Skip to main content

Machine Translation and Teaching Writing

by Betsy Gilliland |

What Is Machine Translation?

Machine translation (MT) can be defined as the use of a computer to convert text from one language to another. Though electronic dictionaries have been in use for several decades, they generally serve the same purpose as a print dictionary, providing one or two words that directly translate from another single word. They cannot accommodate idiomatic usages or collocations. In contrast with electronic dictionaries, MT can process multiparagraph text and take into account collocations and other phrase-level structures.

Language teachers worldwide have noticed their students producing more and more sophisticated sounding texts, often well above the writers’ actual language proficiency. In many of these cases, the students have simply fed into an MT platform a text they wrote in their first language (L1) and then copied the output into their assignment. They may not even understand the text that was produced.

Some writing teachers have responded to recent developments in MT by resorting to handwritten assignments completed during class time. I believe that this approach is detrimental to our students’ long-term learning needs for writing in a second language (L2). Instead, as writing teachers, we need to help our students understand why they should develop skills for writing without assistance, as well as when and how it is useful to employ MT in their writing process. In this blog post, I focus specifically on the use of MT platforms for L2 writing. Other TESOL Blog articles have recently addressed other forms of generative AI, like ChatGPT (see previous TESOL Blog posts for discussions about ChatGPT).

Machine Translation Platforms

You are probably already familiar with at least one or two MT platforms, and you may already use some of them in your own work or leisure activities. In order of the number of languages included, here are a few of the better known and emerging MT platforms:

  • Google Translate is probably the best known MT platform, allowing translation among 133 different world languages (including many African and South Asian languages). It also integrates with a smartphone app that can translate text seen through the phone’s camera.
  • Yandex Translate claims to offer text-level translation for 100+ languages, including some indigenous languages of Russia and the former Soviet Union; it also provides a direct link to Google Translate, allowing users to cross-check translations.
  • DeepL can translate at the text level in 27 languages (mostly European, but also some more commonly spoken Asian languages). The tool also allows for translation of uploaded documents. A related DeepL tool provides corrective feedback on text written in English or German (this seems to be similar to Grammarly’s main function). Many reports say DeepL is the most natural sounding output for the languages it can translate.
  • PROMT.One says it can do text-level translation of 20+ languages (although a quick test found it has extremely limited understanding of one of those languages, Uzbek, to the point that it couldn’t even accurately translate basic phrases).
  • Reverso offers text-level translation of 26 languages, mostly European. It appears to be similar in output to the other MT platforms listed here.

The latest generation of MT platforms work with neural AI, based on an enormous corpus of text found on the internet. It translates by comparing thousands of documents and, according to Jolley and Maimone, “leverag[ing] artificial neural networks to teach itself to accurately translate entire sentences.” Google Translate shifted to neural MT in 2016, and users saw enormous improvements in the quality of texts, to the point that much of its output now sounds like natural human writing. In a post on the Bridge Universe site, Dorothy Zemach explains how MT works better for some types of text over others:

Both DeepL and Google Translate do better with what John McRae, past professor of Language in Literature Studies at the University of Nottingham, calls referential language, as opposed to representational language. Referential language is largely transactional. Information is requested or presented, and meanings are literal. Representational language, on the other hand, engages the imagination and makes use of figurative language.

Zemach explains that this difference means MT can be more useful when applied to texts like menus or websites compared with creative literature like poetry or novels.

Why Should We Be Concerned as Writing Teachers?

First of all, students are using MT, whether we allow it or not. Google Translate is now installed on over a billion mobile phones worldwide and continues to expand its data sources, accessibility, and languages. Other MT apps similarly are expanding their capacities and adding new languages all the time. They are increasingly accurate and natural sounding (although still not 100%). Two studies (Lee, 2020; Tsai, 2019) found that most of the errors in the L2 produced by MT were actually due to errors in the L1 text (spelling and punctuation), suggesting that MT is more accurate than it might seem.

One Thai colleague told me about how her students use Google Translate on texts written in Thai, followed by DeepL to polish the English, instead of writing directly in English as directed by their course assignments. She said that given the effort her students invested in multiple rounds of MT, they might have just written and revised the text directly in English! As teachers, we need to be aware of what our students are doing and how we might work with them rather than just forbid MT.

Further, strategic use of MT can actually save time, depending on the task. In the real world, professionals are usually judged on the product they submit, not how they got to that point. If MT helps them create a more job-appropriate product, then they should learn how to use it smartly rather than avoid using it. For example, writing texts like emails that are quite formulaic and primarily convey information can be done much more quickly by writing in the L1, using MT to translate, and then proofreading (and postediting) the translation.

In the language classroom, MT into the first language can give readers access to information written in languages they don’t know or are not confident reading. In many academic situations, learners may need to access sources that are not available in their L1; they may also want to compare perspectives from authors writing in their L1 and in other languages. 

Ways to Incorporate Machine Translation in the Second Language Writing Class

MT can be used in many different writing class activities. Here are a few suggestions:

    • Pre-edit in the L1: Students write in their L1 with awareness of English grammar and text structure so that MT can be more idiomatic and accurate. A Japanese colleague told me how she teaches her general education technology major students to pre-edit texts they have written in Japanese. Based on their understanding of English sentence structure, they write Japanese texts that Google Translate can turn into smooth-sounding English. She acknowledges that her students are unlikely to need to write directly in English or translate without a computer, and as such, has worked to help them understand ways to maximize the efficiency and accuracy of their use of MT.
    • Postedit MT Output: During postediting, students review the MT output and revise it based on their understanding of English, evaluating it in terms of appropriate phrasing and word choice for the genre. Both phases are useful if the final goal is to produce polished, natural sounding texts.
    • Review MT Output for New Concepts: In reviewing MT output, students can identify a set of new lexical items and research them (looking up meaning, connotations, collocations, and other aspects of words or phrases). They can then make note of these in a vocabulary journal and try to use them in other writing tasks.
    • Utilize MT for Short Pieces of Text: MT can also be used as a tool with shorter pieces of text. When writing, learners can use MT to look up sentences and phrases they don’t know, rather than resorting to word-by-word translation through a dictionary (a less useful practice, because it doesn’t take into account the way language works in phrases and other idiomatic chunks).
    • Check for Clear Expression: MT can also provide feedback in real time when teachers or tutors are unavailable. Learners can compare what they wrote in the L2 with an MT translation into their L1 to see if it reflects what they meant, for example. Translating the L1 text back into the L2 can then give them something to compare with what they thought they had written and revise for clearer expression. They might also use these reverse translations to learn new structures or identify ways that learned rules have been applied.

It is nevertheless important to help students understand practical and ethical reasons why they should use MT with caution. Instead of assuming it produces perfect, idiomatic translations, they need to learn how to use it smartly and how to evaluate the accuracy and appropriateness of the output. In class situations, they need to understand why we want them to write directly in English rather than using MT — for learning and practicing structures, developing understanding of different registers and genres, or recognizing cultural ways of writing. Lessons might also help learners determine the advantages and disadvantages of using MT and to compare those with use of AI chatbots like ChatGPT, which generate text (but are still quite unreliable in terms of the veracity of what they produce).

We are still learning how MT works and where its limitations are. This seems like a perfect time to start discussing its potential with our students.


About the author

Betsy Gilliland

Betsy Gilliland is an associate professor in the Department of Second Language Studies at the University of Hawaiʻi Mānoa, where she teaches courses on second language writing, teacher research, and qualitative research methods to undergraduate and graduate students. Co-editor of the Journal of Response to Writing, she was chair of the TESOL Second Language Writing Interest Section (2019-2020) and has published in TESOL Journal, Journal of Second Language Writing, and ELT Journal, among others. She was a Fulbright Scholar at the Universidad de Atacama (Chile) in 2018.

comments powered by Disqus

This website uses cookies. A cookie is a small piece of code that gives your computer a unique identity, but it does not contain any information that allows us to identify you personally. For more information on how TESOL International Association uses cookies, please read our privacy policy. Most browsers automatically accept cookies, but if you prefer, you can opt out by changing your browser settings.