1. A multilingual setting in the technological era: translation

1Multilingual language professionals play a key role in today’s society. In fact, one such type of multilingual communicative activity, i.e. written translation, has become increasingly indispensable in a growing area of expertise domains. At a time when English, Chinese and Spanish are gaining momentum in different parts of the world (Siegel 2018, 491), and at a time when many other languages are becoming extinct (Siegel 2018, 382ff), the idea might grow that in a foreseeable future only a few languages will be used. However, the exact opposite seems to be taking place, and mainly because of globalisation and modern communication means, the worldwide volume of translation work has been increasing tremendously. While the internationalisation and localisation of products such as software and websites already strengthened the Globalisation, Internationalisation, Localisation and Translation (GILT) industry (Fernández Costales 2012), the optimistic trend in the European language economic activity sector, or language industry as they call themselves, has continued. With their annual Language Industry Survey, a collaboration of different European stakeholder associations (EMT, EUATC, FIT Europe, GALA and LIND) under the leadership of ELIA aims 'to establish the mood of the industry' in order to complement other surveys with the 'dimensions of perception and trust, which determine to a great extent the actions of industry stakeholders' (ELIA & al. 2018, 3). They established two types among their 866 respondents (translation companies and independent professionals) and reported that there is continued growth in the industry. Although both groups to a similar extent mentioned activities within the highly specialised domains such as finance, medical, ICT and telecom, the domain of translation activity showed a few differences as well. The translation volume ratio produced by the translation companies was somewhat larger in the domains of traditional manufacturing and automotive industries than that of the independent professionals, while the latter turned out to show larger volume ratios than the companies in the 'humanity' areas of legal translation, government, tourism, marketing and literature (Figure 1).

Figure 1. Customer type (Source: ELIA & al. 2018, 5).

Figure 1. Customer type (Source: ELIA & al. 2018, 5).

2From the survey they also learnt that the industry is expecting continued growth and 'an ever-increasing role of technology and automation' and aims to increase awareness of the value that translations offer to the market (ELIA & al. 2017, 26). Although most companies are small and the average size may even be shrinking, more respondents than last year reported a sales figure of € 250k - € 1m, and the sales figures of almost one third of respondents were even higher. The trend is worldwide: Common Sense Advisory research by DePalma & al. (2016) showed continued growth for language service and technology providers, who are now also offering global content strategies.

3Translation has therefore become a multilingual research topic that has given rise to a flourishing field of translation studies in recent years and is warranted a place in this special issue of the Revue française de Linguistique Appliquée on multilingualism. While it is impossible for this contribution to encompass all aspects of translation that have been covered by the different branches of translation studies, this contribution first briefly presents a linguistic approach that has been adopted in a product-oriented, or text-oriented, area of translation studies. After sketching Toury’s norms of adequacy and acceptability, examples are presented of both recent source text-oriented and target audience-oriented corpus-based issues and studies. These are related to the so-called and highly contested notion of translation universals, and to explicitation, in particular. Secondly, recent inquiries into the translation process will be presented. Some of them investigate the translator’s cognitive activity (in terms of effort) and apply methods of keystroke logging and eye-tracking. Others are to be situated within what has been called the 'technological turn' in Translation Studies (Cronin 2010). They have been inspired by the digital era and technological advances in the translation setting of recent decades that have urged translators to incorporate a range of Computer-Aided Translation (CAT) tools into their daily translation workflow. These CAT-tools, including translation memories, terminological databases, localization software and machine translation plug-ins, have been shown to improve translation efficiency and consistency (Fernández Costales 2009) and optimise translation project management.

2. Multilingualism in the translation product: Linguistic issues

4Translations have been known to reveal themselves as the products of a multilingual activity in various ways. In particular, the influence from the language of the source text on the translation itself has been noted by many translation scholars and has led Toury to stipulate a scientific law of interference as an explanatory hypothesis as follows 'in translation, phenomena pertaining to the make-up of the source text tend to be transferred into the target text' (1995, 72). An example is the Dutch near-word-for-word and translation Zijn pensioen kocht alleen goedkope dingen for His pension only bought cheap things. Such a translation does not fulfill what Toury calls the norm of acceptability. Interference often reveals itself when cognates are translated (NL pensioen and EN pension) especially when they turn out to be false friends (EN preservatives versus FR préservatifs). While the area of lexis is the obvious one to even predict instances of interference, especially with non-professional translators. Whether the latter are translation trainees, fan or volunteering translators, interference can also be detected at the syntactic level with grammatical words such as demonstratives. The English-Spanish P-Actres parallel corpus, for instance, revealed that the use of demonstrative pronouns in translated Spanish texts was different from that in the Corpus compiled by the Spanish Royal Academy CREA (Labrador 2011). In particular, the neutral forms esto, eso and aquello are used more frequently in translated Spanish texts. A typical example would be is this true? when translated into Spanish as ¿es eso verdad?: this literal would sound, however, 'a little awkward or even stilted' (Labrador 2011, 81) than the more neutral ¿es verdad?. At the same time, the distant masculine in the plural and the distant feminine both in the singular and the plural aquella, aquellos and aquellas, are used less frequently than in non-translated Spanish texts

5This kind of influence is traditionally seen by many as undesirable or translationese. Veisbergs (2016) has recently summarised several aspects of translationese: it does not only include linguistic characteristics that do not typically occur in original target language texts and translation solution types that have been identified as instances of universals of translation, such as explicitation, simplification or normalisation, but it also reveals individual traits of the translator him- or herself (Veisbergs 2016, 27). Since translationese is multi-language-specific, however, because it depends both on the languages involved in the translation and on the translation direction from a language X into a language Y, he proposes to see translationese as the collection of 'typical traits of translation language of a specific language pair' (Veisbergs 2016, 28). On this view, translationese is different from 'translatorese', a term which would then be reserved to refer to a translator’s idiolectal impact on the text: his or her individual solutions to translation problems, personal preferences and features of style.

6The traditional linguistic approach to translation with individual case studies about one translator, one translation, with one language pair in one translation direction, however, changed crucially when it introduced corpus studies following Baker (1993). This allowed for broader generalisations, and, in particular, focus was directed at differences between language features in translations produced in a particular language as compared to texts produced originally in that specific language. A recent study by Vermeire (2018), for instance, confirmed the difference in the presence of that in complement clauses between original English and translated English that was already discovered by Olohan (2002). The latter pointed out that the translated English texts in the Translational English Corpus (2018) at the time contained a higher frequency of that in complement clauses than in non-translated texts. Vermeire found similar results about sentences such as He insisted that / Ø I repeat that after him or In the painting you see that / Ø the Tsar was the only blue-blooded guest in the Dutch Parallel Corpus (2018). It showed that in cases where the translator has the option to insert the complementiser that in the English translation of a Dutch sentence, they usually did so, while writers of original text omit it. In Dutch, however, the complementiser in those sentences is obligatory :

7*Hij stond erop ik dat na hem herhaalde.

8 Hij stond erop dat ik dat na hem herhaalde.

9*In het schilderij zie je de tsaar de enige gast van adel was.

10 In het schiderij zie je dat de tsaar de enige gast van adel was.

11By applying the Multifactorial Prediction and Deviation Analysis (MuPDAR) approach (fitting a general linear mixed-effects model for the that instances in original English data and subsequently testing it on those from the translated data), the study revealed a slight tendency towards a higher frequency of the that-complementiser in translated English. In contrastt to expectations, however, this difference could not be explained in terms of interference or priming, but in terms of both the cognitive complexity hypothesis and the risk-aversion hypothesis. Findings like these have, indeed, given rise to various hypotheses that could explain the differences detected in the corpora (see next section).

12The norm of acceptability, however, is not the only one that is basic in translation: translators also aim at producing translations that are adequate, of 'constituting a representation [...] of a text already existing in some other language, belonging to a different culture and occupying a definable position within it' (Toury 1995/2012, 69). In the past, deviations from this norm of adequacy have been called translation shifts (Catford 1965) and translations were scrutinised at various levels of linguistic analysis in the previous century. Such deviations have also been considered undesirable (Xu 2008), but as they are also unavoidable in bridging the gap between two different languages, they are now also referred to as differences between source and target texts (Chesterman 2007).

13Caught between the two languages and their features, translators have been shown in translation studies to constantly navigate between these norms. One of the translation scholars who aimed at describing this multilingual activity tension at the level of the text and its purpose, the so-called skopos (Nord 1997), is Lux’s inquiry into the differences between British and German patient information leaflets (Lux 2016). The different national legislations regarding patient information leaflets (PIL) have led to two different prioritised skopoi of the same text type: she suggested that the British PILs aim at involving the doctor to ensure patient compliance, while the German ones reveal protection of the pharmaceutical industry from liability claims. This also showed in a macro-structural analysis of the text genre.

14In literary translation studies, too, translation solutions for the multilingual tension brought about by a source text have been amply discussed. Not only do such studies describe the textual differences between the two linguistic versions of a text, they also involve the socio-politico-historical contexts of the source texts and the translations in their discussions. In particular, scholars have pointed out the extent to which certain translation choices lead to what could be considered as manipulative translation. Ben-Ari (2002), for example, identified some features that testified of anti-Christian considerations in the Hebrew translations of Lewis Wallace’s Ben-Hur:A tale of the Christ. Usually, translators just omitted undesirable passages, but they also did rewrite the source text into a text that was more acceptable from an ideological point of view.

15Obviously, the tension created by the two norms in the multilingual activity of translation is also a hard obstacle to be overcome in the pedagogical area of translation training. Here, a recent study inquired into translation trainees’ contrastive pragmalinguistic competence that is required in translation. In particular, Castagnoli (2016) showed how the different preferences that languages display with respect to interclausal linkage posed difficulties for trainees. Their English-to-Italian translations of the same source text in a parallel corpus were analysed and compared to a comparable corpus of Italian original texts. Their translation behaviour revealed both regularity in copying the connecting patterns of the source text, but also some variability, where translation learners normalised their translation, that is, they reverted to cohesive patterns that were more in line with the comparable corpus texts.

16While all of the translation scholars mentioned above investigated translated texts and pursued text-oriented issues comparing certain language phenomena in the translations to source texts or comparable texts in the target language, other translation scholars took up the hypotheses and went to search for some explanations of those phenomena in both cognitive and sociological fields, both of which have been found to contribute to knowledge about multilingualism in an individual and in society respectively. The next section will, therefore, first present some of the hypotheses as formulated in the cognitive branch of translation studies, while the following section will describe the recent technological advances and their impact on the translation process.

3. Multilingualism in the translation process of an individual translator: Cognitive studies

17One of the pioneers in cognitive translation studies is Tirkkonen-Condit (2005; 2008), who showed how multilingualism in a translator’s brain can influence the process of translation. In her investigations by means of keyboard logging to capture processing elements empirically and measurably, she detected that almost half of the revision activities that bilingual translators carried out while typing their target text consisted of changing first literal translations. These findings led her to adopt Krashen’s Monitor Hypothesis (1982/2009) in the translation process, which posited a distinction between acquisition and learning of a second language: the former produces the utterances and leads to fluency, while the latter monitors or edits those utterances, which enhances acceptability in that second language. Hence, Tirkkonen-Condit hypothesised that translators for a large part do first translate literally by default, but then monitor their own process by revising their first choices.

18More inquiries into the cognitive aspects of translation were carried out by Halverson, who focused on lexical and syntactic patterns in translated language. She posited the 'gravitational pull hypothesis' (Halverson 2003, 2017), according to which translators will be more inclined to choose those linguistic items – whether lexical or syntactic – that are more salient to them. Given that this type of salience is determined by the translator’s own lexical and syntactic experiences, which are multilingual rather than monolingual, the texts produced will show features that deviate from monolingual text productions. Hence, the overrepresentation of certain items in a translation corpus, such as basic verbs like see, take, or go, can be explained.

19Basing themselves on previous psycholinguistic findings that there is much cross-linguistic overlap and 'influence of a bilingual’s first language (L1) on the processing of a second language (L2) or vice versa', Vandepitte and Hartsuiker (2011) introduced the notion of 'priming' into Translation Studies as a psycholinguistic explanation for translational interference. While priming occurs when something is easily repeated after it has just been encountered, the presence of the source text sentence that triggered the cognitive L2 lexicon and syntax of a translator was seen as having a similar type of influence on the production of translated sentences. In other words, they presented 'priming' as an explanation for the occurrences of literal translations in a translation process, such as the rather unacceptable sentence Muziek nam hem mee rond de wereld as a literal translation for the English Music took him around the world.

20Basing herself on four comparable corpora that were controlled for register, Kruger (in press), in contrast, found evidence against the source-language transfer hypothesis as the main explanation of the translation feature that was studied. After applying a multivariate analysis to her results from an investigation into the frequency rate of the conjunction that in a corpus of translated English from Afrikaans, and from a comparison of the translation data with those in reference corpora of written British English, Afrikaans English and Afrikaans, she found that there was a much higher frequency rate of the complementiser that in translated language than in either the Afrikaans or English comparable corpora. Instead, she concluded that the risk-avoidance behaviour of translators (as hypothesised by Pym (2015)) was the likeliest explanation for the phenomenon, since an inquiry into the complexity of the context of the appearance of that in translated English did not confirm the hypothesis that the more complex contexts would trigger more occurrences of explicit that than in less complex contexts (Wulff & al. 2014): the corpus showed that translators also avoided zero complementation and inserted that in contexts of low complexity. Kruger recognises, however, that the comparable corpus design is less stringent than a parallel corpus design, where the occurrence of that could be compared to the structures in source text utterances and provide stronger evidence regarding the transfer hypothesis. Perhaps more importantly, however, it should be added that confirmation or falsification of a process-oriented explanation like the transfer hypothesis is more likely to be valid if the analysis is carried out on the basis of process data rather than the corpora data that were used by Kruger: the latter will indeed never reveal what the translator actually did before they monitored their very first translation solutions or which additional reviewing was carried out by the editor before publishing the translation. In order to arrive at the explanatory level of research, it will be necessary to collect enough process data.

4. Multilingualism in the translation process in the digital era: Technological studies

21Translators’ cognitive activities, however, do not take place in a vacuum but in a concrete setting. Whether they are working on their own at home, or surrounded by other translators, language experts or employees of a company, the source text and a word processing tool are not the only devices that they employ. To investigate professional translators’ work practices, and, in particular, the role that modern technology nowadays plays, Van den Bergh & al. (2015) used complementary research methods, such as surveys among language professionals, semi-structured interviews with translation companies and contextual inquiries, where translators and revisers were observed while working. Their respondents (N=181) originated from a total of 47 different countries, with a predominance in Europe (75%), and smaller representations in North-America (10%), Asia (7%), South-America (4%) and Africa (4%). During the sessions, the users were interrupted to discuss specific aspects about their workflow. More than 75% of the 181 respondents indicated that they use a translation environment tool (TEnT) in their daily work, with SDL Trados being the most popular tool (71%). They used TEnTs to ensure consistency, save time, increase productivity and improve the general translation quality. While more than 50% of the translators pre-translated the source segments by means of a translation memory (TM), only 10% of them used machine translation (MT). Translation memory systems store translation units (i.e. source and target language utterances in pairs) in a linguistic database, called a translation memory. By accumulating all translations within the translation memory, translators do not have to translate identical or almost identical sentences twice, since their previous translation solution will appear automatically. The screenshot in figure 2 shows the translation editor of the translation software SDL Trados Studio where the left bottom screen lists the segments of the English source text to be translated underneath each other, the middle column shows the percentage of overlap between these source segments and the segments present in the translation memory, and the right column is dedicated to the target translation (in Spanish in this case). The top level window shows the translation suggestions from the database: for segment 4 there is a similar segment in the database, which overlaps for 88% with the segment to be translated, and the differences between the source segments are indicated as well (words that are different are stricken).

Figure 2. Screenshot of translation results int he SDL Trados Studio interface.

Figure 2. Screenshot of translation results int he SDL Trados Studio interface.

22More than half of the language professionals (51%) also included terminology management tools in their work, while only 14% used automatic term extraction functionalities. Reasons mentioned for not managing terminology were the complexity of terminology management theory and tools, perceived lack of added value and its time-consuming nature.

23From this limited study it could be concluded that the use of translation memories was well integrated in the multilingual translation workflow, whereas translators were less familiar with automatic terminology extraction and post-editing of machine translation output. The state-of-the-art and latest developments of the latter two technologies will therefore be elaborated upon in the following sections.

4.1. Terminology management and extraction

24Terminology management is supposed to enhance translators’ consistency and productivity. Current translation environment tools already present many useful features to facilitate a correct and consistent use of terminology. By means of terminology recognition, terms are highlighted in a segment to be translated, and autosuggestions appear while typing. This process is illustrated in Figure 3, showing a source segment containing the term 'Spelling Shecker' (at the bottom of the screen) which is indicater with a red line by Trados, while the corresponding translation suggestion for the term is shown in the right upper screen

Figure 3. Screenshot of term recognition in the SDL Trados Studio interface.

Figure 3. Screenshot of term recognition in the SDL Trados Studio interface.

25These tools can also be integrated with external multilingual terminology databases, such as IATE (InterActive Terminology for Europe) or EuroTermBank. Translators, however, often find the translation suggestions coming from these general term bases useless for very specialised types of text, and they consequently switch off their terminological plugins. To acquire domain-specific terminology, and prepare terminology collections for specific projects, automatic terminology extraction appears to be a more appropriate solution.

26State-of-the-art term extraction tools apply a hybrid methodology combining linguistic and statistical information (Daille 1994; Macken & al. 2013). Large monolingual domain-specific corpora are first linguistically preprocessed to perform automatic tokenisation, lemmatisation and Part-of-Speech tagging. Based on the resulting linguistic information, a list of candidate terms is generated by means of predefined syntactic patterns (e.g. Noun + Noun, Adjective + Noun). As this linguistic approach overgenerates, and considers, for instance, all noun phrases as valid terms, a statistical filtering step is added to remove terms based on the principles of termhood and unithood (Kageura & Umino 1996). Termhood filters measure the domain-specificity of a given term by comparing its relative frequency in the domain-specific corpus with a general reference corpus, assuming that domain-specific terms appear more frequently in specialised monolingual corpora. Unithood filters, in contrast, measure the lexical cohesion between the different parts of multiword terms, in order to remove random sequences of words from the term list and only keep 'real' multiword terms referring to one semantical unit. In a final step, term candidates from the two or more languages involved in a translation are linked by means of statistical word alignment (e.g. Giza++) performed on sentence-aligned bilingual corpora. Figure 4 shows an excerpt of a bilingual terminology list extracted from a medical Dutch-French sentence-aligned corpus

Figure 4. Excerpt of a bilingual Dutch-French terminology list.

Figure 4. Excerpt of a bilingual Dutch-French terminology list.

27Recently, research on bilingual terminology extraction has shifted from parallel corpora (sentence-aligned translations) to comparable corpora (Heyman & al. 2017). Since comparable corpora, on the one hand, are collections of texts in different languages about the same topic without being translations of each other, they pose much difficulty for bilingual term extraction: it is, indeed, impossible to know where to look for term translation equivalents, or even to know whether equivalents are available (Rigouts Terryn & al. 2018). On the other hand, however, it is much easier to compile these corpora for various languages, as they do not require human aligned translations. Whereas they were almost non-existing in the past, interpreter-specific terminology management systems (TMS), too, have become increasingly popular in recent years. The recent systems differ from the more traditional TMS in the sense that they offer advanced functions to sort and filter by subject and to search intuitively in terminology bases while interpreting simultaneously (Rütten 2017). An example of a booth-specific tool is InterpretBank, which assists the interpreter during the various phases of the interpreter’s workflow: interpreters can create and edit glossaries, train vocabulary and carry out sophisticated booth searches for terminology retrieval.

28Despite the existence of these advanced tools, conference interpreters often use more generic tools, such as Google Sheets, to share and update their glossaries. The latter tools lack advanced terminology management functions, but offer easy access to cloud-based glossaries that can be easily consulted and edited. As such, whereas interpreting and its preparation was always considered an individual task, it has now been shifting to a collaborative task, where interpreters share the workload to prepare the glossaries of increasingly more technical conferences (Rütten 2017).

5. Conclusion

34This brief survey of recent studies covering bilingual and multilingual aspects of translation has also revealed some of the high degree of the multidisciplinarity in the field: (text) linguistics, literary studies, cognition studies, market studies, and language technology. However, this is not an exhaustive review article and it should be regarded as illustrative of the field of translation studies only. Nevertheless, it shows how dynamic and rapidly evolving the world of professional translation has become and how translation studies are keeping up in their use of numerous innovative methods and discussions of hypothesis formation.

35Still, many aspects of multilingualism need to be discovered: although the field of language technology and the development of modern communication platforms has enabled the translation market to offer translations of the same source text into different languages in a much shorter period than before, empirical data of the volumes involved in this activity or the challenges posed by such forms of what could be called 'simultaneous multilingual written translation' are, for example, missing. Equally undiscovered is the high number of text volumes that companies and organisations are producing in multiple languages about a similar topic, often by means of translation: here, too, translation studies could explore this rich data field and develop support.