Google Translate may transform teaching of international students

Anyone who used Google Translate in its early days will no doubt recall the frustrating – if sometimes amusing – gibberish that it often generated. However, the platform has recently introduced significant improvements to the way it translates to and from English that could have far-reaching effects in academia.

The new neural network approach uses statistical probabilities based on huge banks of human-translated texts. It could become so good that it allows students at anglophone universities who are not native speakers to write assignments in their native tongues and then submit Google-translated versions in English. The coming perfection of machine translation also has the potential to bring about broader changes related to language teaching and learning.

According to the British Council, about 2 billion people will speak English by 2020, the majority of whom will be non-native speakers, helped to proficiency by an army of English teachers, estimated to number about 12 million. Naturally, before entering the profession, many prospective English-language teachers attend courses offered by universities and teacher training colleges on language teaching methodology and second-language acquisition. For instance, in a worldwide study carried out in 2014, Qing Shao and I found that there were well over 200 master of arts in teaching English to speakers of other languages (MATESOL) programmes. However, key components of these courses – namely, reading and writing – could be made redundant in the eyes of students if they know that, rather than incurring the effort and expense of learning English vocabulary and grammar, they can just use Google Translate instead.

In light of this, my colleague Becky Leung and I conducted a small-scale experiment in a Hong Kong primary school, which was published in the open access journal English for Specific Purposes. Briefly, one class of students wrote a composition in their second language, English; then, a few days later, the same students wrote in their native Chinese on the same topic. After digitising all the students’ scripts, we used Google Translate to generate English versions of the Chinese scripts. We then randomly shuffled the two sets of scripts together and asked a dozen English teachers – who were unaware of the experiment’s nature – to grade them for grammar, vocabulary and comprehensibility.

Confirming the recent advance in machine translation, the teachers scored the Google-translated scripts higher than those written in English. Moreover, in subsequent interviews, most of the dozen teachers we spoke to claimed to have had no idea that half the scripts had been machine-translated. The upshot appears to be that Google Translate has reached such a level of accuracy that it can generate reasonably good translations from Chinese to English, at least at a level normal for 11-year-olds.

We suspect that one of the reasons why the scores for the translated scripts were higher relates to the more sophisticated ideas the students were able to convey through their native language. We were able to isolate quite a number of instances where individual students had made the same point in Chinese that they had previously written in English. However, the Google version often proved to be more nuanced and to have fewer grammatical errors. Also, when writing in one’s native tongue, there may be less of a cognitive load. For example, a student wrote in English: “Students will get tired in class.” In the translated version, they wrote: “Students will doze off in class.” In another example, a student wrote in English “…use less cost for electric things”, while the Google-translated version from Chinese was “reduce spending on utilities”. We presume the latter versions of these examples came closer to what the students really wanted to convey.

Although our study was small and confined to a very specific age group and native language, it could be a precursor of things to come. Numerous studies have already shown that non-native English-speaking university students are using Google Translate for their assignments; as the service continues to improve, it is easy to predict that this usage will increase.

Students in places such as Hong Kong and Singapore, where English is taught as a second language, might still feel it is worth the effort to improve their reading and writing. However, students in countries such as China, Japan and South Korea, where English is taught as a foreign language (EFL), might feel rather differently given how onerous they often find learning a language that is very different from their own. Indeed, their reluctance to learn might not be confined to reading and writing. New apps that allow instant translation of spoken language have recently become available, too.

Nor is this merely a concern for universities and colleges that offer English teacher training programmes. All anglophone universities that have overseas students need to start seriously considering what all this means for their assessment practices. Will they still require applicants to attain certain IELTS or TOEFL scores? Will they accept Google-translated essays? Will they embrace seminars mediated by electronic voice interpreters? Will they permit to graduate individuals who lack the ability to understand what is written on their certificates?

No doubt the answers will partly depend on what the students themselves expect – which, in turn, will depend on how deeply Google Translate penetrates the modi operandi of international companies. But if the future does indeed involve humankind working ever more closely with computers, it might be that interpersonal connections in natural language become a skill for which market returns rapidly diminish.

Paul Stapleton is an associate professor of education at the University of Hong Kong.