Vous êtes sur la page 1sur 10

English for Specic Purposes 37 (2015) 112121

Contents lists available at ScienceDirect

English for Specic Purposes

journal homepage: http://ees.elsevier.com/esp/default.asp

Friend or foe? Google Translate in language for academic

Michael Groves a,1, Klaus Mundt b, *

Centre for English Language Education, University of Nottingham, Malaysia Campus, Jalan Broga, Semenyih, Selangor, 43500, Malaysia
Centre for English Language Education, School of Education, The University of Nottingham, Dearing Building, Jubilee Campus, Wollaton
Road, Nottingham, NG8 1BB, United Kingdom

a r t i c l e i n f o

a b s t r a c t

Article history:
Available online 2 October 2014

A recent development in digital technology, machine translation (MT), is improving in its

ability to translate with grammatical and lexical accuracy, and is also becoming increasingly available for students of language for academic purposes. Given the acceptance of
other digital technology for teaching and learning, it seems likely that machine translation
will become a tool students will rely on to complete their assignments in a second language. This would have implications for the community of practice of academic language
teaching. In this study students were asked to submit an essay in their rst language and
this was then translated into English through a web-based translation engine. The
resulting English text was analysed for grammatical error. The analysis found that the
translation engine was far from able to produce error-free text however, judging in
relation to international testing standards, the level of accuracy is approaching the minimum needed for university admission at many institutions. Thus, this paper sets out to
argue, based on the assumption that MT will continue to improve, that this technology will
have a profound inuence on the teaching of Languages for Academic Purposes, and with
imaginative use, will allow this inuence to be positive for both the students and their
2014 Elsevier Ltd. All rights reserved.

Language for academic purposes
Machine translation
Academic literacy

1. Introduction
An expert writer of academic English writes at a number of levels. The writer needs to take well-formed words, and use
them to construct well-formed sentences, and then link these sentences together into clear, coherent and cohesive paragraphs. Beyond this, the writer then needs to align the writing with the generic expectations of the intended audience
(Swales, 1998), as well as the stylistic conventions of the discipline that are commonly attached to specic genres (Biber,
2006). So the writer aligns the writing with the expectations of the communities of practice and enquiry in the eld in
which the writing takes place (Hyland, 2004).
Technological developments have long helped writers of academic English with many of these issues, either directly or
indirectly. For example, corpus analysis has made much stronger and more robust descriptions of academic English as a genre,

* Corresponding author. Tel.: 44 115 82 32193.

E-mail address: klaus.mundt@nottingham.ac.uk (K. Mundt).
Tel.: 60 3 8924 8692.
0889-4906/ 2014 Elsevier Ltd. All rights reserved.

M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121


as well as the sub-genres within it than traditional linguistic work possibly could have (Peacock, 2002; Reid, 1992; Silver,
2003). Another example of benecial uses of digital technology is that the wide availability of published papers on the
Web has allowed students and researchers to examine the style and structure of vast swathes of academic writing. Further,
freely available tools, such as Lextutor (www.lextutor.ca) and concordance tools have allowed students of academic English to
examine their own interlanguage through various lenses and improve it (Cobb, 2010). This is well documented in various
threads in the literature (Adolphs, 2006; Bruce, 2010).
In addition, new technologies have even entered the toolkit of all academic writers with little or no fanfare in the EAP
literature. The spellchecker has enabled non-native and native speakers alike to write English that is less prone to error, and the
inclusion of an effective autocorrect facility removes many mistakes without the user even being aware of their existence. Some
word processors also have grammar checkers and the more recent development of the format consistency checker, signied by
blue underlining in Microsoft Word. However, the success of these features has been less impressive (Vernon, 2000).
On the other hand, there is another technology that could overtake and replace these features web-based machine
translation (MT), most noticeably Google Translate. This is a free, web-based service that translates between a variety of
languages. It is also available on mobile devices as an app. Google Translate is a statistics-based translation tool, which means
that the system calculates probabilities of various translations of a phrase being correct, rather than aiming for word-for-word
translation. It also has a level of interactivity with the end user, with users being able to correct the original translation, and
this information being absorbed into the database.
Machine translation by means of computers dates back to punch card systems in the 1940s. Since then it has experienced
several setbacks and signicant advances (Hutchins, 2000), and with developments in Articial Intelligence might well be
poised to reach a degree of signicant sophistication (Newton, 1992). Early machine translations were often of poor, even
impenetrable, quality (Komeli, Hendavalan, & Rahimi, 2011), and are still far from perfect. However, the use of MT tools is
becoming more and more widespread. Examples include the socio-political empowerment of minority language communities in Canada (Bowker, 2008), the use of spoken machine translation for non-English speakers in the British healthcare
system (Somers & Lovel, 2006) and screening of the gist of news reports by US intelligence agencies (Koehn, 2010). Despite the
fact that the quality of the translation is often regarded as poor in comparison to human translations, the use of MT is now
reaching a much wider audience than before (Hutchins, 2006), and the development of more sophisticated MT options is
receiving more substantial attention from policymakers (Bellos, 2012).
Several MT applications are available, both as professional software and web-based freeware, that can cater to different
clients and specic purposes (Austermhl, 2001). This study focusses on only one of those, namely Google Translate, since,
while a comparison of the quality of different MT tools would be interesting, it goes beyond the remit of this paper.
Once MT has become an effective tool, it has the potential to have a profound impact on the eld of language teaching.
After all, why would a potential student go to the effort and expense of learning a foreign language if she is able to produce an
acceptable L2 text from her own L1 writing, instantly and with no nancial cost? It is therefore vital to understand the ability
of MT to create sophisticated academic text, and the present study takes a step towards that goal.
The aim of this study is to examine text which has been translated from Malay (Bahasa Melayu) and written Chinese
through the Google Translate engine into English. The study will only investigate the grammatical accuracy of the translation,
without examining the larger discoursal and epistemological features of the writing. It aims to discover if Google Translate has
the ability to produce stretches of grammatically correct, communicative English.
Studies on this topic are scarce, probably due to the lack of imminent implications for current EAP practice. However, it
would be short-sighted to ignore this emerging technology and the changes it may bring with it. The present papers
contribution to considerations related to academic language teaching and learning is to highlight these possible changes and
to hopefully trigger more specic discussion on the benets and the threats MT can bring, so EAP remains well prepared to
address the needs of its community of practice.
2. Technological advances and their use in EAP
Technological advances have always played a signicant role in second language teaching and acquisition, and they have
generally been accepted as valuable tools in the classroom, for autonomous practice, for tutorstudent communication and
for research. The fact alone that most EAP courses these days seem to provide additional learning resources for their students
on online platforms indicates that digital technology has rmly established itself as an integral and valued part of EAP.
In the course of this development, the face of the classroom has been altered as well. For instance, traditional learner
dictionaries have largely given way to electronic dictionaries accessed through mobile phones, tablet computers or laptops,
which, in turn, allow students to access other applications as well (Mehta, 2012). The quality of these dictionaries and other
learning aids seems questionable on occasion, so the EAP practitioners role has been extended to that of quality assurance, or
at the very least to that of a guide to reliable online resources. After all, in the plethora of resources available to the student,
there are signicant differences in quality and reliability, and it seems part of the EAP tutors remit to stay abreast of recent
developments so as to be able to recommend appropriate ones to their students (Chapelle, 2003; Mehta, 2012).
It would seem that the willingness with which useful tools have been accepted into EAP practice may lead to other tools
nding their way into the classroom as well; some of these, for example established data-driven tools, such as concordancers,
have since their inception been used extensively in teaching, research and materials development (Oakey, 2010). Other
technological applications may not receive such a warm welcome. Given the availability of online resources to the students in


M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121

the classroom, it is more than likely that students will not stop at the use of online dictionaries. It seems inevitable that
students will also utilise applications that will allow them not only to understand the meaning of single lexical items but also
to render entire stretches of text into their L1 or into English. Unless the use of online technology is banned from the
classroom, it seems that students use of online translation tools is inevitable and this may be cause for concern.
This paper does not adopt the position of one of the specic streams of thought discussed by Chapelle (2003, p.10), namely the
technologist, the social pragmatist or the critical analyst position. Rather, it places itself in a more neutral analytical space. Thus
we do not advocate the growing inuence of technology as a welcome and positive development, but neither do we perceive
technology as inextricably connected to ideology and commerce. In that sense, this study takes place from a perspective that
could be termed analytical realism namely the investigation of research data to illuminate the space between technology and
language teaching/learning without the ideological bias of the three schools of thought mentioned above.
2.1. Mechanics and quality of machine translation
According to Cancedda, Dymetman, Foster, and Goutte (2009), ideas and ambitious plans for MT stretch back to the 1940s.
Traditionally, MT had been based on the idea of parsing and generations that is, the computer understanding the underlying
meaning of a text in the source language, and then applying a new set of rules belonging to the target language. However, by
the 1980s this emphasis had shifted to statistics-based translation practise, and this is still the dominant paradigm.
Currently, the reliability of MT is still questionable, and it seems advisable not to rely on it as an absolute communication
tool, but, if used, to ensure that the content of a translation is veried. In that, as Scherf (1992) points out, MT remains a device
of support, not one that eliminates the necessity to acquire a degree of prociency in a language the writer wants to
communicate in. In addition, Sheppard (2011) acknowledges the usefulness of Google Translate, especially when the cost is
compared to the use of professional translation services, but describes the many aws of the translated text, from syntactic
problems to issues of style and identity. She also points out that there is a major issue with condentiality, since the text is
stored on the Google servers which, in the academic context, would have serious ethical implications. Other empirical
studies in the use of MT have come to similar conclusions. Costa-juss, Farrs, and Pons (2012) found that in the eld of
medicine, the translation program was unable to produce acceptable translations all of the time. They also noted a major
difference between the Romance and Germanic languages, for which the translation program could produce acceptable, as
dened by the researchers, translations up to 80% of the time, and other languages, such as Basque, for which these could only
be produced 27% of the time. Meanwhile, Kirchoff, Turner, Axelrod, and Saavedra (2011) found that when health literature
was translated between Spanish and English, the quality was unacceptable unless post-editing by a human translator took
place. They also found that the most common error types were morphological and in word sense. Both of the above studies
came to the conclusion that MT was only truly effective when used with a human post-editor.
One of the reasons that MT is not yet sufciently developed to be authoritative in producing accurate renditions of a text is
that it is not able to evaluate the context of a stretch of text that a human translator or writer will take into account.
Kaltenbacher (2000) identies this as the cause for a relatively large number of syntactical errors in MT, in particular across
languages in which the grammar differs signicantly. This nds support in a recent study by van Rensburg, Snyman, and Lotz
(2012). In this study it was found that Google translations of academic texts between English and Afrikaans were no match for
those created by human translators, indicating that MT is not yet sufciently sophisticated to create acceptable target texts.
In the case of Google Translate, the most widely accessible translation tool that is currently available, Bellos (2012) points
out that, due the enormous amount of data it relies on, Google Translate is an exceptionally well-developed translation tool
that quite frequently produces acceptable target texts. However, he also points out that the language pairs that Google
Translate handles rely heavily on what translations have been produced prior to a new translation attempt. This means that
Google Translate draws cross references between two corpora that of the source language and that of the target language.
The more similar texts exist in both languages, the more precise the translation can be because it can utilise recurring linguistic bundles in both languages. This in turn means that Google Translate is probably much more reliable when it comes to
translations between the dominant language English and one that is also frequently used for the same purpose. In other
words, it seems more likely to produce an accurate translation between English and French academic texts, both languages
that have a long tradition of academic genres and a long history of translated seminal works, than to produce an accurate
translation between English and a language that has not had such an extensive academic text production and exchange.
Austermhl (2011) discusses the issue of the quality of MT in some depth and from a number of perspectives. He described
the idea that MT is not aiming for high quality translation. Instead, it is aiming for usable translation. He also suggests that
the perceived rise in the quality of machine translation is in fact due to a lowering of expectations, suggesting that the
foreseeable future lies in machines and humans working together in order to create polished stretches of the target language.
However, he does not take into account the situation of many students studying in a medium where English is not their rst
language they understand that their own English is not of a high quality, and may believe that using a free translation
service such as Google is the better option for them, despite the drawbacks.
On the other hand, Google, with some justication, claim great success for their translation engine. In a recent blog post
(Google, 2012), they claim that they translate in any one day the same amount of text found in 1 million books. Put another
way, the output of the translation engine in one day is equivalent to the output of all the worlds professional translators in a
year thus becoming responsible for the majority of the worlds translations. However, it is not clear whether this translation
is receptive (in order to understand something in a foreign language) or productive (in order to generate something in a

M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121


foreign language). If a user is using the tool to read something in a foreign language, there will be a far greater tolerance of
error in terms of sentence structure. This is because the emphasis will be only on personal understanding of the text.
However, if the user is expecting the program to create text to be read by the reader of the target language, there is likely to be
a greater need, or at least expectation of, grammatical accuracy.
This is the reason why monitoring developments in the accuracy of Google Translate is so interesting and certainly a
valuable aspect of language teaching. It seems likely that one of three things will happen. Either English (in one of its
standard forms or in a lingua franca form) will be used increasingly as the primary language of communication in academia,
which means that there will be a growing demand for EAP. Or there will be a stronger reliance on local languages to deepen or
establish local academic traditions and identities. In the latter case, the use of communication aids is likely to become more
widespread, which in turn would mean that Google Translate would become increasingly reliable making it more attractive to
EAP learners. In a third scenario, a different language would take over from English as an international lingua franca, again
increasing the likelihood of machine translation being used.

3. Methodology
3.1. Analytical framework
What is arguably of the greatest interest to the EAP community of practice is whether or not Google Translate is at present
able to accurately render a source text into the target language of English. To assess this, two approaches come to mind. One
proposed by Colina (2009) investigates the issue from a perspective of translation quality. This approach has been previously
employed by van Rensburg et al. (2012) to investigate the quality of translations by Google Translate of different academic
genres from Afrikaans into English and from English into Afrikaans. However, van Rensburg, Snyman and Lotzs study
focussed on the translation competence of Google Translate, and therefore on the overall quality of the translation, while the
present study is more interested in the linguistic accuracy of the translation product. Hence, Colinas assessment model does
not seem appropriate for the present purposes.
We will instead make use of the taxonomy of error types introduced by Ferris, Liu, Sinha, and Senna (2013). This seems
more suitable, as we intend to treat the translation like we would treat student writing in English, for this is how a text would
be treated in EAP contexts. Ferris et al.s taxonomy (see Table 1 for a more detailed description) allows for detailed error
analysis and thus a measurable assessment of the translations linguistic accuracy. This will in turn allow for an assessment of
its acceptability as an EAP student text. It will also allow for conclusions to be drawn regarding which areas already show a
degree of sophistication and which the translation software at present does not seem to be able to cope with.
Empirical research that subjects Google translations to analysis of specic linguistic features, to date, is hard to nd. Most
studies seem concerned with an assessment of the translation quality and efciency. That is why most studies seem content
with more general ndings and comparisons between human and machine translations (e.g., Aiken & Balan, 2011; Garcia,
2010); yet, it has been demonstrated that Google Translate seems to produce more accurate results when working with
European languages, while it seems to struggle with Asian languages (Aiken & Balan, 2011). Research that investigates MT

Table 1
Error coding (adapted from Ferris et al., 2013).


Example from translations


Verb tense
Verb phrase
Word form
Word order
Wrong word
Word choice
Sentence structure
Missing word


Pronoun reference unclear

Pronoun incorrect
Run on





one that measured the level of ability in several ways

the individual who failed the exam ignored or looked down upon by society
A student test detects only the ability to say yes or memory
same way to learn the memorizing
Examination, especially in Malaysia plays an important role
Activities such as off-site is very dominant
Abuse and misunderstanding among students on examinations should be eliminated
students will focus on such topics only
Support parents and teachers are required so that they can be overcome
Examination is considered something very high
Learning aspects such as, music and art, can not be measured
How can the ideological principles Specically implement them?
Third, teachers and students too expect students exam results
This result is that parents do not ignore and less affection on them
First, the examination has been highly benecial to students but also students to study a
topic that will be tested only on the exam
and students will focus on such topics
Teachers will also place high expectations on him
I believe that in order to test the ability of the method to detect the candidates more harm
than good, in other words, the examination system is not a good way to test students abilities
In addition, people who have a bias to the students who got poor marks from students who
get higher scores.
College entrance examination system for screening system, especially in the eyes of their talents


M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121

from the perspective of English language acquisition and EAP is, to our knowledge, not available, and more intricate
frameworks that look at sentence level accuracy, that is those that would lend themselves to be used as marking/feedback
criteria for second language text production, as utilised in the present study, do not seem to have been used in MT research.
Of course, we expressly recognise here that EAP deals with much more than sentence level accuracy. Indeed it would seem
that the focus of EAP lies elsewhere, namely on discourse features, considerations of sociolinguistic competence with regard to
different disciplines and the fostering of autonomous learning strategies (Alexander, Argent, & Spencer, 2008; Bruce, 2010).
However, we argue that linguistic accuracy will always factor in the clarity of student writing. It is for this reason that the
present study investigates linguistic accuracy in such detail, so as to be able to make inferences about the translations overall
success. It is also in the areas of lexical and grammatical accuracy that we anticipate the most impressive advances in translation
technology in the near future because it seems easier to establish equivalencies at word, collocational and sentence level across
languages based on corpus data than to establish metadiscoursal equivalencies, provided they even exist in a language pair.
Thus, it seems appropriate, given the present state-of-the-art of machine translation, to investigate text at this basic level.
3.2. Study procedure
The students involved were enrolled on a pre-university foundation course at a UK branch campus in Malaysia. The course
balances input in the area of EAP and also more content-based social science modules. The emphasis of the EAP modules is on
discourse level content, rather than sentence level grammar.
After ethical approval was granted, the students were asked to write a short essay in their rst language with the title,
Exams are the best way to assess students: Discuss. The essays were written in their own time and then emailed to the
researchers, who translated them through the Google Translate interface. The students were asked to write as they would at
school not to try and apply what they had learned in previous EAP classes. This was done so that the students would not try
to hybridise their writing, using, for example, organisational patterns they had been taught as part of EAP, but not in their
rst language. This exercise took place in August 2013. In total ve scripts were translated, made up of 1,523 words in Malay
and 744 words in Chinese.
Certain words were not translated by the program. In such cases a native speaker of the source language was consulted. If it
was a matter of a clear typographical error in the original, then the original was changed to the correct version and the text
was re-translated. The three instances where this did not solve the issue probably illustrate lexical limitations in the program.
The translations were then independently coded for errors by two researchers, and an agreed count of errors was arrived at.
Agreement between both researchers was 92 per cent on average. Both researchers are experienced tutors of EAP, both in the
UK and overseas. Both hold Masters degrees in a related eld and are actively engaged in assessing EAP courses.
The errors were coded using an adapted model from Ferris et al. (2013), which can be seen in Table 1. This framework was
chosen because it was based on EAP corrective feedback in other words, areas that experienced EAP practitioners deemed
important when correcting student writing. Therefore the focus was on the communicative effect of the writing.
Table 1 shows the coding categories, and examples from the translated scripts. The framework has been changed slightly.
In the original there were two categories, one for run on sentence and the other for comma splice. This model only uses run
on sentence for the sake of simplicity, as, arguably, the two are quite similar. In addition, a category of unclear was added, in
case the intended meaning was impenetrable.
4. Results
The results were mixed. Table 2 shows the number of errors in each script.
It is clear from the data presented in Table 2 that the translation program was more accurate when translating Malay than
Chinese, with roughly half the number of errors per 1,000 words, (64.3 and 134.4, respectively). It is worth noting that the
number of errors is lower than the number of sentences in Script 2, meaning that the translation program was able to
translate some sentences without apparent error.
Table 3 shows the aggregated counts between the two languages that were measured. It clearly shows that the program is
more able to produce grammatical, correct English from Malay than from Chinese. In addition, it shows that the Malay scripts
were averaging fewer than one error per sentence, thus suggesting that the program is able to produce intermittent stretches of
grammatically correct English. One possible reason for this is the relative strength of English in Malaysia there is a larger
number of documents in both Malay and English online leading to a larger pool of potential translations for the Google engine.
Table 2
Errors per script.





Errors per 100 words

Errors per sentence








M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121


Table 3
Aggregated error counts by language.
Source language

Total words

Total sentences

Total errors

Errors per 1000 words

Errors per sentence







In terms of the classications of errors counted, Table 4 shows some interesting results. Firstly, the most common forms of
error were to do with word choice, followed by sentence structure and missing words. This would suggest that the program
struggles with the subtle distinctions of meaning between words in the target language as well as the original language. In
addition, it seems to be unable to parse certain structures of the original language effectively and may be left to translate word
for word, which can lead to a lack of clarity in the text produced.
Table 5 provides evidence from the scripts of instances where Google Translate struggled to render the text appropriately,
illustrating clear limitations. The sample error categories chosen here are arguably those that can signicantly affect reader
comprehension; and, thus, the lack of translation accuracy can have a signicant impact on the quality of the target text.
However, it is also interesting to note some of the areas where the program did relatively well. There were virtually no
spelling errors this is not surprising, given the dichotomous nature of spelling accuracy, and the lack of ambiguity involved.
What is more interesting is the fact that areas like pronoun use, preposition use and article use were handled well by the
translation software. These are areas with which EFL learners often struggle. Also, for example the article as it is used in
English does not exist in Malay or Chinese. It would thus seem that the software was here able to compensate such syntactical
gaps between the languages.
It is also illuminating to look at some of the sentences that the program produced which are not only grammatically correct
but also written in a convincing academic style, as illustrated in Table 6.
These sentences clearly show the ability of the program to translate relatively long stretches into clear and formal English.
However, also apparent is a certain cultural imprint in the discourse the writers produced with the help of the translation
engine. For example, emphasis on the importance of parental support to enhance examination results is more likely to be seen
in a Confucian educational context than a Socratic one. In addition, the Chinese writer of script 4 started the essay with a
discussion of the Tang Dynasty, and how their inuence was still apparent in modern-day China. This supports the idea that
the program is at present unable to translate beyond the sentence in other words, to deal in a meaningful way with
concepts that may be culturally or generically inappropriate to the target eld of writing.
Overall, then, the Google translation service was able to produce stretches of clear and accurate English. However, this
accuracy was patchy, and at times led to a breakdown of clarity.

5. Discussion
Even though Google Translate was able to translate a number of stretches of Malay or Chinese into grammatically correct
English, it was far from producing overall convincing results. There were errors in all areas of Ferris et al.s (2013) taxonomy,
and the majority of these were to do with sentence structure and word choice. The scripts, while communicatively relatively
Table 4
Error classications.
Student script
Missing word
Pronoun incorrect
Run on
Sentence structure
Verb phrase
Verb tense
Word choice
Word form
Word order
Wrong word
Pronoun reference unclear
Grand total






Grand total














M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121

Table 5
Samples of awed translations.
Error category



Word choice

Therefore, this forum would like to discuss the disadvantages of a comprehensive examination.
NB: The purpose statement of the text
. such a large population base in the kingdom of exam system is a more equitable and appropriate
educational detection system.
NB: One might speculate that kingdom could be replaced by realm and detection might be replaced
by assessment however, the software opted for other choices.
From educational institutions to the religious institution, one that measured the level of ability in several
ways, for example, examinations.
NB: While there is more than one issue in this sentence, the main one seems to be that it is an incomplete
structure without a main clause.
. some talents and abilities is not possible in an exam in which can be tested.
NB: Apart from the issue in subjectverb agreement, the use of the relative clause affects the
sentence structure overall.
However, no doubt slip showing the performance of the examination and an important achievement
as a measure of credibility in the workplace and so on.
In the extreme case, but even common sense are also ignored, focusing on university students lack
knowledge of life news uncommon, students lose your wallet cry by reminding parents to go rst to
know of cases is not just a special case of an alarm

Sentence structure



Table 6
Examples of accurate translations.


For example, the teacher will inform the students about the topics that will be tested in the exam and students
will focus on such topics only.
There are times when students pay bribes to teachers to pass exams.
Finally, moral support from parents can give students a passion for outstanding results in examinations.
Intelligence or excellence of a person can not be assessed through exams.
Therefore, people should change their perspective and open their minds to look from another angle.
Overall, the authorities should recognize the seriousness of this problem.
On the other hand, the emperor also made use of the examination system to complete his statist ideology [.]
[.] in other words, the examination system is not a good way to test students abilities.
[.] the stage is a dancers platform to express their talents and abilities


clear, were riddled with error. The translation program, then, is not yet able to be used as a substitute for a professional
translator or language prociency. The level of error is far too high to create what could be called a polished or professional
standard of language.
However, it is important to bear in mind that, while universities would normally expect their students to strive for
excellence in their skills of expression, they are also pragmatic understanding that grammatical accuracy supports
communicative competence, but does not dictate it. In their public descriptors, IELTS denes a 6.0 in their grammatical
category as makes some errors in grammar and punctuation but they rarely reduce communication (IELTS, n.d.). It denes
8.0 as the majority of sentences without error showing that even at a very high level, IELTS would tolerate certain levels of
grammatical error. Therefore, it is important to see these results not through the lens of a native speaker. Rather, we should
see them in comparison with an academic writer who is on the margins of acceptable competence. The question is whether
the translate program can perform as well as, or better, than a student who has just managed to scrape though IELTS.
The answer is from a grammatical point of view no longer unequivocally no, but also not yet denitely yes. There is
clearly a long way to go before the program can create long stretches of accurate text, and a large number of errors still feature
in the translations. The evidence presented in this paper shows that there is a difference in accuracy when Google Translate is
used in Malay and Chinese, and it is not unreasonable to assume that other languages will show different levels of accuracy.
However, it would be short-sighted to deny the future possibility that this or other translation programs will be able to
translate between languages to a level with greater accuracy than a large number of students who are studying in their second
language. Therefore, and again from a purely grammatical point of view, such students would be better off translating their
work through Google. A student would have no incentive to go through the long and frustrating process of writing in a foreign
language and then proofreading the text.
However, this approach would not necessarily help the student in the long term. Google Translate is (currently) only able
to translate at a lexico-grammatical level. The program is unable to help students align their writing to the norms and expectations of the wider discourse community. The translation program operates at a level of shallow literacy, not deep
literacy (Davies, 2007). There are many other issues from basic organisation and avoidance of fallacy to hedging and the
epistemological balance of academic writing for different disciplines which Google Translate is simply unable to deal with in
its present form. Thus, as commonly agreed in the literature (e.g., Austermhl, 2001; Kirchoff et al., 2011; Koehn, 2010; Quah,
2006), the use of MT necessitates post-editing to achieve truly high-quality outcomes.

M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121


It is also important to bear in mind that there were certain limitations in this study. Chief among these is the limited
amount of data analysed. It is important to analyse the output of a larger amount of translated text in order to discover more
widely generalisable patterns. In addition, a framework specically designed for this analysis would have allowed more
precise analysis of the output. For example, a framework where the type of sentence structure errors could have been coded
would have provided a more nuanced analysis.
6. Implications and perspectives
It seems counterintuitive to the EAP practitioner to welcome machine translation tools into their territory, as these could
be perceived to undermine the very act of teaching and learning a foreign language. The act of translation and bilingual
teaching is not very common in EAP, where student cohorts usually consist of members with a wide variety of cultural and
linguistic backgrounds. So the rapid developments in machine translation and their potential use by EAP learners might
indeed be perceived as a threat to what is currently held as best practice in EAP teaching.
Certainly, such concerns are not unfounded, as the sanctioned use of translation tools may undermine the actual language
acquisition process or even the need to learn another language in the rst place, potentially leaving examinations that forbid
the use of such devices the main incentive for students to actually learn the language. Reading and writing could be conducted
in L1, as translation services could well extend into the area of electronic books so that entire passages could be rendered into
an acceptable version in another language.
Before examining these issues in detail, it is worth considering the ways in which Google Translate may be used. Firstly, we
can divide the use into short and long stretches of text. With the short use, writers treat the translation engine much like they
would use a dictionary entering one or two words into the engine. The longer use involves entering stretches of text, from
paragraphs to whole essays.
The next distinction is between use for production and use for reception. In use for reception, a learner would take a text in
the target language, and translate it for his/her own understanding. For example, a Chinese student at a UK university might
take a journal article in English, and translate it into Chinese for easier comprehension. In use for production, the same
student would write an essay in Chinese and then use the translation engine to change it into an English essay for submission.
For short use, the translation work functions very much like a dictionary the weakness being that it only offers one
translation at a time, making the translation of an English word like bank highly problematic, whether it is translated into or
out of English (cf. Austermhl, 2001). A longer translation however, is likely to work better, translating for comprehension
since, as long as the student is able to understand, the grammatical errors are a distraction, not a major cause of failure
(Austermhl, 2001). It is at the point when the translation engine is used to create large stretches of text for assessment that
there will be an issue with grammatical accuracy and this may seriously affect the outcomes, given the likely impact on the
evaluation of the text. As we have seen, such an approach would be highly problematic for the student. While they might
produce a text with tolerable grammatical accuracy, their writing would be awed in a number of other areas. This is likely to
include the mechanics of referencing and citation, organisation, construction of the argument, and such like.
On the other hand, it seems highly likely that the grammatical accuracy of the translations will only improve in the future.
Simply by adding more texts to its database, Google is increasing the probability of each translation being correct. Thus, as we
move into the future, the grammatical accuracy of the translations will increase. This then leaves the EAP teaching community
with two choices: either to deny or to embrace this new technology. Obstructing it would, in our opinion, not be constructive.
Technologies are adopted or rejected by the users, and any authorities attempting to regulate their use in order to conform to
a previously held world view is bound to fail. This has been seen again and again, from the use of mobile phones in schools to
issues with IP and le sharing.
Certainly, a point to be made in favour of banning the use of tools such as Google Translate from EAP is that one of the
traditional core functions of presessional EAP is to develop and assess the students ability to cope with the language demands
of an English-speaking academic environment. Thus, one could argue that a translated text is no longer the students own
work. This is an argument not to be dismissed lightly. However, EAP seems to be experiencing a gradual shift towards deep
disciplinary literacy (as illustrated in practice, for instance, by the continuing rise of English for Specic Academic Purposes)
and away from the over-generalised teaching and assessment of allegedly universal academic language (e.g., through ad hoc
essays based on topics claimed to be of general interest a debate that goes far beyond the parameters of this paper). It
would thus seem that the intellectual achievement of demonstrating the ability to create a deep, logical and critical argument
(e.g., as usually manifested in written assignments as mode of assessment on EAP courses) may outweigh points such as
grammatical accuracy in assessing a students suitability to become a member of their intended academic community. This is
a point that may not be palatable to many EAP practitioners with a more traditional view of what should be assessed but we
must concede that the text (translated or not) is still produced and owned by the student, and is, thus, their achievement;
after all, in academic practice, we seem to be content to nd translations of academic research and philosophy into English
just as acceptable as reading the original without questioning the ownership of the text.
Whichever attitudes are adopted by the EAP teaching community, the scenario that seems to unfold is still a way off,
however. Machine translation will not replace language acquisition in the near future. Indeed, the teaching and learning of
academic language will not become superuous, as academic language is the means of communication and exchange between researchers worldwide, which involves interpersonal communication, coupled with crucial non-universal interpersonal and intercultural communicative competences that will be hard to develop for machines. Further, it would seem, as


M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121

Brown and Duguid (2000) point out, that the speed at which technological advances are developed will not coincide with the
speed at which these advances are integrated into teaching.
However, should the profession embrace the use of machine translation, it will bring a large number of exciting potential
developments. Firstly, the use of translation will allow a move away from the sentence level work needed in a large number of
EAP classrooms. Students will be able to express themselves, even if they have a very limited level of English. This will allow
EAP instructors to shift the emphasis away from the low level mechanics of English, and begin to focus on the deep literacies.
Concepts such as academic identities and construction of knowledge will be able to be embedded earlier in the curriculum,
leading to a more solid adoption of them by the students.
In addition, the translation engines might even be used as one of the growing number of tools in teaching the language.
Strategic students would want to check the output of the translations it is unlikely that they would want to place blind faith
in this output. As pointed out previously, machine translation in its current state is only able to produce texts of limited
quality. Thus, while they would not need to be able to generate such grammatically complex sentences, students will need to
be able to check it for accuracy, cohesion and quality of translation. This process could be exploited in the classroom to
enhance teaching and learning.
Further, even if MT were to reduce the numbers of those who want to acquire the English language as a means of
communication in their academic endeavours, it seems far less likely that MT will be able to cope with specic discourse
features across languages. It will not insert linking devices into a translation. It will not add hedging devices where the
writers L1 text is too assertive for the English context, and it will not rearrange text to full the expectations of Anglophone
academic tutors and publishers. While probably considered good style, for example, in commonly elaborate and elegant
French academic writing (Siepmann, 2006), to the English reader very elaborate sentence structures may appear convoluted.
MT would probably render such structures as 1:1 equivalent structure into the target language, unaware that this may not be
appropriate. Thus, even if MT nds acceptance in EAP practice, students will still require sophisticated knowledge of how
academic English works, deep literacy, as Davies (2007, p. 51) calls it, so as to be able to verify the suitability of their
translated texts and to appropriately amend them where necessary. This means that MT is not likely to threaten the status of
EAP; and it will not replace human input (Quah, 2006). Instead it may change the face of EAP and that is why a good
awareness of the tools at our students disposal is of such crucial signicance. It will allow us to stay abreast of technological
developments and the needs effective EAP teaching must address.
There is a parallel to be drawn with the teaching of mathematics and the introduction of the electronic calculator. The
calculator did not remove the need for the teaching of maths instead it allowed students to go further, quicker. It minimised
the need for endless practice of long division, and cumbersome and slow aids such as slide rules and logarithmic tables.
However, students and teachers use the calculator to assist the students fundamental understanding of maths, and their
ability to apply it in the real world.
It is for these reasons that we believe that the use of MT may fundamentally change the face of the teaching of English for
Academic Purposes. We predict that this will involve a shift away from sentence level grammatical teaching and a shift
towards the deeper understanding of the functions and metafunctions of language as used for academic purposes.
7. Conclusion
When considering the implications of Google Translate, we have made three assumptions. The rst of these is that the
grammatical quality of the translation will continue to improve, as the Google database grows. Secondly, we assume that it
will remain free at the point of access for the user and become available on an increasing number of platforms, from
computers to tablets and mobile phones. Our third assumption is that if the technology is available to, and useful for, students,
they will use it whatever their EAP instructors/counsellors advise.
The current study has been exploratory and small-scale in nature. However, with the above assumptions in mind, some
tentative conclusions, worthy of further investigation, can be drawn. It is clear that Google Translate has the potential to make
a large difference to the EAP teaching community. However, it is not to be compared to robots in factories making the factory
workers redundant. As long as we accept this technology and try to work with it, not against it, it has the potential to make the
teaching of EAP a much more exploratory and critical activity. This will free students and teachers to examine issues of
epistemology, and a deeper understanding of what academic expression actually is.

Adolphs, S. (2006). Introducing electronic text analysis: A practical guide for language and literary studies. London: Routledge.
Aiken, M., & Balan, S. (2011). An analysis of Google Translate accuracy. Translation Journal, 16(2). (online) Available at http://www.bokorlang.com/journal/
56google.htm (03.05.14).
Alexander, O., Argent, S., & Spencer, J. (2008). EAP essentials. Reading, UK: Garnet.
Austermhl, F. (2001). Electronic tools for translators. Manchester, UK: St. Jerome Publishing.
Austermhl, F. (2011). Of clouds and crowds: Current developments in translation technology. (online) Available at http://www.t21n.com/homepage/articles.
php (30.04.14).
Bellos, D. (2012). Is that a sh in your ear? London: Penguin.
Biber, D. (2006). University language: A corpus-based study of spoken and written registers. London: John Benjamins.

M. Groves, K. Mundt / English for Specic Purposes 37 (2015) 112121


Bowker, L. (2008). Ofcial languages minority communities, machine translation, and translator education: Reections on the status quo and considerations
for the future. TTR Traduction, Terminologie, Rdaction, 21(2), 15-61. (online) Available at http://www.erudit.org/revue/ttr/2008/v21/n2/037491ar.html?
Brown, J. S., & Duguid, P. (2000). The social life of information. Boston: Harvard Business School Press.
Bruce, I. (2010). Textual and discoursal resources used in the essay genre in sociology and English. Journal of English for Academic Purposes, 9(3), 153-166.
(online) Available at http://www.sciencedirect.com/science/article/pii/S1475158510000238.
Cancedda, N., Dymetman, M., Foster, G., & Goutte, C. (2009). A statistical machine learning primer. In C. Goutte, N. Cancedda, & M. Dymetman (Eds.),
Learning machine translation (pp. 1-38). Cambridge, MA: MIT Press.
Chapelle, C. (2003). English language learning and technology. Lectures on applied linguistics in the age of information and communication technology. Philadelphia, PA: John Benjamins.
Cobb, T. (2010). Instructional uses of linguistic technologies. Procedia Social and Behavioral Sciences, 3, 14-23. (online) Available at http://www.
Colina, S. (2009). Further evidence for a functionalist approach to translation quality evaluation. Target, 21(2), 215-244. (online) Available at http://www.jbeplatform.com/content/journals/10.1075/target.21.2.02col.
Costa-juss, M., Farrs, M., & Pons, J. (2012). Machine translation in medicine: A quality analysis of statistical machine translation in the medical domain.
Advanced Research in Scientic Areas. (online) Available at www.arsa-conf.com/ (Accessed 08.05.14).
Davies, C. (2007). What can technology do for/to English? In A. Adams, & S. Brindley (Eds.), Teaching secondary English with ICT (pp. 50-66) Maidenhead:
Open University Press.
Ferris, D., Liu, H., Sinha, A., & Senna, M. (2013). Written corrective feedback for individual L2 writers. Journal of Second Language Writing, 22(3), 307-329.
(online) Available at http://www.sciencedirect.com/science/article/pii/S1060374312000811.
Garcia, I. (2010). Is machine translation ready yet? Target: International Journal on Translation Studies, 22(1), 7-21. (online) Available at http://www.jbeplatform.com/content/journals/10.1075/target.22.1.02gar.
Google. (2012). Breaking down language barriers. http://googletranslate.blogspot.co.uk/2012/04/breaking-down-language-barriersix-years.html (03.02.13).
Hutchins, W. J. (2000). The rst decades of machine translation: Overview, chronology, sources. In W. J. Hutchins (Ed.), Early years in machine translation (pp.
1-16). Amsterdam: John Benjamins.
Hutchins, W. J. (2006). Machine translation: History of research and use. In K. Brown (Ed.), Encyclopaedia of languages and linguistics (2nd ed.), (Vol. 7, pp.
375-383). Oxford: Elsevier.
Hyland, K. (2004). Disciplinary discourses: Social interactions in academic writing. Ann Arbor, MI: University of Michigan Press.
IELTS (n.d.). IELTS Task 2 Writing band descriptors (public version) (online). Available at http://www.ielts.org/PDF/UOBDs_WritingT2.pdf (03.02.13).
Kaltenbacher, M. (2000). Aspects of universal grammar in human versus machine translation. In A. Chesterman, N. San Savador, & Y. Gambier (Eds.),
Translation in context (pp. 221-230). Amsterdam: John Benjamins.
Kirchoff, K., Turner, M., Axelrod, A., & Saavedra, F. (2011). Application of statistical machine translation to public health information: A feasibility study.
Journal of the American Informatics Association, 18(4), 473-478. (online) Available at http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3128406/.
Koehn, P. (2010). Statistical machine translation. Cambridge: Cambridge University Press.
Komeli, Z., Hendavalan, J. A. F., & Rahimi, A. (2011). An investigation of the translation problems incurred by English-to-Persian machine translations:
Padideh, Pars, and Google Softwares. Procedia Social and Behavioral Sciences, 28, 1079-1082. (online) Available at http://www.sciencedirect.com/
Mehta, N. K. (2012). Mobile phone technology in English Teaching: Causes & concerns. The Modern Journal of Applied Linguistics, 4(2), 82-92. (online)
Available at http://www.mjal.org/removedproles/2013/Mobile%20Phone%20Technology.pdf.
Newton, J. (1992). Introduction and overview. In J. Newton (Ed.), Computers in translation A practical appraisal (pp. 1-13). London: Routledge.
Oakey, D. (2010). Using corpus and web language data to create EAP teaching materials. In C. Ho, K. Anderson, & A. Leong (Eds.), Transforming literacies and
language: Multimodality and literacy in the new media age (pp. 167-184). London: Continuum International.
Peacock, M. (2002). Communicative moves in the discussion section of research articles. System, 30(4), 479-497. (online) Available at http://www.
Quah, C. K. (2006). Translation and technology. Basingstoke: Palgrave Macmillan.
Reid, J. (1992). A computer text analysis of four cohesion devices in English discourse by native and nonnative writers. Journal of Second Language Writing,
1(2), 79-107. (online) Available at http://www.sciencedirect.com/science/article/pii/106037439290010M.
van Rensburg, A., Snyman, C., & Lotz, S. (2012). Applying Google Translate in a higher education environment: Translation products assessed. Southern
African Linguistics and Applied Language Studies, 3(4), 511-524. (online) Available at http://www.tandfonline.com/doi/abs/10.2989/16073614.2012.
Scherf, W. (1992). Training, talent, and technology. In C. Dollerup, & A. Loddegaard (Eds.), Teaching translation and interpreting (pp. 153-160). Amsterdam:
John Benjamins.
Sheppard, F. (2011). Medical writing in English: The problem with Google Translate. La Presse Mdicale, 40(6), 565-566. (online) Available at http://www.
Siepmann, D. (2006). Academic writing and culture: An overview of differences between English, French and German. Meta: Translators Journal, 51(1), 131150. (online) Available at http://www.erudit.org/revue/meta/2006/v/n1/012998ar.html.
Silver, M. (2003). The stance of stance: A critical look at ways stance is expressed and modeled in academic discourse. Journal of English for Academic
Purposes, 2(4), 359-374. (online) Available at http://www.sciencedirect.com/science/article/pii/S1475158503000511.
Somers, J., & Lovel, H. (2006). Can AAC technology facilitate communication for patients with limited English?. (online) Available at http://personalpages.
manchester.ac.uk/staff/harold.somers/ESRCnal.pdf (30.04.14).
Swales, J. (1998). Genre analysis: English in academic and research settings. Cambridge: Cambridge University Press.
Vernon, A. (2000). Computerized grammar checkers 2000: Capabilities, limitations, and pedagogical possibilities. Computers and Composition, 17(3), 329349. (online) Available at http://ac.els-cdn.com/S8755461500000384/1-s2.0-S8755461500000384-main.pdf?_tida3331a86-43e6-11e4-830000000aacb35d&acdnat1411562038_854e326556c97a98368b7f20d5a5b368.
Michael Groves holds an MA in Applied Linguistics and ELT from the University of Nottingham. He is currently head of the Centre for English Language
Education at the University of Nottinghams Malaysia campus. He is particularly interested in investigating the development of students academic literacy,
autonomy, and how these are related.
Klaus Mundt holds a PGCert in Teaching EAP, an MA in Southeast Asian Studies and an MA in Applied Linguistics and ELT and has been involved in language
teaching for over a decade. He teaches on the insessional EAP programme and the MA CETI at the University of Nottingham.