99

Teaching Post-editing: A Proposal for Course ContentSharon O’Brien

SALISDublin City University

Glasnevin, Dublin 9Ireland

sharon.obrien@dcu.ie

AbstractThere is a growing demand for translation. Tomeet this demand, many translation companiesare introducing a hybrid technology solutioncombining translation memory and machinetranslation. However, few trainee translatorsreceive training in machine translation post-editing. This paper asks the question: Whyshould translator training programmes teachpost-editing skills? Is post-editing the same astranslation and traditional revision? The skill-sets required of a post-editor are listed and theusual list of skills is extended. An outline for acourse in post-editing, divided into theoreticaland practical components, is proposed. Finally,the question of when such a course should begiven to trainee translators is addressed.

1. Why teach post-editing?

1.1 Growing DemandThe global market for translation wasvalued at around $13 billion in 2000, and agrowth to around $22.7 billion by the endof 2005 has been predicted.1 Thisincreasing demand has led to an increase inthe use of translation aids, includingterminology management tools, translationmemory (TM) and machine translation(MT) technology. At a recent conferenceon multilingual communication, leadingtranslation companies reported that theyare now testing and implementing a hybridTM-MT technology solution to meet thegrowing demand for translation.2

1 Allied Business Intelligence (1998).2 At the Society for Automotive EngineersMultilingual Communication TOPTECSymposium, which took place from October 3-4 2002 in Nashville, three companies presented

Vasconcellos and Léon (1985:122)claim that a full-time, trained post-editor,working on-screen, can produce polished,standard quality output at a rate of betweentwo and three times faster then traditionaltranslation (i.e. 4,000 to 10,000 words perday). These figures suggest that MT andpost-editing are viable solutions to meetthe growing translation demand. Yet, howmany translation professionals havereceived formal training in post-editingtechniques? In the English translation ofKrings’s study of post-editing, the editor,Geoffrey S. Koby, suggests that “thetranslator must be trained in post-editing”(Krings and Koby, 2001:12). To the bestknowledge of the current author, there arefew translator training programmesoffering full courses on post-editing at thetime of writing.

1.2 Post-editing skills developedgradually

Vasconcellos (1986a:145) maintains thatpost-editing skills are developed gradually.The level of comfort with post-editing isgreatly increased after 100,000 words (1month of full-time post-editing). Somers(1997:201) also reports that it is recognisedby many that post-editing is a skill thatneeds to be “honed”. Companies wishingto implement machine translationtechnology would therefore benefit iftranslation graduates were already “com-

solutions combining machine translation andtranslation memory solutions (Bowne GlobalSolutions, SDL International and TelelinguaSoftware).

100

fortable” with post-editing. Additionally,post-editing skills would give translatorsan extra boost when it comes to findingemployment opportunities.

1.3 Teaching post-editing meanstranslators will embrace MT

Translators who do not have post-editingskills are frequently hostile to machinetranslation technology. Common argu-ments against MT include a dislike forcorrecting repetitive errors that a humantranslator would never make, a fear oflosing language proficiency by workingwith poor MT output and a dislike ofhaving one’s freedom of expressionlimited (Wagner, 1985:213). However,translators who embrace post-editing oftenreport that their day-to-day work becomesmuch more interesting.3 Drawing on herexperience of implementing Systran, Ryan(1988) maintained that the more and theearlier the translator was involved with theimplementation of machine translation, thefaster a usable system can be developed.Ryan correctly points out that “in an agewhen some systems can already translateover a million words in an hour, the timethat it takes to run a translation becomesinsignificant; the cost-effectiveness of theMT system must be measured largely bythe effectiveness of the post-editingprocess” (Ryan, ibid:131). Involvement ofa translator, therefore, improves thechances of success for machine translation.Senez (1998b:293) confirms this effectwhen she reports that a translator involvedwith an MT project eventually “no longerfeels threatened by the machine, but haslearned to reap as much benefit as possiblefrom what the computer gives him”.

1.4 ConclusionsPost-editing skills should be taughtbecause:

3 Personal opinion expressed by members ofthe Luxembourg-based EuropeanCommission’s Spanish translation departmentin September 2002.

• it would help meet the increasingdemand for translation and for fasterproduction times;

• post-editing skills are different fromtranslation skills and we cannotassume that a qualified translator willbe a successful post-editor (c.f. 2.0below);

• it would produce graduates who arealready “comfortable” with post-editing and who are more ready to beproductive in a machine translationenvironment upon graduation;

• and it could improve the uptake ofmachine translation technology byimproving translators’ perceptions ofMT and its capabilities.

2. Who is the Target Audience?Having established that there are advan-tages to teaching post-editing skills,consideration should be given to thequestion “who are the target recipients ofthis teaching?”. It is logical to assume thattrainee translators should be the primarytarget audience for post-editing training.However, this assumption encompasses anunderlying assumption that translation andtraditional revision are similar to post-editing and that translators are the bestcandidates for post-editing. It is interestingto consider whether, firstly, translating andpost-editing are in fact similar activitiesand, secondly, whether translator trainingtransfers the necessary skills to anindividual for post-editing?

2.1 The cognitive viewpointKrings and Koby’s (2001:360) uniquestudy on post-editing demonstrates thatcognitive processes relating to source-textcomprehension during translation and post-editing differ. Also, Krings and Kobyconclude that traditional translation is asignificantly less linear process than post-editing (ibid:498). Therefore, there isreasonable evidence to suggest that post-editing differs from translation from acognitive point of view.

101

2.2 The practical viewpointPost-editing and translation differ on thepractical level. Translation usuallyinvolves one source text and the creationof one target text to a level of publishablequality. Post-editing, on the other hand,involves two source texts, i.e. the textauthored in the source language and theraw MT output, which a translator uses tohelp produce a final version. The taskrequirements also differ. The usualrequirement from the translation process isto produce a target text that meets highquality criteria, whereas post-editingrequirements can range from gisting tohigh-quality publication quality.

2.3 Post-editing and traditionalediting or revision

McElhaney and Vasconcellos (1986) claimthat post-editing is different fromtraditional revision of translation. Forexample, unlike traditional revision, thepost-editor is more or less assured that nopassages have been skipped and there areno spelling errors. They also point out thatwhile misconstructions will be present inMT as well as Human Translation (HT),misconstructions in the former will likelybe more local than in the latter (ibid:141).Löffler-Laurian (1985:71) also drawsattention to the fact that the types of errorsthat occur in machine translation aredifferent from those that occur in humantranslation.

2.4 Translation and post-editingobjectives

Translator training focuses on accuracyand equivalence. Where specialisedtranslation is the main focus, the traineetranslator is taught to be as accurate aspossible where terminology and meaningis concerned and to aim for cultural andtextual equivalence. The trainee translatoris taught to produce texts suitable forpublication. It is for this reason thattranslator training, in the traditional sense,can act as a hindrance to post-editingwhere the aims are frequently different. On

the subject of differences betweentranslating and post-editing, Senez (1998a)says: “A translator will always strive todisguise the fact that the text has beentranslated. In the case of post-editing, it isenough for the text to conform to the basicrules of the target language, even if itclosely follows the source text” (pages notnumbered).

2.5 What are the similarities?Where translation and post-editing do notdiffer is in the requirement for ascertainingthe target audience’s needs. Translationtraining programmes train translators toexamine the expectations of the sourcelanguage audience and to compare these tothe expectations of the target languageaudience and to translate accordingly.Post-editors need to perform this task too.

2.6 ConclusionsWe have seen evidence that post-editing isnot the same as translation or traditionalrevision. In fact, some of the demands ofpost-editing are contrary to the skills andobjectives of translators and probablyrepresent one of the reasons why MTimplementation has failed in the past.Nevertheless, McElhaney and Vascon-cellos (ibid:142) believe that there arestrong arguments in favour of trainingtranslators as post-editors. They argue thata translator is best able to identifylinguistic errors, has a fund of knowledgeabout the cross-language transfer of con-cepts, and has the technical resources attheir disposal to work efficiently.

The conclusion, then, is that translatorsshould be trained as post-editors. However,this type of training should be optionalrather than compulsory. Qualification for amodule on post-editing should be madedependent on the strengths and personalityof each student, if possible. The skill-setsrequired are outlined in section 3 below.

102

3. What skill-set does a post-editor need?

According to Johnson and Whitelock(1987), post-editing is a highly skilled taskwhere the post-editor ought to be an expertin the subject area, the target language, thetext-type and contrastive knowledge: “Ineffect, the post-editor should be at least asskilled in all of these domains as theoriginal translator” (ibid:140).4

Wagner (1987:76) lists excellentknowledge of the source language, perfectcommand of the target language, special-ised subject knowledge, word-processingexperience and tolerance as the essentialskills of a post-editor.

Vasconcellos (1986a:136-138) elabo-rates on the need for word-processingskills, listing full key proficiency,efficiency in cursor positioning, effectiveuse of search and replace functions andability to use macros as essential for theskill set of a post-editor. Knowledge ofterminology coding for machine translationis mentioned later in the same article(ibid:142). In Vasconcellos (1986b), theentire paper is dedicated to the significanceof text linguistic knowledge for effectivepost-editing.

In addition to these tangible skills,several authors and MT practitioners list apositive predisposition towards MT as anessential quality for a post-editor(Vasconcellos and Léon, 1985:135;Somers, 1997:201; Wagner 1987:73).Wagner (ibid) reports that translators whoare forced to post-edit will not be asefficient as those who have volunteered.She also suggests that “a certain amount ofconfidence in one’s own translation abilityand technical expertise is essential for thistype of work” (ibid:204).

There are few differences between theskills mentioned above and those

4 By “contrastive knowledge”, the authors referto both SL and TL components that mapbetween texts and deep representations of“interface structures” (Johnson and Whitelock,1987:137).

demanded of a professional translator.However, ability to use macros, to codedictionaries for MT, and a positive attitudetowards MT are three attributes required ofa post-editor that are not usually demandedof a translator.

This author would argue that severalother skills are required for successfulpost-editing and these are addressed in thefollowing sections.

3.1 Knowledge of MTKnowledge of MT technology in generalwould go a long way towards helping thepost-editor understand what is going on inthe so-called “black-box” and why certainerrors occur consistently. Understandingthe history of MT development, its currentstatus and future prospects would ensurethat the post-editor had an appreciation forthe technology, its limitations and how itmight improve in the future.

3.2 Terminology Management SkillsWhile most trainee translators are taughtthe theory and practicalities of terminologymanagement, the trainee post-editor wouldbenefit from an extensive course inmachine translation dictionary coding andterm base management. In any onetranslation environment, multiple tools canbe used to store and retrieve terminologyboth for source and target text production.This presents challenges when terms haveto be used across multiple tools andprocesses. Trainee post-editors not onlyneed to know how to code MTdictionaries, but they also need to knowhow to manage term bases. This requiresknowledge of multiple term managementtools and terminology exchange formats,which are only emerging at this time (see,for example, the OLIF, TBX, SALT andXLT initiatives.5

5 For information on OLIF seehttp://www.olif.net/; on TBX and SALT see:http://www.opentag.com/tbx.htm; on XLT seehttp://www.ttt.org/oscar/xlt/dxlt.html. (Allwebsites last checked on October 12, 2002).

103

3.3 Pre-editing/Controlled Languageskills

It has been documented on numerousoccasions that authoring source text usingcontrolled language rules improves MToutput.6 A drawback to this approach isthat authors are unwilling to be constrainedby controlled language rules. Analternative solution is to use anintermediate editor who has the necessaryskills to apply CL rules to a text before it issubmitted to MT. Being an expert in bothsource language and target language makesthe post-editor a good candidate for thisjob. There is also a significant incentive,i.e. it reduces the time spent on cleaning uptedious and non-sensical errors in multipletarget language versions! Therefore,knowledge of controlled languages andcontrolled authoring tools would benefitpost-editors.

3.4 Programming skillsVasconcellos (1986a:136) mentions usingmacros as a necessary skill for post-editors. In the current author’s opinion, apost-editor is an ideal candidate for writingmacros to automatically clean-up textssince s/he has extensive experience ofcommonly occurring errors. These macrosare the first step towards the concept of anautomatic post-editing tool, as suggestedby Ryan (1988), Knight and Chander(1994), Allen and Hogan (2000). Ifequipped with programming skills, thepost-editor could develop his or her ownprogramme for automatically correctingconsistent errors for specific languagepairs, text types and MT systems.

3.5 Text linguistic skillsAs mentioned above, Vasconcellos(1986b) outlines the importance ofknowledge of theme and rheme and otherlanguage-specific text type norms for post-editing. A good grounding in textlinguistics would therefore seem to be of

6 See Adriaens et al. 1996, Mitamura et al.1998, Adriaens et al. 2000.

benefit to post-editors. This knowledgecould be applied not only for post-editingbut also for programming macros andautomatic post-editing modules.

4 Proposed outline for a coursemodule in post-editing

We have so far established the need forteaching post-editing, the requirement thatpost-editing be taught to translationstudents and the core skills required. Theadditional skills of knowledge of MTsystems, terminology management,controlled language, programming and textlinguistic skills have been added to thislist. In this section, a course outline thataddresses these skill-sets will be proposed.

Since we have established that the bestcandidate for this type of training is atranslation student, we will assume that thestudent has acquired certain skills beforetaking the module in post-editing. If this isnot the case, then these skills would haveto be taught prior to the post-editingmodule: specialised translation skills;basic linguistics; basic terminologymanagement; IT skills; an introduction tolanguage technology (focussing ontranslation memory tools). An assumptionis also made that the student has excellentsource and target language skills.

The module in post-editing is dividedinto two, with a focus on theoretical issuesin the first half and a focus on practicalissues in the second half.

4.1 Theoretical ComponentThe theoretical content will contain thefollowing subjects:

• Introduction to Post-editing• Introduction to Machine Translation

Technology• Introduction to Controlled Language

Authoring• Advanced Terminology Management• Advanced Text Linguistics• Basic Programming Skills

104

The Introduction to Post-editing wouldaddress the concept of post-editing. Forexample, why do we need post-editing,how does it differ from translation andrevision, what levels of post-editing exist,how do we determine user requirements,what technology can we use for post-editing, can we classify typical post-editingerrors, and so on?

The Introduction to MachineTranslation Technology should cover thehistory of MT, MT system types,description of commercial MT systems,evaluation methodologies, current state ofthe art, including integration withtranslation memory tools, and futureprospects.

The Introduction to ControlledLanguage Authoring should include ahistory of Controlled Languages, adescription of CL tools, evaluationmethodologies for CL tools, current stateof the art, integration with authoring andMT tools, and future prospects.

Advanced Terminology Managementwould aim to build on the basicterminology management skills the studentbrings to the course, by discussing thestrengths and weaknesses of terminologymanagement tools, dictionary coding forMT, and, most importantly, terminologyexchange between tools using terminologyexchange standards such as XLT, OLIF,and TBX.

Advanced Text Linguistics would buildon the basic linguistic skills of the studentby introducing them to the standards oftextuality, text type classification, and theuse of corpus linguistics and corpusanalysis tools for analysing text types.

Basic Programming Skills wouldintroduce the student to the basics ofprogramming and would then instruct thestudent in macro programming and in aprogramming language suitable for NaturalLanguage Processing, for example Perl.

4.2 Practical ComponentSince post-editing is a practical skill andone of the objectives of teaching post-

editing is to allow a student to acquire the“comfort” factor Vasconcellos talks aboutbefore being recruited, practical experienceof post-editing would form a majorcomponent in this module. According toVasconcellos (1986a:145), a post-editor atPAHO (the Pan-American HealthOrganisation) post-edits 100, 000 words,or almost one full working month, beforethat level of comfort is reached. While itmay not be possible for a student to attainthis goal, especially considering theworkload from the theoretical componentof a programme, the student should beencouraged to practise post-editing bothwithin and outside course hours. Post-editing of different text types fromdifferent MT systems should be carriedout. If the student has more than one targetlanguage, post-editing into multiple targetlanguages would be desirable. Also, sincepost-editing requirements sometimes varybetween producing a text for informationpurposes and for publication purposes,students would practise these different“levels” of post-editing.

Practical experience with at least twocommercially-available MT systems wouldalso form part of the practical component.Students would be required to submit textsfor translation to the MT system and toanalyse and compare the results whensystem settings have been changed, user-specific terminology has been coded, and,where possible, linguistic rules have beenaltered. Students would also be required toinvestigate the pros and cons of a MTsystem’s integration with a translationmemory tool.

To gain practical experience ofterminology management tools, dictionarycoding utilities, and terminology exchangeformats, students could be instructed tocreate a term data base using a specificterminology management tool (let’s callthis term management tool A), to code thatterminology using a machine translationdictionary coding tool and then to exportthe terminology from the MT dictionary toa second term management tool (called

105

term management tool B) using differentterminology exchange standards.

Practical experience of controlledauthoring tools could be gained in thefollowing manner: Students are asked tocheck and edit a text in the sourcelanguage using a CL tool and to submit thecontrolled and uncontrolled texts to anumber of MT systems. Post-editing ofboth versions would then reveal the prosand cons of controlled authoring formachine translation.

Corpus Analysis practical experiencecould be gained by compiling parallelcorpora, tagging them, and analysing themfor specific text linguistic features such astheme/rheme structure, voice, tense,cohesive ties, etc. using corpus analysissoftware such as Wordsmith tools.

Finally, students would acquirepractical programming skills by writingmacros to automatically apply commonchanges in target texts. They could alsoapply the programming language skillslearned in the theoretical component of thecourse by designing a rudimentaryautomatic post-editing application.

When designing a course in anacademic environment, it is usual tospecify how many hours are dedicated toeach component of the course, whether acomponent is considered core or optional,what weighting a component has incomparison to other components and howeach component will be assessed. Sinceevery academic institution differs, there islittle point in specifying this informationhere. Instead, a general suggestion is madethat equal status be given to the theoreticaland practical components and thatassessment be carried out on a continuousbasis, using practical and written methodsof assessment.

The reader will most likelyacknowledge that the proposed courseoutline is quite extensive and that, unlessthis module was offered as a stand-aloneprogramme, such as a graduate certificateor diploma, it would be difficult to coverall components adequately in the time-

frame of an academic semester. Theproposed outline is an ideal, from theauthor’s point of view. The module could,of course, be split over two semesters.

5 At what stage should post-editing be introduced?

Since translator training programmes arestructured differently in every institutionthat offers this type of programme, it isimpossible to say exactly where a courseon post-editing fits in. However, as hasalready been mentioned, successful post-editing requires a high level of confidencein the post-editor’s own work (Wagner,1987:204). In the introduction to his bookon traditional revising and editing, Mossop(2001) reports his finding thatundergraduate students are rarely ready forself-revision or revision of others’ workuntil after they have completed apracticum. This, too, suggests thatexperience and confidence are necessaryingredients for the task of post-editing.

In the proposal outline above, it wasassumed that students would take a post-editing course only if they had some pre-requisite skills, i.e. excellent languageskills, specialised translation skills, basiclinguistics, basic terminology manage-ment, IT skills and an introduction tolanguage technology. Since experience,confidence in one’s own work, and anumber of pre-requisite skills are requiredbefore taking a course in post-editing, thissuggests that such a course should only beoffered in the last part of an under-graduate translator training programme, or,even more ideally, in a post-graduateprogramme because the students enrolledin the latter type of programme are morelikely to be experienced and to have moreconfidence in their own work.

6 SummaryCurrent industry trends seem to suggestthat machine translation will form anincreasing part of the technology solutionsput in place to meet the growing demandfor translation. If this turns out to be true,

106

an growing number of translators will haveto deal with machine translation output.Let the educators of translators preparefuture generations for this by teachingstudents about machine translation andpost-editing. This paper outlines the skill-sets required and proposes course contentand structure.

ReferencesAdriaens, Geert, Jeffrey Allen, Arendse Bernth,

Kurt Godden, Teruko Mitamura, EricNyberg, Rick Wocjik, Rémi Zajac (2000),Third International Workshop onControlled Language Applications, CLAW2000, Seattle, Washington.

Adriaens, Geert, Roger Havenith, Rick Wocjikand Bruno Tersago (1996), FirstInternational Workshop on ControlledLanguage Applications, CLAW 96, Centrefor Computational Linguistics, Leuven,Belgium.

Allen, Jeffrey and Christopher Hogan (2000),“Toward the Development of a Post editingModule for Raw Machine TranslationOutput: A Controlled LanguagePerspective”, in Adrieans et al. (2000), 62-71.

Allied Business Intelligence, Inc. (1998),Language Translation: World MarketOverview, Current Developments andCompetitive Assessment, Oyster Bay, NewYork.

Johnson, Roderick L. and Peter Whitelock(1987), “Machine Translation as an ExpertTask”, in S. Nirenburg (ed) MachineTranslation: Theoretical and Methodo-logical Issues, Cambridge University press,Cambridge, 136-144.

Knight, Kevin and Ishwar Chander (1994),Automated Post-editing of Documents, 12th

National Conference on ArtificialIntelligence, Seattle, Washington, 779-784.

Krings, Hans P. and Geoffrey S. Koby (eds)(2001), Repairing Texts: EmpiricalInvestigations of Machine-TranslationPost-Editing Processes, Kent StateUniversity Press, Kent, Ohio.

Löffler-Laurian, Anne-Marie (1985),Traduction automatique et style, Babel 31,70-76.

McElhaney, Terrance and Muriel Vasconcellos(1988), “The Translator and the Postediting

Experience”, in Vasconcellos (1988), 140-148.

Mitamura, Teruko, Eric Nyberg, GeertAdriaens, Linda Schmandt, Rick Wocjick,Rémi Zajac (1998), Second InternationalWorkshop on Controlled LanguageApplications, CLAW 98, LanguageTechnologies Institute, Carnegie MellonUniversity, Pittsburgh, Pennsylvania.

Mossop, Brian (2001), Revising and Editing forTranslators, St. Jerome Publishing,Manchester, UK.

Ryan, Joann P. (1988), The Role of theTranslator in Making an MT System Work:Perspective of a Developer, in Vasconcellos(1988), 127-132.

Senez, Dorothy (1998a), “Post-Editing Servicefor Machine Translation Users at theEuropean Commission”, Translating andthe Computer 20, London.

Senez, Dorothy (1998b), “The MachineTranslation Help Desk and the Post-EditingService”, Terminologie et Traduction 1,289-195.

Somers, Harold (1997), “A Practical Approachto Using Machine Translation Software:‘Post-editing the Source Text’”, TheTranslator 3, 193-212.

Vasconcellos, Muriel (1986a), “Post-editingOn-screen: Machine Translation fromSpanish into English”, Proceedings ofTranslating and the Computer 8, London.

Vasconcellos, Muriel, (1986b), “FunctionalConsiderations in the Post-editing ofMachine Translation Output: Dealing withV(S)O versus SVO”, Computers andTranslation 1, 21-38.

Vasconcellos, Muriel (1988) (ed), Technologyas Translation Strategy, AmericanTranslators Association ScholarlyMonograph Series, Vol. II, State Universityof New York at Binghamton (SUNY),

Vasconcellos, Muriel and Marjorie Léon(1985), “SPANAM and ENGSPAN:Machine Translation at the Pan AmericanHealth Organization”, ComputationalLinguistics 11, 122-136.

Wagner, Emma (1985), “Post-Editing Systran– A Challenge for Commission Translator”,Terminologie et Traduction 3.

Wagner, Emma (1987), “Post-editing –Practical Considerations”, in ITIConference I: The Business of Translatingand Interpreting, London, 71-78.