Translation strategy decision support at the European Commission Martin Karlberg and Werner Grünewald Directorate-General for Translation, European Commission, L-2920 Luxembourg Abstract Purpose The paper presents a decision support (simulation) tool for matching translation supply and demand of European and other international multilingual translation services. Design/methodology/approach The tool takes multiple constraints into account, including the availability of internal translation staff, translation quality control requirements, translator work rate, or available freelance translation budget. The model can be used to underpin strategic decisions concerning e.g. an increased scope of translation or changes in internal translation staffing. Findings The tool has been applied successfully on data from the European Commission s Directorate-General for Translation (DGT) in a study commissioned by DGT top management who wanted to assess the impact of increasing the scope of the texts accepted for translation. It is found that an increased scope could, for some languages, be absorbed by in-house translation staff already in place but for other languages, DGT would have to rely on freelance translation to a greater extent. As a consequence, the European Commission would have to increase its budget for freelance translation. Research limitations By its very nature, any model is a simplified representation of reality. While the tool developed is sophisticated enough to accommodate multiple constraints, it is based on a global view of translation supply and demand. There is thus room for developing an even more complex model based on microsimulation. In principle, such a model could better mimic reality by incorporating aspects such as the fluctuation over time of translation demand as well as the workload of individual translators. Practical implications The paper provides European and other international public administrations having multilingual translation services with a simulation tool for ex ante assessment of the impact of strategic decisions. Social implications Translation services make up a relatively large part of the European Union public administration, and translation into all Official Languages is an integral part of the European legislative process. Without an adequate balance between translation demand and translation resources, the European Commission service would not be able to provide legislative texts in all Official Languages. Thereby, the European legislative process would be slowed down severely. Application of the tool presented in this paper reduces the risk of such a mismatch between demand and resources by facilitating strategic decision-making underpinned by quantitative evidence. Originality/value This paper provides a structured translation strategy decision support tool, which, in comparison to previous ad hoc models, takes a number of constraints inherent to a multilingual translation service into account. Keywords: Translation strategy, decision support systems, simulation, European Union. 1(22)
1. Introduction One of the principles of the European Union (EU) concerns the equal treatment of all official languages. This principle has already been fixed in the very first piece of legislation approved by the then Council of the European Economic Community (Council of the European Economic Community, 1958). It stipulates that any citizen or company can approach any of the European Institutions in any of the official languages, and has the right to receive a reply in the same language. This principle also means that any piece of legislation has to be made available in all official languages of the EU. When this principle was established in 1958, there were just four official languages. Following different rounds of enlargement, the number of official languages has meanwhile increased almost fivefold to 23 (European Commission, 2009, p. 59). With a number of countries being candidates to join the EU, there will be further enlargements, which will further increase the number of official languages. In addition to this drastic increase in official languages, the responsibilities of the different institutions have also increased. Taken together, this has resulted in a steadily increasing demand for translation. Taking the European Commission (EC) with the largest translation service of all European institutions as an example, its translation output exceeded 1.7 million pages in 2010. The translation service of the EC, the Directorate General for Translation (DGT), currently has some 2 500 staff members out of which about 1 600 are translators while the rest are management and support staff. Although DGT employs about 10 % of all Commission officials (European Commission, 2011) and has a budget well exceeding 10 million EUR for external (freelance) translation at its disposal, these resources are still not enough to satisfy all translation needs of the EC. On 1 May 2004, 10 new Member States joined the EU, bringing a total of nine additional official languages with them, leading to a steep increase of translation demand. In response to this need, the Commission s first ever Demand Management Strategy (European Commission, 2004) was adopted. This strategy distinguishes between core documents (such as legal acts) for which translations must be provided into all official languages, documents which must be made available in the three procedural languages of the Commission (English, French and German), and other documents (such as working papers, minutes of meetings, Commission internal documents or speeches) which are translated more on an ad-hoc basis, if at all. Apart from the sheer volume of translation, the distribution of translation requests over time also poses a challenge. DGT has constant internal translation resources and can use external resources to deal with request peaks to some extent but the distribution of translation requests over time is fluctuating considerably a not very surprising fact as translation requests in a political organisation like the European Commission are themselves a function of political developments, and such developments are often neither foreseeable nor easily predictable. Moreover, in spite of the low profile of the translation activity in comparison to the more visible EU policy areas, the potential consequences of failure are grave. As all language versions of EU legislation are official, it cannot be adopted unless it is available in all official languages. Thus, if there is a shortage of translation capacity for a single official language, the entire EU legislative process risks grinding to a halt. In order to prevent translation capacity shortfalls, careful planning must be applied, both on the micro level, i.e. concerning how to manage individual translation requests, and on the macro level, i.e. concerning the general translation strategy for matching translation demand with Commission translation resources. 2(22)
DGT is constantly striving to improve its planning and management capacities to better cope with unforeseen challenges. In this paper, we address the issue matching demand and resources for translation at the macro level, and present a tool developed for this purpose. First, the key demand, management, resource and capacity concepts used as model parameters are presented in Section 2. Thereafter, Section 3 addresses the decision support tool and how it could be used to assess the impact of changes in various parameters. In Section 4, it is demonstrated how the tool could be used to assess the impact of an increased scope of translation. A concluding discussion is held in Section 5. 2. Concepts In order for a simulation model to be set up, a number of parameters need to be defined. Given the multitude of symbols introduced, we list all symbols used in this paper in the Annex for ease of reference. First, we introduce the key concept of language () in Section 2.1; in principle, demand for translation into one language could only be met by a translator having that language as its mother tongue. Thereafter, we proceed to define the concept of (in-house or outsource) execution (EXE) in Section 2.2; outsourced, i.e. freelance, translation is far less demanding in terms of in-house human resources (but does on the other hand entail an increased expenditure). In the same section, we also define the concept of task type (TTP), it is through tasks of various types that translation demand turns into workload of DGT translators. In Section 2.2, we also introduce the quality control level (QCL) concept, which determines at which rate translations are quality controlled. These four concepts (, QCL, EXE and TTP) are subsequently used as indices for the concepts defined in the sequel; for instance, and QCL are used as indices for demand in order to distinguish between the demand for (say) quality control level 1 translation into German and (say) quality control level 2 translation into English. For all concepts, we present values heavily influenced by the actual values either values observed for DGT in previous years, or benchmark values defined by DGT senior management. We introduce the concept of demand (DMD) in Section 2.3. This demand has to be met by internal human resources (i.e. the DGT translators) and financial resources for external translation (i.e. the freelance/outsourcing budget); we therefore introduce the concepts of internal resources (RSI) and external resources (RSE) in Section 2.4. In order to see how far these resources would go in meeting demand, we somehow have to determine how many pages a translator could translate and/or quality control, and how many pages of freelance translation the outsourcing budget could cover. This is covered by Section 2.5, in which we set a benchmark for the annual translator capacity, i.e. the work rate (WRT), and define the supplementary task difficulty (TSD) which is used to treat translation and quality control jointly. In the same section, we define the price per externalised page (PRE) to determine how much the outsourcing budget could render in terms of pages of freelane translation. Finally, we use Section 2.6 to present how translation demand is managed via the outsourcing rate (OSR) and the task rate (TSR), which is the rate at which quality control tasks are performed. 2.1. Languages An obvious key element of translation in a multilingual context is the language for demand as well as for resources. In terms of demand, the documents to translate both have a source language (i.e. the language of the original document which is to be translated) and one or more target languages (i.e. the 3(22)
language(s) into which the document is to be translated). With 23 official languages, there are 23 22=506 different possible combinations of source and target language. However, as could be seen from Figure 1 (which contains rounded figures, meaning that language combinations for which only a few hundred pages are translated per year are not presented), the number of language combinations actually employed in practice for a substantial number of pages is considerably smaller. EN 120 000 20 000 110 000 DE 15 000 20 000 7 000 60 000-70 000 per language 10 000-20 000 per language EL ES IT NL PT PL 5 000-10 000 per language 2 000-6 000 per language FR 60 000 per language 4 000-7 000 per language BG CS DA ET FI HU LT LV MT RO SK SL SV 4000 per language 6 000 GA Figure 1. Annual translation production (number of pages) of DGT by source and target language; based on the actual production in recent years. Note: The figures are rounded. The 23 official languages of the EU (and their two-letter abbreviations used throughout this paper) are: Bulgarian (BG), Czech (CS), Danish (DA), Dutch (NL), English (EN), Estonian (ET), Finnish (FI), French (FR), German (DE), Greek (EL), Hungarian (HU), Irish (GA), Italian (IT), Latvian (LV), Lithuanian (LT), Maltese (MT), Polish (PL), Portuguese (PT), Romanian (RO), Slovak (SK), Slovenian (SL), Spanish (ES) and Swedish (SV). As translation involving languages other than the official languages (e.g. Russian and Chinese) is marginal, the overview is limited to the official languages of the EU. We see from Figure 1 that: English (EN) is the main drafting language of the EC, with 60 000 pages of English originals being translated into each other official language each year. As French (FR) and German (DE) are procedural languages of the EC, the translation production for these languages is even higher (with more than 100 000 pages of translation from EN into each of these languages each year). French is also sometimes used as a drafting language, but less frequently so, with a few thousand pages of French originals being translated into all other official languages each year Translation from the other ( non-procedural ) official languages mainly takes place whenever the EC receives incoming correspondence in that language; the document is 4(22)
then translated into English or French; for most languages, this concerns a few thousand pages per year. Irish (GA) has a special status, and only a few thousand pages are translated from EN into GA each year. The source language distribution is thus rather skewed, and completely dominated by EN (with FR a distant second). From a macro perspective, not much information is lost if this aspect is disregarded. On the other hand, the target language distribution is far more equitable, with a demand volume of 60 000 pages per year for most languages. In the sequel, we will therefore focus on the target language aspect. This distribution is also reflected in the distribution of language skills of translators. Concerning the source language, each translator has very good passive knowledge of at least two foreign languages including at least one language out of English and French meaning that their ability to cover original texts in the predominant drafting languages (English and French) is very good. Although Pavlović (2007) points out that reverse translation (i.e. translation from one s mother tongue into another language) is commonplace, DGT adheres to the philosophy of Newmark (1988), who according to Pavlović states that translat[ing] into your language of habitual use [ ] is the only way you can translate naturally, accurately and with maximum effectiveness. Thus, given the high linguistic quality requirements for the translated text, translators are typically only translating into their working language (i.e. their mother tongue), meaning that the distribution of staff with respect to their working language is far more even. DGT is organised accordingly, with one language department per official language (subdivided into translation units), meaning that each translation assignment for a particular language is managed by the corresponding language department. To summarise, with the source language distribution dominated by EN and FR, and with virtually all DGT translators having a good passive knowledge of one or both of these languages, the key aspect to focus on when matching translation demand and resources is the target language of the texts and the working language of the translators. In the sequel, the term language () is therefore limited to the 23 official languages of the European Union, and will mainly be used to distinguish between documents with respect to their target language, and between translators with respect to their working language. 2.2. Execution, translator tasks and document categorisation For each document arriving, the translation unit manager has to decide whether a document should be translated by the in-house translators working in his/her unit, or whether it should be outsourced. This decision depends on numerous aspects (the current/foreseen availability and specialisation of in-house translators, as well as the nature of the particular document in question to mention but a few). In the sequel, we reflect this via the execution (EXE) aspect of a document. If a document is translated in-house, execution is said to be internal (EXE=INT), whereas execution is said to be external (EXE=EXT) for outsourced documents. Obviously, documents that are translated in-house at DGT have to be translated by somebody. In practice, this takes place by assigning a task having the task type (TTP) translation (TTP=TRA) to a DGT translator. However, TRA is not the only kind of task befalling DGT translators. Often, translations have to be revised, i.e. thoroughly checked against the original text by a fellow translator. This gives rise to task of the type revision of internal translation 5(22)
(TTP =REI) for documents that have been translated in-house (EXE=INT), as well as tasks of the type revision of external translation (TTP=REX) for documents that have been outsourced (EXE=EXT). The documents translated by DGT are of various types (legal texts, correspondence with citizens, web sites, press releases, brochures, technical reports etc.) and concern numerous policy areas (agriculture, competition, environment, energy etc.). For most document types, accuracy of the translation is primordial, whereas the impact of a less than perfect translation is not equally severe for other document types. This is reflected in the quality control level (QCL) of the document (European Commission, 2007). The highest quality control standards are applied for quality control level 1 (QCL=1). Among these documents, which are routinely revised, we have for instance legislative texts - here, a faithful rendition of the original legislative text is of essence, since all language versions of EU legislation are equally authentic the argument that (say) the Hungarian version of a directive is merely a translation of the English original doesn t hold water in court (however, as described by Solan (2007) the European Court of Justice generally looks to the legislative purpose in interpreting statute rather than basing its judgements on literal application of texts with obvious translation errors ). Less exacting quality control standards are applied for quality control level 2 (QCL=2). These documents are less frequently revised; an example of such documents is incoming correspondence, for which it only is important that the general meaning of the letter is conveyed, and more subtle aspects such as the exact wording and the style of the translation is of secondary importance. As will be shown below, execution (EXE), quality control level (QCL) and task type (TTP) all are necessary aspects by which the model parameters are defined. 2.3. Demand Over the last years, DGT has translated between 1.5 and 2 million pages annually, with one page being defined as 1 500 characters of text (today, a typical single-space A4 page often contains twice as many characters as that, but the concept of a page actually predates the modern information technology era, and corresponds to one A4 page of double-spaced typewriter text). Whereas the annual translation production volume varies with a few thousand pages between languages in practice, the simulations have been carried out under the assumption that with the exception of the procedural languages (English, French and German), for which more documents are translated (European Commission, 2006, 14-19) and Irish for which only key documents are translated (Council of the European Union, 2005), all official languages are structurally similar. This is reflected in Figure 2, in which the presumed baseline annual translation demand to be met would be the same (approximately 70 000 pages) across all non-procedural languages (except Irish). From Figure 2, it could also be seen that for the procedural languages in general, and for English in particular, the number of quality control level 2 documents is higher. This is due to the types of documents that are translated only into the procedural languages and not into the other languages. For instance, there is plenty of incoming correspondence from administrations, enterprises, organisations and citizens of the EU Member States which is translated into English only so that Commission staff not proficient in the language of the Member State can understand the content of what has been submitted. 6(22)
DMD,QCL n. of pages 200 000 180 000 160 000 140 000 120 000 100 000 80 000 60 000 40 000 20 000 0 QCL 2 QCL 1 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 2. Annual translation demand (DMD) in number of pages by language () and Quality Control Level (QCL). 2.4. Resources Based on DGT data for the past few years, it is reasonable to assume the translation volume is rather similar across the non-procedural languages, and that the differences observed for a particular year are mainly attributable to random fluctuations. In other words, the best guess for the demand of a particular language is not the exact demand observed in the past year, but rather the average demand across all non-procedural languages. Hence, the demand volumes are similar across all languages in Figure 2. In contrast, the actual number of translator posts attached to each department differs across languages and any observed differences are likely to persist (or decrease gradually) over the years. The presumed internal resources (RSI) presented in Figure 3 are thus based on actual staffing which varies across otherwise structurally similar language departments. RSI n. of translators 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0 DE EN FR BG CS DA EL ES ET FI HU IT LT LV MT NL PL PT RO SK SL SV GA Figure 3. DGT internal resources (RSI) in terms of the number of full-time translator posts by language (). 7(22)
To achieve an equitable workload for units with less staff (and to manage workload peaks), DGT also has external resources (RSE) in the form of a budget for external translation. The presumed value of RSE, based on previous budgets, is 17 500 000 EUR. 2.5. Capacity Translators have a wide variety of tasks to carry out. Apart from the obvious translation (TRA) task, there are also, as described in Section 2.2 above, various quality control tasks. Other tasks include work which in not necessarily attributable to a specific translation request, such as improvement of terminology databases. For simplicity, we presume in the current model that the volume of general work (such as terminology) is equally distributed across all language departments (meaning that it will not affect the below capacity calculations). On the other hand, we do account for the quality control tasks in our capacity calculations. As there are various task type, we bring them all together by converting tasks into translation page equivalents, applying different degrees of task difficulty (TSD) to different tasks. In Table 1, the values established by DGT senior management (European Commission, 2010) are presented. Trivially, one page of translation (TRA) obviously corresponds to one translation page equivalent ; more importantly, it is assumed that revision of internal translation (REI) and revision of external translation (REX) have a difficulty of one-third (0.333), meaning that 30 pages of revision is regarded as requiring the same amount of effort as 10 pages of translation. Table 1. Task difficulty (TSD) by task type (TTP) TTP TSD TTP TRA 1 REI 0.333 REX 0.333 Now that we have a means of converting the tasks befalling a translator in a year into one single value (the number of translation page equivalents), we need to know how high the capacity of an average translator is. We refer to this capacity, i.e. the number of translation page equivalents that a translator could execute in a year, as the translator work rate (WRT). Here, we will use the target established by DGT senior management (European Commission, 2010), which works out to a work rate (WRT) of 1 250 page equivalents per year and translator (see Figure 4). 8(22)
WRT page eqs. per annum 2000 1800 1600 1400 1200 1000 800 600 400 200 0 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 4. Translator work rate (WRT) in terms of the number of translation page equivalents per translator and year by language (). In order to correctly assess the capacity for external translation, we need to make some assumption on the price per page of translation. In Figure 5, we present the price per page of freelance translation (based on actual amounts charged in 2010) used in this paper. PRE EUR/page 100 90 80 70 60 50 40 30 20 10 0 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 5. Average price (EUR) per page of freelance translation (PRE) by language. 9(22)
2.6. Management As noted above, the decision whether to outsource the translation of a particular document depends on numerous aspects. However, in our model, which is defined on a macro level, specific decisions are not considered, but rather the average outsourcing rate (OSR). As illustrated in Figure 6, this rate is assumed to depend only on the language () involved and the document quality control level (QCL; introduced in Section 2.2 above). OSR,QCL 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% QCL 2 QCL 1 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 6. Baseline outsourcing rate (OSR) by language () and Quality Control Level (QCL) Similarly, the decision on whether or not to conduct quality control for a particular document depends on several aspects as well and in our model, we do not consider individual documents, only the average control rate (TSR). Furthermore, while the model allows it to depend on the language (), the execution (EXE) and the quality control level (QCL) of a document, we do (as could be seen from Table 2) in essence only let it depend on the quality control level, with 90% quality control rate for quality control level 1 and 40% quality control rate for quality control level 2. (The task rate of 100% for TRA is just a technical devise in order to address the fact that there is always a task of the type TRA for any internal translation). Table 2. Task rate (TSR) by language (), Quality Control Level (QCL), execution (EXE) and task type (TTP) QCL EXE TTP TSR,QCL,EXE,TTP (Any) (Any) 1 1 INT INT TRA REI 100% 90% (Any) 1 EXT REX 90% (Any) 2 INT TRA 100% (Any) 2 INT REI 40% (Any) 2 EXT REX 40% 10(22)
3. A decision support tool With the key demand, resource, capacity and management elements defined in Section 2 above, we now turn our attention to how all these concepts could be brought together to render impact estimates which could underpin the strategic decisions of the translation service. This is done by means of a model in which most of the elements are kept constant and a few independent X elements are used as model parameters which are changed the remaining Y elements are set up to depend on the X elements. We could then assess the impact of the X elements on the Y elements. In Section 3.1, we present the general model setup used for determining how changes in demand would affect the outsourcing rate, and in Section 3.2 the algorithms needed for implementing this model, as deployed in the 2010 simulation study conducted at DGT, is described in greater detail. Finally, other possible variants, using other independent variables, are presented in Section 3.3 in order to demonstrate that the model could be applied in many other situations, and not only in the case of changing translation demand. 3.1. Increased demand At the European Commission, an increase in the translation demand could occur for a number of reasons. For instance, the general level of political or legislative activity of the EC could increase. Alternatively, at a constant level of activity, the scope for translation could increase, with additional documents being translated into all official languages. If demand increases, DGT has to be address it somehow with the resources at hand. Here, we present a procedure for doing so by means of increasing the outsourcing, if internal capacity is not sufficient. We start by assuming that the situation to address is an increase in demand for translation from DMD to DMD. We further assume that most aspects would remain unchanged: The increased demand would take place in a situation of zero headcount growth; i.e. the internal resources (RSI) would stay the same. The quality control requirements would remain, meaning that the quality control rates (TSR) would stay the same. The difficulty (TSD) and translator capacity (WRT) would remain unchanged, as would the freelance translation prices (PRE). With most of the parameters fixed, the demand increase would have the following effect: Once all of the internal capacity has been used up, the outsourcing rate (OSR) would have to increase from OSR to OSR. By comparing the freelance budget to the cost of freelance translation resulting from this new set of outsourcing rates OSR, we find out whether there would be a budgetary surplus or a deficit. 3.2. Implementation There are several steps involved in the implementation of the model. However, taken by itself, each step is, with few exceptions, rather straightforward. The model therefore lends itself very well to implementation in an IT tool accommodating algorithms with several 11(22)
simple steps. In order to enable the interested reader to try out this model, we have therefore chosen to present all steps involved in detail. First, we calculate the internal capacity for translation and quality control in terms of page equivalents per year for each language: ICP =RSI WRT. (1) For notational convenience, we split up DMD into in-house translation volume (TRI): and outsourced translation volume (TRO): TRI,QCL =DMD,QCL [1 OSR,QCL ] (2) TRO,QCL =DMD,QCL OSR,QCL. (3) Thereafter, we calculate the internal capacity need, i.e. how many page equivalents are requested for each language during the year: ICN ( ) = 2 QCL= 1 TRI (, QCL) TTP TSR + TRO (, QCL, INT, TTP ) (, QCL) (, QCL, EXT, TTP ) TTP TSR. (4) For each language, we then end up with an internal capacity balance: ICB = ICP ICN. (5) For languages where ICB >0, we have a capacity surplus, and need not undertake any further action in order to be able to cope with the demand. Conversely, for languages where ICB <0, we have a capacity deficit, which needs to be addressed in order for all translation demand to be met. As mentioned in Section 3.1 above, we will address this by means of increasing the outsourcing. In principle, a capacity deficit of (say) 1 000 pages equivalent would be resolved by increasing outsourcing by 1 000 pages. However, the operational reality is a bit more complex. If one page of translation is outsourced, one does indeed not have to do the translation work in-house, and so there is no in-house translation to revise either. However, on the other hand, there will be one page of outsourced translation which might have to be revised. In order for us to know by exactly how much the outsourcing volume needs to be increased, we therefore calculate the net internal capacity gain obtained from outsourcing one page of translation as GNO,QCL =TSR,QCL,INT,TRA TSD TRA +TSR,QCL,INT,REI TSD REI TSR,QCL,EXT,REX TSD REX. (6) For each language, and each quality control level, we then obtain the potential capacity gain from outsourcing of all translation currently conducted in-house as PCG,QCL =TRI,QCL GNO,QCL. (7) 12(22)
Generally, quality control level 2 documents are more suitable for outsourcing (as could be seen from the current outsourcing rate; see Figure 6). Therefore, for each deficit language, we start out by assessing if outsourcing quality control level 2 documents would suffice to eliminate the deficit, by setting the new outsourcing rate of quality control level 2 documents as follows: TRO ( PCG + ICB ) TRO, 2 ICB GNO, 2 if, 2 > 0 = DMD,2 otherwise.,2 (8) If necessary, i.e. if all quality control level 2 documents were outsourced (meaning that there is still a need for capacity) we also increase the outsourcing rate of quality control level 1 documents by setting TRO,1 if ( PCG,2 + ICB ) > 0 TRO,1 = DMD,1 if ( PCG,1 + PCG,2 + ICB ) < 0 (9) TRO,1 ICB GNO, 1 otherwise. To summarise, these adjustments have three possible outcomes; it could suffice to meet the deficit by outsourcing quality control level 2 documents, or we might also have to outsource quality control level 1 documents, or, in the worst case, we would adjust the outsourcing rate to 100% but still have an internal capacity deficit (this is possible since outsourced documents also consume internal capacity via the quality control task REX). For languages with 100% outsourcing, the remaining capacity deficit is calculated as 2 IC B = ICP DMD QCL TSR.,, QCL, EXT, TTP (10) QCL= 1 TTP This third case is rather far from the current operational reality of DGT, but might be a completely realistic case for organisations which mainly provide translations by means of outsourcing, using in-house linguistic staff mainly for quality control of the freelance production. We define the outsourcing shift for each language and quality control level as: OSH,QCL = TRO TRO. (11),QCL,QCL Thereafter, we assess whether it is possible to cope with this increased outsourcing within the frame of existing budgetary resources by computing the outsourcing budget balance as the difference between the outsourcing budget and the overall cost of outsourcing: OBB = RSE QCL= 1 The impact of the increase in demand could then be summarised via 2 TRO, PRE. (12) QCL 13(22)
the internal resource surplus { :ICB >0} ICB WRT (13) the internal resource deficit { :ICB < 0} (14) the increased outsourcing for quality control level 1 OSH, 1 = OSH,1 (15) the increased outsourcing for quality control level 2 OSH, 2 = OSH,2 (16) the freelance budget surplus or deficit OBB. 3.3. Other situations The model is not limited to assessment of the impact of changes in translation demand; it could also accommodate situations in which other aspects of the translation service change. For instance, the model could assess the impact of: staffing changes (via RSI); changes in work rate (via WRT); changed quality control requirements (via TSR): changed outsourcing prices (via PRE). In each of these cases, the procedure described in 3.2 above would be applied mutatis mutandis to render capacity surplus/deficit, outsourcing change and freelance budget balance. 4. Application of the decision support tool As previously mentioned, the algorithm presented in Section 3 above has been deployed in practice at DGT, using the values presented in Section 2 above as the baseline scenario. In 2010, DGT top management wished to assess the impact of increasing the scope of the texts accepted for translation, and commissioned a simulation study for this purpose. A number of scenarios were examined, all involving expansion of the package usually translated in connection with new legislative proposals. Below, we present how the decision support tool described in Section 3 above was applied to one of these demand increase scenarios. 4.1. Increased demand In the scenario studied here, there would be an additional 9 000 pages of texts associated to legislation (hence having quality control level 1, i.e. more stringent quality control criteria) to translate from English into all official languages except Irish (in Figure 7, this increase is illustrated in the white bar segments). 14(22)
DMD (,QCL) n. of pages 200 000 180 000 160 000 140 000 120 000 100 000 80 000 60 000 40 000 20 000 0 QCL 2 QCL 1 - increase QCL 1 - baseline DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 7. Increased annual translation demand (DMD ) in number of pages by language () and Quality Control Level (QCL). 4.2. Establishing the internal capacity balance By multiplying the workrate (WRT) and the number of translator posts (RSI), as in (1), the internal capacity (ICP) is obtained (see Figure 8). ICP page eqs./year 200 000 175 000 150 000 125 000 100 000 75 000 50 000 25 000 0 DE EN FR BG CS DA EL ES ET FI HU IT LT LV MT NL PL PT RO SK SL SV GA Figure 8. Annual internal capacity (ICP) in number of page equivalents by language. Applying (4), we calculate the internal capacity need based on the demand (DMD), the outsourcing rate (OSR), the task rate (TSR) and the task difficulty (TSD). 15(22)
ICN page eqs./year 200 000 175 000 150 000 125 000 100 000 75 000 50 000 25 000 0 DE EN FR BG CS DA EL ES ET FI HU IT LT LV MT NL PL PT RO SK SL SV GA Figure 9. Annual internal capacity need (ICN) in number of page equivalents by language (). By taking the difference between the capacity (Figure 8) and the need (Figure 9), the internal capacity balance (5) is obtained (see Figure 10). It should be noted that both of the languages (ES and PT) for which there is excess capacity belong to the group of languages for which the annual translation volume is somewhat higher than for the typical non-procedural language (see Figure 1). It is thus possible that the presumed demand (see Figure 7) is a bit low for these languages; if the baseline parameters were adapted to incorporate a higher demand for those non-procedural languages, there would be no excess capacity for these languages. 100 000 ICB page eqs./year 75 000 50 000 25 000 0-25 000-50 000-75 000-100 000 DE EN FR BG CS DA EL ES ET FI HU IT LT LV MT NL PL PT RO SK SL SV GA Figure 10. Internal capacity balance (ICB) in number of page equivalents by language. 4.3. Meeting the capacity deficit through outsourcing Applying (6), we have that for each language, the outsourcing gain is 16(22)
for QC level 1 documents and GNO,1 =100% 1.0+90% 0.333 90% 0.333 = 1 GNO,2 =100% 1.0+40% 0.333 40% 0.333 = 1 for QC level 2 documents. Thus, since the quality control rates for internal and external translation coincide, and since the same difficulty is assigned to revision of internal translation (REI) as to revision of external translation (REX), the net internal capacity gain of outsourcing one page of translation is exactly one page. In other situations, the outsourcing gain could differ from 1. For instance, if outsourced translations could be expected to be of subpar quality, and thus are revised more frequently (TSR,QCL,EXT,REX > TSR,QCL,INT,REI ) and are more difficult to revise (TSD REX > TSD REI ), the net capacity gain from outsourcing might be far less than one page, since the outsourcing of one page would entail an increased quality control workload. As the net capacity gain per outsourced page has been established at one page equivalent, applying (7) renders a potential capacity gain (see Figure 11) identical to the in-house translation volume. PCG,QCL p. eq./year 200 000 180 000 160 000 140 000 120 000 100 000 80 000 60 000 40 000 20 000 0 QCL 2 QCL 1 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 11. Potential internal capacity gain (PCG) from outsourcing currently internally translated documents expressed in terms of the number of page equivalents per language () and quality control level (QCL). Comparing the grey segments (potential capacity gain from 100% outsourcing of QC level 2 documents) of Figure 11 to the instances of negative internal capacity balance (Figure 10), we see that they rarely suffice to absorb the entire negative capacity balance, and that some increase in outsourcing of QC level 1 therefore is necessary. This is illustrated in Figure 12, where we see that after applying (8), (9) and (11), an increase in QC level 2 outsourcing is sufficient for for DE, EN, FR, FI and CS to meet the internal capacity deficit, whereas for all 17(22)
other languages with capacity deficits, an outsourcing increase must also take place for QC level 1 documents. OSH,QCL n. of pages 200 000 180 000 160 000 140 000 120 000 100 000 80 000 60 000 40 000 20 000 0 QCL 2 QCL 1 DE EN FR DA EL ES FI IT NL PT SV CS ET HU LT LV MT PL SK SL BG RO GA Figure 12. Increase in outsourcing (OSH) necessary to meet the increased demand. Expressed in terms of the additional number of pages outsourced per year (by language and quality control level). 4.4. Impact summary The impact of the demand increase could now be summarised as follows: applying (13) to the data of Figure 10, we have that the internal resource surplus would be approximately 12 posts applying (14) we see that there would be no internal resource deficit applying (15) and (16) to the data of Figure 12, we see that there would be an additional outsourcing of approximately 130 000 QC level 1 pages and an approximately 50 000 QC level 2 pages. applying (17), the outsourcing budget deficit is estimated to be approximately 5 400 000 EUR. 5. Discussion The tool presented in this paper was developed in the European Commission s Directorate General for Translation in 2010 in the course of internal reflections on possible modifications of the European Commission's Translation Strategy. No such tool was available when the previous strategies (European Commission, 2004 and 2006) were established. It would be interesting to see to which extent the strategy would differ if the tool had already been available in 2004. The recent internal discussions on possible changes to the existing strategy have shown that such a tool can point to interesting and not always directly visible features of such complex (translation management) systems and help the discussion on more strategic considerations instead of minute technical details. Examples of such features are the limiting character of the language () with the lowest relative capacity, which has as an effect that excess internal capacity in one language can coexist with internal capacity deficits in another language. 18(22)
In its concrete application the tool has not only been used for one-shot estimations (point estimations). In fact, a whole set of scenarios within a certain range of values for individual variables has been tested, such as the effect of an increase of demand in the range from 5 000 to 30 000 pages with steps of 1 000 (2 000 or 5 000 or any other appropriate multiple of) pages. Such ranges need not be restricted to one variable, but could be set for two or more variables at the same time. An example would be to combine the range for an increase of demand with a second range for changes in staffing. The combination of each value from the first range with each value from the second (or any further) range would give a whole set of possible effects for the variable under investigation which could be summarised in a kind of effect interval (empirical confidence interval). Assuming that not all values of such a range have the same probability to occur, one can combine the values of the ranges with their expected probability to happen. The result is again an empirical confidence interval indicating the more or less likely effects of the assumptions about the basic variables such as an increase in demand or changes in outsourcing. As applied currently, the tool does not reflect the full complexity of DGT s translation process. Certain important aspects are only partially or not at all covered. An evident example concerns the thematic specialisation of translators. It is fact that many of DGT s translations are of a very technical nature, and it is generally accepted that translators are the more efficient, the more they are familiar with a certain area (agriculture, competition, energy etc.). If the requests for translations in a very technical area increase above the limit of available expertise, either the work rate of the language departments goes down or the outsourcing rate must increase. The effects would be the same if many more translations from other, non-official languages such as Arabic, Chinese or Russian are requested or not enough support staff is available so that translators have to deal increasingly with administrative tasks. Another case leading to shortcomings in specific language departments is to deviate from the principle to translate into the mother tongue, and to ask translators from other language departments to translate from their mother tongue into another language (so called reverse translation, also referred to by Pavlović (2007) as L2 translation ). It should be noted that certain of these complexities could be accommodated within the scope of the current model. For instance, the issue of thematic specialisation could be covered by extending all parameters which currently depend on language () to also depend on the thematic specialisation. It could of course be argued that the application of a tool such as the one presented unduly reduces the complexity of the texts translated by a translation service to mere numbers (since there is no page which is exactly equal to any other page in terms of complexity and difficulty) and that it de-humanises the translators by converting them into a faceless internal resources commodity. Still, it has be borne in mind that by its very nature, any model is a simplified representation of a more complex reality, and the challenge is to determine which of the more complex aspects should be retained. In theory, one could go even one step further and try to simulate individual translation requests. One could also attempt to expand the model to cover management and support staff, who also are part of the translation workflow. The result would be a fairly complex micro-simulation model, the application of which would not only lead to estimations of the variables presented above but also of other variables for example of the workload pressure at the level of units for specific calendar days. The additional information yielded by such a complex model would have to be weighted against the advantage of the relative simplicity of the current model, which could be deployed rather rapidly to generate impact estimates for a number of interesting scenarios, thereby helping 19(22)
DGT management to fulfil its duty to ascertain that the European Commission s legislative activity does not grind to a halt. While the tool has been developed for use within the European Commission s Directorate-General for Translation, it is sufficiently general for application in similar settings as well, with certain adaptations. For instance, while the quality control level (QCL) concept is specific to DGT, this concept could be used to represent any other document categorisation which, together with the language combination concerned, determines the outsourcing and quality control rates. Possible users comprise translation services of other EU institutions, translation services of international institutions such as the United Nations or national translation services of countries with multilingual administrations. References Council of the European Economic Community 1958 Regulation 1/1958 determining the languages to be used by the European Economic Community, Official Journal of the European Communities L 17, pp. 385-386. Council of the European Union 2005 Regulation 920/2005 amending Regulation No 1 of 15 April 1958 determining the language to be used by the European Economic Community and Regulation No 1 of 15 April 1958 determining the language to be used by the European Atomic Energy Community and introducing temporary derogation measures from those Regulations, Official Journal of the European Union, L156, pp. 3-4. European Commission 2004 Matching supply and demand for translation, Communication SEC(2004)638. Brussels: European Commission. 2006 Translation in the Commission Responding to the challenges in 2007 and beyond, Communication SEC(2006) 1489. Brussels: European Commission. 2007 Translation Quality Control, DGT internal note KJL/MBR D(2007) of 17 December 2007. Brussels: European Commission. 2009 La traduction à la Commission: 1958-2010. Brussels: European Commission. 2010 Efficient management of translation workload operational guidelines for Heads of Department and Heads of Unit. DGT internal note dgt.c.dir(2010) 983858. Luxembourg: European Commission. 2011 Distribution of officials and temporary agents by Directorates General and function groups, European Commission, Directorate-General of Human Resources (http://ec.europa.eu/civil_service/docs/europa_sp2_bs_catsexe_x_dg_en.pdf, version of 3 March 2011). Newmark, P. 1988 A textbook of translation. London: Prentice Hall. Pavlović, N. 2007 Directionality in translation and interpreting practice. Report on a questionnaire survey in Croatia, Forum 5(2), 79-99. Solan, L. M. 2007 Statutory Interpretation in the EU: The Augustinian Approach, Brooklyn Law School Legal Studies Research Paper No. 78. New York: Brooklyn Law School Legal Studies. (http://ssrn.com/abstract=998167). 20(22)
Annex. List of symbols Symbol DMD,QCL EXE GNO,QCL ICB ICP ICP OBB OSH,QCL Description/possible values Demand (number of pages) Annual demand for translation of texts having quality control level QCL into language Execution In-house (INT), outsourced (EXT) Internal capacity gain from outsourcing (number of page equivalents) Number of page equivalents of internal capacity freed if a quality control level QCL page to be translated into language is outsourced instead of translated in-house introduced in (6) Internal capacity balance (number of page equivalents) The difference between the available internal capacity and the internal capacity need introduced in (5) Internal capacity (number of page equivalents) Number of page equivalents which could be executed in-house by the translators attached to language department introduced in (1) Internal capacity need (number of page equivalents) The annual number of page equivalents needed to execute all tasks related to in-house and outsourced translation introduced in (4) Language The 23 official languages of the EU (and their two-letter abbreviations used throughout this paper) are: Bulgarian (BG), Czech (CS), Danish (DA), Dutch (NL), English (EN), Estonian (ET), Finnish (FI), French (FR), German (DE), Greek (EL), Hungarian (HU), Irish (GA), Italian (IT), Latvian (LV), Lithuanian (LT), Maltese (MT), Polish (PL), Portuguese (PT), Romanian (RO), Slovak (SK), Slovenian (SL), Spanish (ES) and Swedish (SV). Outsourcing budget balance (EUR) Annual differences between the outsourcing budget and the financial resources needed for covering the necessary outsourcing volume introduced in (12) Outsourcing shift (number of pages) Increased number of outsourced pages (following application of the model to an increased demand DMD ) for quality control level QCL pages to be translated into language introduced in (11) 21(22)
Symbol Description/possible values OSR,QCL Outsourcing rate (%) Proportion of the translation demand for language and quality control level QCL which is met by means of freelance translation PCG,QCL PRE QCL RSE RSI Potential capacity gain (number of page equivalents) Total number of freed internal capacity if all quality control level QCL pages to be translated into language are outsourced introduced in (7) Price per externally translated page (EUR) Average price charged for freelance translation of one page into language. Quality control level Quality control level 1 (QCL=1) with high quality control standards Quality control level 2 (QCL=2) with less exacting standards External resources (EUR) Annual outsourcing/freelance translation budget Internal resources (number of full-time translator posts) Number of translator posts attached to language department TRI,QCL In-house translation volume (number of pages) introduced in (2) TRO,QCL Outsourced translation volume (number of pages) introduced in (3) TSD TTP Task difficulty (number of page equivalents per page) Task difficulty for task of the type TTP TSR,QCL,EXE,TTP Task rate (%) Proportion of the in-house (EXE=INT) or outsourced (EXE=EXT) pages of translation into language with quality control level QCL for which a task of type TTP is executed TTP WRT Task type Translation (TRA), revision of internal translation (REI), revision of external translation (REX) Work rate (number of page equivalents) Average number of page equivalents that could be executed per year by a translator attached to department 22(22)