Aa& e M 000`@ `ppP @p   HH $ @dHHHH̀̀̀ff@  /d Footnote TableFootnote**.\t.\t/ - :;,.!?$ _l bVTOCHeadingHeading2   EquationVariables3.0aiCDIJccd\ddee  <$lastpagenum><$monthname> <$daynum>, <$year>"<$monthnum>/<$daynum>/<$shortyear>;<$monthname> <$daynum>, <$year> <$hour>:<$minute00> <$ampm>"<$monthnum>/<$daynum>/<$shortyear><$monthname> <$daynum>, <$year>"<$monthnum>/<$daynum>/<$shortyear> <$fullfilename> <$filename> <$paratext[Title]> <$paratext[Heading1]> <$curpagenum> <$marker1> <$marker2> (Continued)+ (Sheet <$tblsheetnum> of <$tblsheetcount>)Heading & Page <$paratext> on page<$pagenum>Pagepage<$pagenum>See Heading & Page%See <$paratext> on page<$pagenum>. Table & Page7Table<$paranumonly>, <$paratext>, on page<$pagenum> Table All7Table<$paranumonly>, <$paratext>, on page<$pagenum>Table Number & Page'Table<$paranumonly> on page<$pagenum> A  A-A00Headings<4HTMLh(d(e(f(g(h``"(ib"))a))`)a))a)b%))a))cb'))b))ab))c))`))`))`)`))`))`))`))c))`)`)`)`$`)`)`"`"`"`))`))`))`))`))`))`))`))`))`))`))`))ae))a)))c4))a))a))b.))cl))cm))cp))cq)cr))cs))b5))ct))cu))cv))cw))cccb=))b>))c!c!c!c!bC))bD))c#c!ccbI))ccc!c!c!bO))bP))c!c!c!bT))bY))d!d !d !d!d!d!d!d!bd))be))d!d!d!bi))d!d !d"!d$!bn))d&!bp))d(!d*!d^d`dbbv))bw))dddf!dh!b{))b|))dj!dl!dn!dp!dr!dt!b))dddb))d!d!d!d!d!d!ddb))b))ddd!d!d!d!d!d!h!!d!h!!h!!e0!!e3!e5!e7!e9!e;!e=!i!!eA!eC!eE!eG!eI!eK!eM!i!!eQ!eS!eU!eW!eY!e[!e]!h!!ea!ec!ee!eg!ei!h!!h!!ez!!i!!i !!e!!i!!i!!e!!eeee!e!e!e!e!e!eeeeee!e!e!e!e!e!e!e!e!e!e!e!e!e!f!f_!fa!fc!fe!fg!fi!fk!fm!fo!fq!fs!fu!fw!fy!f{!f}!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!f!g!g!g!g!g !g !g !g!g!g!g!g!g!g!g!g!g!!g#!g%!g'!g)!g+!g-!g/!g1!g3!g5!g7!g9!g;!g=!g?!gA!gC!gL!!gN!!g!!g!!h!!h!!h(!!h*!!h!h!h!in!ip&ir!it!iv%ix!0d(kd(l  H$H$Ge H$(mH$l Hw!HwGmPage M#N Hw(nHwl HHˆ$HHˆG e HHˆ(oHHˆl H$'H$ W` H$(p H$ l Hw*Hw WlPage O#P Hw(q Hw l HHˆ-HHˆG e HHˆ(r HHˆ ld(s HmR(tHmRHRHRFootnote Hr@(uHr@HzHz Single LineH(vFootnote `^  HD(xHDHH Double LineH(y Double Line`b`dH(| Single Line`gHZ(~ TableFootnote EGxR(EGxREPwEPw TableFootnoted( HHˆ>HHˆƒ++ d1Bar Hillel and Machine Translation: Then and Now !,`Sergei Nirenburg :`Computing Research Laboratory $`New Mexico State University %`Las Cruces, NM 88003, USA &`sergei@crl.nmsu.edu  ` ! [The name of Yehoshua Bar Hillel is known to every student of machine translation. The late Xphilosopher richly deserves this recognition: he had been a central figure in the early Zdevelopment of the field and contributed what should be considered the first set of sober `assessments for MT at the time of the widespread gung-ho attitude to its prospects. The fame is cprobably due to the correctness of Bar Hillels forecasts of the fields future stumbling blocks. cStill, what is generally remembered of Bar Hillels contributions to MT is but a small fraction of @(his ideas and opinions on the subject. ` ! XThere is an unsettling tendency among my colleagues in MT, natural language processing, _linguistics and AI tacitly to assume that their predecessors in the field, not having at their adisposal either the latest machines or the latest theories, were somehow naive and incomplete. dHistory of the subject starts for them with the dissertation work of their thesis advisor. For such _people, Bar Hillel may only have the distinction of being the earliest widely quoted author in bMT. In reality, reading Bar Hillel can be very instructive for todays scholars, especially as he _excelled in recognizing and assessing the intellectual evolution of entire fields (philosophy, @Zlogic, linguistics, MT) and pointing out lacunae in the applicability of their findings. #` !" cThis paper, first, takes the reader on a brief tour of Bar Hillels writings about MT and, second, _attempts an assessment of the current state of MT research in the spirit of Bar Hillel (though @,without any claims for similar profundity). ` UTUT`1. Then  ^Bar Hillels central argument can be presented as the following sequence of brief theses. The @0summaries are illustrated by a few quotations. ` a* b1. Bar Hillel believed that, in order to achieve fully automatic high-quality machine translation [(FAHQMT), machines must be able to process meaning. The task of instructing a machine how ato translate from one language it does not and will not understand into another language it does enot and will not understand presents a real challenge for structural linguists, in that their thesis Wthat language can be exhaustively described in non-referential terms undergoes here an aexperimentum crucis. If, in a translation programme, some step has to be taken which directly or bindirectly depends on the machines ability to understand the text on which it operates, then the bmachine will simply be unable to make the step, and the whole operation will come to a full stop. _(I have in mind present day machines that do not possess a semantic organ. The situation might Xchange in the not too distant future.) Some Linguistic Problems Connected With Machine mTranslation. sAspects of Languagev, p.308. In this passage, Bar Hillel addresses translation without HHˆ(HHˆ!ld`!! HHˆ`HHˆ‚..!*`treatment of meaning. "Non-referential" means "uninterpreted" that is, free of meaning. It now 0cseems that for the purpose of computer-aided translation the semantic structure of sentences to be itranslated has to be exhibited. The Outlook for Computational Semantics. sAspects of Languagev, @p.358. +` ! ^2. The study of meaning in language is the realm of semantics, and there were at the time two ]major traditions that could be drawn upon to provide solutions for MT. These were linguistic @"semantics and logical semantics. '` !, Z3. Contemporary linguistic semantics (epitomized by the work of Katz and Fodor, 1963) was `found largely unpromising by Bar Hillel because a) it was not adequately formalized; and b) its \apparatus neglected large parts of semantics such as non-compositional phenomena, and every Vother phenomenon for whose description it is necessary to rely overtly on "background _information," the knowledge of the world. Note that Bar Hillel actually saw something positive \in Katz and Fodors program, in that he believed that linguistic semantics can overcome its _shortcomings by paying attention to issues not developed by Katz and Fodor: It is very likely cthat their shortcomings will further the field much more than their actual positive achievements. @LReview of The Structure of Language. sAspects of Languagev, p.175 (` !) ^4. Contemporary logical semantics was dismissed by Bar Hillel in the context of MT because it afocused on artificial languages. ... [Rudolf Carnap] is thinking mostly in terms of constructed @]languages... The Outlook for Computational Semantics, sAspects of Languagev, p.359. -` !. `5. Bar Hillel believed that treatment of meaning can only be based on a system of logic: first, ^because for him only hypotheses formulated as logical theories had any scientific status and, @Psecond, because inference rules necessary for MT could only be based on logic. /` !0 f6. At the same time, he considered such logical systems sunattainablev because they could not @cwork directly on natural language, using instead one of a number of artificial logical notations. !4 \...The evaluation of arguments presented in a natural language should have been one of the dmajor worries... of logic since its beginnings. However, ... the actual development of formal logic ftook a different course. It seems that ... the almost general attitude of all formal logicians was to cregard such an evaluation process as a two-stage affair. In the first stage, the original language aformulation had to be rephrased, without loss, in a normalized idiom, while in the second stage, athese normalized formulations would be put through the grindstone of the formal logic evaluator. Y... Without substantial progress in the first stage even the incredible progress made by Tmathematical logic in our time will not help us much in solving our total problem. iArgumentation in Natural Language, sAspects of Languagev, pp. 202-203. Bar Hillel criticized the amethodology of logical semanticists because they spent all their time on a partial task, without ]concern for complete coverage of language phenomena. One major prejudice... is the tendency _to assign truth values to indicative sentences in natural languages and to look at those cases Xwhere such a procedure seems to be somehow wrong... Argumentation in Natural Language, cAspects of Languagev, p. 203. There were in Bar Hillels time no extant proposals about how to b_translate natural language into a formal language amenable to processing by logic. The reason, ]according to Bar Hillel, was that such a translation process would have to rely on knowledge @about the world. HHˆ`HHˆ$ lda$$ HHˆa"HHˆf,,$1` 12 _7. Thus, acquiring world knowledge became a precondition for the success of an entire sequence aof enterprises culminating with machine translation. Bar Hillel considered this task infeasible, Yand this was the ultimate reason for his well-known pessimism about MT and computational Xsemantics. It seems now quite certain ... that with all the progress made in hardware, Zprogramming techniques and linguistic insight, the quality of fully autonomous mecahnical dtranslation, even when restricted to scientific or technological material, will never approach that Vof qualified human translators and that therefore MT will only under very exceptional `circumstances be able to compete with human translation. This pessimistic evaluation is based cupon various considerations, only one of which will be presented here, and even this, for obvious Zreasons, only very shortly and therefore dogmatically. Expert human translators use their \background knowledge, mostly subconsciously, in order to resolve syntactical and semantical Sambiguities which machines will have either to leave unresolved or resolve by some Zmechanical rule which will every so often result in a wrong translation. The Future of @AMachine Translation, sLanguage and Informationv, p.182. 7` 6 cBar Hillels famous example concerned the following text: sLittle John was looking for his toy fbox. Finally he found it. The box was in the pen. John was very happy.v Why is it that a machine " ]... is ... powerless to determine the meaning of pen in our sample sentence within the given _paragraph? The explanation is extremely simple, and it is nothing short of amazing that, to my Xknowledge, this point has never been made before, in the context of MT... What makes an \intelligent human reader grasp this meaning so unhesitatingly is ... his knowledge [of] the _relative sizes of pens, in the sense of writing implements, toy boxes and pens in the sense of ^playpens... Whenever I offered this argument to one of my colleagues working on MT, his first ^reaction was: But why not envisage a system which will put this knowledge at the disposal of dthe translation machine? Understandable as this reaction is, it is very easy to show its futility. ]What such a suggestion amounts to, if taken seriously, is the requirement that a translation ^machine should not only be supplied with a dictionary but also with a universal encyclopedia. aThis is surely utterly chimerical and hardly deserves any further discussion. Nonfeasibility of @3FAHQMT, sLanguage and Informationv, p.176.  ` ! ^8. The above line of reasoning led Bar Hillel to the conclusion that FAHQMT should not be the \stated goal of MT researchers, as significant practical advances could be obtained for less ]demanding objectives. ...[T]here is really no need at all to compromise in the direction of `reducing the reliability of the machine output. True enough, a smooth machine translation looks fimpressive... [but] it is much safer to compromise in the other direction. Let us be satisfied with a \machine output which will every so often be neither unique nor smooth, which every so often `will present the post-editor with a multiplicity of renderings among which he will have to take dhis choice, or with a text which, if it is unique, will not be grammatical. ... Let the machine ... cprovide the post-editor with all possible help, present him with as many possible renderings as he `can digest without becoming confused by the embarrass de richesse ... but never let the machine _make decisions by itself on purely frequential reasons even if these frequencies can be relied @Jupon. Aims and Methods of MT. sLanguage and Informationv. p.171. A3` HHˆa"HHˆ!'##lda'' HHˆa%HHˆ{--' UTUT`2. Now  \Our MT research group at CRL shares all of Bar Hillels premises, including that concerning 0 [unattainability of a complete database of world knowledge. We believe, however, that world _models are useful and feasible even when they are not complete or provably correct. This seems `to be in the spirit of Bar Hillels own opinion concerning competence and performance theories: _I have already voiced ... my misgivings over the conception of Chomsky and others that a more `or less complete development of a theory of competence is a prerequisite for the development of ba theory of performance. That this could not be so can be seen simply from the fact that the very Uadequacy of a particular theory of competence can only be determined on the basis of Xperformance, with or without theory Review of John Lyons Introduction to Theoretical @8Linguistics. sAspects of Languagev, pp371-72. ` !  XOur approach to FAHQMT-oriented research (e.g., Nirenburg et al., 1992, Onyshkevych and ZNirenburg, 1995) is based on the centrality of meaning and on a logical system underlying [meaning. The logical system is interpreted with the help of an ontology, a world knowledge @model. ` ! YHowever, at present, MT research and development at CRL and elsewhere, incorporates work Xwhose objectives cover the entire spectrum between theoretical and empirical linguistic Sapproaches, fully-automatic MT and machine-aided human translation tools and, most `prominently, between rule-based and corpus-based approaches, where the latter label applies ito systems which treat language as an artifact (the sum of all texts written in it or at least all texts Zavailable in electronic form) and apply general-purpose statistics-based methods to solve \problems such as translation. The debate between the rule-based and corpus-based approaches bhas enlivened the field of computational linguistics since about 1990. One possible assessment of @.the state of affairs in this debate follows. ` ! eMachine translation has been a fashionable field for at least forty years of its fifty-year history. ]The reasons for this vary from R&D glory to commercial payoff. Over the years, an impressive Zvariety of methods have been used as the basis for translation programs. The problem has, ahowever, proved so complex that the quality of the final result has not correlated significantly @Uwith the method chosen. Rather, it correlated with the amount of descriptive work on 8` language that was carried out. ` ! cOf course, MT research has brought about significant side benefits. Entire scientific fields were `created largely due to MT efforts: witness the nascence of computational linguistics. Often, MT `was used as an application of choice for a variety of workers to test and attempt to corrborate @_their theories of language and of human thinking capacity. It is characteristic that the final !? creport of the Eurotra project listed as its major success the creation of computational-linguistic Yinfrastructure in the countries of the European Community deemphasizing the fact that no ^realistic MT system was built under its auspices. Many factors contributed to the lack of the ^engineering achievement in this project, among them the relative lack of accent in Eurotra on [actual description and system building, with preference given to designing detailed formal @^specifications of (largely syntactic) levels of analysis and their corresponding formalisms. AG` HHˆa%HHˆ$*&&ldbU** HHˆbV(HHˆ‚..*H dIs the Eurotra case prototypical for the entire field of MT? One of the problems with the field has 0_been that the descriptive work is, frankly, rather monotonous and boring. This is why attempts `were made either to make it less boring (by adding an independently motivated theoretical angle @;to the descriptive work) or to try to avoid it altogether. M` !N \The latter objective was made manifest in a) attempts to use AI learning techniques or more Xpractical semi-automatic procedures for knowledge acquisition and b) the application of Zstatistical methods for establishing cross-linguistic correspondences in lieu of language @adescription work. The former solution made itself manifest in viewing MT as a testbed for ones !S \favorite linguistic or computational-linguistic theories, such as the currently fashionable Yprinciple-based approach to syntax. Machine translation is indeed a tempting avenue of Ycomputational inquiry into modeling human mental and language processes, and a number of Zapproaches to NLP in AI dabbled in MT as a potential application. Knowledge-based MT is a @&direct offshoot of the AI tradition. Y` !Z eThe most remarkable feature of the statistical methods in MT is that they are not at all specific to `their subject matter --- the same techniques applied to processing language could and are used, @1for example, in the studies of the human genome. ^` !_ _The current R&D-oriented MT approaches, whether rule-based or statistical or hybrid, are based [on imported ideas. At the same time, the best systems on the market cannot boast much by \way of technological or scientific advances. Instead, they rely on brawn: huge, handcrafted ddictionaries and grammars and a plethora of specialized translation routines. All of us are curious `to see how well the R&D approaches will work once sufficient resources are allocated for one or bmore of them to reach the status of a product. The question is: what kind of imported techniques ]shows the most promise? The answer is not clearly obvious and is determined by sociological @P(read: the vagaries of funding) as well as scientific and technological trends. h` !i _The major scientific (or methodological) trend in the field is experimenting with how well the astatistics-oriented methods will advance the state of the art in MT without the need for massive @manual knowledge acquisition. ` !m cThe major technological trend in the field looking for the best ways of mixing the statistical and _the rule-based methods. This author has been an early advocate of mixing such methods at the \level of their final results, a method called multi-engine MT. Other approaches seek a more @ainvolved interaction, with statistics used not only during the process of MT but also to support r`Gdevelopment of background resources (i.e., dictionaries and grammars). ` !t bThe major sociological trend, at least in the US, is the emphasis on a regimen of evaluations and [competitions among MT (and, more broadly, NLP) systems. This promotes rigor and discipline Yas well as conformity and search for local solutions, which are not necessarily the most Xpromising ones in the long run. Approaches that show a steady improvement are rewarded. @5Approaches with long gestation periods are punished. z` a{ aEmphasis on mixed approaches is, for non-statisticians, a rearguard regrouping action, while for astatisticians (witness the evolution of the claims and practices of the Candide IBM MT group), a HHˆbX(HHˆ'-))ldb-- HHˆb+HHˆ‚..-{@Esearch for any avenue for improving the rather modest final results. ` ! QThe knowledge-based and linguistics-based methods will do good by regrouping and aconcentrating on those tasks and situations in which statistical approaches fail to deliver. One Ymust, however, remember the lesson of computer chess: at present, the best chess-playing csystems are not terribly knowledgeable about chess strategy and tactics but they consistently beat YAI-based programs and compete on equal terms with grandmasters. The $64,000 question is: @Whow much more complex is human translation ability compared to the human chess-playing ! aability? That is, for how long will there be an opportunity to study language use through MT? If @[statistical methods succeed, rule-based MT may go the way of the AI-based chess programs. 5` !  YMT seems to be too complex a task to be fully accountable for by the current statistical Yprocessing methods, even though these methods do not aspire to building representational \models of human language capacity and rely only on the input-output behavior of such models `(in MT, a text and its translation). In the final analysis, the open-endedness of language will ]become the stumbling block for these methods conceptually, just as logistically, the chronic ^shortage of resources (bilingual corpora) may precipitate the swing of the pendulum of MT R&D @Kfashion back to the mentalist camp from its current behaviorist direction. ` ! [How long will this take? If history is any guide, such swings come roughly every 30 years: \mentalism was in scientific ascendancy between 1960 and 1990, while behaviorism reigned, at cleast in the US, for about thirty years prior to that. Of course, we cannot be certain that we are \witnessing this pendulum swing and not some other, unconnected development. Time will show. bA more intriguing thought is that, just possibly, the rule-based/corpus-based dichotomy is not as ^important as we currently think. Maybe the real problem of MT as technology is that it is not Zgenerally understood how difficult the problem actually is. The confident claims, made by Tnewcomers to MT (including this author some fifteen years ago), help stoke the high aexpectations of getting the desired result with a modest expenditure. At the current level of MT `R&D, either the expectations should be lowered or the time scale of getting the results must be @significantly extended. 9` !: ^The above list of opinions, though quite current in 1996, would not have been entirely out of _place around 1960, the time at which Bar Hillels celebrated assessments of MT were published. dI think the entire enterprise of MT will continue to be reasonably successful (recurring outlandish Xclaims coming from confident newcomers notwithstanding) as long as at least some of Bar @2Hillels legacy of self-assessment is preserved. ;` <` References =` >`VBar Hillel, Y. 1964. xLanguage and Informationv. Reading, MA: Addison Wesley. @`GBar Hillel, Y. 1970. xAspects of Languagev. Jerusalem: Magnes. A`eKatz, J. and J. Fodor. 1963. The Structure of a Semantic Theory. sLanguagev, 39, pp.170-210. B VNirenburg, S., J.Carbonell, M.Tomita and K.Goodman. 1992. xMachine Translation: A @?Knowledge-Based Approach.v San Mateo, CA: Morgan Kaufmann. C WOnyshkevych, B. and S. Nirenburg. 1995. A Lexicon for Knowledge-Based MT. sMachine B@ Translationv, 10, pp. 5-57. HHˆb+HHˆ*,,ldc00 $$c.$$g1-0G@mR $$c.$$//l} c.4 /GD eHeadings Table } c. /GE e } c. /GF e }lc.15l/GI eHeading Level }c.46/GJ eParagraph Format }c.57/GK e Comments }lc.68l/GL e2 }c.79/GOUTe Heading1 }c.8./GP e dc<< $$c:$$ƒ*<mqu }ydgjX\`=@CUROLIFGQ@mXWVUTS $$c:$$4;; l} c2c@ 3GR eCharacter Macros } c2 3GT e } c2 3GU e }Hc2=AH3GV e Character }c2@B3GW e Replace With }c2AC3GX e Comments }Hc2BDH3G[ e }c2CE3G\ e¢ }c2DU3G] e }Hd2KGH3G` e }d 2FH3Ga e... }d 2G3Gb e }Hd 2NJH3Gc e }d2IK3Gd e- }d2JF3Ge e }Hd2QMH3 Gf e }d2LN3 Gg e-- }d2MI3 Gj e }Hd2TPH3 Gk e }d2OQ3 Gl e° }d2PL3 Gn e }Hd2WSH3 Go e }d!2RT3 Gp e® }d#2SO3 Gq e }Hd%2EVH3 Gs e }d'2UW3 Gu e© }d)2VR3 Gv e } d]2l\ 3 Gw eGeneral Macros } d_2 3 Gx e } da2 3 Gy e } dc2 3 G| e }hde2X]h3G} e Macro Name }hdg2\^h3G~ e Replace With }hdi2]_h3G eHead }dk2^`3G e Comments }hdm2_ah3G e }hdo2`bh3G e }hdq2ach3G e }ds2b=3G e } d2|g 3 G eCross-Reference Macros } d2 3 G e } d2 3 G e }d2dh3 G e Macro Name }d2gi3 G e Replace With }d2hj3 G e Comments }d2ik3 G e See Also }d2jl3 G eSee <$paratext> }d2kX3 G e } d2q 3 G eSystem Macros } d2 3 G e } d2 3 G e } d2 3 G e }hd2mrh3 G e Macro Name }hd2qsh3 G e Replace With }hd2rth3 G eHead }hd2suh3 G e Comments }h:d2tvh:3 G e StartOfDoc }h:d2uwh:3 G e }h:d2vxh::3 " e ���e <$defaulttitle>  ���e A!e }h:d2wh:3 G e }he22zh3 G# eEndOfLastSubDoc }he42y{h3 G$ e }he62z|h3 G% e }he82{dh3 G& e }h:e:2~h:3 G' eStartOfLastSubDoc }h:e<2}h:3 G( e }h:e>2~h::3 G e )���e <$defaulttitle> E���e AFe }h:e@2yh:3 G* e }heB2h3 G+ eEndOfFirstSubDoc }heD2h3 G, e }heF2h3 G- e }heH2}h3 G. e }h:eJ2 h:3 G/ eStartOfFirstSubDoc }h:eL2h:3 G0 e }h:eN2h::3 D e 1���e <$defaulttitle> B���e ACe }h:eP2h:3 G2 e }heR2 h3 G3 e EndOfSubDoc }heT2 h3 G4 e }heV2 h3 G5 e }heX2 h3 G6 e }h:eZ2h:3 G7 eStartOfSubDoc }h:e\2 h:3 G8 e }h:e^2h::3 A e 9���e <$defaulttitle> ?���e A@e }h:e`2 h:3 G: e }heb2xh3 G; e EndOfDoc }hed2h3 G< e }hef2h3 G= e }heh2 h3 G> e } e:9 ; GH eHTML Options Table } e: ; GI e } e: ; GJ e }e:; GK eControl }e:; GL eValue }He:H; GM e Comments }e:; GN e Image Format }e:; GO e0001IMAGGIF GIF }He:(H; GP e } e:# ; GQ eHTML Mapping Table } e: ; GR e } e: ; GS e } e: ; GT e } e: ; GU e }.e:$.; !GV eFrameMaker Source Item }~e:#&~; !GW e HTML Item }6e:6; !GX e }6.e:$'6.; !gY %Include EAuto# }.e:&).; !GZ e Comments }e:; "G[ e }He:'*H; "G\ eElement }6e:)-6; "g] %New Web EPage? }6e:6; "G^ e }e:; "G_ e }e:*.; #G` eP:Body }He:-/H; #Ga eP }6e:.06; #Gb eN }6e:/16; #Gc eN }e:0#; #Gd e df44 $$f2$$JmH4mqu }ydgjX\`=@CUROLIF$$f2$$<33 l}f^:>6; $Ge eX:Table Number & Page }Hf`:57H; $Gf e See Also }6fb:686; $Gg eN }6fd:796; $Gh eN }ff:8; $Gi e }fh:C;; %Gj e X:Table All }Hfj::<H; %Gk e See Also }6fl:;=6; %Gl eN }6fn:<>6; %Gm eN }fp:=5; %Gn e }fr:H@; &Go eX:Table & Page }Hft:?AH; &Gp e See Also }6fv:@B6; &Gq eN }6fx:AC6; &Gr eN }fz:B:; &Gs e }f|:ME; 'Gt eX:See Heading & Page }Hf~:DFH; 'Gu e See Also }6f:EG6; 'Gv eN }6f:FH6; 'Gw eN }f:G?; 'Gx e }f:RJ; (Gy eX:Page }Hf:IKH; (Gz e See Also }6f:JL6; (G{ eN }6f:KM6; (G| eN }f:LD; (G} e }f:WO; )G~ eX:Heading & Page }Hf:NPH; )G e See Also }6f:OQ6; )G eN }6f:PR6; )G eN }f:QI; )G e }f:\T; *G eC:EquationVariables }Hf:SUH; *G eEM }6f:TV6; *G eN }6f:UW6; *G eN }f:VN; *G e }f:aY; +G e C:Emphasis }Hf:XZH; +G eEM }6f:Y[6; +G eN }6f:Z\6; +G eN }f:[S; +G e }f:f^; ,G eP:Title }Hf:]_H; ,G eH* }6f:^`6; ,G eN }6f:_a6; ,G eN }f:`X; ,G e },f:kc,; -G e P:TableTitle }H,f:bdH,,; - eLI ^e Parent = OL A_e Depth = 0 }6,f:ce6,; -G eN }6,f:df6,; -G eN },f:e],; -G e }f:ph; .G eP:TableFootnote }Hf:giH; .G eP }6f:hj6; .G eN }6f:ik6; .G eN }f:jb; .G e },f:um,; /G e P:Numbered1 }H,f:lnH,,; / eLI \e Parent = OL A]e Depth = 0 }6,f:mo6,; /G eN }6,f:np6,; /G eN },f:og,; /G e },f:zr,; 0G! e P:Numbered }H,f:qsH,,; 0" eLI Ze Parent = OL A[e Depth = 0 }6,f:rt6,; 0G# eN }6,f:su6,; 0G$ eN },f:tl,; 0G% e }f:w; 1G& eP:Mapping Table Title }Hf:vxH; 1G' eP }6f:wy6; 1G( eN }6f:xz6; 1G) eN }f:yq; 1G* e }f:|; 2G+ eP:Mapping Table Cell }Hf:{}H; 2G, eP }6f:|~6; 2G- eN }6f:}6; 2G. eN }f:~v; 2G/ e }f: ; 3G0 e P:Indented }Hf:H; 3G1 eP }6f:6; 3G2 eN }6f:6; 3G3 eN }f:{; 3G4 e }f:; 4G5 eP:HeadingRunIn }Hg:H; 4G6 eP }6g:6; 4G7 eN }6g: 6; 4G8 eN }g:; 4G9 e }g: ; 5G: e P:Heading2 }Hg : H; 5G; eH* }6g : 6; 5G< eN }6g: 6; 5G= eN }g: ; 5G> e }g:; 6G? e P:Heading1 }Hg:H; 6G@ eH* }6g:6; 6GA eN }6g:6; 6GB eN }g: ; 6GC e }g:; 7GD e P:Footnote }Hg:H; 7GE eP }6g :6; 7GF eN }6g":6; 7GG eN }g$:; 7GH e }g&:"; 8GI eP:CellHeading }Hg(:H; 8GJ eP }6g*:6; 8GK eN }6g,:6; 8GL eN }g.:; 8GM e }g0:'; 9GN e P:CellBody }Hg2: H; 9GO eP }6g4:!6; 9GP eN }6g6: "6; 9GQ eN }g8:!; 9GR e },g::1$,; :GS e P:Bulleted }H,g<:#%H,,; :T eLI Xe Parent = UL AYe Depth = 0 }6,g>:$&6,; :GU eN }6,g@:%'6,; :GV eN },gB:&,; :GW e }h:); ;G` e!Copy Files Imported by Reference }h:(*; ;Ga eN }Hh:)H; ;Gb e }lim.0,l/<Gc e1 }io.+-/<Gd eTitle }iq.,/<Ge e }lis.9/l/=Gf e3 }iu..0/=Gge Heading2 }iw./+/=Gh e dLeftdRightd  Referenceddd"d%d (d+d .Headingsd :HTMLd 2HTML&@ H    h  Body. @  H    h  Bulleted\t. f@ CellBody. f@  CellHeading. f@  Footnote. f@T Heading1Body. f@T  Heading2Body. @P  H    h  HeadingRunInBody. @ H    h  Indented. @  H    h  Numbered.\t. @A  H    h  Numbered1.\tNumbered. f@  TableFootnote. f@T   TableTitleT:Table : .  @@P TitleBody. f@T   TableTitleT:Table : . f@  CellHeading. f@ CellBody. f@  CellFooting.  @@P TitleBody. @ H    h  Body. @@    Header. @@    Footer. @@    Header. @@    Footer. f@T Heading1Body. RR1f@ Body. @  H    h  Body. @@   Body. @  H    h  Body. @@   Mapping Table Title. @@  Mapping Table Cell. @@ Mapping Table Title. @@   Mapping Table Cell. @ H    h  Body. @@ Mapping Table Cell. @  H    h  Body. @@  Mapping Table Cell. @@ Mapping Table Cell. @  H    h  Body.   Emphasis  EquationVariables                            ThinMediumDoubleThick@ Very Thin HHHHHFormat B HHHHHFormat AH Mapping TableH Mapping Table <lh pH  hhh   ( hhhh  ;H> $H66= /123/456=/789h 3=>?h3@ABh 3CDEh3FGHh 3IJKh 3L M N h 3O P Q h 3R S T h 3U V W  3X Y Z [  3\]^_3`abc 3def3ghi3jkl 3mnop3qrst:3uvwx3yz{|:3}~3:33    :3 3 ;;;; !;   ! "  ";#!$!%!&!'!!#;(")"*"+","":;-#.#/#0#1#%;5$6$7$8$9$&$;:%;%<%=%>%'%;?&@&A&B&C&(&;D'E'F'G'H')';I(J(K(L(M(*(;N)O)P)Q)R)+);S*T*U*V*W*,*;X+Y+Z+[+\+-+;],^,_,`,a,,.,;b-c-d-e-f-/-;g.h.i.j.k.,0.;l/m/n/o/p/,1/;q0r0s0t0u020;v1w1x1y1z131;{2|2}2~2242;3333353;4444 464; 5 5 5 5575;6666686;7777797;88888:8;99 9!9"9,#9;#:$:%:&:':;(;);*;=/+<,<-<`=4Ak|eZ*R&=