Source First or Target First? Insight into the Order of Reading in Revision Using In-depth Interviews

While revision procedures have been studied in some detail in the literature on revision, the question of the order of reading during bilingual revision has hardly been investigated. This article explores this issue by analysing in-depth interviews conducted in Switzerland with translators in different professional contexts. It discusses the practices described in the interviews and analyses the reasons given by the participants for choosing their method. It shows that both orders of reading are well represented in the dataset: 10 participants read the source first during the bilingual check, 9 read the target first, and only 3 alternate between both orders. No pattern was found according to sector activity or languages, but trends emerge at the level of translation departments. The justifications provided by translators appear very similar: all participants who read the source first stated they seek to better spot discrepancies in meaning, whereas participants who read the target first indicated they want to avoid interferences between the languages or better appreciate the readability or correctness of the target language.


Introduction
Translation revisionor, more simply, revision -can be defined as "revising somebody else's draft translation" (Künzli 2007: 116). It hence refers to the editing of a translation carried out by a human, as opposed to the editing of machine-translation output. Over the last 15 years or so, revision has become a sub-field of Translation Studies in its own right, as shown by the significant development of research in this area (see recent literature reviews: Koponen et al. 2021 andRobert 2018). As such, it raises its own problems and questions. In practice, revision is a common activity in the day-to-day work of translators 1 . However, it can be very different from one context to another, in terms of what is revised and how it is carried out. For example, in multilingual public institutions, e.g., government bodies of official multilingual countries or intergovernmental organisations working with several languages like the EU or the UN, revision may concern every text or only those from junior translators (Drugan 2013: 127-145, Riondel 2021. In the translation industry, the ISO 17100 standard requires that all texts undergo bilingual revision before being delivered to the client (ISO 2015). Even if revision is mainly written and bilingual, other procedures are possible. In some departments, revision is oral (Allain 2010), while some authors favour monolingual revision, i.e., reading of the target text plus ad hoc use of the source text (Mossop 2020, Nord 2018. Revision procedures have been taken up in many descriptive or experimental studies ever since empirical research on revision began flourishing in around 2005. Descriptive studies have been conducted with questionnaires or a combination of questionnaires and interviews. Schnierer (2019: 185-187) pointed out that, in Austrian translation agencies, bilingual revision prevails, sometimes preceded or followed by a monolingual check of the target text, whereas monolingual revision is scarce. For Danish translation agencies, Rasmussen/Schjoldager (2011: 104-105) showed that about two thirds of the respondents (15/22) submit all or almost all translations to a bilingual revision, although the comparison does not always concern the whole text. Scocchera (2017: 13) found that 40% of the Italian literary revisers surveyed read the source text in full or in part before the revision itself, and the exact same proportion do the same with the target text. Prior to this, Morin-Hernández (2009: 137) revealed that 53% of her French participantsa great majority of which are freelancersdo a bilingual revision, 17% do a bilingual check preceded by a reading of the source text, while 16% do a monolingual revision and 2% do a monolingual revision preceded by a reading of the source text. In a double survey of Belgian agencies, Robert (2008: 8-12) found a predominance of two-steps procedures (bilingual revision, followed or preceded by a reading of the target text), and little use of bilingual revision alone or monolingual revision. Experimental studies have sought to determine which procedure leads to the better revised translation. Brunette et al. (2005) compared monolingual and bilingual revision. The most comprehensive study was conducted by Robert (2012Robert ( , 2013Robert ( , 2014 and Robert/Van Waes (2014), who measured the effect of different revision procedures on time, error detection and the quality of the revised translation. Overall, the experimental research indicates that bilingual revision produces better results than monolingual revision.
In real contexts, revision is often carried out in a single step, with bilingual revision (Horguelin/Brunette 1998, Morin-Hernández 2009, Schnierer 2019. During bilingual revision, there are two possible methods: starting by reading either a part of the target text or a part of the source text. The order of reading has been investigated in three evaluative studies and two descriptive studies (one survey and one eye-tracking study), and the topic is also addressed by Mossop (2020) in his handbook. In a qualitative analysis linking up process (using TAP) and product (analysis of the revised translation), Künzli (2006: 13) recommended that revisers start with the target text, since it tends to produce better results. Later on, he conducted a quantitative analysis on the same data set, linking the quality of the output with post-task interview data (Künzli 2009). He found that the revisers who started reading the target performed significantly better in one of the three texts of the experiment. With the same kind of data (product data and post-task interview data), Ipsen/Dam (2016) showed that participants who are likely to detect the highest number of errors in a draft translation all take the target text as a point of departure. As far as theoretical literature is concerned, only Mossop (2020) suggests a modus operandi. He strongly advocates beginning with the target text, arguing that this method enables the reviser to "have a golden opportunity to see the translation from the user's point of view" (Mossop 2020: 176). In a descriptive study that included a questionnaire on revision and post-editing, Girletti (2022) found that there was equal representation for starting with the source and starting with the target among respondents who carry out segment-by-segment bilingual revision (the vast majority). In an eyetracking experiment with student translators, Huang (2018) showed that, depending on the different phases of the revision task, almost all to all participants started by reading the source, but the quality of the outcome was not analysed because of the descriptive nature of the study. It is also worth mentioning another study that uses eye-tracking technology to examine revision. Schaeffer et al. (2019) measured the eye movement of students and professional translators while revising, without any specific reference to the order of reading. They revealed that professionals perform better than students and demonstrate more strategic reading behaviour: they use the source text only when it matters.
The question of reading order also applies in post-editing which "can be defined as editing and correcting machine-translation output" (Koponen/Mossop/Robert/Scocchera 2021: 2). The order of reading has been even less studied in post-editing than in revision. Volkart et al. (forthcoming) analysed the behaviour of translation students when the source or the target segment is shown first. While there is no statistically significant difference in the time spent and the number of errors corrected, there is a difference when it comes to optional modifications, which are greater when students see the source segment first. In a survey, Girletti (2022) showed that a majority of respondents (55%) adopt a segment-by-segment reading with the source as a starting point as their main strategy, while 34% read segment-by-segment with the target as a starting point. Beyond these two very recent studies, eye-tracking has also been used in earlier studies to investigate postediting behaviour, but not reading order itself. In two exploratory studies, Carl et al. (2011: 140) argued that post-editing certainly begins with the reading of a target-language segment, which is then compared with the corresponding source-language segment, while Mesa-Lao (2014: 236) stated that "post-editors generally only refer to the source text after reading the target text and before or after editing the MT output". In both cases, the target segment therefore seems to be the starting point, not the source segment. Finally, post-editing teachers were surveyed about the recommendations they give to their students with regard to the order of reading (Ginovart Cid/Colominas Ventura 2021). Half do not recommend any particular method (49%), while a third favour starting with the source (33%) and about a sixth advise students to start with the target (18%). No conclusions can be drawn from the above since the available results point in different directions.
This overview of the literature gives rise to three observations. First, although revision procedures have resulted in a number of studies, the order of reading in bilingual revision remains largely unknown. Second, research seeking to give recommendations suggests starting by reading the target, even though there is little evidence to support this recommendation; at the same time, two descriptive studies show that starting with the source is common. Third, reading order has been slightly more studied in revision than in post-editing, where no tendency emerges. This article further explores the issue of the order of reading in revision with interview data collected among professional translators working in different contexts in Switzerland. It aims at answering the following research question: what order of reading do translators adopt during bilingual revision and how do they justify their choice?

Method
This article originates from an extensive research project dedicated to revision and translatorreviser communication, for which 45 in-depth, semi-structured interviews were conducted in public multilingual institutions, private companies with in-house translation departments and translation agencies, as well as with freelance translators. The translation departments were chosen to represent the different aspects of specialised translation in Switzerland and to encompass the diversity of revision policies (e.g., horizontal model of revision, in which revision is advisory, with the final say by the translator, and vertical model of revision, in which revisions are corrections; for a discussion of these two models in institutions, see Lafeber 2018: 74-76).
The 22 interviews analysed here come from a larger pool of interviews with translators who revise (n=28) and heads of department who manage revision (n=10). (The original data set also includes 4 translators who do not revise, 2 project managers and 1 proofreader). In 26 interviews, the issue of the order of reading was brought up. Four of these interviews have been left out from this analysis due to ambiguous statements.
The interviews took place between March 2019 and October 2020. Access to the field was first negotiated with the department heads, after which other staff were interviewed (the freelancer was contacted through a professional association). Participants were recruited via email. Within each department, an effort was made to meet participants with the most varied profiles, as a search for diversity is a major factor that boosts validity in qualitative approaches (Flick 2018, Saldaña 2011). All interviews were face-to-face, with two exceptions, which took place via videoconference, due to pandemic restrictions. Interviews were mainly conducted in French (20 out of the 22 analysed here), some were in German (2/22). They were conducted in a mode inspired by the concept of "responsive interviewing", as defined by Rubin and Rubin (2012), which is characterised by the involvement of the interviewer in the conversation and their responsiveness to what the interviewee says. Interviews drew on a general interview guide customised for the situation (see Appendix for an example of an interview guide). They were transcribed verbatim by the researcher according to transcription and spelling conventions developed for the study, which adhere to a denaturalised approach, i.e., full transcription without involuntary vocalisation and with some grammatical corrections (Oliver et al. 2005). The transcripts were then coded with the software QDA Miner Lite, following the "Template Analysis" approach (King 2004(King , 2012. Template Analysis combines deductivity and inductivity: the template, i.e., codebook, takes the interview guide as a starting point, but evolves with the coding. According to King, Template Analysis "works particularly well when the aim is to compare the perspectives of different groups of staff within a specific context" (King 2004: 257).
This study adopts a qualitative approach and is based on interview data, which have benefits and limitations. Qualitative research "is a means for exploring and understanding the meaning individuals or groups ascribe to a social or human problem" (Creswell 2009: 4). As with other qualitative methods, in-depth interviewing helps produce rich, contextual data (Miles et al. 2014, Rubin/Rubin 2012. Here, "contextual" means that data refer to actual situations and are embedded in lengthy discussions. As interviews provide extensive personal accounts, they are useful to understand motivation and justification (the latter being one of the focuses of this article). On the other hand, interview data are indirect. Hence, interviews do not provide information on what people are actually doing, but what they think they are doing, which has its own value. Perceptions and practices are both part of reality, and perceptions provide information on the meaning that practices have for the agents (Olivier de Sardan 2008: 321).
This study received ethical approval from the Ethical Review Board of the Faculty of Translation and Interpreting of the University of Geneva. The participants were given an Information and Consent Form, which they had to sign before the start of the interview. They could choose not to be recorded and their right to withdraw from the study was clearly stated. Data are stored on a server at the University of Geneva and recordings will be deleted when the study has been completed.

Results
The results section is divided into four sub-sections. First, the frequency of the different methods is described for the data set and subgroups. Second, the particular methods adopted by three participants are discussed in more detail. Third, the reasons provided by the interviewees are analysed. Finally, a small subsection is devoted to the notion of influence, which was a recurrent topic in the participants' discourse.
The data set consists of 22 participants: 10 translators from the Swiss Confederation, 4 revisers 2 from an intergovernmental organisation based in Geneva, 6 in-house translators working for two different companies operating in the financial sector, 1 translator in a medium-size Swiss translation agency, and 1 freelancer. 15 participants are French-speaking translators (10 work mainly from German, 5 from English), 5 are German-speaking, and 2 are Italian-speaking.

Frequency of the different methods
Target first and source first are equally represented in the data, while other methods, called hybrid, are scarce (see Table 1). Of the 22 participants, 10 read the source first during the bilingual check, 9 read the target first, while 3 use a hybrid method (see section 3.2 for a discussion of the last category). The results do not differ in the two main sub-categories of the data set: in the institutional sector, 6 participants begin with the target and 5 with the source (plus 3 hybrid), whereas, in the private sector, 3 begin with the target and 5 with the source. Nor are there any fundamental changes from one target language to another: of the 15 French-speaking interviewees, 7 read target first, 5 source first, and 3 use a hybrid method; in the less represented languages, the 2 German-speaking translators use the target as point of departure and 3 use the source, and the 2 Italian-speaking translators read the source first. As far as patterns at the departmental level is concerned, certain tendencies can be observed. The study mostly took place in translation departments, which are designated with the two first letters of a participant's code (e.g., A2, B1, C1, C2). In department B1, the target-first method is predominant (3 target vs. 1 source) and in department C1, source-first is the only method described (3 source vs. 0 target). Even if it may be a sign of patterning, it cannot lead to any definitive conclusions, because of the small number of interviews in these departments and the absence of information on where participants derive their practice from. Department A2 presents a different scenario. Here, data are available for a higher number of participants (n=6) and the order of reading has been discussed among members of the team. On the whole, 3 translators read the source first, 2 read the target first, and 1 alternates between reading target and source. Among the 3 who begin with the source are the head of the department and his deputy. Both deemed this method to be the best way to proceed and said that they recommend it. This department is characterised by a high level of communication and collaboration: frequent questions to colleagues, post-revision talk between translator and reviser, shared coffee breaks, continuous training (including on revision). Given the expressed willingness of the managers to recommend a method and the lively exchanges within the team, it is not surprising that the order of reading was discussed. These discussions seem to have an effect, as demonstrated by the person who alternates between methods. This translator reported that she used to start with the target but changed her practice after a team discussion in which colleagues encouraged her to begin with the source. Now her practice varies: sometimes she starts with the source, sometimes with the target. The two people who indicated that they begin with the target are the two who joined the team most recently. It is possible that the transfer of norms will continue and that they will change their approach over time.
Another conclusion that can be drawn from Table 1 is that translators are inclined to choose one method and use it in all situations, as indicated by the low number of participants who use a hybrid method. This contrasts with the variety of approaches observed for the whole data set, in which both orders of reading are well-represented. Hybrid cases are dealt with in detail in the following sub-section.

Three alternative practices and a natural tendency to start with the target
The situation of the three people who said they alternate between reading the source and the target first is worth examining, because they can be regarded as extreme cases, which are of great significance in qualitative research. Analysing extreme cases provides a means of giving a full account of the phenomenon under study or, to put it in Becker's words "to refine the portrait of the whole-to offer, in the end, a convincing representation of its complexity and diversity" (Becker 1998: 284).
The first person who varies her practice has already been mentioned in the previous sub-section. While she used to read the target first, she now alternates as a result of team discussions. Although she did not specify the criteria for choosing between the two methods, she said that, in her opinion, when starting with the target, the translation is less put into question, whereas when starting with the source, one compares it with one's mental translation and tends to intervene more in the draft translation.
The second person indicated that he has no preference between the two orders of reading, so he alternates following an a-b-b-a-a pattern: as long as he does not encounter any problems, he reads the source and verifies it with the target; he then continues with the target text and reads a new sentence, then checks it against the source, and so on. In his view, this method is automatic, and allows him to go faster and is less tiring for his eyes. He described the texts he has to revise as high quality, which may be why he uses this particular method, which is certainly most feasible when the reviser only has to occasionally stop reading to edit the text.
The third person indicated that her approach differs according to her confidence in the translator's ability to do a good job, in particular their understanding of the source language and ability to work meticulously. When she has some doubts, she reads the source first, then the target, and possibly the source again. When she thinks the translator has done a good job, she reads the target first, then the source. She added that she spontaneously reads the target first.
That person was not the only one to mention a natural tendency to read the target first. At the level of the whole data set, two people who begin with the source mentioned this tendency as well. The first said that she intuitively or automatically tends to look at the target first but refrains from doing so because she believes it is better to start by reading the source. The second participant stated several times that his standard practice is to read the source first. Yet, at one point in the interview, he indicated that when he is absorbed with the text and the translation is good, he sometimes switches for a while and spends some time on the target, before verifying the transfer with the source text. These statements suggest that reading the target first might be more natural.

Reasons for choosing to read the target or the source first
This section describes the justifications given by participants who said they have a preference between target and source. Table 2 shows the breakdown of the rationale by reading order. Of the 19 participants who said they began with the target or the source, 7 did not provide any justification. The high proportion of lack of justification originates in the method used (semistructured interviews). Participants where not specifically asked to justify their revision procedure and the content analysed here is drawn from long answers in which participants described how they proceed in revision in general. The missing data are not problematic since the analysis aims to develop a deeper understanding of the order of reading in revision, rather than estimate proportions. Of the 12 for whom there is a justification, 8 begin with the target and 4 with the source.

Participant
Reading order Rationale A1f05 Target Interference A2f07 Target None A2f08 Target Interference + Readability B1f09 Target Readability + being in the reader's position B1f11 Target Readability B1f12 Target Readability C2d02 Target Interference C2d03 Target Interference Ef01 Target Correctness A2f03 Source Meaning A2f06 Source None A2fR Source Meaning A5iR Source None B1f10 Source None C1d02 Source None C1d03 Source None C1f01 Source Meaning C2i01 Source Meaning D2d01 Source None The justifications given by those who begin by reading the target can be divided into two categories. A first group of participants seem to be particularly attentive to interferences between languages. One translator simply said that he starts with the target "in order not to be influenced [by the source text]" 3 . Another indicated that she starts with the target "because [she] know [s] there are interferences between the two" 4 . A third said that she starts with the target to detach herself from the source language, so that she can evaluate the target sentence independently, because if she looks at the source, she is "so focused on the structure and then can no longer, like, switch to another language" 5 . A second group of participants pay special attention to the readability of the target text and the correctness of the language. One stated that she starts with the target and tends to "stop at what seems to be poorly worded or strange" 6 . Another participant simply said that he "tend[s] to read the translation first, to see if it reads well" 7 before saying that he "obviously" compares it with the source, afterwards. A third said that she "really need[s] to be immersed in French [her target language] to see if the sentence is correct in French" 8 .
Another interviewee linked the issue of idiomaticity to the position of the reviser. She said that she starts with the target to see if there is "something [that] doesn't sound quite right", adding that sometimes "as a reader, it isn't working for me" 9 . She emphasised that "the reviser is actually a reader in the first place". 10 This participant's effort to put herself in the reader's shoes echoes some of the literature. Mossop (2020) encourages revisers to put themselves in the position of the final reader and Brunette (2007) puts great emphasis on the reviser as the first reader of a translation, as the reviser has more distance from the text and works with bigger units of meaning than the translator.
Interferences between languages and the fluidity of the translated text are discussed in the following quote from an interview with an experienced translator working for the Swiss Confederation, who mainly translates from German into French: The second part of the quotation makes it clear that concerns about source language interference are linked to the readability of the target text. To her, it is easier to assess the readability of a text and, hence, to avoid interferences when reading the target first. Second, having fresh eyes relates to the idea of aiming to distance oneself from the source text, and, in doing so, getting closer to the position of the final reader. Finally, she says that she starts with the target in order not to be influenced, a point that will be addressed in section 3.4.
The four interviewees who read the source first and justified their practice substantiate it on the grounds of meaning transfer. One translator stated that she reads the source first "in order not to be influenced by possible transfer errors or the like" 12 . A second said that she always starts with the source, because "otherwise you start to be influenced by the translation", further stating that she "really want[s] to see [things] from the point of view of the source language, and then compare [them]" 13 . A third commented that starting with the source allowed her to see "right away if the translator has understood the sentence in the same way", pointing out that she "like[s] to know what it was about before seeing the translation" 14 .
The following quote illustrates the view that it is easier to detect possible discrepancies in meaning by reading the source first. The interviewee is the head of department A2, mentioned in section 3.1, where most translations are German-French. He compares revision with the scenario of being in a corridor with different doors (each representing different interpretations), and checking if the translator has selected the correct door, i.e., the one that conveys the meaning of the source text: some read the French before the German and I'm in the camp that says you may go faster, but you get a bit lulled to sleep by a well-written text in French. And when you're in a corridor of 20 doors and it's just next door, it's very challenging... it's more challenging to see when you've read the French beforehand. Whereas when you read the German, without any other influence... because we're in German-French... you read the source and you understand, you stage the text, you put it into perspective. Then, when it's just next door, it jumps out at you 15 .
According to the interviewee, reading the target first goes faster, but it presents a higher risk of being fooled by the draft translation, which is why he advocates for reading the source first. In this quote, the notion of influence is again evoked. This recurrent topic is addressed in the next section.

The influence of the first version that is read: a recurrent theme
The concept of influence is recurrent in the rationale put forward by translators to justify the way they revise. It is present in the two long quotes analysed in the previous section, as well as in the narratives of four other participants. The argument of not being influenced is present in both groups of translatorsthose who start with the target and those who start with the source. This might seem contradictory but is not if one considers what one is trying not to be influenced by. When starting with the target, it is a matter of not being influenced by the source text, i.e., to avoid interference and to examine the wording without any reference to the source language. When starting with the source, it is about not being influenced by the translation, i.e., starting by understanding the meaning of the text to be translated before comparing it with the meaning of the draft translation. As bilingual revision must be grounded in both source and target text, it would be irrelevant to try to free oneself entirely from the influence of one of the texts. In other words, the influence of both texts must be accepted, although one of them is initially set aside. It might therefore be better to refrain from using the term "influence".

Discussion
This section discusses the respective proportions of the two main methods (source first and target first), the low number of hybrid methods, and the justification given by participants for choosing one of the two orders of reading.
As shown in section 3.1, both orders of reading are well represented in the data set: 10 participants read the source first, while 9 read the target first. In other words, a high proportion of translators start by reading the source text during bilingual revision. This contrasts with the literature: the most quoted textbook on revision firmly advocates for starting with the target (Mossop 2020: 176-177), and there is empirical evidence suggesting that this procedure brings better results (Ipsen/Dam 2016, Künzli 2009). Nevertheless, it must be stressed that empirical evidence is scarce. In the one experiment, participants were students (Ipsen/Dam 2016), andSchaeffer et al. (2019) convincingly proved that professionals do not behave the same way as students during revision. In the other experiment, no significant effect on the quality of the final text was found for two of the three texts (Künzli 2009). The results of the present study are more in line with Girletti (2022), who found that starting with the target is as common as starting with the source, and Huang (2018), who put forward that, depending on the different phases of the revision task, very few to no participants started by reading the target. Regarding the latter, the conditions of the experiment must be taken into account when analysing the results: participants were students, the texts were very short, and the reading behaviour was not linked to the quality of the outcome.
All in all, there seems to be a gap between the recommendation in the literatureread target firstand the practices reported (this study and Girletti 2022) or observed (Huang 2018), where starting with the source seems to be more common. This gap is difficult to explain. It may lie in the validity of the results: Huang (2018) and Ipsen/Dam (2016) studied students' performance, and the present study and Girletti (2022) are based on indirect data. Until further evidence is available, it must be acknowledged that (i) translators use both orders of reading and (ii) the method that brings the best results remains unclear, either in absolute terms or in certain conditions, e.g., for different kind of texts, with different time limits or with different draft-translation qualities.
Another striking result of this study is that very few participants use both methods. Only three participants alternate between the two reading orders, compared to the 19 who stated that they only use one. This stability in the order of reading could suggest that translators consider that one method is preferable (some said so expressly). This result is in line with Künzli (2009: 300), who argued that systematic reading behaviour may have a greater effect than the order of reading.
As far as justifications for starting with the target or source are concerned, they are both very disparate from one group to the other, and very similar within the two groups. The main reasons for starting with the target are to avoid interferences between languages and better assess the readability of the target text. The secondary reasons are to verify the correctness of the language and put oneself in the reader's shoes. As for starting with the source, the only reason put forward by participants is to better spot transfer errors. This uniformity must be tempered by the low number of participants who explained why they read the source first (n=4). These reasons can be considered to be quite predictable. It seems logical that reading the target first makes it easier to assess the elements specific to the target language (readability, correctness), whereas reading the source enables a better comparison of the meaning of both texts. The same was argued almost 40 years ago in an old institutional resource (Secrétariat d'Etat du Canada 1985), which recommends starting by reading the whole source or target text but leaves the reviser free to choose which method to apply. The advantages of each procedure listed in the guide are quite similar to those mentioned by the participants of this study: reading the source first helps the reviser gain their own insight into the text, while reading the target first helps the reviser correct language errors as well as spot the ambiguities, incoherencies, and obscure, incomprehensible and unnatural passages. The stability of arguments over time is noteworthy, as is the correspondence between reasons given in a Canadian practical guide and those expressed by translation professionals in Switzerland. These three elementsstability over time, concordance between guide and data, and similarity across countriesare advances that can only be identified by conducting an empirical study like the one described here. The predictability of arguments also points to another result: the subconceptualisation of revision, which the unsurprising nature of arguments would seem to support.

Conclusion
In this article, the reading behaviour during revision of translators working in different translation sectors in Switzerland (n=22) was analysed. The study shows that the two orders of readingsource first and target firstare well represented (source=10, target=9). This contrasts with the literature, which recommends starting by reading a part of the target text, although the recommendation is empirically not very well founded (see previous section). The results of this study can be viewed as putting into question this recommendation or calling for it to be clarified through further research. At the same time, only three participants use both orders of reading, which indicates that the great majority of translators have their routine and behave the same way in all situations. In the data set as a whole, no particular pattern is found in terms of sector activity or languages. At the departmental level, trends seem to emerge, especially in one department, where the head and his deputy advocate for starting by reading the source. The justifications for choosing one method over another do not reveal any major surprises: participants who read the target first said that they wish to avoid interference between the languages or better appreciate the readability or correctness of the target language, while those who read the source first do so in order to better spot discrepancies in meaning. However, the uniformity in the discourse of the two camps is striking: among those who justified using the target language first, all participants referred to interference, readability, or correctness. This uniformity is even greater in the source group, where all participants talked about assessing meaning. The predictability of the reasons could be a sign that the revision has not been sufficiently reflected upon. Another element of homogeneity is the idea of influence, which is present in both groups: for those who start with the target, it is about not being influenced by the formulations of the source language, and for those who begin with the source, it is about not being influenced by the translator's interpretation. As the reviser should be influenced by both texts (source and target), it is advisable not to use this concept. The final result is that reading the target first seemed more natural to several participants, even for some who start by reading the source.
The main limitation of this study lies in the component that deals with practices, which is reliant on the statements of participants. Interviews can only provide access to claimed practices, and it is quite likely that the observation of actual practices would produce other results, as people usually do not behave in (exactly) the same way they claim to. One main difference might be that actual reading behaviours may be more complex than they appear to be in interviews. Participants may have focused on their typical way of behaving and, in doing so, put aside secondary behaviours or micro-variations. Eye-tracking studies can reveal behavioural patterns at the micro level, while long-term observational studies in authentic setting, e.g., ethnographic studies, may reveal whether or not revisers always behave in the same way, as they state in this study, or whether they alternate more than they claim to.
This analysis opens up potential avenues of research. The first is to find out whether one of the two reading orders is preferable (i) in absolute terms or (ii) in certain conditions, for example in different time conditions or for different types of text (where importance of e.g., meaning transfer or smoothness have different weight). Research of this type would be very much in line with studies that have tried to establish which revision procedure produces the best revised translation (Brunette et al. 2005, Ipsen/Dam 2016, Künzli 2009, Robert 2012, Robert/Van Waes 2014. More specifically, it would be interesting to test the hypotheses that can be derived from the justifications put forward by the participants: (i) starting with the source leads to a better detection of errors in meaning, and (ii) starting with the target leads to a better identification of interferences and language errors. Furthermore, other statements from the participants could be investigated, i.e., whether starting with the target is faster or whether starting with the source leads to more intervention in the text (which is one of the results of Volkart et al. (forthcoming) in their study of post-editing). All these goals could be achieved by carrying out an analysis of revised translations produced in an experimental setting that restricts the order of reading, then calculating the proportion of errors corrected in the different categories (meaning, smoothness, correctness…), the time spent, and the total number of interventions. This would mean conducting a study for reading order that is comparable to the one conducted by Robert (2012Robert ( , 2013Robert ( , 2014 and Robert/Van Waes (2014) on revision procedures.