How to use Google Translate in the classroom
Machine translation: cause for concern?
The running joke has always been that MT is just five years away from being perfect and has been for the past fifty years. As the main article suggests, MT can get the main idea across more than 50% of the time for 35 languages, less than 50% of the time for 67 languages, and less than 1% of the time when neither language pair is English. The introduction of Google’s Neural Machine Translation (GNMT) in September 2016 may further narrow the gap between these inconsistencies (Turovsky, 2016). The updated system improves the former example-based system by learning over time and translating more holistically. Rather than individual parts of a sentence, the new system works by translating the whole sentence and working at a more semantic level which takes into account the wider context (ibid). Indeed, as the main article suggests, ‘Sometimes GT offers translations that are indistinguishable from those a person would produce’.
Consolidating this trend, research indicates a correlation between a student’s lack of ability and their recourse to MT, with blind marking suggesting superior results vis-à-vis work produced without MT (Garcia and Pena, 2011). In fact, the accuracy of MT is already snapping at the heels of the minimum entry-level required for international testing standards required to attain entry to university (Groves & Mundt, 2015). The American Council on the Teaching of Foreign Languages (ACTFL) lends support to this news by discussing technologically literate students as, “productive global citizens [who] use appropriate technologies when interpreting messages, interacting with others, and producing written, oral, and visual messages” (White and Heidrich, 2013: 230). Viewed from this perspective, then, and with the advent of earbuds and apps offering real-time translation, what is the incentive to learn a language when the technology will only get better with time? Does this suggest a crisis in the classroom and the death of the language teacher?
In order to answer this question, it is useful to examine one of the bedrocks of learning promulgated by Lev Vygotsky (Vygotsky, 1978); that of scaffolding. As the name suggests, this refers to helping students bridge the gap between what they can achieve on their own, and with the help of a more knowledgeable other (MKO), and various ‘tools’. The former is the dynamic between learner and MKO and termed ‘the zone of proximal development`. The latter could refer to technology such as a paper dictionary to more modern tools such as Google Translate. Scaffolding could be as quotidian an activity as a parent helping their child do their homework by offering just the right amount of information and support. In the modern educational setting, the application of a digital resource or tool such as MT could feasibly serve as a form of digital scaffolding as it builds on the foundations of the student’s existing knowledge and bridges the gap between what is possible alone and through the intermediary of an MKO. Where the metaphor of scaffolding possibly breaks down, however, is when students merely click a button and outsource the application of mental effort to Google. When this is removed from the process, we arguably remove any kind of cognitive engagement, transfer effect to spoken discourse, or real insight into the target language.
The implications of this are perhaps more lucidly explained when we take a look at a few examples of the kind of work written using MT. The first one is an essay recently turned in by a student which read as follows: “My future dreams are twofold. The first one is the multi-headed dog. In particular, I thought that I would not feel lonely, even when I was not at work, because of the fact that the meatballs and the texture were very pleasant”. When discussing this linguistic curio with a Japanese friend, he had a flash of insight which shed light on what might have happened. In Japanese, the word for paw is 肉球 which, when translated directly, produces meat (肉), and ball or sphere (球). While a quick check on my dictionary app gave the correct translation as ‘[sole of the] paw/pad’, the same word entered into GT produced ‘meat ball’. What this suggests to me, is that the student had failed to carry out the kind of post-editing which may have revealed this blunder. More importantly, and for the purposes of the present article, is the fact that GNMT was unable to read the context of the article to ascertain that the subject was not the texture of a foodstuff, but the texture of a dog’s paw.
To give a second example, I decided to use GT to translate a random extract of text from the internet (eiga.com). A quick search on a Japanese film review site gave the following extract discussing the new X-Men film: ‘The potential was one of the greatest Marvel heroes Jean, and I will release the strongest invincible Power by “Awakening” to the road of evil by the accident in space’. This example perhaps suggests that MT is somewhat pushing the envelope and may have met its match as inevitably as the enemies of the X-Men will by the end of the film. While not wanting to anthropomorphize the technology however, it would appear to be ‘struggling’ to ascertain who the ‘strongest invincible Power’ will be released by. A quick look at the original Japanese however offers an insight into what may have gone wrong – there is no pronoun to indicate who the subject is.
Such unfortunate gaffes are perhaps better illuminated when viewed in light of Malinowsky, an anthropologist investigating the culture of people living around Papua New Guinea towards the beginning of the 20th century. Malinowsky noted that a translation was meaningless without a consideration of both the ‘context of situation’ and ‘context of culture’ (in Coffin et al, 2010: 15). Malinowsky’s point was that if you are not socialized into the culture in which the exchange takes place, you will struggle to understand exactly what is happening. Around 100 years later, the claim made for GNMT is that it uses the broader context to help it work out the most appropriate translation. Despite this, it would still appear to have trouble ascertaining the subject of certain sentences when translating from Japanese to English.
Indeed, the kind of problems outlined above are not limited to X-Men reviews and student essays. I was recently tasked with checking the English content for a website which had been translated from Japanese using MT. On this occasion, the suggested translation offered the following: “The garden is around 1 hour and 20 minutes from Osaka, and you can also accompany your pets there”. The reason for this unfortunate wording is perhaps better understood when viewed through the lens of a forthcoming research project I completed earlier this year. In short, MT appears to have problems distinguishing the subject, or pronouns, of certain sentences when translating from Japanese. Japanese is a high context language stemming from the geographical determinism of its closed country policy extending from 1640 to 1854, and a mountainous landscape which has meant that people have traditionally lived in tight communities. For this reason, information such as 私 or ‘I’ is often omitted as it can be understood from the context. Perhaps because of this, MT can at times appears to lack the ability to determine who exactly is doing what to whom. The original Japanese, when translated into English yet keeping the same word order, would look something like this, ‘beloved pets can also be accompanied facilities are also there’. What starts out as a relative clause in Japanese, becomes a fairly coherent declarative sentence in English. Where it has obviously ‘had trouble’, is in determining the power relationship of the participants in the sentence. From a human perspective, the idea of humans accompanying their pets is obviously one that GT wasn’t able to determine given the information in the sentence.
Viewed from the context of this brief example, it could feasibly be argued that my role was to serve as the MKO and scaffold the gap between what MT can achieve on its own, and where it falls short given the fact that it hasn’t been socialized into the human world. The question then becomes one of examining whether MT can learn this fact about the world in the same way a child socialized into the world would do. John Searle used the Chinese Room thought experiment to underscore a similar point (Hodges, 2019). Searle asks us to imagine that computers using artificial intelligence have got to the point where they are able to pass the Turing Test (ibid). In other words, they can converse to such a degree as to convince a Chinese speaker that they are conversing with another Chinese native. Searle’s point is that the computer has no intrinsic understanding of the language it is producing and is akin to a person sat inside a black box, merely simulating understanding by sorting Chinese characters using a manual. In the current climate of students submitting work written in their native tongue before being machine translated, the metaphor has a further application. Perhaps employers hiring so-called ‘technologically literate students’ should take note that their dissertation on the X-Men series may hide a darker secret – that the author may merely be simulating understanding in the same way the man in Searle`s black box sorts characters according to a manual.
Reading the air
Translating a language can be difficult enough even when using a professional interpreter. As an example, grammatical metaphor (which differs from the colloquial form of metaphor) refers to the phenomenon where words are used in a non- literal or ‘proper’ sense (Halliday, 1985). As ‘native speakers’ of English, we would take for granted that an interrogative clause such as ‘do you think you might be able to do that by six?’ might perhaps be the more appropriate choice over the imperative ‘do that by six’ as the former is likely to result in a degree of resentment. While these kinds of rules are learned by being socialized into the world, they invite the question of whether MT may too be able to understand the difference between the literal and the figurative.
Japanese in particular, is a notoriously non-confrontational language, and the use of ‘難しい’ (it’s difficult) should immediately set alarm bells ringing for anyone engaged in business negotiations, as it is usually used to mean a flat ‘no’. In this example, grammatical metaphor is invoked to hedge the more direct declarative clause ‘that’s not possible’, to something more nuanced, indirect, and tacit. In this way, the term 腹芸 or ‘stomach art’, is used to refer to the skill of inferring meaning rather than having to spell something out explicitly. It is this tacit aspect of language – where being too direct can come off as curt and uneducated – which can cause confusion to even professional translators and can lead to whispers that you are a 空気を読めない人, or someone who lacks the ability to ‘read the air’.
Shariatmadari (2019) illustrates how Richard Nixon ran into this very problem when meeting the Japanese prime minister Eisaku Satō to persuade him to restrict textile imports. Satō’s response was to look upwards and say 善処します, which was translated by the interpreter as ‘I will do my best’. Shariatmadari however, contends that the phrase actually means, ‘no way’ ‘to most Japanese’ (ibid). Considering the fact that this ‘misinterpretation’ caused Nixon to brand Satō a liar after he did nothing, underscores the importance of context and multiple interpretations. Indeed, while Shariatmadari ‘s interpretation may be valid in some situations, a Japanese associate of mine suggests that it is extreme, and not practical in conversation. According to him, it is closer in meaning to, ‘I will take appropriate action soon for the matter with slight meaning that I can’t tell when and how I manage or accomplish it’ (sic). What should be clear however, is that without active engagement with the nuances and culture of the target language, we need to understand the context in which they are used. The question then, is to ask the extent to which we can trust MT when even humans fluent in either language struggle to determine the true intent behind an exchange which can lead to all out trade war.
The centaur language learner
In spite of these reservations, I would argue that the question of how to use MT depends on how you view it. For a tourist with no knowledge of the local language and little desire to learn it, it will arguably allow you to communicate further than the oft-quoted myth that 70% of communication is non-verbal (Keohane, 2008). For the student wishing to immerse themselves in the culture and language of the target language however, merely copying and pasting your essay into Google Translate and submitting without any kind of reflection is perhaps not the royal road to language learning. In order to examine this question then, it is perhaps useful to draw parallels with the world of professional chess. One version of the modern game which players can agree to play is known as cyborg chess, centaur chess or Ivanov chess. Rather than blindly trusting the chess algorithm, moves are made after examining and scrutinizing the possibilities it offers. Whether the algorithm constitutes a more knowledgeable other is debatable. However, it should be pointed out that the program Deep Blue was able to beat the world champion Gary Kasparov in 1997 after he claimed to be so confident that he rejected a 60-40 split of $500,000 between winner and loser for an all or nothing tournament (Latson, 2015).
In this way, the idea of the language learner as a kind of centaur holds water as long as one key element is adhered to – that translations offered are scrutinized and accepted as skeptically as the centaur chess player. In order for any meaningful kind of scaffolding to take place, the language learner must also learn to read the air, consider the context of culture and situation, and the tenor of the participants. Researchers such as Lee (2019) have used the approach of getting students to first write a draft in their own language, translate it using their own knowledge of the target language, and then use MT as a comparison text to help them compare, contrast and edit their final version further. Lee found that this technique allowed for a reduction in errors, helped students think of writing as a process, and to view MT positively. My own view on this approach is that it would allow the teacher to avoid the ‘lazy’ approach of students simply clicking the mouse button to translate, before submitting without a moment’s reflection. There has been many a time where a student has read out a presentation as though they are seeing the English for the first time. This is the wrong way to use MT.
In conclusion then, the question of how to use MT becomes clearer when we consider the extent to which we want to immerse ourselves in the study of the target language. For the traveller heading to Japan for two weeks who barely knows more than konnichiwa, I would say embrace it and make the most of it. For those who want to enhance their own knowledge of the target language and have a second opinion, use it with a pinch of salt.
Coffin, C. Lillis, T. O’Halloran, K. (2010) Applied Linguistics Methods A Reader. Routledge.
Garcia, I., & Pena, M.I. (2011). Machine translation-assisted language learning: writing for beginners, Computer Assisted Language Learning, 24:5, 471-487, DOI: 10.1080/09588221.2011.582687.
Groves, M., & Mundt, K. (2015). Friend or foe? Google Translate in language for academic purposes. English for Specific Purposes, 37, 112–121. Available at: https://www.sciencedirect.com/science/article/pii/S088949061400060X#!). [Accessed 12 July, 2019.]
Halliday, M. (1985). Introduction to functional grammar. London: Arnold.
Hodges, A. (2019). Available at: https://www.turing.org.uk/scrapbook/test.html [Accessed 15 July, 2019].
Keohane, K. (20018). Debunking the “communication is 70% nonverbal” myth… WordPress.com [available at]: https://kevinkeohane.wordpress.com/2008/09/03/debunking-the-communication-is-70-nonverbal-myth/ [accessed 10 July, 2019].
Latson, J. (2015). Did Deep Blue Beat Kasparov Because of a System Glitch? www.time.com Available at: https://time.com/3705316/deep-blue-kasparov/ [accessed 9 July, 2019].
Lee, Michelle. (2019). The impact of using machine translation on EFL students’ writing. Computer Assisted Language Learning. 1-19. 10.1080/09588221.2018.1553186. [accessed 17 July, 2019].
Shariatmadari, D. (2019). Language wars: the 19 greatest linguistic spats of all time. TheGuardianOnline. [Available at https://www.theguardian.com/science/2019/jun/17/language-wars-18-greatest-linguistic-spats] [accessed 22 June, 2019].
Turovsky, B. (November 15, 2016), “Found in translation: More accurate, fluent sentences in Google Translate”, Google Blog, [accessed 15 July, 2019.]
Vygotsky, I. (1978). Mind in society: Development of higher psychological processes. Cambridge. Harvard University Press.
White, K. D., & Heidrich, E. (2013). Our Policies, Their Text: German Language Students’ Strategies with and Beliefs about Web‐Based Machine Translation. Die Unterrichtspraxis/Teaching German, 46(2), 230-250.