Self-repair as a norm-related strategy in simultaneous interpreting and its implications for gendered approaches to interpreting


 This paper analyses a possible gendered manifestation of norms in interpreting. It focuses on the use of self-repair, a textual expression of the norm, by male and female interpreters. Two research questions are examined: (1) whether the extent to which self-repairs occur in interpreting is gendered and (2) whether gender influences the way in which the output is repaired using editing terms. Considering the literature on gender and norm-compliance, female interpreters are expected to produce more self-repairs and editing terms than male interpreters. The research is based on the 2008 subcorpus of EPICG with French source speeches and their English and Dutch interpretations. The interpreters’ self-repairs were manually identified and statistically compared. Regarding the first question, it appears that gender influences the use of self-repairs in interpreting. As for the second one, statistical analysis reveals language-based patterns: in the English booth, women use significantly more editing terms than men. The French/Dutch subcorpus yields no significant difference. However, women seem to also use apologies as editing terms.


Introduction
This paper is part of a broader research project on gender differences in simultaneous interpreting and focuses on a possible gender-based approach towards norms. Research on quality in interpreting has shown that interpreters hold fairly consistent views on what good quality interpretation is, but that gender and age effects are nonetheless observable in survey data (Pöchhacker and Zwischenberger 2010). Perception of quality is determined by norms and so is the interpreting activity itself (Schlesinger 1989;Harris 1990;Schjoldager 1995;Garzone 2002). The question that thus arises is whether norm-adherence is mediated by gender, and whether such differences have an effect on interpreters' performance.
Experimental and corpus-based research has produced some evidence of gender patterns in interpreters' performance (Cecot 2001;Defrancq 2016, 2017;Russo 2016;2018), but it is unclear if these patterns manifest different attitudes towards norms or different norm prioritizations. Magnifico and Defrancq (2017) suggest that the tendency of female interpreters to translate face-threatening acts more straightforwardly than male interpreters could be due to stricter compliance with interpreting norms, but they do not explore the matter further.
In this paper, we endeavour to find ways to investigate norm-related properties of simultaneous interpreting and to find out whether these manifest gendered tendencies. We do so on the basis of corpus data, as these offer us the opportunity to study norm-related properties in authentic contexts, in which norm-related properties are obviously most likely to surface. By doing so, we wish to contribute to research focusing on factors that influence interpreters' performance and, in particular, to research into interpreting norms. Norms are an important area of study in interpreting, but there is a clear lack of empirical studies operationalizing the effect of norms on interpreters' performances. Much of this paper is therefore devoted to the question of which observable properties of an interpreter's performance can be associated with norm awareness, and we argue that repairs are textual manifestations of norm compliance.
Our study also contributes to the field of gender studies, broadly, and gendered language production, in particular. Research has been carried out on gender patterns in norm compliance, including and prominently in the field of linguistics (Labov 1966(Labov , 1990. With this study we aim to add a new perspective, namely, norm compliance in an extremely challenging linguistic environment by highly trained language professionals who are acutely aware of the norms they are supposed to uphold, but face working conditions that make norm compliance extremely difficult.

Norms in interpreting
Early works on norms in interpreting date from the late 1980s, and the first studies conducted on this subject (Schlesinger 1989;Harris 1990;Schjoldager 1995;Garzone 2002) are inspired by Toury's leading work on translational norms. Toury (1980Toury ( , 1995 studies norms in the context of translation, and states that translation is a norm-governed activity as certain regularities are found in the behaviour of translators. The very act of translating implies that the translator is faced with an initial norm, that is, s/he has to choose between the norm system of the source culture or that of the target culture. Choosing the first norm system (subscribing to source norms) will lead to an adequate translation, whereas following the latter (conforming to the target norms) will determine the acceptability of the translationwhere more shifts away from the source text occur. In addition to this initial norm, Toury (1995,58) observes that the norms applying to translation can be subdivided into two further groups: preliminary norms, mainly induced by translation policy, and operational norms, which influence the decisions made during the translation activity itself.
Toury also states that research into translation norms is based on two major sources, textual sources and extratextual sources: (1) textual: the translated texts themselves, for all kinds of norms, as well as analytical inventories of translations (i.e., 'virtual' texts'), for various preliminary norms; (2) extratextual: semi-theoretical or critical formulations, such as prescriptive 'theories' of translation, statements made by translators, editors, publishers, and other persons involved in or connected with the activity, critical appraisals of individual translations, or the activity of a translator or 'school' for translators, and so forth. (65) Toury's distinction between textual and extratextual norms is interesting as it helps to identify and classify the various norms at work in translation. It would be worthwhile to examine whether the same types of norms can be identified in interpreting.
As translation and interpreting are often seen as two sides of the same coin (Pöchhacker 1995), it is plausible that research into interpreting norms may also draw from textual and extratextual sources. We will therefore classify previous studies on norms in interpreting in two categories: scholars adopting a textual approach, focusing on text production to scrutinize interpreters' work and draw conclusions about interpreters' translating strategies and behaviour in different situations (Barik 1971;Seleskovitch 1975;Lederer 1981;Jansen 1992;Wadensjö 1992;Schjoldager 1995;Diriker 2004;Monacelli 2009); and scholars taking an extratextual approach, formulating statements on interpreting based on elements other than the interpreter's output (Schlesinger 1989;Harris 1990;Gile 1998;Marzocchi 2005;Duflou 2014).
The first interpreting scholars studying norms by means of a textual approach (Barik 1971;Seleskovitch 1975;Lederer 1981) are mainly concerned with the translational norms of adequacy and acceptability. Barik (1971) studies interpreters' deviations from the source text and classifies these deviations as errors depending on the seriousness of the deviation. Seleskovitch (1975) collects texts from an experiment with 13 AIIC interpreters, stressing that interpreters do not translate word-for-word but have to understand the meaning before being able to render it in the other language, implying thereby that interpreters should produce an 'acceptable' translation. This comparative analysis of source and target texts, in the framework of the so-called théorie du sens, can also be found in Lederer (1981).
Interestingly, more recent textual research on norms, while still comparing shifts between the source and target text, tends to contextualize interpreting norms: specific settings have an impact on the way interpreters comply with translational norms. Jansen (1992) demonstrates how an interpreter working in a Dutch courtroom adds elements to accommodate the needs of the defendant. In the same vein, Wadensjö (1992, 74-79) notices that interpreters working in medical or police settings tend to expand their renditions in order to disambiguate. Schjoldager (1995) explores the concept of translational norms in interpreting and assumes that interpreting is a norm-governed activity where norms "help interpreters select appropriate solutions to the problems they meet" (67). She concludes from her empirical investigation carried out on student and professional translators and interpreters that translational norms, such as adequacy, are visible in the interpreting process but that they are sometimes superseded by norms peculiar to simultaneous interpreting, for instance, in the case of capacity saturation. Diriker (2004) can also be situated in the textual approach. She transcribes the performance of two conference interpreters and analyzes shifts in the speaking subject, namely, occurrences where the interpreter infringes the 'norm' requiring that interpreters speak in the first person singular (Harris 1990). She records 58 shifts taking place at specific moments, especially in cases of apologies or mistakes. Monacelli (2009) examines whether interpreters deviate from the conduit norm (i.e., translating what is said) in order to preserve face and thereby shows that different types of norms are at work in simultaneous interpreting.
Surprisingly, the textual approach to norms in interpreting relies heavily on the views of the researchers carrying out the analyses: with a few exceptions, such as Gile (1998) who involves the interpreters themselves in identifying errors, the researchers determine what is an error or a deviation from the norm. This study sets out to avoid this kind of observer's paradox, as explained in Section 4.
The extratextual approach towards interpreting norms follows the same developmental pattern as the textual one. Earlier research in this area is concerned with translational norms (Schlesinger 1989;Harris 1990), while more recent research contextualizes interpreting. As Gile (1998, 100) advocates, research on interpreting norms should be based on 'extratextual sources', i.e. by asking interpreters about norms, by reading […] texts about interpreting, by analysing users' responses and by asking interpreters and noninterpreters to assess target texts and to comment on their fidelity and other characteristics using small corpora. Schlesinger (1989) examines the possibility of extending the notion of translational norms to interpretation. She argues that the study of translational norms in interpreting would first require solving a number of methodological problems, such as the absence of a textual corpus. Besides, she points out that interpreters, unlike translators, only observe a few colleagues at work, thus limiting the potential for norm diffusion. On the other hand, Shlesinger also admits that most interpreters are trained in a limited number of institutions where norms are drilled into them, meaning that most of them follow the same norms. In response to Schlesinger, Harris (1990, 115-118) takes a very prescriptive approach and enumerates norms by which interpreters have to abide, such as the interpreter speaking in the first person or being an "honest spokesperson," meaning that s/he has to translate what is said as accurately as possible. However, it seems that interpreters sometimes deviate from translational norms and the norm of the "honest spokesperson" to adapt their speech to a specific setting (Jansen 1992;Wadensjö 1992;Diriker 2004;Monacelli 2009). These findings could be explained with reference to Skopos theory. Reiss and Vermeer (1984) state that the objective of a translation is first and foremost to fulfill the intended function of the source text in the target culture. Drawing on this theory, Pöchhacker (1995, 36) argues that the skopos in simultaneous interpreting is situated at the level of the conference assignment and that different types of meetings could lead to different interpreting approaches. In other words, the varying needs and expectations of the different user groups could impact the interpreting approach. Garzone (2002) even considers users' expectations as "norm-based, derived from a system of beliefs, partly modified by experience, about what a good interpretation ought to be like." Marzocchi (2005) advocates a "thicker" notion of norms in interpreting which is linked to the ethical discourse in the different settings and which emerges from interpreters' perception of the needs of a specific setting.
Studies in the field of quality in interpreting (Bühler 1986;Kopczynski 1994;Kurz and Pöchhacker 1995;Moser 1995Kurz 2000) have shed some light on the expectations of the different actors in the interpreting process. These studies were mainly conducted using surveys and questionnaires, and aimed at obtaining a ranking of the most important criteria for each group. Bühler (1986) lists fifteen criteria affecting the quality of simultaneous interpretingsuch as sense consistency, logical cohesion, voice, and accentand asks AIIC interpreters to assess their importance. Most of the interpreters ranked 'sense consistency with original message' and 'logical cohesion of utterance' as a top priority while 'native accent' and 'pleasant voice' were at the bottom of the list. Nonetheless, interpreters still gave these criteria a higher score than the end users taking part in the study. The results from a survey of speakers and participants at international conferences (Kopczynski 1994) demonstrate that speakers value the exact rendition of their speech whereas listeners are sensitive to incorrect grammar and unfinished sentences. In the field of TV interpreting, Kurz and Pöchhacker (1995) examine a group of Austrian and German television representatives and observe that in this area, respondents are particularly concerned with criteria such as voice, accent and fluent delivery. Based on 201 interviews with speakers and listeners, Moser (1995,1996) finds, for example, that synchronicity (i.e., avoiding long pauses and lagging) is expected both by speakers and listeners, and that terminological accuracy is considered more important in technical meetings than in general ones. In a later study, Kurz (2000) investigates whether interpreters are aware of their users' expectations, and whether they take them into account. She compares the rankings obtained from interpreters in Bühler's study with the rankings obtained from delegates, as users of interpreting, and concludes that interpreters have higher expectations of their performances than their users. Straniero Sergio (2001, 2003) note that in the field of media interpreting users' expectations have evolved, inducing new norms and forcing interpreters to adapt their behaviour and interpreting strategies to meet the needs of the audience. Straniero Sergio (2003) further observes that the quality of an interpreter's performance in the setting of media interpreting depends on the working conditions and that the norm upon which interpreters are judged is not how correctly but how convincingly they render a speech.
Finally, interpreting norms can be identified through an extratextual approach based on ethnographic data. Based on documentary evidence, interview data and field notes collected during a four-year ethnographic study, Duflou (2016), for instance, studies the process which beginning interpreters have to go through to become EU professional interpreters. Newcomers do not have to internalize norms through explicit instruction, but rather acquire these norms through situated learning within the EU specific working context, by actively taking part in the community of practice, and by reificating the shared practical experience, such as rules or procedures, but also words and expressions, or even a "telling glance or a long silence" (Wenger 1998). She concludes that interpreters not only have to master the necessary cognitive skills, but also have to acquire social and embodied skills to become fully fledged members of the community of practice. In her view, norms are thus also acquired through practice.
Before discussing which norm we intend to observe and how we could operationalize it for a corpus-based study, we would like to address the interaction between gender and norms. As gender differences have already been noticed in the broader field of language (Lakoff 1975;Labov 1990;Tannen 1990;Coates 1993;Holmes 1995;Chambers and Trudgill 1998) and in some aspects of the interpreting activity (Mason 2008;Defrancq 2016, 2017;Russo 2018), the question arises whether gender also influences the perception of and the compliance with norms in interpreting.

Gender and norms
Gender is traditionally viewed as an identity constructed through a socializing process (Lakoff 1975) and through stylized repetition of acts (Butler 1988). This perspective on gender has recently shifted towards a construct which takes other variables such as education, age, sexual orientation, class and race into account (Mills 2003). Gender can also be seen as a determining factor in norm patterns. According to the social role theory (Eagly, Wood, and Diekman 2000), gender roles imply a distribution of men and women's activities based on shared expectations. Thus, social roles can be defined as the activities which men and women are expected to fulfill in a society. These social roles may underpin the norms which determine the behaviour valued for men and women, respectively. In other words, the values associated with a particular gender will lead the individuals identifying with that gender to act in a particular way. In a study on pain tolerance, Pool et al. (2007) find that men typically tolerate more pain in experimental settings than women. They conclude that participants conform to different gender norms, which have an impact on their behaviour: both male and female participants know the social norms and identify with the ideal of men tolerating more pain than women.
In studies of spontaneous language, gender is widely addressed through various lenses, such as gender identity, gender construction and negotiation through discourse, floor management, and power strategies (Bergvall, Bing and Freed 1996;Bergvall, Bing, and Freed 2011). For our purposes, we will focus on the normative aspect, that is, how gendered sensitivity towards norms has been studied from a linguistic perspective. Labov (1966) states that the prestigious variant of a language (i.e., the variant chosen and used by the wealthy upper-class) is considered as the norm to follow in a community sharing the same historical and linguistic background. He finds that women tend to use it more than men. In the same vein, Trudgill (1972) observes that men in Norwich, England, valued the less prestigious variant, concluding that men made a conscious choice for 'covert prestige'. In a later study, Chambers and Trudgill (1998, 84-85) show that women demonstrate a greater range of linguistic variation than men in societies with sharply determined gender roles. It therefore seems that gender influences linguistic behaviour. Women develop sensitivity towards linguistic norms at an early stage and outperform men in applying these norms: in a study conducted on gender differences in spelling achievement, Ruel (1990) finds that girls in grades 1 to 6 outperform boys in written spelling tests.
On the other hand, women are also found to be the leaders of linguistic innovation, which Labov (2001) describes as the Gender Paradox: although women tend to use more standard variants than men, they are also more inclined to adopt new language variants faster than men do. Eckert and McConnell-Ginet (2003) observe that linguistic resources can be viewed as markers of a social category. In this way, the use of standard grammarwhich women are found to use more than menis also linked to other parameters such as refinement or obedience. For instance, the use of nonstandard grammar at school is seen as rebellious. In this respect, the relationship between gender and normativity is multilayered, and linguistic innovation could be seen as a norm infringement.
As interpreting a specific type of language production (Gile 1995), interpreting it is likely to reflect the same mediating effect of gender on norm adherence. In some of the more recent studies on norms in interpreting, gender-based patterns are highlighted. Ng (1992), for instance, observes that women listening to student interpreters underlined the importance of correct grammatical structures and speech levels (i.e., social relationship indicators such as terms of address), whereas men focused on the interpreter's lexical choices and overall fluency. On the other hand,  finds that women rank terminological accuracy higher than men. Based on a survey submitted to interpreters, Pöchhacker and Zwischenberger (2010) conclude that female interpreters are more generous judges than male interpreters, as they give more favourable ratings in their assessment of audio samples. It also appears that female interpreters prioritize a lively intonation and find accuracy more important than male interpreters.
A textual approach to gendered norm compliance is adopted by Mason (2008), who studies errors and omissions in a corpus of authentic court interpretations and their relation to increasing turn length. He concludes that male and female interpreters make similar numbers of errors, but the errors they make are different. Female interpreters omit more items of deference and add more politeness markers than men, while male interpreters more frequently omit the discourse marker well and politeness markers. Mason mainly adduces norm-related factors to explain the differences: males are more drawn to patterns that express social hierarchies, such as items of deference, while females stress group solidarity, for instance, through the use of politeness markers. Mason's approach is interesting, because it lays bare possible hierarchies of norms: female interpreters, for instance, giving priority to the social norm of solidarity over the translation norm of accuracy. However, the observer's paradox is manifest: the selection of evidence in the shape of errors and omissions reflects the analyst's assumptions about norms in interpreting, not the interpreters'. It may very well be that the omission of a deference item or the addition of a politeness marker is not felt to be norm-breaching by interpreters. As Schjoldager (1995) and Chiaro and Nocella (2004) put it, errors could actually be interpreting strategies in the mind of the interpreter. The aforementioned literature assumes that interpreting is an activity subject to various types of norms. These norms are acquired at different stages: translational norms are internalized during training (Harris 1990;Schjoldager 1995;Toury 1995), while other professional norms are learned in authentic settings (Duflou 2016). Other norms derived from expectations seem to depend on other factors such as gender (Ng 1992;Moser-Mercer 1996;Pöchhacker and Zwischenberger 2010). In this respect, a largely unanswered question is what role gender plays when interpreters have to choose between conflicting norms. Gender is constructed through a socialization process, which leads men and women to favour different linguistic and sociolinguistic norms (Labov 1966;Trudgill 1972;Ruel 1990). Consequently, we can expect gender to affect interpreting behaviour.

Interpreting norms and corpora
It appears that most studies on norms in interpreting are based on interviews, questionnaires and anthropological research. These are useful to inform us about interpreters' declarative knowledge of norms, but they are of little help in revealing norm-based patterns in interpretations. Textual approaches, on the other hand, are suited to identify such patterns.
However, the textual studies carried out in the field of interpreting studies remain limited to comparisons of source and target texts, focusing mainly on errors or shifts between the source and the target text. In other words, the patterns revealed are based on the researcher's belief of what is (in)correct, while the interpreter's consideration of what norms to prioritize in the heat of the moment remains unclear. The present study aims to take a textual approach to reveal norm-based patterns, but intends to do so from the interpreter's perspective to understand which norms they apply when working in the booth.
To avoid the observer's paradox, we do not focus on errors as such, but on textual evidence of errors that are corrected by interpreters while interpreting, or, in other words, on 'self-repairs'. The occurrence of a self-repair is proof that the interpreter him/herselfand not the observerjudges the output to be inadequate with respect to a particular norm. We therefore argue that repairs are evidence of norm-driven behaviour in interpreters and can be used to analyse norm compliance across genders.
Considering that many studies of self-repairs in spontaneous language and in interpreting draw on Levelt (1983), we will adopt Levelt's point of view that self-repairs are a speaker's corrections of their own output without external stimulus. Self-repairs typically occur in three stages: (a) the speaker's utterance (reparandum), (b) the interruption of the flow of speech, with or without an editing term and (c) the repair proper, that is, the new utterance. In this context the term 'editing term' is to be understood as linguistic cue uttered by the speaker to signal the listener that s/he edits his/her sentence, and which is generally characterized by hesitation or pausing. The following example taken from our corpus illustrates the process.
(1) visible support and I'm talking here about political support but very concrete repo/ euh support to people on the spot The interpreter utters the sentence 'very concrete repo' (a), then interrupts the flow of speech with the editing term 'euh' (b), and finally repairs the utterance with 'support' (c).
Levelt argues that speakers continuously monitor their inner and overt speech and have the ability to repair utterances when the monitoring system detects either discrepancies with the communicative intention (the intended message) or a failure to meet "criteria or standards of production" (Levelt 1983, 50). The former cases are called "appropriateness repairs" (A-repairs) and "difference repairs" (D-repairs), while the latter are called "error repairs" (E-repairs). It is important to note that 'criteria' and 'standards' being synonyms of 'norms', the latter motivation for self-repairs is thus clearly stated in terms of norms (see also Postma and Kolk 1993;Postma 2000). Based on Levelt's illustrations, these norms seem to fall into two broad categories: norms for well-formedness of linguistic structures, on the one hand, as the monitoring system is said to detect "speech errors" and "syntactic flaws," but also user expectations, on the other hand, as it is also presumed to alert speakers when speech "rate, loudness and other prosodic aspects of speech" (Levelt 1983, 50) are not up to standard. Obviously, these standards can only be based on interlocutors' feedback or presumed lack of satisfaction with the properties of the speech produced. Discrepancies with regard to user expectations are usually not self-repaired but adapted during delivery.
What about interpreters? There is no reason to believe that the normative motivation for self-repairs does not also apply to them. E-repairs are thus considered motivated by awareness of norms and can, therefore, count as textual evidence of norm compliance. On the other hand, according to Levelt, A-repairs and D-repairs are not motivated by norms in the case of speakers, but by an assessment of the extent to which intended message and produced message correspond. However, in the case of interpreters, A-and D-repairs also draw on norms. The crucial difference between speakers and interpreters is that the former have communicative intentions of their own, while interpreters are expected first and foremost to express in another language the message produced by source speakers (Gile 1995). They do not have access to the speakers' communicative intentions directly. They can only compare their output with the source speakers' output.
When a speaker's monitoring system detects a discrepancy between the intended message and the message produced, a repair can be initiated on the basis of the intended message, not on the basis of a norm that was not upheld. In contrast, when interpreters do not express the source speaker's message clearly enough, prompting the initiation of a repair, they can only do so on the basis of a translation norm, that is, after assessing the extent to which the message produced matches the message they received from the source speaker.
Ultimately, this means that A-and D-repairs, alongside E-repairs, can also count as textual evidence of norm compliance in interpreters' output. In other words, the three main categories of self-repairs are indicative of norm-driven behaviour as far as interpreting is concerned.
Besides A-, D-and E-repairs, Levelt (1983) also distinguishes covert or C-repairs and a residual category, R-repairs. Covert repairs are a problematic case, as they do not share some of the structural properties of a self-repair (see further), such as proposing a relevant alternative for a discarded segment. R-repairs are difficult to analyse and do not seem to fit into one of the other categories. Neither C-nor R-repairs seem to challenge the general conclusion that self-repairs in interpreting are fundamentally norm-driven. We will therefore hold that evidence of self-repairs in corpora of interpreting can be used to analyse normcompliance in different groups of interpreters. In Section 7, we will, however, suggest alternative explanations for self-repairs.
Ours is of course not the first study on self-repairs in interpreting. The conclusion that interpreters monitor and correct their own speech was already drawn by Gerver (1969) and since then a number of studies have been conducted on interpreters' diverse strategies for self-repair (Van Besien and Meuleman 2004), on self-repair types in interpreting (Kalina 1998;Petite 2005) and on the interplay of directionality and self-repair (Dailidėnaitė 2009).
Most of these studies, with the notable exception of Kalina (1998), use Levelt's typology, although sometimes in a slightly altered way. All the studies focus on the cognitive load or peculiar challenges in interpreting and their role in inducing or affording self-repairs. Van Besien and Meuleman (2004), for instance, examine in an experimental setting how interpreters handle speakers' errors and speakers' repairs and how this affects their own selfrepairs. Kalina (1998) investigates the necessary conditions for a self-repair by an interpreter, stressing the fact that, besides the detection of a segment that does not meet output requirements and the conviction that a better solution is in store, interpreters also need to ascertain that enough cognitive capacity is available to complete the self-repair. She also draws up a typology of self-repairs carried out by interpreters and studies their frequencies, finding that 'replacement' is the most frequent strategy. The qualitative corpus study by Petite (2005) draws attention to the fact that, although interpreters experience high cognitive load, which self-repairs only increase, self-repairs do occur in interpreting, even in cases where they are superfluous. She also finds that interpreters have a tendency to apply multiple self-repairs for the same problematic segment. Finally, in an experimental study, Dailidėnaitė (2009) analyses whether the types and the frequency of self-repairs are influenced by directionality. Focusing on two language pairs, Lithuanian-English and Turkish-English, she finds that the total number of repairs depends on the language combination. E-repairs are the most frequent in both directions, but the number of A-repairs appears to be higher when interpreting into the mother tongue. 1 None of the studies takes the perspective of self-repairs as a testimony of norm compliance. Both Petite (2005) and Dailidėnaitė (2009) do notice that interpreters seem to be compelled to repair even though there is no objective need nor sufficient time or cognitive resources for a repair. Two of the studies briefly mention interpreter style as a factor in the frequency of repairs or of certain types of repairs (Van Besien and Meuleman 2004;Petite 2005), stressing that the sample used is too small to draw reliable conclusions but that repair patterns seem to be individually determined.
In the present study, self-repairs in interpreting are taken as evidence for normcompliance. Considering the literature on diverging gender patterns in norm-compliance, our main research question is as follows: Is the extent to which self-repairs occur in interpreting gendered? If self-repairs are first and foremost evidence of norm compliance by interpreters, and given what is known about gender differences in norm compliance, female interpreters are expected to repair more utterances than their male colleagues.
If, as the literature on linguistic norm-compliance by male and female genders suggests, women are more norm-conscious and norm-compliant, more self-repairs are to be expected, particularly of the E-type. Also, in interpreting, if female interpreters value accuracy higher 1 At this stage, a short comment on the categorisation of self-repairs is in order. In the previous section, we argued that Levelt's appropriateness repairs in interpreting are associated with translation norms. As they are not supposed to express their own communicative intentions, interpreters can only "appropriately" express the speaker's message. "Appropriateness" in interpreting, in our view, thus inevitably applies to the relationship between source text and target text. This differs from Petite's and Dailidėnaitė's views, where appropriateness repairs can both be "input" and "output-generated." According to Petite (2005), "input-generated" repairs propose alternatives that resemble the source text more closely, while "output-generated" repairs attend to the needs of the interpreter's audience in terms of processing ease and relevance. The relevance-theoretic framework in which the latter are expressed does not really concern us here; the important point is that, in Petite's view, appropriateness repairs can also be target-oriented only. However, the examples of "outputgenerated" appropriateness repairs Petite adduces, appear to fit the description of error repairs better than the description of appropriateness repairs, as they are mainly concerned with repairs in which a more idiomatic expression or collocation replaces a lesser one. than men, as reported by survey studies, more A-repairs are expected. Overall, female interpreters are thus expected to self-repair significantly more than male interpreters. Within Levelt's framework, it is also evident that different terms may be used as editing terms. As a complementary research question, we will examine whether interpreters' selection of editing terms is mediated by gender.

Data
This study is based on the EPICG (European Parliament Interpreting Corpus Ghent), which is being compiled at Ghent University. The compilation method is described in Bernardini et al. (2018): source speeches and their interpretation made during plenary sessions held at the European Parliament are transcribed from video footage, according to the Valibel norms (Bachy et al. 2007). The corpus thus includes many oral features such as repetitions, hesitations markers, false starts, and so forth. The full corpus currently comprises ca 220,000 tokens in 9 language combinations: French / Dutch / English / German.
For the sake of comparison and cohesion within the broader project on gender differences in simultaneous interpreting of which it is part, the present study is based on the 2008 subcorpus used in two of our previous studies Defrancq 2016, 2017). It comprises 193,000 words and contains 39 speeches in French and 39 interpretations, both in English and in Dutch: 16 speeches interpreted by men and 23 by women for the French/English language pair, and 19 speeches interpreted by men and 20 by women for the French/Dutch language pair. We decided to include two language pairs -French/Dutch and French/Englishto avoid yielding results which could be linguistically biased. Table 1 gives an overview of the number of texts and words in the 2008 subcorpus.

Methodology
As outlined in the previous section, we define self-repairs as interpreters' corrections of their own output without external stimulus occurring in three stages: (a) the interpreter's utterance (4), these interruptions occur at word, phrase or sentence level, respectively (ST = source text; TT = target text): (2) je veux / euh souligner la disposition du Conseil à le faire [ST] 'I want to / euh / underline the willingness of the Council to do it' [EPICG_03. 09.2008_évaluation_jeanpierrejouyet_fr] ik wil graag benadrukken dat de Com/ dat de Raad euh daar graag op in zal gaan [TT] 'I would like to underline that the Com / that the Council will agree to do it' [EPICG_03.09.2008_évaluation_jeanpierrejouyet_en] In example 2, the interpreter translates the main clause and then starts the sub-clause with 'that the com', probably thinking of 'Committee' or 'Commission', but notices that the French speaker refers to the 'Council'. He therefore interrupts the flow of speech in the middle of his word to start anew the sub-clause with 'that the' followed by the self-repaired subject 'council'. In Levelt's (1983) terms, the interpreter corrects an erroneous term, making an E-repair.
( The interpreter applies the interpreting strategy of chunking, which consists of breaking the long French sentence into two English sentences. After translating her first sentence completely, she is not satisfied with the 'we' and replaces the subject by uttering a new independent sentence. As the repair segment matches the source text more closely than the reparandum, example (4) represents a case of an appropriateness repair.
The self-repairs presented above all straightforwardly match the definition proposed in Section 1: the interpreter interrupts the flow of speech, with or without an editing term, and finally produces an alternative version as a repair. However, while identifying the self-repairs in the corpus, we came across more ambiguous occurrences, where the interpreter repeats the same syllable(s) (5), the same word (6), or even the same phrase (7): As it is hard to tell whether these are examples of real self-repairs, spans of stuttering, or emphasis through repetition, we consistently removed such cases from the analysis.
As far as editing terms are concerned, examples (2) and (4) show self-repairs without editing terms, while (3) presents a filled pause (euh). The Dutch and English editing terms found in our corpus data were the following: Dutch of (14 occurrences), English and (11 occurrences), English or (10 occurrences)as in example (8)and English but (1 occurrence). A handful of Dutch apologetic terms, as in (9), were also used as editing terms: pardon and sorry with 2 occurrences each.
(8) laisse une zone grise qui n'est pas entièrement couverte par l'actuelle proposition This pattern was also observed by Petite (2005). The apology appears either as an editing term between the reparandum and the repair or at the end of the segment, as in (9).
In order to investigate hypothesis 2, we counted the self-repairs occurring with an editing term. As the number of occurrences is quite limited, we included the apologies.
We finally analysed the results in different stages: we first compared the total number of self-repairs used by male and female interpreters, overall and by language combination; afterwards, we examined whether there was a gender-based difference in the use of editing terms. At each stage, the chi-square test was conducted to determine whether the observed differences were statistically significant (i.e., p-value less than 0.05).

Results
This section offers an overview of the results and compares them with the formulated hypotheses.

Research question 1
We hypothesized that female interpreters are more likely to repair utterances than male interpreters. Table 2 shows the number of self-repairs produced by male and female interpreters in the Dutch and English booths of the European Parliament. The numbers are expressed in absolute and normalized (/1000 words) frequencies.
Overall, the English and Dutch interpreters in our corpus produced 206 self-repairs for 480 minutes of recording, or approximately one every 2 minutes, which is far below the figures found in Petite (2005): 171 repairs for 80 minutes of recording, or 2.5 self-repairs per minute.

Research question 2
The second research question focused on the use of editing terms when self-repairing. Table   3 shows the number of acknowledged repairs, that is, repairs with an editing term (of, and, or) or an apology (pardon, sorry), overall and by language pair.  The corpus yields interesting results in this respect. Overall, our data do not support the second hypothesis, as there are no significant differences between the number of self-repairs with editing terms produced by male and female interpreters ( 2 =2.40, df=1, p>0.05).
However, the pattern clearly differs according to the language pair. In the French/English subcorpus, male interpreters only acknowledge 2 repairs versus 20 for female interpreters, a significant difference ( 2 =4.34, df=1, p=0.04). In the French/Dutch language combination, female interpreters also acknowledge more repairs than male interpreters (10 versus 8), but the difference is not significant ( 2 =0.001, df=1, p>0.05).
Regarding apologies used as editing terms in particular, it is very striking that the interpreters in the Dutch-speaking corpus also used apologies as editing terms. This strategy was clearly used more often by women (3 occurrences) than by men (1 occurrence).
However, we have too few data in order to draw conclusions.

Discussion
We set out to research the possibility that gender mediates norms in simultaneous interpreting. The few studies already conducted on norms in interpreting take a textual or extratextual approach, but no study so far has adopted a corpus-based approach to find textual manifestations of norm-compliance by interpreters independent from the observer's point of The interpreter replaces the discourse marker actually with another discourse marker in actual fact which is semantically equivalent, but carries a different degree of emphasis.
Semantically, the self-repair is superfluous. In addition, the source text does not seem to offer any pragmatic clue as to why the interpreter chose to add emphasis. The most plausible motivation for the self-repair is that the interpreter buys time in order to process the long sentence in the input. Another possible explanation could be what Straniero Serigo (2007) calls "synonymic pair," that is, a paraphrastic rendition of one item or of larger semantic units. Example (10) illustrates that not all cases in which interpreters replace segments with other segments can be considered instances of self-repair and, therefore, of norm-compliance.
In our analysis, we have not focused specifically on the possible alternative motivations for self-repair, assuming that all cases of self-repair were indicative of norm compliance.
Motivations are, after all, hard to identify with any degree of certainty in a corpus. However, given the very small number of examples such as (10) where an alternative motivation could have been in play, our conclusions can be maintained.
As for the observed gender differences, other accounts than the one based on norm compliance are possible: it could be that the properties of source speeches interpreted by female interpreters are such that more errorsand therefore self-repairsare provoked: higher delivery rates, higher lexical density, less clearly structured texts are known to induce higher cognitive load in interpreters and, as a result, higher number of errors. We have not taken that factor into account in this study, but plan to do so in future research.
The higher frequency of self-repairs among female interpreters could also be the result of particular cognitive abilities instead of a more acute norm-consciousness. There is evidence in the literature that females have an advantage over males in terms of linguistic production tasks (Hyde & Lynn 1988). This might also explain why females may be inclined to self-repair more to produce the most faithful and correct translation possible. This line of research falls outside the scope of this study, but is worth pursuing in the future.
Finally, while analysing the use of editing terms by male and female interpreters, an intriguing difference in booth practices emerged. In the French/English language combination, female interpreters used significantly more editing terms than male interpreters.
The French/Dutch language pair, however, showed no significant difference. On the other hand, a qualitative difference emerged in the Dutch booth: Dutch female interpreters offered apologies on three occasions, while only one apology was found in male interpreting. No apologies were observed in the English booth. These findings are interesting: apologies differ from the other editing terms in that they are a speech act produced by interpreters on their own behalf whereby they risk losing face. Coincidentally, female interpreters in the Dutch booth were already found to add significantly more hedges than male interpreters to source texts in the same interpretation corpus (Magnifico and Defrancq 2017). It also appears from a detailed qualitative study of the hedges that some were used apologetically by interpreters struggling to cope with the incoming information. It thus seems that female interpreters in the Dutch booth are particularly keen on defusing face-threatening aspects of the interpreting activity.

Conclusion
The present paper focused on gender as a possible mediating factor in norm adherence by male and female interpreters. As the literature suggests that men and women do not value the same interpreting norms, and as gender influences norm consciousness and norm compliance in language, it was hypothesized that interpreters would demonstrate gender-based differences regarding norms in their interpretations.
The various norms in interpreting were discussed based on the literature and operationalized for a corpus-based study. This approach led us to study textual evidence of norm-based behaviour produced by the interpreters themselves, namely, self-repairs. The corpus data, drawn from EPICG, allowed us to confirm the first hypothesis: female interpreters self-repair significantly more than male interpreters. To the extent that selfrepairs are an adequate source of information on how gender influences norm adherence, we can conclude that female interpreters take a more normative approach in their interpretations than male interpreters: they significantly more often correct utterances which they judge to breach particular norms. These results seem to be in line with our incidental findings in the field of impoliteness: we assumed that female and male interpreters followed different norms as they respectively tended to translate more unmitigated face-threatening acts and to perform more face work. The higher number of self-repairs produced by female interpreters confirms that the norms followed by male and female interpreters differ.
The second hypothesis, concerning gender differences in the use of editing terms, was not confirmed. Nevertheless, a closer study by language pair yielded interesting results. In the French/English language pair, female interpreters use significantly more editing terms than male interpreters. In the French/Dutch language pair, on the other hand, no significant difference was observed. However, we noticed that Dutch interpreters, unlike English interpreters, also used apologies as editing terms.
In conclusion, the present paper has shed some light on norms in simultaneous interpreting, a field that has hardly been empirically researched in its textual dimension. It concludes that female and male interpreters show different degrees of norm adherence, as is evident from the higher number of self-repairs by female interpreters. However, the study also suggests that more research is needed: corpus data only reveal behavioural tendencies and can only inform us to a very limited extent about motivations for the observed behaviour.