Translation for language purposes : Preliminary results of an experimental study of translation and picture verbalization

The relative merits and drawbacks of translation for lan guage purposes, i.e. translation as a teaching and testing tool in the foreign-language (L2) classroom, are largely un ex plored empirically, and opinions differ widely. Within the research fi eld, scholars often ignore this use of translation or tend to condemn it out of hand. Other authors may be more positive towards translation in the L2 classroom but tend to base their views on personal beliefs and individual teaching experiences. With this paper I hope to con trib ute to the ongoing debate with some useful empirical fuel. The point of departure is a claim implicit in much criticism, namely that translation leads learners to commit errors that they might not otherwise commit. The paper describes the background, subjects, assumptions, pro ce dure, and preliminary results of a small-scale experimental study of L2 translation (Danish into English) and picture verbalization in L2 (English). Ac cord ing to some preliminary results, learners who translated made more and other errors than learners who wrote a comparable picture verbalization. This is probably due to enhanced interference from Danish, their fi rst language (L1).


Introduction
In the Routledge Encyclopedia of Translation Studies, Guy Cook gives an interesting overview of the relation between translation and language teaching.According to this, there is an obvious dis crep an cy between the practice and theory of language teaching: "Despite the widespread popular assumption that translation should play a major and necessary part in the study of a foreign language, twentieth-century theories of language teaching and learning have at best ignored the role of translation, and at worst vil i fi ed it.From the turn of the century onwards almost all infl uential theoretical works on language teach ing have assumed without argument that a new lan guage (L2) should be taught without reference to the student's fi rst language (L1)".(Cook 1998: 117) Thus, whereas many teachers and policy makers apparently be lieve in translation as a useful component of the L2 curriculum, most theorists are inclined either to ignore it or to regard it as an inadequate, even harm ful, teaching tool and an unreliable testing tool (see also Pedersen 1974, Snell-Hornby 1985, Titford & Hieke 1985, Keith & Mason 1987, Sewell & Higgins 1996, and Malmkjaer 1998).When the issue is dis cussed in the literature, arguments are often theoretical, based on sub jec tive views, or derived from individual teaching experiences, with only a few empirical studies to either support or condemn translation for lan guage purposes.
My own investigation is an attempt to contribute with some useful empirical fuel to the ongoing debate, which I fi nd an important one.In the following pages I shall fi rst give a brief overview of the fi eld and then describe the background, subjects, assumptions, procedure, and preliminary results of a small-scale experimental study of L2 translation (Danish into English) and picture ver bal i za tion in L2 (English).

Translation for language purposes: Opinions and stud ies
As already suggested, most scholars who discuss the merits and drawbacks of translation for language purposes do this from the perspective of practical language teaching, and it is hardly surprising therefore that many arguments put forward in the literature in favour of or against translation may be categorized as opinions rather than well-documented facts.I shall now give a brief overview of some of these opinions.The basic argument in favour of translation in the L2 classroom tends to be that translation, being a controlled task, is a time-saving way of teaching and testing L2 profi ciency (e.g.Duff 1989/1992: 7, Sewell 1996: 142; see also Campbell 1998: 58, who sees L2 translation as a controlled variety of L2 writing).A related argument is that translation promotes students' linguistic agility and accuracy, perhaps mainly at the advanced level (Snell-Hornby 1985: 21).It is also suggested that translation is useful as a tool in contrastive studies, making students aware of struc tur al differences between L1 and L2 (e.g. Snell-Hornby 1985), thereby helping them to develop strategies to avoid L1 in ter fer ence in their L2 production (Sørensen 1988 and1990; see also Sørensen 1991).
The basic arguments against translation in the L2 classroom tend to contradict the above-mentioned arguments in its favour: First of all, it is claimed that translation has very little to do with real language practice, the true aim of language teaching and learn ing.Rather than helping stu dents to communicate in L2, trans la tion merely gives them an op por tu ni ty to study the L2 language system (e.g.Coleman 1986 andMarsh 1987).Furthermore, be cause translation in the L2 classroom tends to be a word-for-word exercise, it is often accused of giving students the im pres sion that there is also such a one-to-one relation between languages (e.g.Lado 1964: 53-54), which tends to lead to fl awed and unidiomatic L2 production (e.g.Irons 1998: 26).(For a more detailed dis cus sion of these pros and cons of translation for language purposes, see Schjoldager forthcoming.)It would be unfair to say that the fi eld is completely unexplored em pir ical ly, and I shall now draw attention to the results of a few important studies.Scherer & Wertheimer (1964) is the most extensive empirical study that I have come across1 .In their book, the authors report on a two-year experiment involving 300 beginning students of German at the University of Colorado, in which they investigate the relative merits of the grammar-translation method2 , which includes translation as a teaching tool, and audiolingualism3 .Their hypothesis is that audiolingualism is a superior teaching method.The conclusion is somewhat surprising: after two years of language instruction, there is no signifi cant difference between those who were taught by the audiolingual approach and those who were taught by a "traditional" approach as far as profi ciency in reading and L2 translation (i.e.translation from English into German) is concerned; those who were taught by the traditional approach excel in L1 translation (from Ger man into English), whereas those taught by the audiolingual method only excel as far as speaking is concerned.
In an exploratory study, Berggren (1972) 4 compares two methods of teaching "fairly simple everyday language" (Berggren 1972: 2), namely L2 translation (Swedish into English) and what she refers to as intensive reading in L2 (English), an "accepted form of close reading with a view to language points only" (Berggren 1972: 10).Though appreciating that it may be diffi cult to ascertain the extent to which learners, in her case fi rst-year university students, are infl uenced by the two methods, Berggren (1972:19) concludes that L2 translation does not have a neg a tive effect.Buck (1992) compares the effi ciency of L1 translation (English into Japanese) as a test of L2 comprehension with other tests, and fi nds -much to his surprise and regret -that L1 translation works fairly well.
Other studies are more critical of translation in the L2 classroom.Klein-Braley (1987) investigates the objectivity, reliability, and va lid i ty of L2 translation (German into English) as a testing tool of L2 profi ciency, comparing it with other tests.According to Klein-Braley's (1987: 128) fi ndings, the translation tests did measure language profi ciency, but they were not the best possible pro ce dure available; and as far as validity was concerned, the trans la tion tests always scored lowest.
Larsen (1990) also focusses on translation as a testing tool and ex plores the possible discrepancy between university students' per form anc es in L2 translation (Danish into English) and other tasks, namely summarizing and essay writing.Her fi ndings more or less confi rm her initial impression that translation is no reliable test of L2 profi ciency.Källkvist (1997) supports Klein-Braley's (1987) stance that L2 trans la tion, which in Källkvist's case is translation from Swedish into Eng lish, cannot stand alone as a test of general L2 profi ciency (see also Källkvist 1998).In Källkvist (forthcoming), she com pares the relative effi ciency of L2 translation (Swedish into English) and non-contrastive monolingual exercises (in English) in a course of English grammar, fi nding that those who did translation per formed slightly worse.Källkvist therefore concludes that translation may not be superior to monolingual teaching methods.

Background
What are we to believe?Is translation a useful component of lan guage teaching and testing, or is it not?As we have seen, opin ions differ wide ly, and we still lack necessary evidence to answer this question.For the time being, we must therefore be content to conclude that the issue is undecided: translation may have some value as a teaching tool and it is probably rather unreliable as a testing tool, except perhaps as a test of L2 comprehension (cf.Buck's 1992 study).
Personally I am inclined to agree with this conclusion.As far as transla tion as a teaching tool is concerned, I strongly believe that this may indeed be quite valuable, especially if the grammar-translation type of approach is supplemented (if not replaced) by a functional-com mu ni ca tive one, using for instance Nord's (1997) version of the skopos theory as a conceptual and analytical framework (as suggested in Schjoldager 2002).In this way, the negative aspects of translation, for instance en hanced L1 interference (see section 5. below), may be outbalanced by its positive aspects, especially the fact that translation is a controlled writing task forcing students to venture into new areas of L2.As far as translation as a testing tool is concerned, I am equally certain that it cannot be a reliable tool.I can see translation as a test of translation skills, but I cannot see it as an effi cient test of general L2 profi ciency.First of all, translation is bound to require more than is required for a monolingual L2 production task.Klein-Braley (1987) puts it like this about her own data: "[..] the translation tests measure the construct less adequately than the other, far more economical, test procedures, and there is good ev i dence for sus pect ing that they are two-dimensional, measuring other skills, probably translation skills, in addition to gen er al language pro fi cien cy." (Klein-Braley 1987: 129) (For some interesting papers on how to defi ne, develop and as sess trans la tion competence, see for instance Schäffner & Adab (eds) 2000.)Sec ond ly, and this seems to be implicit in most crit i cism of translation for language purposes, the enhanced risk of L1 interference may possibly lead students to make more and other errors than they might otherwise commit.
In view of the general lack of evidence within the fi eld, I decided to concentrate on one interesting factor, namely that of enhanced L1 in ter fer ence.By means of a simple comparative analysis of errors occurring in L2 translations and picture verbalizations in L2, the aim of my in ves ti ga tion is to answer these research ques tions: Do learners who translate commit more and other errors than learners who write comparable pic ture verbalizations?If they do, may this be due to enhanced L1 in ter fer ence?

Subjects
The subjects of the investigation were third-year secondary-level students (gymnasieelever) from Risskov Amtsgymnasium, spe cial iz ing in languages (Sproglig linje højt niveau), and third-year university stu dents taking the compulsory translation course in the department of English at the University of Aarhus, my em ploy er at the time of data collection (the autumn of 2000).
Subjects were chosen out of larger groups for these char ac ter is tics: They had Danish L1 and English L2; they were brought up in Danish families in Denmark; and they had learned English as a foreign language in Denmark (as opposed to acquiring it abroad).Several had to be ex clud ed from the corpus because they lacked these necessary charac ter is tics.No streaming was carried out and the division into groups, i.e. those who translated and those who did the picture verbalization, was ac ci den tal.

Assumptions
As already mentioned, I assume that learners who translate into L2 may commit errors that are caused by an enhanced infl uence of L1 in ter fer ence in translation.There are at least two reasons for this: (1) Learners who translate from an L1 text are led to rely on processing via L1, and (2) learners tend to be 'hypnotized' by the source text (ST) wording, which prevents them from making full use of their L2 mastery (Larsen 1990: 97).(See also Wilss (1982: 207), who makes a similar point about L1 translation.)I shall refer to errors that are deemed to be the result of an en hanced infl uence of L1 interference in translation as translationinduced errors.
On this basis, I chose to compare translations into English (L2) with picture verbalizations in English.I assume that picture ver bal i za tion and translation are suffi ciently comparable because both are controlled tasks forcing subjects to express a given series of ideas.The comparability of the two tasks is further enhanced by the fact that the ST of my in vesti ga tion is itself a verbalization of the pictures in question -or, rather, a synthesized version of four native-speaker verbalizations, as explained below.On the other hand, it is also assumed that picture verbalization is suffi ciently different from translation because it does not enhance L1 in ter fer ence: (1) Learners who do a picture verbalization are not led to rely on processing via L1, and (2) there is no verbal 'hypnosis' in picture verbalization.

Procedure
First I had to fi nd some suitable pictures.These should present a series of ideas that could be verbalized, and the material should be suffi cient for me to devise a plausible and coherent ST.Var i ous comic strips, which have been used in other investigations, especially within psycholin guistics, did not quite fi t my re quire ments -mainly because they do not offer enough verbal material.After a rather long search, I found some large colour pictures by Pia Thavlov (1996), in a charming children's book about a family outing.Thavlov's pictures were ideal for my pur pose, for two rea sons: (1) the story was told exclusively by means of pictures, and (2) each picture offered many details about the family's various activities, which could be verbalized into a coherent text.
Then a Danish ST (426 words) was constructed on the basis of four native-speaker verbalizations.After some experimentation in a pilot study, the data collection could begin (in the autumn of 2000).For various rea sons of research methodology and man age a bil i ty, subjects did either a translation into English (L2) or a picture verbalization in English.Both groups were given a max i mum time limit, 75 minutes and 65 minutes respectively, so as to enhance the degree of comparability between the two tasks.To make the assignment as plausible as possible, subjects were in struct ed to pretend translating/writing a story for a children's pro gramme on the radio, and, to rule out external factors, I made my sub jects work without dictionaries in an exam-like situation.Ad mit ted ly, most students were not used to be given a brief5 like this for a trans la tion assignment and none of them had been trained to work without dic tion ar ies, but I have no way of assessing the possible impact of this.(For more details about the background and procedure of this in ves ti ga tion, see Schjoldager forth com ing.)

Preliminary results
Before I mention some preliminary results, I should perhaps point out that some (but not all) picture verbalizations differ somewhat in length and contents from the translations, as a few students who did picture verbalizations seem to have written more per son al ized texts, adding for instance some imaginative details.How ev er, as the initial analysis involves no direct comparison be tween the target texts (TTs), the relative comparability between the two kinds of texts will not be discussed fur ther at this stage.Also, it is important to note that the analysis below is based on the oc cur rence of errors of language, as opposed to trans la tion al errors, which would involve a comparison with the ST or pictures.Finally, though at some later stage I intend to employ a panel of col leagues to help me assess the TTs, the analysis below comprises my own tentative marking of errors.1) Are there more errors in the translations?Yes.I found more errors in the translations than in the picture verbalizations, as shown in tables 4 and 5.Both groups share this tendency, though the university students make fewer mistakes in both tasks, which is hardly surprising.2) Is this overrepresentation of errors translation-induced? Maybe.A part from the above-mentioned difference in absolute numbers, so far I have found no evidence that the two tasks produce sig nifi cant ly dif fer ent pat terns.However, though these are very pre lim i nary results, I would like to mention that two categories of errors seem to be more prevalent in the translations, especially in those by the secondary-level students, namely -what I refer to as -non-translation and unEnglish, as ex em pli fi ed in tables 6 and 7. Non-translation occurs when a ST item (L1) has been transferred unchanged into the TT, i.e. a translation or a picture verbalization, and this is deemed to be unwarranted.UnEnglish occurs when a TT item is deemed to be non-standard in the target language (L2).Admittedly, these cat e go ries are somewhat fuzzy, just as the ex pla na tions in tables 6 and 7 must be regarded as intuitive.How ev er, the fact that errors categorized as non-translation and unEnglish seem to be more prevalent in the transla tions indicates that (some of) these errors may be due to enhanced L1 in ter fer ence, i.e. translation-induced.

Conclusion and discussion
In this paper, I have summarized some general opinions in favour of and against translation for language purposes and I have pre sent ed the re sults of a few important empirical studies.Both opin ions and studies form the background of my own study of trans la tion and picture ver bal i za tion, which is the focus of this paper.My preliminary results indicate that the task of translation may invite more errors than picture ver bal i za tion, that certain errors seem to be more prevalent in translation, and that these errors may be translation-induced, i.e. due to enhanced L1 in ter fer ence in trans la tion.This tendency is more marked in the secondary-level group than in the university group.
The bearing on translation as a testing tool of general L2 pro fi cien cy is quite clear: my preliminary results support that trans la tion is rather unreliable.However, since my subjects were asked to do either a transla tion or a picture verbalization, my in ves ti ga tion cannot show how in di vid u al students would have performed in both tasks.It would therefore be interesting to devise a study in which subjects had to perform in both translation and picture ver bal i za tion in comparable ways, so as to as cer tain whether the same students show a similar tendency of making more and other errors in the translations.It would also be interesting to study some inherent differences between the two tasks.My data show that picture verbalizations vary in degree of comparability with the transla tions in terms of length and contents, but my study cannot show why this is.Above I have suggested that the overrepresentation of errors in the translations may be translation induced, but there may well be other important factors, for instance the fact that pic ture verbalization invites more freedom.
My fi ndings are less clear regarding the bearing on translation as a teaching tool.On the one hand, inviting students to commit er rors that they might not otherwise commit may be construed as 'dangerous' be cause such errors may spill over into other L2 tasks.On the other, as many teachers will tell you, students, at least advanced ones, need to be made aware of systematic dif fer enc es between their L1 and L2, and the marking of language errors in L2 translations is one method of doing this, perhaps an effi cient one.
My own investigation draws specifi c attention to one important drawback of translation for language purposes, namely that of trans la tion-induced errors, while other studies document other fragments of the is sue.Some studies show that translation may be quite valuable, and some conclude that it is not (on its own), with rather strong evidence against translation as a testing tool.However, whatever we decide we still lack conclusive evidence: if we decide to abandon this kind of translation without some defi nite proof that it is really as harmful as many scholars claim, we run the risk of throwing out the baby with the bath water; if we decide not to abandon it and condone a long-standing tradition without refl ection, we risk wasting our time and effort and we risk the exclusion of more effi cient teaching and testing methods (cf.Källkvist forth com ing); and, similarly, if we decide to recommend translation for language purposes, say for advanced students and especially within a functional-communicative framework, as I myself am inclined to do, this is risky too because we still do not know if this is appropriate or not.In other words, more extensive, long-term studies are certainly required for us to know the true merits and drawbacks of translation as a teaching and testing tool of L2 profi ciency.

Table 3 .
Distribution of subjects 206

Table 4 .
Total number of errors

Table 5 .
Average number of errors

Table 6 .
Example of non-translation

Table 7 .
Example of unEnglish