog Management framework which focuses system development effort on creating a description of the underlying dialog task Dialog Engine drives the dialog towards its goals, and uses generic conversational strategies to maintain dialog flow and coherence 5 systems built to date spanning various domains and task complexities RavenClaw adapted easily, indicating high versatility and good scalability propertiesT <B $ @`XB   0DD[|)|)XB   0DD[maZdj  Bd + I!  003*-J_R P R RavenClaw-based systems LARRI [Symphony Project, CMU] A multi-modal conversational agent that provides support for F/A-18 aircraft mechanics performing maintenance tasks: Guidance & information browsing domain Tree-based decomposition very well suited in this domain; portions of the dialog task tree are generated dynamically based on the task to be performed Intelligent Procedure Assistant [NASA Ames] Multi-modal system that provides assistance to astronauts on the International Space Station in the execution of procedural tasks and checklists: Guidance & information browsing domain RavenClaw interfaced in Open Agent Architecture (with Gemini inputs / output) BusLine [Let s Go! Project, CMU] Information search interface to Pittsburgh bus schedules: Information exploration domain Static dialog task tree RoomLine [CMU] Assistance for conference room reservation and scheduling within the School of Computer Science at CMU: Information management domain Static dialog task tree TeamTalk [11-741, CMU] Spoken command and control for a team of robots: Command and control domain Challenges: multi-way conversations, (complex) asynchronous behaviors Static dialog task tree :u.u":7h61y<B$0u$$  $-0$u$$!0:$7$$0h$6$$01  $y  $  $ @`B  s *I@ ? ̙33y___PPT10Y+D=' = @B +r0; Z}mBOh+'04R `h  PowerPoint Presentation Dan Bohus Pdbohusu85hMicrosoft PowerPointon@}4@`+&u@+OkG(Q  (a&    """)))UUUMMMBBB999|PP3f333f3333f3ffffff3f̙3ff333f333333333f33333333f33f3ff3f3f3f3333f33̙33333f333333f3333f3ffffff3f33ff3f3f3f3fff3ffffffffff3ffff̙fff3fffff3fff333f3f3ff3ff33f̙̙3̙ff̙̙̙3f̙3f333f3333f3ffffff3f̙3f3f3f333f3333f3ffffff3f̙3f3ffffffffff!___www4a&A q(qHʦ """)))UUUMMMBBB999|PP3f3333f333ff3fffff3f3f̙f3333f3333333333f3333333f3f33ff3f3f3f3333f3333333f3̙33333f333ff3ffffff3f33f3ff3f3f3ffff3fffffffff3fffffff3f̙ffff3ff333f3ff33fff33f3ff̙3f3f3333f333ff3fffff̙̙3̙f̙̙̙3f̙3f3f3333f333ff3fffff3f3f̙3ffffffffff!___wwwýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýzzzzzzzzY^zzzzzzzzzY^zzzzýýýýý^z^zzzzYýýýý   YzzzzzYzYzzýýýýýýzzYzzzzYzz^z^z^zz^zzYzýýýýýýýýzzzzz^zz^zzýýýý288SX2z^zz^zYzzYzz^ýàýààýýýýýýý zzzzzzzzzzzzzzzz^zzzzzýýýýàýýýýýýàQz^zYz^zYz^zýààýýýààý zzzzzzzzzzzzzzzzzzzzzzzzzzz^ààýàý嚚QQQQQQQQQQQQQz^zޠýàýýýàý  zzzzzz^zzzzzzzzzzzzzzzzzzz^zàààýzz^zzz^ýÚýàààÚàý2S882RS2YààÚýYS*****8*************X82Y2SÚÚýàýR2222282222222222S222282SS72Xà嚠zýýàÚýzzzzzSzzzzzzzzzzSzzzzzzSSS72S2Y堚ààÚ%%%%+%+%%+Úz堠zàÚÚ+%+%%%+%%+F%+%%%F+%+%+%%+%%%++%ý堠嚠堠z嚠嚚堠à嚠嚠堚嚚z嚠堠嚠堚堠ýýýýýýýýýýýýýýýýýýýýýýýýýýýýà  ýýýÚÚýýýàýýýýýýàýýýý+%+%$%%+%%ýýýýýýýýàÚýÚýýýàzzzzyzzzýàýuuýýààR8S8ýÚýÚýýàýýýýýýàýýýýààýýýýýýýýýýzzzz嚚zýýýàýzzýÚýýýzYzYzzýýýýýàýýýýýýýýýýýýýýýýýýýýýýýýýY88YYYYYYYYYYzYYYzz^zYzYzzz嚚z嚚zÚààýýý8YY8YYYYzzz嚠Úàààýýý888888YYYYYYzzzzzzz嚚嚚Úàýýý88888Y888Y8Y8YYYYzYYY^zzYzzzYzzzzzzzýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýàýýýýýýýýýýýàýýýýýýýýÚýààýààààààýààààýýýýýýàýýàýààýàýýàààýýýýýýýýýýýýýýýýý8Y8YYYzzYzzzzzý88YYYYY^Yzzzzzzz嚠àý888Y8Yzzzzz嚠ààýýý888YY^zYYzzýýýýýý8Y8YY^zzzzz嚚àýýý8888YYzzzýý88YYYYYYYzYzzz嚚ýýýý8YYYYYYYYzzzzzzzÚýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýý8YYY8YYY8YYzYYYYz^YzYzzYzzzzzzzzzzz嚠ýýýýýýýý888888YYYzzYYzzzzzzzzz嚠Úààýýýýýý8888Y88YYYYzYYzzzz^zzzzzzzzz嚚àÚýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýýý՜.+,0      CustomCarnegie Mellon Universityl {  Times New RomanArialImpact WingdingsArial Unicode MS Courier NewDefault DesignSlide 1Slide 2  Fonts UsedDesign Template Slide Titles_fdbohusdbohus  !"#$%&()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root EntrydO)PicturesLCurrent UserSummaryInformation(dRPowerPoint Document('DocumentSummaryInformation8