<LINK"ber-n*"> <ATSKSWVIUITUEOZTIBYADTEFLJWRFHTEESGHHCOOEE"ETTRRTTI"DhG1"""i5STH"n4S0ba"kiT"le>"r-vrpgai"earl"oot2Duc2B1ed0Oes3"rsC:n-p2oIarNror(idt2eFoin0Onc0tioe"1ld)s"TinS,trcaonnsclautriroenntrevseerabraclhis"ation,strategies,translationunits,affectiveinvolvement,automaticity,professionalism,routineness" Think-aloud protocols in translation research Achievements, limits, future prospects* SilviaBernardini ScuolaSuperiorediLingueModerneperInterpretieTraduttori, UniversityofBologna Overthelastdecade,Think-aloudProtocols(TAPs)havebeenusedexten- sivelyinprocess-orientedTranslationStudies(TS).Theseriousquestions regardingtheexperimentalvalidityofthisresearchmethodologywhen appliedtotranslationhavenonethelessoftenremainedunspoken.This papersurveysthebreakthroughsaswellasthelimitsofthegrowingbodyof literaturedealingwithTAPsinTS,pointsatthenecessitytotakeissuesof experimental,theoreticalandenvironmentalvaliditymoreseriously,and offerssuggestionsforimprovements.Theclaimisthattherisksinvolvedin theadoptionofalaxexperimentalmethodologyinTAPstudies,oftenun- derestimatedinthepast,mayinvalidatenotonlytheresultsobtainedinthe singleprojects,but,crucially,themethodasawhole. Keywords:process-orientedTS,concurrentverbalisation,strategies,transla- tionunits,affectiveinvolvement,automaticity,professionalism,routineness 1. Introduction Interestinempiricalresearchintothetranslationprocesshasgrownsubstan- tially in the last decade, driven by the idea that what goes on in translators’ headswhiletheyaretranslating(versuswhatscholarshadclaimedmightgoon) iscrucialtoanunderstandingoftranslation,andisnotderivablesolelyfroman analysisofthefinalproduct,thetranslatedtext.Thelatterprovidesanincom- plete and often misleading way into the translation process, hiding both successfulstrategiesandproblems. Target 13:2 (2001), 241–263. issn0924–1884/e-issn1569-9986 ©2002JohnBenjaminsPublishingCompany 242 SilviaBernardini Anumberofattemptshavebeenmadeataccessingthetranslator’smind, withvaryingdegreesofsuccess.Onesuchattempthasbeentoaskthetransla- torsthemselvestorevealtheirmentalprocessesinrealtimewhilecarryingout atranslationtask.Suchamethodofdatacollection,knownas“thinkingaloud”, has been imported from the cognitive sciences and applied to translation research, often with little reflection on the consequences inherent in the application of the approach to the new research framework. Theoretical justificationshavebeenimportedwithoutquestioningtheirapplicabilitytothe newsettings,andthevalidityofthemethodasawholehasbeenassumedrather thanproved. We now seem to be witnessing a decrease of interest in Think-aloud Protocols(TAPs),aftertheenthusiasticeffortsofthepasttentofifteenyears.A firstphase,withidentifiablecharacteristics(anecdotalattitudes,methodological compromises, proliferation of classification schemes) is approaching its conclusion; the birth of a second, more mature testing phase will very much dependonthewillofresearchersworkingwithinthisframeworktomoveonto amorerigorousexperimentalmethodology.Theaimofthispaperisprimarily to provide a survey of TAP-based research so far, with its achievements and limits,andthentolookahead,atwhatapotentialsecondphasemightlooklike, wereitevertoseethelight. 2. TAPs:Asurvey 2.1 Thetheoreticalgrounding:Think-aloudinpsychology andcognitivescience ThetheoreticalframeworkforTAPexperimentsisprovidedmainlybythework ofEricssonandSimon1(esp.1993(1984)).Accordingtotheirmodel(“human cognitionisinformationprocessing”),informationiskeptindifferentmemory stores,withvaryingaccessandstoragecapabilities:whereasshort-termmemory (STM) is characterised by easy access and severely limited storage capacity, long-termmemory(LTM)ischaracterisedbymoredifficultaccessandlarger storagecapacity.OnlyinformationpresentinSTM,thatis,informationwhich isbeingheededbythesubject(staticandconscious“knowledgestates”rather than dynamic and unconscious cognitive processes), can be directly accessed andreported.Itfollowsthatthecognitiveprocessestheseknowledgestatesare inputsandoutputsto,aswellasinformationthatisnotcurrentlybeingheeded, <<LLIINNKK""bbeerr--rr33"">> Think-aloudprotocolsintranslationresearch 243 cannot be reported but must be inferred by the analyst on the basis of the verbalisations.Afurtherassumptionofthismodelisthat,forverballyencoded information,whichcanbereportedinthesameformastheoneinwhichitwas heeded,theverbalisationdoesnotinterferewiththecognitiveprocessitself,the onlyeffectofthinking-aloudbeingtoslowdowntheperformance.Theimplica- tionsofthismodelaremanifold.Hereweshallonlyconsidersomeofthemore closelyrelevanttoourdiscussion. First of all, only concurrent verbalisation of thoughts can be claimed to exhaustivelyreflectthementalstatesofasubjectcarryingoutarelativelylong task(letussay,longerthantenseconds).Oncompletionofsuch“long”tasks, partoftheinformationmovesontoLTM,leavingbehindretrievalcuesonlyin STM: in such cases, post hoc verbalisation has been found to be difficult and oftenincomplete(EricssonandSimon1993(1984):xvi).Moreover,rulingout thepossibilitythatasubjectisinterpretingher/hisownthoughtprocessesor even generating them anew, instead of retrieving them from LTM, can be extremelyproblematicunderthesecircumstances,thusmakingresultsvirtually uninterpretable. Secondly, in order to make sure that the reports actually reflect mental stateswithoutdistortingthem,itisimportantthatsubjectsdonotfeeltheyare takingpartinsocialinteraction:albeitobviouslyamuchmorenaturalsituation, conversation involves reworking thoughts to make them conform to socially established norms, a process which might sensibly alter the information attended to. The interaction between subject and experimenter (or between subjects)shouldthereforebeavoidedoratleastreducedtoaminimum.Aplea forenvironmentalvalidityisunsustainableinthiscase:TAPsareeitherstrictly monologicalornotTAPsatall. Thirdly, practice and experience may affect the amount of processing carriedoutinSTM,sothatfewermentalstateswillbeavailableforverbalisation to subjects experienced in a task. This process, known as “automation”, is explainedthus: Before overlearning has occurred, processes have to be interpreted, with substantialfeedbackfromintermediateprocessingstagesinSTM.Overlearning amountstocompilingtheseprocesses,sothatfewertestsareperformedwhen theyarebeingexecuted,hencelessinformationisstoredatintermediatestages inSTM.(EricssonandSimon1993(1984):127) Automatic processes are therefore faster and more efficient than processes which are under conscious control. However, they are also less flexible and moredifficulttomodifyatneed. 244 SilviaBernardini Finally,thismodeltakesintoaccounttheeffectsofpersonalityandpersonal history over the data collected. The amount of relevant information held in LTM cannot possibly be controlled for, as an experimental situation would require,norisitpossibletocontrolfortheamountofknowledgereportedon inrelationtotheperformancegiven.Eventhoughthisisawell-knownproblem in cognitive science research, TAPs are particularly sensitive to it, suggesting thatitwouldbeadvisabletotryandlimittheeffectsofindividualdifferencesas muchaspossible,andtotakethemintoaccountduringtheanalysis,inorderto obtainmorereliableandgeneralisabledata. Summarising: – Concurrentverbalisation,orthinkingaloud,providesdataonthemental statesheededbyindividualscarryingoutatask. – From these states it is then possible to derive information about the relevantmentalprocesses. – Under the right circumstances (verbally encoded information, no social interaction,nointerferences,noinstructiontoanalysethoughts),verbal- isingisassumednottointerferewiththementalprocessesandtoprovide afaithfulaccountofthementalstatesoccurringbetweenthem. – ThegeneralisabilityandtherelevanceofthedataobtainedthroughTAPs, however, is more difficult to assess, and cannot be guaranteed by the modelitself. 2.2 TAPsintranslationstudies 2.2.1 Achievements Viewing translation mainly as a problem-solving process, some TS scholars haveputforwardthesuggestionthatitshouldbepossibletostudyitbymeans of TAPs, and have set up experiments to test this hypothesis. The varying interestsandbackgroundsofthoseinvolvedhaveresultedinalargevarietyof approaches,whichcanonlybrieflyandunsystematicallybesurveyedhere.In this sub-section we focuson achievements, in the following welook at some inadequaciesobservedandsuggestwaysofproceedinginthefuture,ifTAPsare togoonhavingaroleinTS. MostearlyTAPstudieswereconductedwithforeignlanguagelearnersor translatortrainees.Thiswasmainlyduetotheavailabilityofsubjectsandtothe pedagogicconcernsoftheexperimenters.However,thehypothesiswasalsoput forward that the verbalisations produced by professionals would be less informative than those produced by non-professionals, due to their more <<LLIINNKK""bbeerr--rr143">"> Think-aloudprotocolsintranslationresearch 245 “automatised”processingstyle.Weshallhavemoretosayaboutthisissue.In the meantime, let us consider one of the major early concerns of researchers workingwithinthisparadigm,namelytheanalysisof“translationstrategies”.2 We shall then move on to consider more recent foci of attention, such as attentionunits,automaticityofprocessingandaffectivefactors. 2.2.1.1 Strategies. The researchers whose work is surveyed in this sub-section haveeitheravoidedaterminologicaldiscussionofthetermstrategy(forwhich alternative definitions abound in linguistics) and used the term in a rather undefined, everyday sense, or endorsed the definition provided by Lörscher (who, in turn, adapts Færch and Kasper’s (1983) definition), according to whichatranslationstrategyis a potentially conscious procedure for the solution of a problem which an individualisfacedwithwhentranslatingatextsegmentfromonelanguageinto another.(Lörscher1991:76) Lörscherhimself(1986and1991)reportsonacomparativelylargeTAPstudy, in which 48 German learners of English as a foreign language produced 52 translationseitherintoEnglishorintoGerman.Theywereaskedtoproducea spokentranslationofawrittentextwhilethinkingaloudandwerenotallowed tousedictionaries(thiswasmeanttoensurethatalargernumberofproblem- solvingprocesseswouldbepresentintheprotocols).3 Thetranscriptsofthe sessions were then analysed and a number of “translation strategies” were recognised. According to Lörscher each strategy is formed of a sequence of core elements which can be combined in different ways. A translation process, in turn,isformedofaseriesofstrategies,whichcanalsobecombinedindifferent ways.Thegeneralconclusionsdrawnfromthisstudyarethat: – TAPs seem to provide reliable and useful data, provided that the analyst interpretstheminasystematicand“methodologicallycontrolled”way; – Despiteindividualdifferencesandtheinherentvariabilityofthetranslation process, there are regularities that point at the possibility of establishing taxonomiesoftranslationstrategies; – Noevidenceisfoundoftranslation-specificstrategies:instead,generaltext- processingstrategiesareadaptedtothespecifictaskathand. – Asaside-effectofthinkingaloud,thelearners’capacityforproblem-solving seems to increase, suggesting that this research methodology might also haveimportantpedagogicapplications. <<<<LLLLIIIINNNNKKKK""""bbbbeeeerrrr----rrrr116211"">>"">> 246 SilviaBernardini Krings (1986) reports on an experiment in which eight German learners of Frenchasaforeignlanguagetranslatedatexteitherintooroutofthemother tongue. The main focus of attention here is the identification of translation problemsandtranslationstrategiesonthebasisofTAPs.Withregardstothe former,Kringsoffersthefollowinglistof“problemindicators”: – Thesubjects’explicitstatementofproblems – Theuseofreferencebooks – Theunderliningofsource-languagetextpassages – Thesemanticanalysisofsource-languagetextitems – Hesitationphenomenainthesearchforpotentialreplacements – Competingpotentialreplacements – Themonitoringofpotentialreplacements – Specifictranslationprinciples – Themodificationofwrittentarget-languagetexts – Theassessmentofthequalityofthechosentranslation – Paralinguisticornon-linguisticfeatures(Krings1986:267) With regards to translation strategies that subjects resort to when automatic processingbreaksdown,Kringssuggeststhatthesecanbeclassifiedasstrategies ofcomprehension(inferencinganduseofreferenceworks),equivalentretrieval (especially interlingual and intralingual associations), equivalent monitoring (such as comparing source text (ST) and target text (TT)), decision-making (choosing between two competing solutions) and reduction (for instance of markedormetaphoricaltextportions). A more complex classification of strategies is proposed by Gerloff (1986: 252ff.)who,inhermethodologicallyorientedpaperonTAPstudies,describes “text-processingstrategies”as“anymetalinguisticormetacognitivecomments madeorspecificproblem-solvingbehavioursaffected,duringthedecodingand rendering of the translation text”.4 The categories she identifies are problem identification,linguisticanalysis,storageandretrieval,generalsearchandselection, textinferencingandreasoning,textcontextualisation,andtaskmonitoring. In their discussion of the use of lexical search strategies, Mondhal and Jensen(1996)distinguishproductionfromevaluationstrategies.Theformerare further subdivided into achievement strategies and reduction strategies (also discussed by Chesterman 1998). Among achievement strategies, which are characterisedbyanattempttoremainclosetotheST,arespontaneousassocia- tionandreformulation.Amongreductionstrategies,whicharecharacterisedby their inherently remedial nature, are avoidance and unmarked rendering of <<<LLLIIINNNKKK"""bbbeeerrr---rrr111737""">>> Think-aloudprotocolsintranslationresearch 247 markeditems.Finally,evaluationstrategiesinvolve,forinstance,reflectingon theadequacyandacceptabilityoftranslationreplacements. Séguinot(1996)reportsonanothernon-comparativestudyinvolving,this time, two professional translators working together at the same task. The underlyingassumptioninthiscaseisthatthiseverydaysetting(thesubjectsare used to working as a team) would increase the environmental validity of the experiment,withoutlimitingtheexperimentalvalidityoftheresultsobtained. As a result of this study four types of translation strategies are identified as being typical of “professional” translation, namely interpersonal strategies (brainstorming, correction, phatic function), search strategies (dictionaries, worldknowledge,words),inferencingstrategies(rereadingSTandTT,consult- ing) and monitoring strategies (rereading ST and TT, consulting, comparing units).Thistranslatingprocessisfurtherdescribedas“iterative”,proceedingin anon-linearfashionandoperatingonthebasisofsentence-level“translation units”,whichare,however,ofteninterruptedbypausesandhesitations. None of the studies described so far attempt to systematically compare strategies across two groups of subjects. However, finding out what it is that distinguishes professional from non-professional (student or lay-person) behaviourhasalwaysbeenamajorconcernofresearchersinprocess-oriented translationstudies.Onewayofinvestigatingthisissuehasbeentocomparethe performanceoftwogroupsonthesametask. An investigation along these lines is described by Lörscher (1996) who, building on his previous studies (mentioned above), compares the strategies adopted by professional and non-professional translators (foreign-language students).Hepointsoutthat,althoughthetwogroupsdonotdifferqualitative- lyintheiruseoftranslationstrategies,theydodifferquantitatively,i.e.inthe distribution and frequency of the various strategies employed. Furthermore, differencescanbedetectedintheorientationoftheapproach(towardsformin thecaseofnon-professionals,towardssenseinthecaseofprofessionals),inthe sizeoftranslationunits,intheamountofmonitoringoftheTT,andlastlyin theattentiondevotedtostylisticandtypologicaladequacy(greaterforprofes- sionalsinallcases). InthestudyreportedinSéguinot(1991),twosimilartextsweretranslated by students of translation at different levels of proficiency (at the beginning andattheendoftheircoursesinspecialisedtranslation).FrenchandEnglish mother-tongue speakers translated two advertisements from French into English. The main research focus was once again on the — rather loosely defined — notion of strategies. The author suggests that native speakers of <<LLIINNKK""bbeerr--rr198">"> 248 SilviaBernardini English (as well as better students; the two categories are unfortunately not distinguishedclearly)translatingintotheirmothertongueshowmoreefficient monitoring5 and revising strategies, and work more at the textual level, whereas non-native speakers seem to rely more on learned principles and lexical-levelprocesses. BuildingonLörscher’sdefinition,Jääskeläinen(1993)proposesaclassifica- tionoftranslationstrategiesdistinguishingbetweenglobalandlocalstrategies, theformerapplyingtothewholetask(considerationsaboutstyle,readership, etc.), the latter to specific items (i.e. lexical searches). On the basis of this distinction,sheisabletoclaimthatglobalstrategiesaremuchmorefrequently usedbyprofessionalsandsemi-professionals(translatortrainees)thanbynon- professionalsinherstudy.Aftermakingaplan,theformerappeartofollowit systematicallythroughthetask,whereasthelatterseemtoproceedinamore haphazard way. A similar conclusion is reached by Tirkkonen-Condit (1992) who compares the performance of a professional and of a non-professional translator on the same task, and finds that the professional relies more on textual and linguistic knowledge, whereas the non-professional works with smallerunitsandreliesmoreonextra-textualknowledge.Eventhoughthismay beaconsequenceofthelatter’sgreaterfamiliaritywiththesubjectareaasmuch asaconsequenceofherlesserfamiliaritywiththetranslatingtask,theissueof theinterrelationoflinguistic/textualandextra-textualknowledgewouldappear tobeapromisingtopicforfurtherstudy. Letmetryandchartverybrieflythegroundwehavecoveredsofar.Anumberof TAPstudies,especiallyearlyones,havebeenconcernedwiththerecognitionand classification of translation strategies and with the detection of differences betweenprofessionalandnon-professionalstrategies.Anumberofclassificatory schemes have been provided, adopting labels like global/local, reduc- tion/achievement,monitoring(andrevising),search,comprehension,equivalent retrieval, decision making and so on. Besides, it has been suggested that the performanceofprofessionalsdiffersfromthatofnon-professionalswithregards toboththequantityandthequality(orientation)ofthestrategiesadopted. Letusnowturntoconsiderthreeotherissueswhichhavebeeninvestigated bymeansofTAPs,namelytranslation(orattention)units(2.2.1.2),automaticity (2.2.1.3)andaffectivefactors(2.2.1.4). <<<<<<<<<LLLLLLLLLIIIIIIIIINNNNNNNNNKKKKKKKKK"""""""""bbbbbbbbbeeeeeeeeerrrrrrrrr---------rrrrrrrrr11996399979""""""">>>>>>>"">> Think-aloudprotocolsintranslationresearch 249 2.2.1.2 Translationunits. Translation(orattention)unitsaredefinedas Thoseinstancesinthetranslationprocessinwhichthetranslator’s‘unmarked processing’ is interrupted by shifting the focus of attention onto particular task-relevantaspects(Jääskeläinen1990:173,citedinJääskeläinen1993:102) “Unmarkedprocessing”herereferstounproblematicsectionsoftheprotocols inwhichasubjectverbalisesfluentlywhilereadingorwriting.Markedprocess- ing, then, begins with a problem indicator and ends with a solution to the problemoranindicationthattheproblemistemporarilyabandoned. A“unitofanalysis”codingisdescribedbyGerloff(1986),whoidentifies seven levels of analysis, going from morpheme or syllabic unit to discourse. Accordingtomostresearchers,thelengthoftranslationunitsisanindication ofproficiency,withprofessionaltranslatorsworkingwithlargerunits(sentence anddiscourse,orgroup)andmovingmorecomfortablybetweendifferentunit levels. Clearly, this does not mean that a professional translator never stops midwaythroughasentence,butonlythatthesentenceisprocessedasaunit, withmorelocalproblemstackledontheway(Séguinot1996).Thesuggestion canbeputforward,therefore,thatattentionunitsarebetterdefinedinhierar- chicalratherthansequentialterms,withsmallerunitsbeingprocessedwithin larger units. The search for a term or collocation may be embedded in the processing of a whole sentence, without implying a “word unit” or “phrase unit”analysis. 2.2.1.3 Automaticity. Insofarasautomaticityofprocessingisbelievedtoresult fromexperienceandproficiencyinatask(EricssonandSimon1993(1984); Toury 1988), it is not surprising that researchers have tried to determine whethertheperformanceofprofessionalsisrecognizablymoreautomaticthan thatofnon-professionals.Inordertodoso,theyhaveanalysedtheamountof marked processing in professionals’ and non-professionals’ TAPs. The most straightforwardhypothesis(thatprofessionalsverbaliselessthannon-profes- sionals)isnotendorsedbyJääskeläinenandTirkkonen-Condit(1991)orby Jääskeläinen (1996 and 1997), who make a distinction between routine and non-routinetasks/situations.Intheformer,professionalsdotendtoverbalise lessthannon-professionals,whereasinthelattertheamountofverbalisation is not necessarily smaller. Besides, the nature of the verbalisations tends to differ aswell.Theexplanationofferedisthat“whilesomeprocessesbecome automated,otherprocessesareevokedintoconsciousness,i.e.thetranslator becomessensitisedtonewkindsofproblems”(JääskeläinenandTirkkonen- Condit1991:105). <<<<<<<<LLLLLLLLIIIIIIIINNNNNNNNKKKKKKKK""""""""bbbbbbbbeeeeeeeerrrrrrrr--------rrrrrrrr2111198902888""">>>""""">>>>> 250 SilviaBernardini This conclusion is supported by the finding that semi-professionals (translator trainees) show more extensive processing than both professionals andnon-professionals(Jääskeläinen1997).Thismaybebecausetheyareaware oftheproblemsinvolvedbuthavenotyetautomatisedthenecessaryproblem- solvingstrategies.Equally,professionalsareassumedtobebetteratrecognising theneedtoresorttonon-automatic,controlledprocesses(i.e.problemrecogni- tion) than non-professionals. Automatic processes, as we saw above, are typicallyveryefficientbutnotveryflexible,sothatthereisthedanger(pointed out by Wilss 1994:144) “of problems being forced into a certain structure, because it is believed to offer a solution”. A typical example of this danger would be, for instance, the difficulty experienced by non-professionals in overrulingautomaticlexicalassociations(Ivanova1998:102),or“false-friends”, aprocessrequiringhighcontrol. 2.2.1.4 Affective factors. Leavingaside cognitiveissues for amoment, weshall now turn briefly to consider affective factors in translation. These have been investigated, among others, by Kussmaul (1991); Tirkkonen-Condit (1997); Laukkanen(1996);Tirkkonen-ConditandLaukkanen(1996);andJääskeläinen (1997).Theseresearchersagreethataffectivefactorssuchasinvolvementwith thetranslationtask,arelaxedatmosphereandself-confidencecorrelateposi- tivelywithwhattheyregardas‘successfulperformance’.Inroutinetasks,where these three elements are likely to be present, subjects are found to produce bettertranslationsthaninnon-routinesituations,wheretheytendtostickmore to the ST and avoid reduction strategies as much as possible (Laukkanen 1996:266).Thisfindingshouldbecarefullyevaluatedinrelationtothevalidity oftheresultsobtainedinexperimentalconditions,whereaffectivefactorsare likelytoinfluencetheresultsobtainedinunpredictableways. Afurtherwayintothetranslationprocessisofferedbytheevaluations(of self,task,ST,TT)verbalisedbythesubjects.According toTirkkonen-Condit (1997:83), there is a quantitative as well as qualitative difference between professionals and non-professionals in these regards, due to the fact that “consciousness of the motivations and rationale of one’s own performance seemstogrowwithtranslationalexperience”. 2.2.1.5 Conclusion. As can be gleaned from the above discussion, the large amounts of data about the cognitive and affective factors involved in the translationprocesswhichcouldbecollectedbymeansofTAPshavenodoubt favouredamoreempiricalapproachtothestudyoftranslation,andhighlighted