DIRECTIONALITY : TYPES AND FREQUENCY OF REPAIRS IN SIMULTANEOUS INTERPRETATION

The article describes an experimental study of self-corrections in simultaneous interpretation (SI) in relation to directionality while making use of previous studies in this field on cognitive processes and strategies in SI. The paper aims to contribute to the literature regarding repair strategies the interpreter resorts to during SI, and examine their possible link to directionality between the language pairs Lithuanian-English and Turkish-English by way of an experimental study. The experiment was carried out with 8 students each working in two directions, from English (B) into Lithuanian (A) or Turkish (A) and the reverse. The material analysed comprises the types and frequency of self-repairs in SI and the reflections on the retrospective verbal protocols provided by the subjects, prompted by listening to the source texts and their interpretations directly after the interpreting task. The results of the study suggest that the types and frequency of repairs differ greatly when interpreting into A and B. The conclusions of the study present some pedagogical implications for training future interpreters.


SI AS A cogNITIVE procESS
Simultaneous interpretation is a special act of communication where the interpreter receives a message in the source language and conveys that message in the target language, while still listening to the original input.In other words, the interpreter enters into some kind of loop where the production of the original message, its comprehension by the interpreter and the production of the target message are overlapping, i.e. simultaneous, processes.Some researchers like gerver (1976), Moser (1978), andgile (1997) have proposed mental models that describe the interpreting process as a complex cognitive process.gile developed the 'Effort Model of Simultaneous Interpretation' which intends to explain the difficulty of this process (gile 1997, 197).gile's model is based on the concurrence of three efforts: the Listening and Analysis Effort (L), consisting of all comprehension-oriented operations; the production Effort (p), described as 'the set of operations extending from the initial mental representation of the message to be delivered, through speech planning, and up to the implementation of the speech plan' (gile 1997, 198); and the Memory Effort (M), consisting of all the operations related to the use of short term memory.To these three efforts the author adds a fourth one, the coordination Effort (c), which coordinates the other three.
gerver (1976) underscores the interpreter's control through the distribution of attention in the different phases of the activity.Moreover, the author discusses the role of memory in SI and states that 'ear-voice span data suggests that some form of short-term buffer memory' (gerver 1976, 191) helps to interpret and receive the information and at the same time becomes involved in the SI process.gerver also refers to the output buffer memory as the one which helps the interpreter to monitor and correct the output, which is an inseparable part of SI and which will be addressed later in this study.

STrATEgIES IN SI: rEpAIr MEchANISMS
controlling and coordinating the above-mentioned overlapping activities in such a way that effective communication is ensured between the parties requires specific strategic efforts (Kohn & Kalina 1996, 129).The strategies applied by interpreters in order to carry out their overlapping tasks successfully have been the subject of numerous studies, however this article presents a reflection just on a few of them.riccardi, for example, categorizes strategies in four main groups and states that this is the most common categorization: comprehension, production, overall and emergency strategies.'comprehension strategies generally include anticipation, segmentation, selection of information, stalling or waiting' (riccardi 2005, 765), whereas production strategies consist of comprehension, expansion, approximation strategies, generalization, use of prosody elements, etc. overall strategies include monitoring and décalage, and emergency strategies can include omissions, transcoding, etc. (riccardi 2005, 765).Kalina (1998) distinguishes mainly between comprehension and production strategies.Emergency strategies come into play if any one of these strategies fails and the interpreter does not want to jeopardize the essence of the message or the macrostructure of the source text and resorts, for instance, to omission or approximation, or when, as a result of monitoring which is both an overall strategy and an automatism, an error in production is detected and repair strategies have to be applied (see also Kohn & Kalina 1996).
other authors offer other explanation of coping strategies, however, what is important to remember for the sake of this paper is that even though some strategies might become automated in time as a result of their successful repeated use (e.g.Bartlomiejczyk 2006, Kalina 1998, Kohn & Kalina 1996, riccardi 2005), one side of monitoring will always be and has to stay a conscious act so that it can function as a security valve throughout the interpreting process in order to make self-corrections possible.
In trying to ensure communication, which is the ultimate goal of SI, an interpreter, as a result of monitoring, might detect an error in the comprehension or production stage, which s/he fears could risk the understanding of the recipient.This error might or might not be reflected in the output yet.however, if it has, the interpreter could choose to employ repair strategies, i.e. resort to self-correction.As for SI, interpreters do not have time to reflect on the choices and are unable to work out the best possible solutions, therefore self-corrections or repair mechanisms very often become part and parcel of the SI process.
repair mechanisms are one type of the so-called emergency strategies (Kohn & Kalina 1996) and they are also a possible consequence of self-monitoring.According to Kalina the interpreter decides to correct the already produced target language output when she believes that she has not been able to meet the requirements of the given segment, finds or thinks she has found a solution and has extra capacity available to interrupt the interpreting process for the sake of the correction (Kalina 1998, 124).There are different repair mechanisms used by interpreters, for example, post-articulatory appropriateness repairs, when the information needs qualification; post-articulatory error repairs, to correct a mistake; post-articulatory D (different) repairs, when the interpreter realizes that a different arrangement of the word order of the message would be more effective; and mid-articulatory repairs, when the interpreter starts uttering a word and then stops and corrects the mistake.
Kalina (1998) describes different self-correction or repair strategies that she has identified in her research.The most important of these are replacement (replacing an already-produced segment with another, i.e. explicit correction), completion (an already produced segment is not withdrawn and is followed by another segment that is more precise), approximation (one or more segments that bring the interpreter closer to the searched segment are produced; this strategy is defined elsewhere by Kalina (1998, 120) as a strategy that offers more precision or synonyms in order to conceal its corrective nature) and relativation (a less absolute statement is formulated after an absolute statement).
Thus, replacements are the most frequently used among these repair strategies and take place at word, word group and segment level (Kalina 1998, 195-196).This same argument could also be explained using gile's Effort Model.The interpreter might intentionally decide not to correct him/herself because this decision would cost extra processing capacity that would no longer be available for the other vital processes.It may seem that no repair and error that slips by the interpreter unnoticed are the same.The author argues there is a fundamental difference between the two.The process of detecting something going wrong and making a decision not to repair requires some processing capacity and is a strategy in its own right.
further, Bartlomiejczyk (2004, 246) concludes that interpreters tend to notice more mistakes in their own output when interpreting into their native (A) language than when interpreting into their B language.This may be due to the fact that the level of their proficiency is higher in their A language compared to their B language.That, however, is not necessarily reflective of the actual number of errors.

DIrEcTIoNALITy IN SI
one of the most controversial aspects of SI is that of directionality, i.e. whether interpretation into the mother tongue (B into A or 'passive' interpreting) provides a more accurate rendition of the message than interpretation into a foreign language (A into B or 'active' interpreting).opinions seem to be divided as to which combination is best suited for the interpreter, and which allows for a more faithful or accurate interpretation.Whereas some authors argue that passive interpreting is not the most logical approach stating that 'the 'foreign language-tomother tongue' mode of simultaneous interpretation to which most interpreters' schools are traditionally geared is not the most rational or optimal approach' (Denissenko 1989, 157), others such as herbert (1952) and Seleskovitch (1978) consider that SI can only be performed properly into one's mother tongue due to various reasons.Seleskovitch argues that, even if the interpreter has high fluency as a speaker in the B language, when interpreting 'his native-like fluency disappears.his words no longer flow easily and naturally, and his pronunciation and vocabulary reflect the influence of his native language' (Seleskovitch 1978, 79).The author also points out that the native language is best suited to transmit information, since 'only in the A language will the speech production be spontaneous and idiomatic' (Seleskovitch 1968, 43).
According to Donovan (2004), this is basically the perception of the past that SI into B is easier.Denissenko believes that 'a full or near full message gotten across even if in a somewhat stiff, less idiomatic or slightly accented language serves the purpose much better than an elegantly-worded and an impeccably pronounced half message or less' (Denissenko 1989, 157).In his opinion, comprehension is the most important cog in the SI machinery, and 'it can hardly be denied that comprehension in the mother tongue is easier than in an acquired foreign language' (Denissenko 1989, 157).Besides, one's higher command of the mother tongue may turn into a disadvantage because there is 'a wider choice of possible ways and means of conveying the same message' (Denissenko 1989, 157).Therefore, it takes longer to make decisions for delivery since the interpreter has a lot of options to choose from.on the other hand, a B language offers a more restricted choice of ways of conveying the message, so the effort allocated to re-encoding the message into the target language is less burdensome.
What are the reasons behind each of these theories?The preference for the passive interpreting mode is generalized nowadays in most interpreting schools, and even in the international arena interpreter employers prefer interpreters working exclusively into their A language.Why is that so?According to gile (2005, 10), these positions are based on 'a mix of personal experience, ideology and tradition' rather than research.Some empirical studies have been carried out in recent years in order to examine which combination (A into B or B into A) is best suited for SI.Al-Salman and Al-Khanji's (2002) study on directionality in Arabic-English professional interpreters has shown that the majority of the subjects 1 seemed to be more comfortable when interpreting from their A into their B language.Bartlomiejczyk's (2004) survey detected that trainees and professional interpreters differed in their opinion concerning directionality: students preferred A into B interpreting, whereas professionals considered B into A interpreting to be less problematic.When reporting the frequency of specific problems the majority of the students said that when interpreting from A into B the main problems were grammar, intonation and accent-related, whereas in the case of B into A understanding and vocabulary caused most of the problems.The professional interpreters reported that in the case of A into B the interpreter's grammar, intonation, vocabulary and accent were most problematic, whereas in the case of B into A most of the problems were comprehension-related only.This lack of consistency in the results lets us deduce that there are many other variables that need to be considered during SI.
The results obtained by Darò et al. show that A into B interpreting 'makes interpreters commit more errors that are generally considered to be the most serious ones' (Darò et al. 1996, 112).The test revealed that subjects made significantly more errors when doing difficult texts from A into B. As for the error type, false starts, wrong corrections and correct corrections together with slips of the tongue and pauses were analysed in the article as added mistakes.The test revealed that more added mistakes were made in the condition with attention focussed on the input during the SI of difficult texts (dense in content) from A into B as opposed to any other type of condition.This lack of consistency in the results lets us deduce that there are many other variables that need to be considered during SI.
1 Subjects -Interpreting students who took part in the experiment and whose interpreting results the authors of the study were analysing.rEpAIrS: hypoThESES relatively few studies have addressed self-repairs in SI as one of the self-monitoring strategies used by conference interpreters.To the best of the authors' knowledge, no study has investigated the effect of directionality on the type of self-repairs or self-corrections in SI either.
As mentioned above, the paper aims to contribute to the literature regarding repair strategies the interpreter resorts to during SI, and examine their possible link to directionality between the language pairs Lithuanian-English and Turkish-English by way of an experimental study.Therefore, for the purposes of the study the following questions may be raised to initiate deliberations on the topic: a) Does directionality affect the types of corrections made by the interpreter?Does interpreting into one's mother tongue mean that the self-corrections are going to be purely stylistic?Does interpreting into one's B language mean that self corrections are going to be mainly grammatical, lexical and/or pronunciation-related?b) Does directionality affect the frequency of corrections made by the interpreter?Does interpreting into one's A language mean more or fewer self-corrections?firstly, assuming that the interpreter would have fewer problems in comprehension when interpreting from his/her A language into his/her B, the authors thought they could expect fewer corrections related to meaning, as there would be fewer errors regarding accuracy, completeness, etc. (cf.gile 2005), and anticipation strategies would prove successful most of the time.The authors also believed that when working from his/her mother tongue into his/her second language, the interpreter would make more grammatical corrections, and would resort to lexicon-or pronunciation-related repairs more frequently.corrections of language, i.e. style and presentation (prosody, accent, etc.), on the other hand, were expected to be more frequent when the interpreter works into his/her mother tongue as s/he has a better command of the target language, and consequently can detect violations more easily.
regarding the second set of hypotheses resulting from the second group of questions which dealt with the possible link between directionality and frequency of repairs, the authors argued, for the very same reason stated above, i.e. better command of the mother tongue, that because the interpreter has more capac-ity available for monitoring and making corrections, one could expect more repair situations when the interpreter works into his/her A language, despite the probably lower number of errors made compared to that in the output in the foreign language, as directionality requires more available capacity for production (cf.gile 2005).In other words, when interpreting into their mother tongue, the interpreters were expected to make fewer errors in total, but correct a higher number of these errors (higher corrected errors/total number of errors ratio compared to interpreting from A into B).however, as the authors did not investigate the number of errors in this study, the ratio of repair to no-repair situations (elicited from retrospective verbal protocols with the subjects) was used to assess this very hypothesis.Another argument worth noting under this hypothesis was that as the interpreter has a higher linguistic awareness in his/ her mother tongue, s/he would aspire to render the speech in the most accurate, appropriate and faithful, i.e. flawless way, and would have higher standards for the acceptability of his/her interpretation, which would lead to more repairs.however, rather than repairs of errors, these decisions were expected to be more like stylistic 'fine-tuning'.

ENcoDINg of ThE rESuLTS AND cLASSIfIcATIoN of rEpAIrS
In the experiment (Dailidenaite et al. 2007) we opted for a blend of both classifications and have come up with a more complete classification that contains four main categories: when the repair appears, what is being repaired, why it occurs and how it is perceived by the subject.
In terms of the moment they occur in the utterance, repairs can be grouped in the following types: post-articulatory repairs take place after the 'reparandum' (Levelt 1983) has been uttered; mid-articulatory repairs are 'within-word interruptions' (petite 2005, 30), when an interpreter starts uttering a word and then stops and corrects the mistake; covert repairs are, according to petite, pre-articulatory repairs, 'produced in the internal loop, before the utterance stage' (petite 2005, 30).Levelt describes them as 'the case where after the interruption and editing phase, the utterance is continued where it broke off (i.e.zero alteration)' (Levelt 1983, 44).This type also includes cases where an interpreter interrupts him/herself and then adds what Levelt calls an 'editing term' (uh, ehm, etc.), or s/he repeats one or more lexical items.The authors further expanded this clas-sification by adding no-repair situations, i.e. the conscious decision of the interpreter not to repair the utterance since the damage could outweigh the benefit (Bartlomiejczyk 2006, Kalina 1998, Kohn & Kalina 1996, petite 2005), and delayed repairs, where the interpreter might choose to delay the 'solution' so as not to waste precious processing capacity and offer a repair at a certain distance from the original input.An example of this could be using a term tentatively several times, knowing that there is another 'more accurate' option, and producing this option when it is eventually found.
According to what is repaired by a subject, repairs may be classified into the following types: appropriateness repairs take place when an interpreter becomes aware that the expressed information, although correct, 'needs qualification in view of the context of expression' (Levelt 1983, 52).According to Levelt, these repairs seek either ambiguity reduction, terminological precision (moving 'from a less to a more precise item'), or 'coherence with previously used terms or expressions' (Levelt 1989, 459).The authors, seeing that the interpreter is not uttering his/her own words, but someone else's, and that s/he needs to wait and decode the message as it comes along, have considered it necessary to add a new sub-category, i.e. completion repairs, which, as the name implies, complete the information provided beforehand by the interpreter; error repairs are intended to correct a mistake.Based on Levelt (1983), the authors of the study distinguished lexical error repairs, syntactic error repairs where the speaker starts a syntactic construction which is repaired as he/she cannot finish it otherwise, and phonetic error repairs.The authors have also added grammatical error repairs as a separate subgroup in this category; comprehension (error) repairs take place when an interpreter becomes aware that the expressed information is incorrect or unfaithful to the source text.comprehension error repairs usually stem from the incorrect interpretation of the source text.This type of repairs has not been singled out in previous studies.however, the authors of this study considered it important to discuss this type of repairs as they were quite frequent in the study, and different repairs refer to situations when an interpreter realizes that a different arrangement of the message would be more effective.If we examine what motivates the repair, except in the case of no-repair situations, all these repairs could be either input-generated (to bring the output closer to the ST) or output-generated (to increase appropriateness in the TT) (petite 2005, 44).The latter group would also include those repairs which are actually not necessary but are resorted to because of the interpreter's own criteria for acceptability (petite 2005, 37).
The authors also wanted to examine the repairs in the corpus with regard to how fully they are recognised and perceived by the listener.Therefore, they have distinguished among disguised (Kalina 1998), explicit and signalled repairs (petite 2005).
The categories explained above were used in classifying all the repairs found in the examined corpus of interpretation performances.All detected repairs in the interpretations from B into A were transcribed as explained by petite (2005,(33)(34) with the original speaker's input, the interpreter's output, and a gloss.The part in the source text causing the repair, the item to be repaired (reparandum; petite 2005, 35), and the repaired item or segment (reparatum; ibid.) were also included in the transcription.
The corpus analysis was supported by the rVps (retrospective verbal protocols).Notes taken by the experimenters during the experiment, as said, were used to prompt the students when they failed to remember a repair spontaneously.All of these data were brought together, compared and analyzed to arrive at findings which would either verify or invalidate the authors' initial hypotheses.In this paper you will find what the authors found as a result of their quantitative and qualitative investigations with regard to the impact of directionality on the type of repairs, directionality and frequency of repairs.

ThE ExpErIMENT
The experimental set-up included the interpretation of three pre-recorded speeches into both directions for each language pair at one speech rate (130 words/min); notes taken by the experimenters of detectable repairs during interpretation, to be used later on as prompting cues; recordings of the students' interpretations; retrospective verbal protocols (rVps) elicited immediately after each interpretation session of two performances, one into the A language and one into the B; and the recordings of these rVps.The authors wanted to examine the repairs in the corpus with regard to how fully they are recognized and perceived by the listener.

Source material
The source materials for the study were the following: one 10-minute pre-recorded English speech to be interpreted into Lithuanian and Turkish, one 10-minute prerecorded Lithuanian speech to be interpreted into English, and one 10-minute pre-recorded Turkish speech to be interpreted into English.In order to control more variables the same English speech was utilized as the source text (ST) at both locations for the B-into-A exercise, and the Turkish and Lithuanian translations of an English speech agreed upon in advance were recorded for interpretation into English (A-into-B exercise).The original speeches, i.e. the English speech for interpretation from B language into the mother tongue (by uS Senator robert c.Byrd about the uS invasion of Iraq), and the English speech that served as the ST of the translations into Lithuanian and Turkish (by Wesley cragg, Director of the george r. gardiner program in Business Ethics at york university, Toronto, canada about business integrity and ethic codes), were modified for purposes of the study.The original English speech to be interpreted into Lithuanian and Turkish was pre-recorded by a native English speaker who was not known to either the Turkish or the Lithuanian subjects beforehand.Similarly, the speeches to be interpreted into English were also recorded by Turkish and Lithuanian native speakers not known to the subjects, to rule out the accent familiarity variable.
As mentioned above, all the material, including any type of repairs, has been transcribed by the experimenters.Thus, to have the subjects' recollections of particular types of repairs, where a subject stopped the recording or, if a subject failed to recognise the repair, an experimenter would stop the recording and elicit a comment.The subjects were asked to recall what they had been thinking about at the time of repair and what the main triggers were for the repair they made.

Subjects
As explained above, the experiment was carried out at two locations: Vilnius university, Lithuania and the Boğaziçi university in Istanbul, Turkey.four graduate students in their second year from the Department of Translation and Interpretation Studies of Vilnius university for the language pair Lithuanian-English (LT-EN), Lithuanian being their A (mother tongue) and English their B language (retour into English is one of the requirements of the course), and four undergraduate students in their fourth year from the Department of Translation and Interpretation Studies of Boğaziçi university for the language pair Turkish-English (Tr-EN), Turkish being their A and English their B language (retour into English is one of the requirements of the course), took part in the study.
Moreover, both groups of students have been taking SI classes for three semesters, which enabled the authors to control comparability of the students' progress in SI-related skills.Students were preferred over professional conference interpreters at this stage because frequently used interpreting strategies become more automated with experience and are therefore more difficult to identify and verbalize for professional subjects (Kalina 1998, 153).Also, as the experiment itself was expected to bear some pedagogical implications, which may be valuable tools for future interpreter trainers teaching retour, the authors of the paper preferred students over professionals in their research.

Procedure
The course of the experiment included two sessions (one in Turkey and the other in Lithuania), carried out by four (two at each location) professional conference interpreters and teachers of simultaneous interpreting who are at the same time the four authors of the study.Each experimenter worked with two students in two one-on-one sessions, and spent approximately 1,5 hours for each session including the rVps.
Each session started with a short debriefing, where the students were told that their output was going to be recorded and used for the purpose of the study later on.The confidentiality of the results was guaranteed and the subjects were informed that the results of the study would be analyzed for the purpose of an MA paper.They were informed about the length and topic of the speeches.
The students were exposed to their recorded interpretations in both directions immediately after their two performances.After the interpretations, before the subjects were asked to produce the rVps, they were told that the experimenters were looking for and analyzing repairs (self-corrections) in SI in relation to directionality.The students were told once again that the results of the study would be public and available to them.
Then the students were played their own output.The experimenters asked them to stop the recording they were listening to whenever they remembered making a correction or deciding not to make a correction, and try to verbalize the underlying reasons or processes leading to their final product.Notes taken by the experimenters of the corrections they heard while listening to the actual performance and of omissions that could be the result of no-repair decisions were used as prompting tools when the subjects failed to make a comment spontaneously, in which case the experimenter stopped the recording, replayed that part and asked the student whether there was anything s/he remembered about that particular segment.
The aim of letting the student control when to stop the recording was to make sure that as many disguised repairs as possible were identified, and no-repair decisions, which would otherwise go unnoticed, were brought to the attention of the experimenters.Another advantage of this choice was that experimenters could take note of automated repairs that were not remembered by the subjects.All the self-corrections and the recorded rVps were summarized and the results will be reflected upon in the findings of the study.The following quantitative conclusions may be drawn from the study of the whole set of repairs and the comparison of the results among the languages:

Total Repair Count
• Working from EN into LT the subjects have concentrated on error repairs, especially lexical and grammatical, followed by appropriateness completion repairs.• Working from EN into TR the subjects have primarily repaired the appropriateness of their interpretations, especially the terminological precision and the completeness of their renditions.• Working from LT into EN the subjects have concentrated on lexical and phonetic repairs.• Working from TR into EN the subjects have concentrated on lexical error repairs, with a significant number of no-repair situations.• The overall tendency in both languages and in both directions is that repairs were mostly output-generated, post-articulatory and explicit.

QuALITATIVE rESuLTS
The interpretation of the qualitative results of the study is based on the classification of the repairs described above and examples of repairs obtained from the recorded experiment.The discussion of the repairs focussed on the postarticulatory (appropriateness, error and different) or mid-articulatory repairs which account for within-the-word interruption repairs.Covert and no repair cases were also mentioned in the study.The analysis on whether the repairs were input-or output-generated and how they were perceived by the subjects themselves were presented in the study too.The following are the generalised qualitative analysis results: • As in the case of examples in the Lithuanian-English pair, the subjects working in the Turkish-English pair in both directions wanted to make appropriateness repairs at the cost of time despite the fact that their initial renderings were not incorrect.• After recording the results of the experiment, it became obvious that interpreters correct not only lexical or syntactic errors, but also phonetic ones.further, repairs are often based on corrections of grammar.• Judging from the subjects' recalls and the repairs themselves, the interpreters tend to correct more grammatical mistakes when interpreting from language A into language B.
Quite often subjects decide to repair after the utterance because of a phonetic error.As a result of the self-monitoring process, interpreters realise that, unless repaired, a phonetic error might lead to a misunderstanding by the listeners or to greater ambiguity of the message, and this results in post-articulatory error repairs in the flow of interpretation.• Most of the no-repairs in the other direction (B into A) were due to comprehension problems or the speed of the source text.• The signalled repairs are not only more time-consuming, but also quite stressful to interpreters.We can only conclude that they use up more processing capacity and cause more stress in the interpreting process, which might lead to the misinterpretation of the input message that follows.

coNcLuSIoNS
The results of the present study show that directionality does have an effect in the types and frequency of repairs.however, the differences turned out to be more evident in the Turkish subjects than in the Lithuanian ones.Even with such a limited scope (two languages and eight subjects) it was possible to observe different patterns by language, by direction and even by subject.This makes it difficult and even risky to try and extrapolate these results to the larger interpreting population.furthermore, since our subjects were interpreting students, we cannot affirm that these same patterns would also be observable in practitioners.
In the experiment, the number of words was the same in both speeches in an attempt to put equal loads on the processing capacity.The recording speed was also the same.however, the recordings used in the A into B part of the experiment were perceived to be faster.We can therefore recommend an interdisciplinary approach to ensure the equal loads on the processing capacity by considering the facts and criteria other than the number of words (like e.g.lexical density).
In the Turkish-English language pair (A into B) in particular, the number of norepairs has significantly increased due to lack of time as the Turkish words are usually much longer than the English ones.This being said, and always bearing in mind the limitations of our study, the results we have obtained seem to provide answers to our hypotheses: although the overall number of error repairs has been higher than any other type of repair in either direction, the number of appropriateness repairs (stylistic) has been significantly higher when working into the mother tongue.Also, repairs have always been more frequent when working into the A language.And finally the language pair influences the repairs being made (for example, the number of 'different' repairs was found to be higher in Turkish than in Lithuanian).
This experimental study may contribute to a better understanding of the nature of the repairs in terms of language direction, whereas its limited results may help future interpreter trainers to concentrate more on particular deficiencies which are more relevant when working into a foreign language and others which are more evident when working into one's mother tongue.for a more efficient use of repairs, or a less frequent use of repairs, students have to be better trained in coping with the specific difficulties of a given language direction.Exercises focusing on anticipation and chunking could be very helpful, especially for the A into B direction.