Aad  0``0  ` ` PP00`HH $ @d HHHH̀̀̀ff@  d Footnote TableFootnote**.\t.\t/ - :;,.!?1btbuTOCHeading1Heading2   EquationVariables>56677,7E7996@ _Hlt422568980<$lastpagenum><$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 All7Table<$paranumonly>, <$paratext>, on page<$pagenum>Table Number & Page'Table<$paranumonly> on page<$pagenum> &HTML Headings33A55A7799;;==?XA]555555555555555555555;Q;S5 5555555555555555555555555555555555;;6 66666 6 66666666;;6  6"6$6&6(< <6, 6.60626466686:6<6>6@6B6D6F6H6J6L6N6P6R6T6V6X6Z6\6^6`6b6d6f6h6j6l6n6p6r6t6v6x6z6|6~666666666666666666666666>>+>6666666666>6666666> 6666666>6666666>677777 7 7 7777777777!7#7%7'7)7+7.70727476787:7<7>7@7B7D7G7I7K7M7O7Q7S7U7W7Y7[7]7_7a7c7e7g7i7k7m7o7q7s7u7w7y7{7| 77777777777777 777 777 77 77777>*>:>>*:7 7 77777 8=8>8?8?8@8@8@9A9A9?9B9B9?9B9B9B9 B9 B9 B9 B9 B9B9?9C9D9C9E9F9G9G9G9G9G9G9G9D9G9G9 H9!D9"I9#J9$H9%K9&G9'G9(G9)L9*G9+M9,N 9-N 9.N 9/N 90N 91O92P93O94P95Q1. 98R99S9:T9;Q1. 9<U9=T9>U9?U9@T9AU9BU9CT9DU9ET9FQ1. 9GU9HU9IT9JQ1. 9KU9LT9MU9NT9OQ1. 9PU9QT9RQ1. 9SU9TT9UQ1. 9VV9WT9XQ1. 9YU9ZT9[Q1. 9\U9]T9^Q1. 9_P9`?9aW9b?9c?9dX(1) 9eX(1) 9f?9gX(1) 9hX(1) 9iX(1) 9j?9k?9l?9m?9n?9o?9p?9q?9r?9s?9t?9u?9v?9w?9x?9y?9zY9{?9|?9}?9~?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9?9Z9[9?9?9\9?9?9?9?9?9?9?9?9?9?9?9?:?:?:?>>:>)>;<<<<<<<<<<<<<<<<<<<<<<<<<<<====== = ========== ="dq=3dq95d7##d5y HmR5zHmRHRHRFootnote Hr@5{Hr@HzHz Single LineH5| Footnote 5}  HD5~ HDHH Double LineH5 Double Line5 5 H5  Single Line5 HZ5  TableFootnoted5p?? EGxR5EGxREPwEPw TableFootnoted5$$5$$l $$5$$5_5`ejoty~ '{vqlgb]1, &*.26:>BFJNQTWZ]`dhlorux{~W{ @l }$6 5$6 WdHTML Mapping Table }6 56 We }6 56 We }26 526 We }_6 5_6 We }$F.5$F.W dFrameMaker Source Item }F~5F~W d HTML Item }F65F6W e }2F-.52F-. $Include PEAuto# }_F?.5 _F?.W  d Comments }$V5$VW  e }VH5!VHW  dElement }V65 $V6 $New Web PEPage? }2V-52V-W e }_V?5_V?W e }$t5!%$tW dP:Body }tH5$&tHW dP }t65%'t6W dN }2t-5&(2t-W dN }_t?5')_t?W d }$,5(*$,W d P:Bulleted }H,5)+H,,dLI e Parent = UL Qe Depth = 0 }6,5*,6,W dN }2-,5+-2-,W dN }_?,5,._?,W d }$5-/$W d P:CellBody }H5.0HW dP }65/16W dN }2-5022-W dN }_?513_?W  d }$524$W! dP:CellHeading }H535HW" dP }65466W# dN }2-5572-W$ dN }_?568_?W% d }$579$W& d P:Footnote }H58:HW' dP }659;6W( dN }2-5:<2-W) dN }_?5;=_?W* d }$5<>$ W+ d P:Heading1 }H5=?H W, dH* }65>@6 W- dN }2-5?A2- W. dN }_?5@B_? W/ d }$5AC$ W0 d P:Heading2 }H5BDH W1 dH* }65CE6 W2 dN }2-5DF2- W3 dN }_?5EG_? W4 d }$5FH$ W5 dP:HeadingRunIn }H5GIH W6 dH* }65HJ6 W7 dN }2-5IK2- W8 dN }_?5JL_? W9 d }$,5KM$, W: d P:Indented }H,5LNH,, =dP ;e Parent = UL Q<e Depth = 0 }6,6MO6, W> dN }2-,6NP2-, W? dN }_?,6OQ_?, W@ d }$<6PR$< WA dP:Mapping Table Cell }<H6 QS<H WB dP }<66 RT<6 WC dN }2<-6 SU2<- WD dN }_<?6TV_<? WE d }$L6UW$LWF dP:Mapping Table Title }LH6VXLHWG dP }L66WYL6WH dN }2L-6XZ2L-WI dN }_L?6Y[_L?WJ d }$\,6Z\$\,WK d P:Numbered }\H,6[]\H,,NdLI Le Parent = OL QMe Depth = 0 }\6,6!\^\6,WO dN }2\-,6#]_2\-,WP dN }_\?,6%^_\?,WQ d }$b,6'a$b, WR d P:Numbered1 }bH,6)`bbH,, UdLI Se Parent = OL QTe Depth = 0 }b6,6-acb6, WV dN }2b-,6/bd2b-, WW dN }_b?,61ce_b?, WX d }$63df$ WY dP:TableFootnote }H65egH WZ dP }667fh6 W[ dN }2-69gi2- W\ dN }_?6;hj_? W] d }$6=ik$ W^ d P:TableTitle }H6?jlH W_ dH* }66Akm6 W` dN }2-6Cln2- Wa dN }_?6Emo_? Wb d }$6Gnp$ Wc dP:Title }H6IoqH Wd dH* }66Kpr6 We dN }2-6Mqs2- Wf dN }_?6Ort_? Wg d }$6Qsu$ Wh d C:Emphasis }H6StvH Wi dEM }66Uuw6 Wj dN }2-6Wvx2- Wk dN }_?6Ywy_? Wl d }$6[xz$ Wm dC:EquationVariables }H6]y{H Wn dEM }66_z|6 Wo dN }2-6a{}2- Wp dN }_?6c|~_? Wq d }$6e}$ Wr dX:Heading & Page }H6g~H Ws dHeading }66i6 Wt dN }2-6k2- Wu dN }_?6m_? Wv d }$6o$ Ww dX:Page }H6qH Wx dHeading }66s6 Wy dN }2-6u2- Wz dN }_?6w_? W{ d }$6y $ W| dX:See Heading & Page }H6{ H W} d See Also }66} 6 W~ dN }2-6 2- W dN }_?6 _? W d }$6 $ W d X:Table All }H6 H W d Table All }666 W dN }2-62- W dN }_?6_? W d }$6$ W dX:Table Number & Page }H6H $Table PENumber }666 W dN }2-62- W  dN }_?6'_? W  d }$R 60$R "W dHTML Options Table }R 6R "W e }hR 6hR "W e }$b6$b"W dControl }b6b"W dValue }hbH6hbH"W d Comments }$r6$r"W d Image Format }r6r"W d0001IMAGGIF GIF }hrH6 hrH"W d }$6!$"W d!Copy Files Imported by Reference }6 ""W dN }hH6!&hH"W d }l> $l Ge8 }u> #%u Ge Plain Text }H> $H Ge }$$ 6"*$$ " WdSystem Macros }$ 6$ " We }$ 6$ " We }\$ 6\$ " We }$4h6&+$4h"!W d Macro Name }4h6*,4h"!W d Replace With }4h6+-4h"!W  dHead }\4B6,.\4B"!W! d Comments }$Dh6-/$Dh""W" d StartOfDoc }Dh6.0Dh""W# d }Dh6/1Dh""g% <$default���Etitle> }\DB602\DB""W% d }$ph613$ph"#W& d EndOfDoc }ph624ph"#W' d }ph635ph"#W( d }\pB646\pB"#W) d }$h657$h"$W* dStartOfSubDoc }h668h"$W+ d }h679h"$g% <$default���Etitle> }\B68:\B"$W- d }$h69;$h"%W. d EndOfSubDoc }h6:<h"%W/ d }h6;=h"%W0 d }\B6<>\B"%W1 d }$h6=?$h"&W2 dStartOfFirstSubDoc }h6>@h"&W3 d }h6?Ah"&g-% <$default���Etitle> }\B6@B\B"&W5 d }$h6AC$h"'W6 dEndOfFirstSubDoc }h6BDh"'W7 d }h6CEh"'W8 d }\B6DF\B"'W9 d }$h6EG$h"(W: dStartOfLastSubDoc }h6FHh"(W; d }h6GIh"(g0% <$default���Etitle> }\B6H\B"(W= d }$$h6$K$$h%)W> dEndOfLastSubDoc }$h7$JL$h%)W? d }$h7$KM$h%)W@ d }\$B7$LN\$B%)WA d }$: 7$MQ$: %*WBdCross-Reference Macros }: 7 $: %*WCe }D: 7 $D: %*WDe }$J7$NR$J%+WE d Macro Name }J7$QSJ%+WF d Replace With }DJ?7$RTDJ?%+WG d Comments }$Z7$SU$Z%,WH dHeading }Z7$TVZ%,WI d <$paratext> }DZ?7$UWDZ?%,WJ d }$j7$VX$j%-WK d See Also }j7$WYj%-WL dSee <$paratext>. }Dj?7$XZDj?%-WM d }$z7 $Y[$z%.WN d Table All }z7"$Z\z%.gO%Table <$paranumonly>, E <$paratext> }Dz?7$$[]Dz?%.WP d }$7&$\^$%/WQ d Table Number }7($]_%/WR dTable <$paranumonly> }D?7*$^`D?%/WS d }$Z 7-$_d$Z %0WTdGeneral Macros }vZ 7/$vZ %0WUe }Z 71$Z %0WVe }2Z 73$2Z %0WWe }$jev75$`e$jev%1WX d Macro Name }vjev77$dfvjev%1WY d Replace With }jC79$egjC%1WZ dHead }2jQ7;$fh2jQ%1W[ d Comments }$zev7=$gi$zev%2W\ d }vzev7?$hjvzev%2W] d }zC7A$ikzC%2W^ d }2zQ7C$jl2zQ%2W_ d }$ 7F$ko$ %3W`dCharacter Macros }l 7H$l %3Wae } 7J$ %3Wbe }$H7L$lp$H%4Wc d Character }l7N$oql%4Wd d Replace With }?7P$pr?%4We d Comments }$H7R$qs$H%5Wf d }l7T$rtl%5Wg d¢ }?7V$su?%5Wh d }$H7X$tv$H%6Wi d }l7Z$uwl%6Wj d© }?7\$vx?%6Wk d }$H7^$wy$H%7Wl d }l7`$xzl%7Wm d® }?7b$y{?%7Wn d }$H7d$z|$H%8Wo d }l7f${}l%8Wp d° }?7h$|~?%8Wq d }$H7j$}$H%9Wr d }l7l$~l%9Ws d-- }?7n$?%9Wt d }$H7p$$H%:Wu d }l7r$l%:Wv d- }?7t$?%:Ww d }$H7v$$H%;Wx d }l7x$l%;Wy d... }?7z$?%;Wz d d7~$$7$$# l $$7$$#`f`ejoty~ '{vqlgb#]1, &*.26:>BFJNQTWZ]`dhlorux{~d7 $$h7 $$h l $$h7 $$h % W @l }$h 7 $h  <W|dHeadings Table }h 7 h  <W}e }h 7 h  <W~e }$Cl7 $Cl =W dHeading Level }Cu7 Cu =W dParagraph Format }CH7 CH =W d Comments }$Sl7 $Sl >W d1 }Su7 Su >W dTitle }SH7 SH >W d }$il7 $il ?W d2 }iu7 iu ?W UTUTd Heading1 }iH7 iH ?W d }${l7 ${l @W d3 }{u7 {u @W  d Heading2 }{H7 {H @W  d }$l7 $l AW  d4 }u7 u AW d HeadingRunIn }H7 H AW  d }$l7  $l BW d4 }u7 !u BW  d TableTitle }H7 H BW d $$7$$]I#]1, &*.26:>BFJNQTWZ]`dhlorux{~$$7$$&""ld7&& $$7$$$pJ&JNQTWZ]`dhlorux{~$$7$$$#%%l}$<7($< CW d P:Blockquote }<H7')<H CW dP }<67(*<6 CW dN }2<-7)+2<- CW dN }_<?7*{_<? CW d }<\-"DG1e P:Plain Text }H<,.H"DG2eH* }6<-/6"DG3eN }-<.0-"DG4eN }?</?"DG5e }<aY"EG6e P:heading 5 HHˆ7HHˆ 3G e HHˆ7;HHˆ22l HHˆ7HHˆ 5G e HHˆ77HHˆ44l HUV 7HUV 7W` HUV 795HUV 66l H$ 7H$ 9W ` H$ 77H$ 88l HUV 7HUV ;G!e HUV 7=3HUV ::l H$ 7H$ =G"e H$ 7;H$ <<l HHˆ7HHˆ8$$?$z+z` %%u“`0Title: Where Interlingua Can Make a Difference &z+f` 'f`HAuthors: Keith J. Miller (MITRE Corporation and Georgetown University) (`Phone: (703) 883-6920 )`Fax: (703) 883-1279 *`Email: keith@mitre.org +`Mailing Address: ,`The MITRE Corporation -x+m`!Washington C 3  Center .'%wh`1820 Dolley Madison Boulevard /`McLean, VA 22102-3481 0z+J ` 1J `>David M. Zajic (MITRE Corporation and University of Maryland) 2`Phone: (410) 850-5815 3`Fax: (410) 850-5459 4`Email: dzajic@mitre.org 5`Mailing Address: 6`The Mitre Corporation 7`Commons Corporate Center 8`7467 Ridge Rd, Suite #140 9`Hanover MD, 21076 :` ; 2-` <u_ eAbstract:  Furnished with an English text and its equivalent in 13 foreign languages, we set out 2%n _to determine the potential for improving the quality of translation results by an Interlingual Z(IL) approach to machine translation (MT) as compared to transfer-based systems. In this Wpaper, we analyze the errors made by two commercial transfer-based MT systems, provide \an observational classification of the errors, and group the errors according to whether or Znot an Interlingual approach would improve system output. Then, for several examples, we \identify the prerequisites for an IL representation, and provide a sample IL representation ]as a straw man for discussion at the workshop. The idea is to 1) look at cases that might ^prove to be particularly complex with respect to meaning representation, and 2) to illustrate @Wcases in which an IL system could compensate for weaknesses in transfer-based systems. = 2-,` S>`SSubject area keywords: Interlingual Machine Translation, LCS, Machine Translation HHˆ7HHˆC>> l~H¯97DH¯E `z+z%p Although we list only two subclasses of Untranslated Words, this is far from a complete treatment of even this pyr(seemingly simple) class of errors. For a more thorough treatment of untranslated words in transfer-based MT sysE"tems, see (Reeder & Loehr, 1998). d:CC HHˆ:AHHˆ‚99C?uˆe(Where Interlingua Can Make a Difference @z+fe!Keith J. Miller & David M. Zajic Afm* keith@mitre.org dzajic@mitre.org BeThe MITRE Corporation Ce De E JTe Abstract Fz+U6>e Gvȁj^؋%Furnished with an English text and its equivalent in 13 foreign languages, we set out to determine the potential for improving the quality of trans0h lation results by an Interlingual (IL) approach to machine translation (MT) as compared to transfer-based systems. In this paper, we analyze the errors made by two commercial transfer-based MT systems, provide an observational classification of the errors, and group the errors according to whether or not an Interlingual approach would improve system output. Then, for several examples, we identify the prerequisites for an IL representation, and provide a sample IL representation as a straw man for discussion at the workshop. The idea is to 1) look at cases that might prove to be particularly complex with respect to meaning representation, and 2) to illustrate cases in which an IL system could compensate for weakE"nesses in transfer-based systems. Hz+e I%ue Introduction JJ=e K#Scº %iThough the debate between proponents of transfer-based MT and Interlingual MT is far from being settled, 0Πnlit is not a point of contention that a successful Interlingual MT system would provide certain advantages. iFrom an engineering perspective, the principal benefit of an IL system is the reduction in the number of cmodules necessary to enable translation between a new language pair. From a research perspective, gattempts to design a workable IL are desirable in that they provide a clearer view of the relationship ibetween a languages surface forms and the underlying meaning representations of those forms. They also mprovide an environment for the testing of ideas and theories that elucidate the universality of the semantic E$structure of the worlds languages. Lz+!>Xe MoAG,z%]Our interest in this workshop stems from practical concerns. We are currently involved in a 08^aMachine Translation project, CyberTrans, which provides translation services to a wide community bof users. By providing a common interface to two large commercial transfer-based systems and sev`eral smaller translation systems, CyberTrans gives this community easy access to a considerable fsuite of Machine Translation tools. The maintenance of separate domain-specific lexicons for each of bthese systems, however, is a large task. In an effort to make this task more manageable, we have bundertaken a project known as the Lexicon Service Bureau (LSB). As described in (Miller & Zajic, c1998), the LSB project has as its goal to provide a unified lexical resource from which the system-gspecific lexicons can be generated for the transfer-based systems. The design for the LSB calls for a dsingle lexicon for each language handled by the translation systems, with links between monolingual blexicons providing the information necessary to generate the system-specific bilingual lexicons. bSince source language (SL) to target language (TL) mappings are not all be one-to-one, the design gcalls for links to be made by means of a pivot language (or IL). Thus, we are interested in the deter`mination of what such a pivot must entail. Furthermore, we are interested in discovering which `types of MT problems are particularly suited to an IL approach. That is, what are the areas in ^which an IL approach to MT could solve problems currently encountered by the state-of-the-art EBtransfer based MT systems that we currently provide to our users? NJ=s>e O%u!eOur goals and procedure Pz+ le Q#Sc%lSince we are concerned with the potential benefits to be gained in an operational setting in which analysts p!N hare interested primarily in viewing English translations of foreign language documents, our approach to jthis workshop was likely somewhat different from that of other participants. We began with the text from rthe  UNESCO Courier  that served as a basis of the workshop (Otero, 1997). This text was available in 13 elanguages, of which we chose to focus on the English, French, Spanish, Russian, and German versions, hprincipally for practical reasons. We also had access to two transfer-based MT systems (Systran and an kolder version of Globalink), a system that provided English glosses of the foreign language texts, and, in lsome cases human translations of the texts. As a method of pointing to areas in which the transfer systems gneeded improvement, we submitted each of the four foreign language texts to the two transfer-based sysktems (for translation into English). We then attempted to locate errors/disfluencies in the output of the HHˆ:AHHˆ?FBB ld: FF HHˆ: DHHˆ€33@@FQ#ScSitransfer-based MT systems. From the list of errors, we developed an observational classification of the 0i ierror types, and partitioned these classes into two groups: those containing errors that an IL approach lwould correct, and those containing errors that would not be ameliorated by use of an IL. This grouping of merror classes will be presented in the next section of this paper. Furthermore, we compared the various lanhguage versions of the document with each other and with the English translation/gloss of the document. jThe identification of the error classes and the cross-lingual comparison of the documents pointed to many kinteresting cases in which an Interlingual approach (or at the very least, an approach with a deeper semanltic representation) would improve the capability of the MT system to provide a fluent English translation. hFinally, we chose several examples from the large pool we had collected. For the selected examples, we Eepropose an Interlingual representation, based on Dorrs (1993) LCS, to be discussed at the workshop. Rz+}le S €eClassification of Errors Tz+e U#ScR%fWe submitted each of the foreign language documents to the two-transfer-based systems for translation 9Ekinto English. For each error encountered in the output, we entered the following information into a grid: Ve#the original source language text !W%ethe output of each of the MT systems (For each segment containing an error by either MT system, the houtput of both systems was entered into the grid. It is interesting to note that the two systems often E_made errors on the same segment; sometimes the errors were the same, but often they were not.) Xe/the human translation or gloss of the segment YeDthe corresponding segment from the English version of the document Ze+a slot for a classification of the error. [z+ Re \#Scl*%mClassification of the errors was then done based on all of the factors listed above, as well as the authors 0!Rnknowledge of the systems in question. The result is the observational classification of errors below. We do knot claim that this is a complete classification of errors made by transfer-based MT systems, or even that qthis is the correct classification. The purpose of the classification is merely to separate those problems that E9are amenable to an IL treatment from those that are not. ]z+P=Je ^#Sc["eSWe identified four classes of errors that would not be remedied by an IL approach: _g҇eUntranslated Words avcuX-yWe witnessed two distinct subclasses of this class  . First, there were untranslated words due to errors in the )z+vsource text. The Spanish text, for example, contained hyphenated words in the middle of a line of running text (e.g. *infe rior  for  inferior ), presumably an artifact of the conversion of the document from a previous format. Secondly, )qthere were, not surprisingly, gaps in the MT systems lexicons (unknown words). Effects of this ranged from the Tzsystems attempting to decompose the unknown word and translate its constituent parts (e.g.  micro-entreprise      micro *+ wentreprise , and then translated as microphone company in the French document) to the system leaving the word in ) EHthe source language (e.g.  unterstzt  in the German document). be c#ScϻeAnalysis Problems 1dۢ%cIn certain cases, there were problems with the analysis of the source language. Based on the data lavailable to us in the grid mentioned in the previous section, these errors were either due to inability to brecognize and handle proper nouns (in various documents, the authors name, Mara Otero, was varicously translated as Marrying Otero, Maria Hill, and Maria Knoll), inadequate morphological processhing, or incomplete coverage of the grammar. As an example of the last of these, consider the following E example: ez+"Ye f-Ye/Sp: En cinco aos BancoSol se ha convertido ge4Gls: In five years BancoSol itself has converted he Aie.Human : In five years BancoSol has become HHˆ: DHHˆCIEE ld: II HHˆ:GHHˆx66Ijz+ze5Systran: Into five BancoSol years one has become . kye l#ScQ%qWe can assume that Systran analyzed BancoSol as a modifier of  aos , and then, left with no subject for 0igthe following verb, inserted the generic pronoun one. An Interlingual representation would not help ECin cases in which the syntactic analysis has similarly gone askew. mz+@e n#ScL6e$Failure to recognize Phrasal Units 1oX%mThis is similar to the first class in that it is a problem that points to gaps in the MT systems lexicons. qFor instance, Systran translated the French phrase   but non-lucratif  (to/with goal non-lucrative) as z+o:E p#Scz%kwith nonlucrative goal, rather than as non-profit. There is the possibility that an Interlingua could 0ehandle gaps in the phrasal lexicon for fixed phrases of which the meaning is compositional, however, EEan Interlingua is a powerful tool, suited for more complex problems. qz+e r#SceInherently Ambiguous Source 1s@ %iThis class is similar to the second class in that the correct semantic analysis hinges on the ability to kresolve surface-level ambiguity; however, this class refers more to the level of lexical ambiguity. It is ctruly a borderline call as to whether this type of problem can be handled by using an Interlingual capproach. The determination will depend on how strong one assumes the analysis modules to be. If dthe assumption is that all ambiguity would be removed during analysis, thus arriving at an unambigudous IL representation, then an appropriate translation would be produced. Consider, though the folglowing example from the German text. On the morphological level,  Darlehen  is ambiguous with jrespect to number. There is nothing else in the source language sentence that indicates whether it is to hbe singular or plural; a human is able to disambiguate based only on world knowledge. If this is to be hhandled by an IL system, it would be by virtue of access to an extensive knowledge base and the ability E+to reason out relevant inferences from it. tz+7e u#ScCi%iMore interestingly, we discovered that six of our classes would benefit from an Interlingual treatment. OOEThey were the following. vz+Ze w#Scf6_eIncorrect Word Sense 1xr%fNaturally, since an IL representation of a source language sentence is intended to be unambiguous and `meaning-preserving, cases involving some problem with word sense would benefit from an IL treathment. This is a vast class, and includes issues such as polysemy, more specifically problems with prepgositional meaning, and fine-grained sense distinctions. For example, the French document contains the bcglsubheading  Un succs clatant  (lit. a success bursting), which was translated as A Vivid Suce (in loc (cities) ?e' (United States))))))))) @e (providingly)) Ae (currently))))) Bz+e C!ScE%`To clarify the sense of this representation, we offer a rough human translation of it. ACCION 0,?mInternational is identificationally an organization. The organization has the characteristics of being U.S.-nbased, private, non-profit, and that it participates as agent in a cause event. In the representation of the lcause event it is referred to as *head*, meaning that the object modified participates in the IL which modipfies it. In this case, the organization causes technical assistance to to go possessionally to a network. The nnetwork is composed of institutions, which are in the conjunction of countries and cities. The countries are lfurther modified as being thirteen and in Latin America; the cities are six and in the United States. The jorganization causes this to happen in a manner described as providingly and the entire event is further E#modified as happening currently. !D%dNote that there is no confusion in this representation about what modifies what, owing to the tree istructure. However, there is some confusion about what the modifiers mean. It should be understood that jthe atomic words appearing in the LCS stand for concepts rather then lexical items, however how this done pis not clear. For instance, that the countries are thirteen could mean that they are thirteen in number or kthat they are thirteen years old, or that they are number thirteen in some context-specific ordering. Obvimously, the handling of nominals in this system is still an issue. In the future a system of qualia based on E6[Pustejovsky 95] may be incorporated into the LCS IL. Ez+uQe F!Sc)%eNext we consider a pair of translation equivalent texts. Note that the English text is broken into E2two sentences, and thus represented by two CLCSs. Gz+e H"e Example 2 !I%tBanco Solidario, S.A., or BancoSol, grew out of a non-profit joint venture created in 1986 by prominent members of pthe Bolivian business community and ACCION International. The latter brought with them leadership and seed capE=ital, while the former provided technology and methodology. J 1AeLCS 2a KGe (cause nil Le (go ident Me (Banco Solidario Ne (be ident (*head*) (S.A.)) Oe# (be ident (*head*) (BancoSol))) Pe" (toward ident (Banco Solidario) Qe (at ident (Banco Solidario) Re (grown)))) Se (from ident (*head*) Te (venture (joint) (non-profit) Ue ((cause Ve (conj We! (member (of poss (*head*) Xe (community (business) Ye (Bolivian)))) Ze (ACCION International)) [e (go exist (*head*) \e (toward exist (*head*) ]e& (at exist (*head*) (EXIST))))) ^e (creatingly) _e# (in temp (*head*) (1986))))) A`eLCS 2b HHˆ:PHHˆOUQQ ld:UU HHˆ:SHHˆ€OEEUa 1A1e(conj bQe (cause (ACCION International) ce (go loc (conj (leadership) de (capital (seed)))) ee (bringingly)) fe* (cause (community (business) (Bolivian)) ge* (go poss (conj technology methodology)) he (providingly))) ie (co loc (*arg1*) (*head*))) je (managingly))) ke (at temp (*head*) le (of poss (BancoSol) me! (beginning (in temp (*head*) ne (1986)))))) oz+zܦe pܥe Example 3 !q%iEn sus comienzos, en 1986, BancoSol era una asociacin sin fines de lucro adminstrada conjuntamente por vACCIN Internacional que se encargaba de la gestin y proporcionaba el capital incial y por representantes de los Egcrculos bolivianos de los negocios, que suministraban apoyo logstico y su conocimiento del terreno. r 1AXe(be ident (BancoSol) sZe (at ident (BancoSol) te (organization (non-profit) ue (go loc ve (conj we (ACCION International xe (go loc (*head*) ye (toward loc (*head*) ze* (co loc (*head*) (management?)))) {e (be-in-charge-ingly)) |e (cause (*head*) }e# (go ident (nil? BancoSol?) ~e( (toward ident (nil? BancoSol?) e% (at ident (nil? BancoSol?) e (provided)))) e (with instr (*head*) e (capital (initial))))) e (representatives e (of poss (*head*) e* (community (business) (Bolivian))) e (cause (*head*) e$ (go ident (nil? BancoSol?) e) (toward ident (nil? BancoSol?)  e& (at ident (nil? BancoSol)  e (provided))))  e (with instr (*head*)  e (conj  e (help (logistical)) e (knowledge e) (of poss (terrain)))))))))) e (toward loc (*arg1*) e! (co loc (*arg1*) (*head*))) e (managingly)) e(at temp (beginning)) e(at temp 1986))) 2-N%YOne might expect that the LCS representations of translation equivalent texts should be 0cquite similar. The vast divergence between these IL representations should not be construed as a afailure of the representation, however. These are mappings into IL of sentences that are equivaalent in the mind of a human translator, who performed many semantic level transformations on the E*content before generating the final form. z+ e SceFuture Work  3#+H%mThe process of creating CLCS representations for sentences is still a difficult and unclear process even for ehumans. We will continue to model sentences from source texts as CLCS and examine the properties of kthese representations, both to discover ways in which the representation may be strengthened and to better funderstand the process, so that we can ultimately design systems which can automatically create CLCSs jfrom text. In addition, further study of divergent CLCSs representing semantically close sentences, both gbetween two languages and within one language, will aid in the process of discovering constructs which ELtruly represent semantic content divorced from the details of surface form. _z+~$2e HHˆ: SHHˆRXTT ld:!XX HHˆ:"VHHˆޒ\X z+ze %u“e References z+fe feNDorr, Bonnie, Machine Translation: a View from the Lexicon, MIT Press, 1993. e ebGruber, Jeffrey S. Studies in Lexical Relations. Doctoral Dissertation, MIT, Cambridge MA, 1965  e !eLJackendoff, Ray S. Semantics and Cognition. MIT Press, Cambridge, MA, 1983 "e !#%nLuperFoy, Susann and Miller, Keith, The Use of Pegs Computational Discourse Framework as an Interlingua RepreEksentation, AMTA SIG-IL First Workshop on Interlinguas (held at MT Summit VI), San Diego, California, 1997. $e %ehMiller, Keith and David Zajic, Lexicons as Gold: Mining, Embellishment, and Reuse, AMTA-98 (to appear). &e !'%iOtero, Mara, Latin America: ACCION Speaks Louder Than Words,  UNESCO Courier , January, 1977. E (eZPustejovsky, James, The Generative Lexicon, Massachusetts Institute of Technology, 1995 )e *ePReeder, Florence and Dan Loehr, Finding the Right Words, AMTA '98 (to appear). W# ڒ\e HHˆ:$VHHˆUWW l}H<1ZH"EG7eH* }6<Y[6"EG8eN }-<Z\-"EG9eN }?<[,?"EG:e }<^"FG;e P:heading 4 }H<]_H"FG<eH* }6<^`6"FG=eN }-<_a-"FG>eN }?<`1?"FG?e }<kc GG@e P:heading 3 }H<bdH GGAeH* }6<ce6 GGBeN }-<df- GGCeN }?<e? GGDe }<ph HGEe P:heading 1 }H<giH HGFeH* }6<hj6 HGGeN }-<ik- HGHeN }?<jb? HGIe }<um IGJeP:footnote text }H<lnH IGKeP }6<mo6 IGLeN }-=np- IGMeN }?=og? IGNe }=zr JGOeP:Body Text 3 }H=qsH JGPeP }6= rt6 JGQeN }-= su- JGReN }?= tl? JGSe }=w KGTeP:Body Text 2 }H=vxH KGUeP }6=wy6 KGVeN }-=xz- KGWeN }?=yq? KGXe }=+| LGYe P:Body Text }H={}H LGZeP }6=|~6 LG[eN }-=}- LG\eN }?=!~v? LG]e }l> l MG$e3 }u> u MG,e heading 5 }H> #H MG4e }l> l NG<e7 }u> u NGe heading 4 }H> H NGe }l>  l OGe6 }u> u OGe heading 3 }H> H OG+e }l> ! l PG,e5 }u> u  PG.e heading 1 }H> H PG/e dLeftdRightd Referenced dHTMLdHTMLd Headingsd HTMLd$HTMLd Ad Dd Gd Jd MdPdSdV @@ Mapping Table Title. @@ Body. @@ Mapping Table Cell. @@ Header. @@ Footer. f@T   HeadingRunInBody. @@  Mapping Table Cell. @@  Mapping Table Cell. @@   Mapping Table Cell. f@  Body@@ Mapping Table Cell. f@  Blockquote. f@   heading 1. f@  Plain Text. f@  Body Text. f@   heading 4. f@  CellFooting. f@ Bodyf@  Bulleted\tf@ CellBodyf@ CellBodyf@  CellHeadingf@  CellHeadingf@  Footnotef@T Heading1Body f@T  Heading2Bodyf@   heading 3. f@  Indentedf@  Numbered.\tf@E  Numbered1.\tNumberedf@  TableFootnotef@T   TableTitleT:Table : f@T   TableTitleT:Table :  f@P TitleBodyf@  $Hl Dh footnote text. f@   $Hl Dh Body Text 2. f@  $Hl Dh Body Text 3. f@   $Hl Dh heading 5. @@  Mapping Table Cell. @@  Mapping Table Cell. @@ Mapping Table Cell. f@ Body. f@   $Hl DhTitle. f@  $Hl DhBody. $Hf@  $l DhBody. f@  $l DhBody. $Hf@  $Hl DhBody. f@   Blockquote. f@   heading 1. f@   Blockquote. f@  Title. f@ Body. f@  Plain Text. f@ Body. f@  Body Text. f@   heading 4. f@   heading 3. f@   heading 3. f@  Hl Dh Body Text . f@  $Hl Dh Body Text. f@  $Hl Dh Body Text. f@  Hl Dh Body Text. . f@  $Hl Dh footnote text. f@  Body Text. f@  $Hl Dh Body Text. f@  $Hl Dh Body Text.  f@  $Hl Dh Body Text. f@   $Hl Dh Body Text 2. f@  $Hl DhBody () . $f@  $Hl DhBody. f@  $Hl Dh Plain Text. f@ # $Hl Dh Body Text 3. f@   $Hl Dh heading 5.  Emphasis EquationVariables )  ڝ [  ڝ ڝ) )  )  ڝ  ڝ ڝ )  )  )  )  ڝ ) ڝ i_mF )  )  i_mFi_mF ڝ ڝ ڝ  )  ڝ ڝ tu  ڝ tu  w uo&   tu ThinMediumDoubleThick@ Very Thin "HHHHHFormat A "HHHHHFormat BH Mapping Tablez>ŽDH6-? Hz  )hhhB_ N*/?_ 02evevCQ p3;H?) <luHP$6z $Fz$Vz !"#$tz$%&'($z,)*+,-$z./012$z34567$z 89:;<$z = > ? @ A $z B C D E F $z G H I J K $z, L M N O P $<z Q R S T U $Lz VWXYZ$\z,[\]^_$bz, `abcd$z efghi$z jklmn$z opqrs$z tuvwx$z yz{|}$z ~$z $z     $z  $zC $R "$b"$r"$" !")M #$%$$z !"& ' ( ) $4z ""*!+!,!-!$Dz!#"."/"0"1"$pz"$"2#3#4#5#$z#%"6$7$8$9$$z$&":%;%<%=%$z%'">&?&@&A&$z&("B'C'D'E'$z')"F(G(H(I($$z(%J)K)L)M)$:_ +%N*O*P*$J_*,%Q+R+S+$Z_+-%T,U,V,$j_,.%W-X-Y-$z_-/%Z.[.\.$_.%]/^/_/$Z_ 1%`0a0b0c0$j_02%d1e1f1g1$z_1%h2i2j2k2$ 4%l3m3n3$35%o4p4q4$46%r5s5t5$57%u6v6w6$68%x7y7z7$79%{8|8}8$8:%~999$9;%:::$:%;;;$) =  <<<$C)<> ===$S)=? >>>$i)>@ ???${)?A @@@$)@B AAA$)AP B B!B$<zL 'C(C)C*C+CzE",D-D.D/D0DzFD"1EYEZE[E\EzGE"]F^F_F`FaFzHF bGcGdGeGfGzIG gHhHiHjHkHzJH lImInIoIpIzKI qJrJsJtJuJzLJ vKwKxKyKzKzCK {L|L}L~LL)N MMM)OM NNN)PN  O O O)BO  P PPCommentd BlackT!WhiteddARedddGreendd BluedCyandMagentad YellowHeader/Footer $1Header/Footer $1Header/Footer $2Header/Footer $2IndexIndexCommentCommentSubjectSubjectAuthorAuthorGlossaryGlossaryEquationEquation Hypertext Hypertext  Cross-Ref Cross-Ref Conditional TextConditional TextPositionFMPrivatePositionFMPrivateRangeEndFMPrivateRangeEndFMPrivate HTML Macro HTML Macro Times-Roman FrameRoman Times-Bold FrameRoman FrameRoman FrameRoman FrameRomanHelvetica-Bold FrameRoman FrameRomanSymbol FrameRoman FrameRomanTimes Helvetica Symbol TimesNewRoman CourierNew RegularRegular BoldRegularItalic1Kko㓟Ά2g#k?V0sQe@53ga~GT:\a5U9?9 N НO3#1J{N0!.xǰiH Pw |Gѳ\M鷉N8- KL2T!;f7!"ƾ'q.ij$땇,b.?h$