Controlled translation as a new translation scenario
Contents
1. The Evolution of CASL Controlled Authoring at General Motors Proceedings of the 3 International Workshop on Controlled Languages Applications CLAW 2000 14 19 Hofmann C and T Mehnert 2000 Multilingual Information Management at Schneider Automation in Sprung R C ed Translating into Success Cutting edge strategies for going multilingual in a global age Amsterdam Philadelphia John Benjamins Kamprath C et al 1998 Controlled Language for Multilingual Document Production Experience with Caterpillar Technical English in Proceedings of the 2 International Workshop on Controlled Languages Application CLAW 98 1 12 Means L and K Godden 1996 The Controlled Automotive Service Language CASL Project in Proceedings of the 1 International Workshop on Controlled Languages Applications 106 114 Mitamura T and E Nyberg 1995 Controlled English for Knowledge Based Machine Translation Experience with the KANT System Proceedings of the International Conference on Theoretical and Methodological Issues in Machine Translation TM 23 Mitamura T 1999 Controlled Language for Multilingual Machine Translation in Proceedings of Machine Translation Summit VII Singapore September 13th 17th O Brien S 2002 Multilingual Communication in the Automobile Industry in Localisation Focus University of Limerick Localisation Research Centre Dec 2002 27 Also available at http www ctts dcu ie report ht2. Translating and the Computer 26 November 2004 London Aslib 2004 Controlled Translation as a New Translation Scenario Training the Future User Celia Rico Enrique Torrej n Departamento de Traducci n e Interpretaci n Universidad Europea de Madrid Villaviciosa de Od n 28670 Madrid Espa a celia rico ti fil uem es enrique torrejon uem es Abstract In a paper presented at the 6th EAMT Workshop Teaching Machine Translation held in Manchester in 2002 under the European Association for Machine Translation framework the authors introduced the concept of controlled translation as a new teaching scenario in an attempt to respond to the challenges that the translation industry is facing namely shorter turnaround time multiplicity of languages and a highly competitive environment combined with the broadening use of machine translation MT systems translation memory TM software controlled languages and post editing The present paper adapts the teaching scenario explained then and develops a_ controlled translation course in the automotive field for senior translation students focusing on the translation needs of the automotive industry as discussed during the 2002 TOPTEC Symposium and reported by O Brien 2002 As shown below three main areas are catered for in the course effective terminology management increasing use of hybrid automation models comprising TM and MT and rising demand of translators specialized in th
3. across any similar report on the European sector but in any case we claim the objective is still valid it is highly important to increase the level of awareness and knowledge of translation software including machine translation and translation memories so that 12 The right translation here is que est fijado 21 the process of documentation generation can seamlessly fit into the decentralized model of the industry Training objectives we have already mentioned the particularities and complexities of the language used in automotive texts As in all specialized fields the work of the professional translator depends on his her ability to discover the intricacies of language and his her capacity to offer suitable solutions To this the automotive sector adds the need for a formal training in pre and post editing skills as well as a thorough understanding and extensive practice in quality standards such as J2450 Functionality objectives translators trained under this programme should be ready to start work in the real world This is why a key aspect in the course is extensive practice with real automotive service and user s manuals Keeping in mind these three objectives as the strategic framework the formal training program on controlled translation for the automotive industry that we propose should mainly focus on three axes 1 Tools Use of translation tools such as MT systems like ENGSPAN Systran
4. automotive industry in order to objectively measure the translation quality of automotive service information Pan American Health Organization http www paho org english AGS MT Machine_Trans htm May 2004 e Experiment with the advanced features of ENGSPAN which allows users to customize the system by modifying the grammatical coding of lexical entries according to parsing information e Learn how to post edit the raw MT output generated by ENGSPAN following recommendations for minimal post editing principles and practice these principles extensively e Perform evaluations of the translation quality using SAE J2450 In the sections that follow we explain in detail each of the steps in the course 2 2 Using PAHO s ENGSPAN Machine Translation For the controlled translation course at UEM the Pan American Health Organization s PAHO ENGSPAN English to Spanish MT system was selected This system is an application which runs under Windows 95 98 NT 2000 either in standalone or networked versions The graphical user interfaces provide easy access to all the features of the system especially dictionaries and transfer rules and the system can be invoked from within MS Word ENGSPAN is very suitable to translate technical and scientific documentation as well as administrative information training manuals and official correspondence among others In addition it easily handles MS Word files Saved as rtf files XML HTML SGML and text
5. frontal de la solera de puerta bajo la abertura de la puerta delantera Coloque los anexos en el extremo frontal y en el extremo posterior del elevador hidr ulico Table 5 SAE J2450 for evaluation post edited text 3 Formal Training Program in Controlled Translation for the Automotive Industry Following from our experience in the course described in the previous sections our proposal for a formal training program in controlled translation for the automotive industry rests on Allen s technology integration triangle 1999 whereby it is essential to maintain a balance of the 3 sides of the triangle technology training and functionality in order to successfully integrate new language technologies into the workplace At UEM we try to achieve this balance by carefully defining the objectives of the Controlled Translation Course for the three sides of the triangle in answer to one key question what skills is the automotive industry calling for Technological objectives considering that technology is already used in many other translation fields it should be natural to introduce it with similar success in the automotive sector According to Woyde 2002 there is little awareness of popular translation tools in the automotive industry in the United States which means that translation suppliers find it difficult to help them successfully implement these technologies into their publishing workflows 40 So far we have not come
6. just as important as the engineering components O Brien 2002 impels the automotive industry to abandon the silo perspective of product development marketing communication technical writing translation and product support Hofmann and Mehnert 2000 60 The documentation project s life cycle must comply in this context with the demands of a wider production environment which considers multilingual information management in the form of information objects IO which according to Hofmann and Mehnert is a collection of information identified as a unit and defined by its communicative purpose the specific user it is addressed to the business entity it represents a product line or a corporate function the information it provides in a specific format and for a target audience and some publishing restrictions 61 Translation is then an essential part in the information cycle depending on other parts of the cycle and feeding key input to them The importance of then effective handling of multilingual documentation in this industry is demonstrated by an active interest for the various aspects involved in translation the adoption of translation quality metrics such as SAE J2450 a concern for exploring cost effective methods for the translation of large volumes of documentation either by using machine translation and translation memories sometimes a combination of the two or by using some kind of authoring tools that allow for the impl
7. or Promt combined with TM software like DVX Trados or SDLX in the technical translation process of automotive documentation There are multiple combinations that can be selected for the classroom but in any case the selected software should allow dictionary customization terminology management and pre and post editing facilities if possible This axis ensures that by the end of the course students are knowledgeable about translation technologies and have access to ready to use customized translation tools Process Extensive practice of the iterative process whereby the five main steps involved in the controlled translation process listed in the figure below are mastered with special emphasis on J2450 based evaluation of sample texts from service and user s manuals This axis ensures the skills needed in technology driven localization processes common in the automotive industry as well as the practical knowledge in the application of translation standards Topics Thorough coverage of a wide range of technical topics from the automotive service and user s manuals ranging from more basic topics such as gasoline engines diesel engines ignition systems steering systems brakes automatic and manual transmission differentials etc to more technology related ones such as Electronic Stability Program ESP CVT Continuous Variable Transmission EBD Electronic Brake force Distribution ASR Anti Slip Regulation traction control or hydro
8. remolque para evitar posibles da os al veh culo durante la operaci n de remolque Table 3 Post edited text After applying all the possible minimal post editing rules to the text the students are ready to evaluate the final translation using SAE J2450 quality metric as explained in the next section 2 7 Applying SAE J2450 quality metric A training course on controlled translation for the automotive industry must necessarily include practice in translation quality assessment using SAE J2450 quality metric Ideally as Sirena 2004 points out the implementation of quality assurance QA as an integral part of the translation process should lead not only to significant improvement on quality but also to dramatic improvements in cost and turnaround time but before any translator can reach General Motors reported standards on QA the learning curve is steep and requires consistent dedication and effort as the experience in our classroom described below will show The main goal in training our students in the application of SAE J2450 is that they get acquainted with this metric understand the error categories and are able to apply the evaluation criteria with as much efficiency as possible Students also learn that what really matters is not so much the style but the identification of problems that can affect the ability of users to understand information However for translation students this is a controversial aspect since even w
9. a e Masculine which converts a word and its determiner to masculine by substituting the final a with an o and the determiner a with el Masculine Plural which changes feminine plural words to masculine Feminine Plural which converts masculine plural to feminine e Delete El which erases the subsequent determiner el la los las to the right of the cursor it can also delete the determiner in the contracted forms del and al e Mente Adverb which changes an adjective to its corresponding adverb form with the ending mente Add diacritics macro which adds accents to the vowel closest to the cursor and a tilde to the n closest to the cursor resulting in the letter The translation toolbar also include macros for preparing the text which students will submit to the MT system In general these macros make sure that the text format is correct and suitable for the system for instance there is a Block translation button which allows highlighting a fragment of text which we do not want to submit to the system and there is a Clean text file button which takes care of deleting extra spaces and hard returns Students will use these macros before rewriting the text in controlled language Submission of the working text to ENGSPAN can be carried out by means of two other buttons available at the translation toolbar e the Translate into Spanish button which calls up a window where students can type text to quickly find out w
10. a better understanding of how an MT system works and how it can be improved This task may sound a little off track for the future translator but we think our students need to know the MT system internally so that they can be trained for translation positions in large corporations where the use of machine translation technologies is already in place and close collaboration with computational linguists is a common practice It is also a good expertise for future translators to be able to use and customize MT systems if they want to have a competitive advantage in the freelance market When customizing ENGSPAN s dictionaries our students need to be aware of two different types of source entries simple words and group words which are further classified into substitution units analysis units and transfer units e Simple words include uninflected single words full forms of irregular or ambiguous words and microglossary entries for single words Their maximum length is 30 characters 12 e Substitution units SU include multi word expressions maximum five words that can be regarded as a single word and therefore their individual words will not be analyzed independently These units could also form other bigger units e Analysis units AU include multi word expressions like substitution units but their individual words can still be analyzed independently by the MT system These units are used whenever one of the multi word expressions are am
11. according to the Spanish orthography PostG7 Fix any other linguistic error that has not been fixed by the previous guidelines These post editing guidelines are formulated in such a way as to avoid misinterpretation when they need to be applied to the raw MT output Our experience in the classroom tells us that the students try to over postedit any error they find by claiming the application of PostG7 This over postediting is the most difficult to avoid and it takes the students a long time and some extensive practice to refrain from it Some of the over postediting cases we discourage are the removal of repetitive words either by deleting them or replacing them with pronouns in order to avoid a repetitive style the changes in the output because of subjective interpretation of the source text the conflation of two source sentences into one output sentence and the split of one source sentence into two or more output sentences The insistence on keeping a one to one correspondence between source sentence and target sentence is not gratuitous Often times machine translation systems are used in the industry and translation companies in coordination with translation memories like DVX or Trados Keeping that correspondence facilitates the alignment process and guarantees better exact and fuzzy matches when using translation memories As we explained in section 2 2 ENGSPAN features very convenient post editing macros available in the translation tool
12. alters the meaning of the sentence possibly causing misinterpretation Error category 7 in SAE J2450 translation metric 2 Interesting enough Sirena 2003 also reports this excessive use of the miscellaneous category for marking stylistic issues when sending sample texts to translation suppliers 11 That is to say the right parse is the stative reading and the wrong parse os the active reading The translations of these two structures are clearly distinct in Spanish by means of the verbs ester and ser respectively 20 Source text The board on lift attachment LM4086 0200 set at front end of vehicle should be set on the front of the sill under the front door opening Position attachments at front and rear end of board on lift Controlled language version The board on lift attachment LM4086 0200 WHICH IS set at THE front end of THE vehicle should be set on the front of the sill under the front door opening Position THE attachments at THE front END and AT THE rear end of THE board on lift Raw output El acoplamiento para los elevadores hidr ulicos _M4086 0200 QUE SE FIJA al final frontal del vehiculo debe fijarse en el frente de la solera de puerta bajo la abertura de la puerta delantera Coloque los anexos al extremo frontal y al final trasero del elevador hidraulico Post edited version El acoplamiento para los elevadores hidr ulicos LM4086 0200 QUE SE FIJA en el extremo frontal del veh culo debe fijarse en la parte
13. at meaning either since it lacks the verb should that is the source text is somewhat vague and does not highlight the importance of the recommendation PostG7 therefore detects vagueness in the source text that can be remedied with a new controlled language version Likewise in the same fragment below the application of PostG1 which corrects the translation of the preposition at en instead of a is in fact raising a flag about the vagueness of the source sentence where the expression at dealer actually means available at the dealership or available at your automobile dealership This vagueness is then removed when submitting a second controlled language version to the MT system Source text Towing is in accordance with Towing Procedure Manual at dealer Controlled language version THE TOWING OPERATION is in accordance with THE Towing Procedure Manual at the dealer Raw output La OPERACION de remolque es en conformidad con EL Manual para el Procedimiento de Remolque AL distribuidor Post edited version La operaci n de remolque DEBE ESTAR en conformidad con el Manual para el Procedimiento de Remolque EN EL distribuidor New controlled language version The towing operation SHOULD BE in accordance with the Towing Procedure Manual WHICH IS AVAILABLE at the dealer Table 4 SAE J2450 as a predictable instrument In the second step of training in SAE J2450 students work in groups to evaluate Spanish MT output that has been alr
14. bar which can be activated in Word Some of these macros help in the application of two of the post editing guidelines by facilitating a quick manipulation of the output text and avoiding involuntary deletion of words or time consuming copying and pasting manipulation For example by means of the Switch Right and Switch Left macros students can quickly apply PostG2 when fixing the wrong linear order of words In the example below students use that macro to change the position of the adjective adecuado so that it modifies equipo not remolque Likewise the students use the same macro to change the position of the adjective posible from post nominal to pre nominal position Another widely used macro is the Plural macro that changes the word where the cursor is placed to the plural form In the fragment below the noun phrase da o posible can quickly and effortlessly be changed into the plural form da os posibles by clicking on the aforementioned macro 17 Source text It is necessary to use proper towing equipment to avoid possible damage to the vehicle during towing operation Controlled language version It is necessary to use the proper towing equipment in order to avoid any possible damage to the vehicle during the towing operation Raw output Es necesario usar el equipo para remolque adecuado para evitar da o posible al veh culo durante la operaci n de remolque Post edited version Es necesario usar el equipo adecuado para
15. biguous for instance when codifying multi nouns containing ing forms Analysis units help therefore the MT system to disambiguate the source text Transfer units TU are transfer rules which select different translations from the ones codified in the main entries when certain context is present in the text The words that are codified in TU do not have to be contiguous like in AU and SU For instance the different translation of a verb depending on the direct object is codified by means of TU When reviewing the source document after applying the pre editing guidelines the students will have to review the automotive vocabulary in the document They can easily do so by highlighting the word or expressions and clicking on the macro that calls up the vocabulary browser or the macro that invokes the MT system and returns the translation of the highlighted segment If the term is not in the dictionary nor in the automotive microglossary or if the translation of the segment is incorrect the student then has to open the update lexicon browser and codify the new words expressions or new meanings according to the various types of entries explained above CAUTION All the applicable local laws regarding the towing operation must be obeyed It is necessary to use the proper towing equipment in order to avoid possible damage to the vehicle during the towing operation Use caution when you are towing the vehicle The towing operation is in accordance wi
16. ch if they are applied consistently can produce an objective value for comparison purposes The corresponding post editing guidelines can be considered as minimal post editing guidelines according to the classification by Allen 2003 in the sense that we try to remind students that the process of post editing needs to be as controlled and mechanical as possible in order to avoid unnecessary post editing usually because of stylistic and subjective judgments The minimal post editing guidelines are the following PostG1 Fix any wrong term in the text either a technical automotive term or non technical term Also correct any inconsistent use of the same term PostG2 Fix any syntactic error that consists in a wrong part of speech an incorrect phrase structure or a wrong linear order of words and phrases SAE J2450 is available at http www lisa org useful 2001 J2450Practice pdf October 2004 The next section is devoted to the use of this metric in the translation course and the difficulties encountered when trying to apply the rules consistently 16 PostG3 Add any missing text paragraph sentence phrase word which should have been translated PostG4 Fix any error that consists in wrong morphological form gender number person tense aspect or voice and wrong formal agreement number gender or person PostG5 Fix any misspelling error according to the Spanish orthography PostG6 Fix any punctuation error
17. e automotive domain 1 Introduction Controlled Translation has slowly emerged in the past few years as a new business scenario which features the use of controlled languages pre editing machine translation MT translation memory TM software and post editing processes in order to speed up and standardize the multilingual production of technical documentation This new scenario is increasingly being adopted by industries such as telecommunications and software localization and offers an excellent framework that meets the translation needs of the automotive industry Controlled languages CLs were first used in the aeronautic industry to tackle the increasing complexity of the aircraft s technical documentation Farrington 1996 and since then they have been introduced in other areas such as heavy equipment machinery Kamprath et al 1998 Whenever CLs are used the readability of the documents is improved by imposing clear and direct writing syntactic and lexical ambiguities are reduced by applying grammatical and lexical constraints and the translatability of the text is increased making it amenable to MT Mitamura amp Nyberg 1995 The resulting effect is consistency in the style of the documents the reusability of texts and the corresponding savings in authoring and translation processes along with higher customer satisfaction because of better documentation and translation The automotive industry also followed the path initiat
18. eady postedited by students in other group This activity leads to fruitful discussion on quality rating and selection of error categories simulating to a certain extent the difficulties professional translators and editors encounter when applying the metric in real world projects When inviting the students to discuss the selection of error 3 After application of the post editing guidelines PostG1 and PostG7 19 categories with their peers we place emphasis on sticking to the description of those categories for evaluating a translation error The experience in the classroom regarding this second step tells us that our students usually select the miscellaneous category when in doubt or because of stylistic considerations However when properly instructed open discussion with their peers facilitates the students learning process and the extensive practice on various texts covering different automotive topics also reinforces the application of the two secondary subcategories from the metric minor versus serious which assign different weight values to the error categories Even though the students necessarily base the application of these subcategories on a judgement call our experience in the controlled translation course shows that it is possible to reach a consensus with peer students for the right assignment of weight values When reviewing the post edited version of the fragment below students in several groups realized that the post ed
19. ed by the aeronautic and heavy equipment industry General Motors for instance started in 1993 the CASL Project Controlled Automotive Service Language for technical documentation Godden 2000 and the Swedish manufacturer Scania also explored the feasibility of defining a controlled Swedish for truck maintenance documentation Sagvall amp Almqvist 1996 Pre editing technical texts is also part of the working environment of Controlled Translation Translators need to have the skills to adapt text to the CL guidelines so that the translatability of the source text is improved Pre editing guidelines can be often found in the literature on CLs when grammatical and lexical constraints are described In section 2 3 below we provide the complete set of pre editing rules applied by our students in the controlled translation course Another key element in Controlled Translation is that introduced by the use of TM and MT systems in combination with CLs Successful experiences in this direction are Caterpillar s translation environment Rintanen amp Zetzsche 2002 and General Motors CASL Godden 2000 By combining the capabilities of TM and MT productivity is increased and consistency is maintained in all documentation produced The last element in Controlled Translation is Post editing PE an activity mainly associated to MT Clearly since MT provides almost 100 accuracy only under certain restricted circumstances limited vocabulary and
20. ementation of a controlled language as a previous step to translation and the application of the latest localisation tools and standards If this is matched with the use of terminology management tools the result is consistency of use and accuracy across the documentation material Rico 2003 In an attempt to meet these needs and adapting the teaching scenario explained in Torrej n and Rico 2002 at UEM Universidad Europea de Madrid we developed a controlled translation course for senior translation students which specifically places emphasis on the use of PAHO s ENGSPAN MT system as the core technology for Controlled Translation and suitable for the automotive industry as a test bed In this training program the students e Understand the multilingual requirements and needs of the automotive industry by reviewing the different document types the complexity of technical terminology and the targeted audience e Study existing Controlled Language rules from different industries such as Caterpillar Technical English and AECMA Simplified English in order to adapt them and come up with suitable rules for the technical documentation used in the automotive industry e Compile specialized automotive bilingual glossaries to feed ENGSPAN s English to Spanish dictionaries placing special emphasis on the disambiguation of multiple noun compounds describing procedures and parts SAE J2450 Translation Quality Metric is the standard adopted by the
21. ent lexical entries and TU which provide better translations of words in context the next step is to submit the controlled language version of the automotive text to the MT system and to work on the raw output in two ways by applying 14 the post editing guidelines which are discussed in the next section and by applying the SAE J2450 quality metric as explained in section 2 7 EA PAHO MTS Da tionary Update Fie Edt Testing Add wiew Hep o z Seagch mods Fides RETA F Seach Ouecnow Engish lo Spareh 7 E EE fal News Ey i Mody Ses wor Delle Lock ir Enaish Suite a i a 1 1 ob caution ee Tobita iin D ha beh Fue ise Teil Bate ot es fo ann Patt ox laa 3 ti 1 3 fl cau ror euni Find or ie A Vocabulary 420 I Merge Autornalndly Rude in on ondere ze T PAHO Core Enty Hot par ol a set i l When USE har baer pared as VERB and ds DEJECT it aiuti fRax spered E Rean a Ss oy fice dob onion wae 1Cobi cipaialle ares Zi wre 111 ob drug a e m z umer Nore L tre 11 abi tobaces i e use TT obi capes sre eee ton fase ure 11 0b caution i Faison Hone Si use 12 codal E r A uted 15 codal R Voica Mone Z utad f 6 tupel Fees hicro Eny Automotive r used eed fat ttntntntntnnennnnnnnnnnnnnnnunten sis ST Es uted 14 type ginn jii gb oy r RSF EL beatin Machne Tarek 1 Span MTS Dicbonaey s SOME EA ates us Figure 2 Dictionar
22. ents in the controlled translation course heavily use these macros when reviewing the translation of the automotive text submitted to the system The general purpose post editing macros include among others a Search and Replace macro to do specific or global replaces of strings a Browse Dictionaries macro which provides access to the system dictionaries very useful Switch Right and Switch Left macros which move a selected word either to the right or to the left and can be repeatedly used until the selected word s are placed in the right order Delete this word macro which deletes the word where the cursor is placed and Lower case and Upper case macros which convert the word where the cursor is placed to lower case and upper case respectively Except for a set of core entries which cannot be modified in any way The Spanish post editing macros provide access mainly to quick repairs of morphological features as well as deletion of unnecessary determiners Combined with the general purpose ones students can effortlessly review the raw output and apply some of the post editing guidelines explained in section 2 6 The Spanish post editing macros are the following e Plural which changes to plural the word where the cursor is located e Singular which changes to singular the plural word where the cursor is placed Feminine which converts a word and its determiner to feminine by replacing the ending o to a and the determiner el to l
23. files ENGSPAN can be easily customized for the controlled translation course because of two main features e Access to dictionaries by means of their associated graphical user interfaces and availability of several grammars which can be selected according to text type The translation toolbar for MS Word which includes very handy post editing macros and macros for preparing the input text for gaining access to the MT system and to the generated output files Dictionaries and Grammar Selection ENGSPAN features an English source dictionary and Spanish target dictionary containing approximately 90 000 entries It also contain 13 microglossaries or specialized dictionaries for specific subject matters such as medicine finance environment equipment agriculture information technology law radiation and pharmaceutical industries The translator can define another 6 microglossaries for specific subject matters In the controlled translation course students create a microglossary for the automotive industry When running the ENGSPAN system the translator can specify up to 5 microglossaries in order of the priority to be used by the system In addition the translator can add a specific client s terminology if need be ENGSPAN s dictionaries are accompanied by two graphical user interfaces that allow browsing and updating the source and target entries of all dictionaries and microglossaries In order to display all the translations of a so
24. gen based and hybrid engines 13 It could also be easily extended to other industries especially those where use of very rich technical terminology is critical for the localization process of technical documentation 22 among others This axis ensures that by the end of the course students have a sizeable controlled translation portfolio of automotive texts that they can add to their curriculum as proof of experience and savvy in automotive technical literature 4 Conclusion The controlled translation course described here has been running for two years at the UEM Internal evaluation of students satisfaction has revealed that still further practice is needed in SAE J2450 quality metrics as well as in the various strategies for customising ENGSPAN s dictionaries The course is improved for each new edition with input from students and close attention to the industry s needs Acknowledgements We would like to thank Julia Aymerich Senior Computational Linguist at PAHO for her comments and help in the different aspects of ENGSPAN functionalities covered here 5 References Allen J 2003 Post editing in H Somers ed Computers and Translation a Translator s Guide Benjamins Amsterdam 297 317 Farrington G 1996 AECMA Simplified English an Overview of the International Aircraft Maintenance Language Proceedings of the 1 International Workshop on Controlled Languages Applications CLAW 96 1 21 Godden K 2000
25. grammar use of a sublanguage all other output must be post edited one way or another As Allen 2003 states PE ranges from browsing gisting to full PE depending on translation motivation i e whether it is simply the process of translating to understand inbound translation or the process of translating to communicate outbound translation These two factors determine different degrees of PE MT with no post editing for content browsing rapid PE for perishable information and urgent texts allowing only the correction of the most blatant and significant errors partial PE where the post editor decides which amount of changes to make in view of the client reader audience and full PE where MT combined with the use of CL produces in specific industrial projects a faster output than translating the entire document without any computer aided translation assistance 2 A Controlled Translation Course at Universidad Europea de Madrid 2 1 Rationale Translation Needs in the Automotive Industry Two of the biggest challenges the automotive industry is facing today are decentralization and lack of standardization of the authoring process The first is related to organizational aspects and the second to the use of languages In any case both are interrelated and have an effect in the translation process and its output In order to reduce costs parts and components are shared between platforms for manufacturing purposes which means that d
26. hat the output will look like or if the student has previously selected a fragment of the input text the button provides the output generated by ENGSPAN and allows replacing the original with that output e the Translation button which will invoke ENGSPAN and translate the whole document open in MS Word The procedure will ask the student to verify the target language perform a spell checking and save the document as an rtf file For the purposes of the controlled translation course the first button becomes very convenient since it allows students to expeditiously verify any rewrite the text at hand as well as to detect any new word not included in ENGSPAN s dictionaries or any grammatical structure which may generate erroneous translations Thus it provides students quick verifications of the impact of the rewriting guidelines they are using and it also highlights further work which needs to be done on the system dictionaries Finally the translation toolbar features a set of buttons which provide access to the output files generated by ENGSPAN once the input text has been submitted namely e the Open Not Found Words List button which opens up a file containing all the words from the input text which are not in ENGSPAN s dictionaries This file is very convenient for the students when they are ready to provide source and target entries for these words via the interface for updating the dictionaries e the Open last SBS button which ope
27. he leaded gasoline with the right octane rating PreG10b Repeat the adjective when modifying a coordination of nouns for example the sentence Use recommended tools and lubricants should be rewritten as Use the recommended tools and the recommended lubricants PreG11 Repeat prepositions in the coordination of prepositional phrases For example the sentence Keep flammable material off the exhaust pipe and three way catalyst should be rewritten as Keep the flammable material off the exhaust pipe and off the three way catalyst Once provided with an input text students should be able to identify grammar issues that require pre editing and apply the guidelines consistently As part of the controlled translation course students are assigned different samples of automotive manuals covering a wide range of technical topics which they have to rewrite in controlled language In the assignments they have to check each paragraph and add comments on what pre editing guidelines they apply These samples will be used in the subsequent parts of the course namely customization of ENGSPAN dictionaries and post editing of MT output so that by the end of the course students have a sizeable portfolio of controlled translation samples as well as customized ENGSPAN versions for the automotive industry An example of controlled language assignment is shown in the table below the left column contains a sample of original text from a repair manual and the right c
28. hen told to stick to SAE J2450 error categories they find some difficulty in reaching consensus about what the right error category should be or whether or not an error is stylistic in nature We divide training in SAE J2450 into two sequential steps at two different stages of the controlled translation process because as Sirena points out QA is not an add on step at the end of the translation process but a key to improve the entire translation process when 18 implemented as an integral part of this process The first step consists in evaluating the raw output from the MT system on an individual basis so that all the students understand what goes inside the system and what is more important to isolate typical problems which could be resolved by better application of controlled language rules or more accurate codification of ENGSPAN s dictionaries In a sense SAE J2450 is used here as a predictable instrument in order to correct MT system inefficiencies before they create other problems later on in the localization process It can also be used to detect other types of problems in the source text like misspellings ambiguities and vagueness For example in the fragment below Table 4 the raw output does not convey the right meaning of conformance with a given procedure by means of the verb debe which is added in the post edited version because of PostG7 However in closer inspection it turns out that the source text does not convey th
29. ion since it is not listed in the dictionary It therefore needs to be codified as a single word As for the expression Use caution the translation rendered by the system Usar cautela is not used in the automotive manuals The correct translation in this case Prestar atencion is codified by means of a TU see Figure 2 for the TU that codifies Use caution Gi Fad MrS Dutionary Opine ee en E fee it ds awe So 2 amp E 5 Seaieh mode iPhiye Stal vill seach Directign j Enghsh to Spari cjg ardi How Eray t mot tear EES Deen Lokty JEngish Sauce xy English Source rtean aye 2 sabem ino as paw DEJ JS F stems _ Patol Speach New T Pniel n Seuce IPAnD hal tas vents FET nema word Pat of Speech Transigvon 7 zeng Moun persion xj f aystem Houn hd j asen a ae Gemtanbe Featoeg AF Whe OS Me egoi Ll sleemag conhat _ Ure semantic haatuer ot heahurod a a a Eee sheenng tole A sletling whe z streng piluta 7 i When reversing tour a4 Speci ion bubela Corot meetan wji Naber fone Mibe Unspecred Tense Present paleme ST G e 1 ed y gt Maro Engey FAulomolve i g paap o f cade 7 al Finaj E 39 od EL 7 ARPA mastene transis en Benccea2ets si Figure 1 Dictionary entry of steering system as an AU Once the students have finished the codification of all the missing vocabulary or multiword expressions requiring independ
30. ited version overlooked several translation errors from the raw output which were neither corrected nor assigned weight values One of those errors has to do with the translation of the relative clause which is set at the front end in the first sentence The students pointed out how the MT system misparsed the verb structure is set as the present tense in passive voice of the verb set when it should be parsed as the present tense in active voice followed by the adjective set A lively discussion ensued regarding whether this was a syntactic error J2450 s error category 2 because of a wrong part of speech or whether it was an error category 4 because of incorrect morphological form present tense fija versus past participle fijado Other students even claimed that it should be regarded as error category 7 or miscellaneous because it did not fit any of the other error categories And yet another discussion was devoted to ascertain whether this case should be considered a serious versus minor error As can be seen applying the J2450 s error categories is not always as straightforward as it seems Extensive practice in the classroom ensures that the students follow the same criteria for error classification as much as possible In the end the verdict in this case rested upon the following decision the right error category is wrong morphological form active voice versus stative voice and it should be considered serious because it significantly
31. ml January 2003 Rico Perez C 2003 Overcoming the Language Barrier Paths that Converge in Technology and Translation in P rez Gonzalez L ed Speaking in Tongues Languages Across Contexts and Users Universidad de Valencia 185 200 Rintanen K and J Zetzsche 2002 Integrating Translation Tools in Document Creation available at _http Awww internationalwriters com dejavu Integrating_tools html October 2004 Sagvall A and I Almqvist 1996 Defining Scania Swedish a Controlled Language for Truck Maintenance Proceedings of the 1 International Workshop on Controlled Languages Applications CLAW 96 Sirena D 2004 Mission Impossible Improve Quality Price and Speed At the Same Time Using SAE J2450 to Do the Impossible in The LISA Newsletter Volume XIII 2 2 Torrej n Diaz E and C Rico P rez 2002 Controlled Translation A New Teaching Scenario Tailor made for the Translation Industry in 6th EAMT Workshop Teaching Machine Translation November 14th 15th 2002 European Association for Machine Translation 107 116 Woyde R 2002 Translation Needs in Auto Manufacturing Multilingual Computing 13 2 39 42 24
32. ns up a file containing all the sentences of the submitted input text with corresponding output for each sentence side by side This file is especially handy for use in the classroom because students can work in groups and discuss errors in the output as well as possible remedies by means of pre editing or post editing guidelines e the Open last raw button which opens up the file containing the machine translated output of the submitted text preserving the original format in rtf When accessing the file students are requested to save it as a Word document This file provides the text which students need to post edit according to the guidelines explained in section 2 6 2 3 Adopting Controlled Language Rules The first step in the controlled translation course of automotive texts is the rewriting in controlled language of the legacy texts in order to facilitate the readability as well as the translatability of the technical documentation Students at UEM get familiarized with controlled English by extensively practicing the rewriting of real automotive service and repair manuals It is essential that the students gradually assimilate the pre editing guidelines so that in the end they will be able to detect rewriting issues automatically Needless to say we require from them a very good knowledge of the English grammar in SBS stands for Side By Side order to grasp parsing issues such as scope of coordination reduced relative clauses or use
33. ocumentation processes and technology have to be shared as well making translation of this documentation a complex process Consider for instance the documentation issues that might arise when Saab s transmission is based on a GM North American platform As a result of the decentralization of companies communication regarding the production of documentation and its translation is sometimes impeded Woyde 2002 If we add to this scenario a series of language related aspects the result is certainly not much promising These refer to the following a the automotive industry produces large volumes of highly detailed documentation in a wide variety of formats b their information must be available in a wide array of languages c documentation must address a diverse audience d most of the times it has been written for the English speaking community with no translation needs in mind and e the quality of the language used is poor usually tending to be verbose complex and ambiguous Means amp Godden 1996 In order to face these multiple challenges the automotive industry ideally calls for a controlled documentation process which nevertheless provides flexibility and cost effectiveness and acts responsively to the needs of the industry This should be an environment where information is faster to produce easier to use and cheaper to translate In fact the complexity of producing documentation where the engineering of information is
34. of gerunds among others In order to do the rewriting in Controlled English the students are introduced to a defined set of pre editing guidelines which they must apply as consistently as possible so that statistics of the guidelines and their frequency per sentence can be kept for comparison purposes Some pre editing guidelines closely follow the lexical and grammatical constraints from controlled languages in the automotive industry such as CASL Controlled Authoring from General Motors or from the controlled languages used in the heavy equipment and aeronautic industries like KANT Controlled English and Boeing Technical English Other guidelines have been formulated after carefully studying the style of the automotive texts collected for the controlled translation course The pre editing guidelines PreG are the following PreG1 Keep sentences as short as possible not more than 20 words PreG2 Avoid multiple coordination of sentences which could result in ambiguous reading PreG3 Insert determiners a the some whenever possible PreG4 Insert that which in order to in subordinate clauses whenever possible PreG5 Try to avoid anaphoric pronouns such as it them they ones PreG6 Try to avoid elliptical constructions such as dropped direct objects and dropped subjects in coordinations PreG7a Rewrite when while before after followed by present participle ing form by either inserting you are before the ing form or b
35. olumn contains the controlled language version of the same text which students should be able to produce The rewritten additions are in upper case Original text Controlled language version 1 These operations should be done on a level 1 These operations should be done on a level surface surface 2 When removing a heavy component such as the 2 When YOU ARE removing a heavy component Engine or transaxle transmission be careful not to such as the engine or THE transaxle transmission 10 lose your balance and drop them be careful not to lose your balance and drop THE ENGINE OR THE TRANSAXLE TRANSMISSION 3 Also do not allow them to strike adjacent parts 3 Also do not allow THE ENGINE OR THE especially the brake tubes and master cylinder RANSAXLE TRANSMISSION to strike THE adjacent parts especially the brake tubes and THE master cylinder 4 Before starting repairs which do not require 4 Before starting THE repairs which do not require battery power THE battery power 5 Turn off ignition switch 5 Turn off THE ignition switch 6 Disconnect the negative battery terminal 6 Disconnect the negative battery terminal 7 If the battery terminals are disconnected recorded 7 If the battery terminals are disconnected THE memory of radio and each control unit is erased recorded memory of THE radio and OF each control unit is erased 8 To prevent serious burns 8 INORDER TO prevent serious burns 9 Avoid contact with hot me
36. r site However the task of merging the two was too time consuming for the purposes of the course and we decided to use online monolingual dictionaries in both languages as well as printed dictionaries as sources of reference for creating new entries in ENGSPAN dictionaries 2 5 Customizing ENGSPAN dictionaries The next step in the Controlled Translation Course is learning how to use ENGSPAN dictionaries in order to add automotive vocabulary One of the main advantages of ENGSPAN is the fact that its dictionaries both main one and microglossaries can be customized for a specific subject area Customization in our course mainly involves either coding new lexical entries in a specific microglossary Automotive if the source entry has different syntactic or semantic features from the automotive term or adding a new translation related to the automotive field in the existing lexical entry of the main dictionary Customization can also involve modifying the syntactic and lexical transfer rules that are triggered in the dictionary entries themselves This other customization is rarer in our course The results of this customization work are better generation of Spanish syntactic structures and better selection of target translation depending on the surrounding words and context Students are therefore exposed to the internal mechanisms of MT system and get acquainted with the manipulation of the various features in the lexical entries This way they gain
37. tal parts 9 Avoid THE contact with THE hot metal parts 10 Do not remove the radiator cap when the engine 10 Do not remove the radiator cap when the engine is hot is hot 11 Dispose of drained oil or the solvent used for 11 Dispose of THE drained oil or the solvent WHICH cleaning parts in an appropriate manner IS used for cleaning THE parts in an appropriate manner Table1 Sample text and controlled language version from a service manual Notice how the controlled language version is larger than the original text Some of the pre editing guidelines which have been applied to the text are the following PreG3 in sentences 2 3 4 5 7 9 11 PreG4 in sentence 8 PreG5 in sentences 2 and 3 PreG7a in sentence 2 PreG9 in sentence 11 and PreG11 in sentence 7 2 4 Compiling Automotive Glossaries for ENGSPAN Once students are familiar with the controlled languages rules and they have understood the importance of pre editing a text to make it amenable to MT the next step is compiling automotive glossaries they can later use for customizing ENGSPAN dictionary 11 Ideally this task should be automated as much as possible so that all compiled entries could be imported into the MT system Nevertheless we discovered that it was difficult to find online English Spanish automotive dictionaries Students were able to find for instance the English glossaries of Audi in their English website and the Spanish glossary in the company s mirro
38. th the Towing Procedure Manual at the dealer Always attach the safety chains before towing the vehicle When you are towing the vehicle make sure that the transmission the steering system and the power train are in good order If any unit is damaged the dollies must be used Never tow an automatic transmission model or an HCVT model from the rear or backwards with the four wheels on the ground This may cause serious damage and expensive damage to the transmission Nissan recommends that the towing dollies be used when you are towing your vehicle and the vehicle is placed on a flat bed truck as shown Table 2 Sample text with new automotive vocabulary 13 For example if students check the vocabulary in the sample text above Table 2 they will have to make sure that the underlined words and expressions are already in the main dictionary or the Automotive microglossary After running the MT system by invoking the corresponding macro in the toolbar they will realize that the translations rendered by system are compositional word by word For example multi nouns like steering system towing operation or power train are incorrectly translated These multi nouns need to be codified as AU because of the ambiguity of the ing forms or the ambiguity of the word power which can be a verb a noun and an adjective see Figure 1 for the dictionary entry of steering system as an AU Regarding the word dolly the system will not return a translat
39. that have been translated using Systran MT system and are consequently most useful for the controlled translation scenario we explain in our course The type of information on those texts is considered perishable that is for rapid consumption within the Commission departments and does not require a perfectly polished final translation The recommendations are 1 As much of the output translation as possible should be kept therefore neither delete nor rewrite too much output text Repetitions in the output are fine therefore do not delete repetitive words If words are nonsensical or plainly wrong correct them Time permitting correct also ambiguous words if any Once the students understand these recommendations and they are aware of what they mean they are introduced to a set of more specific post editing guidelines which can correct the type of errors detected by the J2450 Translation Quality Metric from the Society of Automotive Engineering SAE As we pointed out in Torrej n and Rico 2002 this metric is widely used in the automotive industry for quantifying the quality of the translations of automotive service information regardless of source language target language human translation or machine translation Therefore it perfectly suits the technical texts we use in the translation course It consists of seven error categories with two different numeric weights serious and minor and two metarules for ambiguous cases whi
40. urce term in the interface for updating entries a tree control is used and related source entries are shown in pop up windows Powerful search options enable the translator to quickly look for a specific term Depending on the part of speech of the source word different boxes specifying morphosyntactic and semantic codes that describe the linguistic characteristics of the source word are available for the translator to change On the other hand ENGSPAN also allows translators to select special grammars when submitting a text The available grammars cover texts such as document abstracts letters instruction manuals reports resolutions surveys speeches and news articles among others For the purposes of the controlled translation course students are requested to select the grammar for instruction manuals This grammar will be able to handle transitive verbs without direct objects and count nouns in the singular form without determiners However students are requested to rewrite these cases according to the pre editing guidelines explained in section 2 3 The grammar for instruction manual will also give higher preference to the imperative verbs when parsing technical texts Translation toolbar for MS Word one of the main reasons to select ENGSPAN as machine translation system for controlled translation is its post editing macros for general purposes and for Spanish available in the translation toolbar which can be invoked from MS Word The stud
41. y entry for Use caution as a TU 2 6 Post editing of ENGSPAN s Output Once the students have customized the ENGSPAN s dictionaries they are ready to submit the automotive text to the MT system and proceed to check the raw output Before doing so students are explained the principles of post editing according to Allen 2003 It is very important to explain to the future translators that the activity of post editing the raw output of a MT system is not synonymous of correcting every single mistake in the output and rewriting the output as if the translator himself herself were drafting the translation from scratch In fact translating and post editing need to be kept apart in the same way translating and editing a human translation are independent albeit related activities According to our experience in the classroom this difference needs to be repeatedly explained because students are at first reluctant to accept any mechanization of the translation process and complain about the lack of creativity The emphasis is therefore on highlighting the flexibility of the translator s roles which the adoption of translation technologies imposes on the translation practice 15 In order to give an example of the use of post editing in the real world of translation services we first explain the rapid post editing recommendations used at the European Commission Translation Service as reported by Wagner 1985 Those recommendations are applied to texts
42. y rewriting the whole present progressive tense as simple present for example when towing could be rewritten as when you are towing or when you tow PreG7b Rewrite when where if followed by a past participle by inserting the appropriate subject between the conjunction and the past participle for example Use approved sealants when required should be rewritten as Use the approved sealants when these sealants are required PreG 8 Try to avoid phrasal verbs or keep the adverb preposition next to the verb for instance rewrite the sentence Turn the engine on as Turn on the engine or better yet Start the engine PreGQ Try to avoid adjectives past participles and present participles ing form in post nominal position Rewrite them as relative clauses whenever possible For example Use only the lubricants specified in this manual should be rewritten as Use only the Adapted from Torrej n and Rico 2002 lubricants which are specified in this manual lf ing forms have an adverbial rather than adjectival value then rewrite them as full sentences For example Continued refuelling may cause fuel overflow resulting in fuel spray should be rewritten as The continued refuelling may cause the fuel overflow This may result in fuel spray PreG10a Repeat the noun when modified by coordinated adjectives for example the sentence Use unleaded or leaded gasoline with the right octane rating should be rewritten as Use the unleaded gasoline or t
Download Pdf Manuals
Related Search