Why You Shouldn’t Use Machine Translation for Research Papers (and What to Do Instead)

As technology advances, more and more ESL researchers are experimenting with machine translation tools like Google Translate and DeepL to get their papers published.

However, these same researchers are finding that these MT tools, on their own, are not good enough to produce papers that academic journals will accept.

Although machine translation tools have severe drawbacks, this doesn’t mean that authors who rely on them to translate their papers into English are doomed to face rejections from their target journals.

As a proofreader, I’ve seen hundreds of papers whose English versions were produced to some extent by AI, and their authors have succeeded in getting published in their target journals.

Read on to learn why your AI-translated papers are getting rejected and what you can do about it.

Machine Translation Shortcomings

MT tools can cause many problems when used to translate entire papers without the author’s input. But I will focus on the three most significant ones that I’ve seen at least once in the majority of papers I’ve edited in the last few years.

MT tools don’t know your paper’s context.

There is a difference between information (which MT tools have an abundance of) and knowledge (which MT tools lack). And because these tools lack the expertise of the authors who use them, they often default to phrases that are so generic that they don’t really mean anything.

For instance, I recently came across this sentence:

All the key phrases are generic: renewable energy technologies, sustainable agriculture, and broader goals of the Kingdom’s Vision 2030 initiative.

A human would have named the specific technologies, and they would have used a more specific term than “sustainable agriculture.” They also would have highlighted which goals of Vision 2030 the technologies would help achieve (and how).

MT tools make false claims.

Even worse than being vague, MT tools can change the meanings of some sentences, especially if they include terminology that has one meaning in your field of study and another, more general, meaning that you did not intend.

For example, according to a recent study, Google Translate made a significant mistake.

Because Google Translate assumed that “negative” carried its everyday meaning instead of its clinical meaning, the translation says the opposite of the author’s intended meaning.

MT tools have a weak writing style.

Journals expect authors to write formally and clearly. This is something MT tools often fail at, as they tend to throw in words and phrases that are too casual for a research article.

Among the most common informal phrases I’ve seen in papers produced using MT tools are:

• A little

• A lot

• And so on

• Basically

• But (to start a sentence)

• Could potentially

• Get (instead of “obtain” or “acquire”)

• Etc. at the end of a list beginning with “such as.”

• Nowadays

When you submit a paper that contains too many of these phrases, it can give the impression that you are unprofessional, which may lead to a rejection, even if your research would add value to the literature.

Alternatives to Machine Translation for Writing Research Papers

If you’re tempted to use a machine translation tool to translate your paper but are concerned about the negative consequences, there are several viable alternatives. These vary in terms of cost and the quality of the writing.

Professional Translation Services

This is the most expensive option, but it is the best choice if you’ve written your paper in full in another language.

Being human, professional translators can understand your paper’s context and will be familiar with your field’s jargon (or they can look it up if they’re not). Because of this, they can ensure that you avoid many of the problems that occur when you use machine translation.

To receive the best service possible, look for a translator who specializes in academic writing (and, ideally, in your subject), rather than business or marketing copy. Such a translator will be familiar with the structure and tone expected by academic journals.

Hiring an English Editor After Machine Translation

Many authors translate their papers using MT tools, then hire an English language editor to check and improve their prose.

This option is typically much more affordable than hiring a translator. For example, according to thewritingpoint’s Prices page, it can cost CA$1,118.84 to have a 5000-word paper translated into English.

By contrast, it would cost only CA$125.00 to hire an English language editor at Magnum Proofreading to revise the same paper after it’s been translated using an MT tool.

However, a language editor can’t fix every error created by an MT tool. For example, editors do not typically include fact-checking in their services. This means that if the meaning of a sentence was altered during a translation, an editor won’t restore its original meaning.

So, if you decide to go with this option, be sure to check your paper carefully before sending it to an editor.

Peer Review

The main benefit of this option is that asking your peers or a writing group for feedback on a translated paper costs nothing. And while they might not have the expertise of a native English speaker, they can at least point out confusing words and sentences that you may be able to fix without hiring a professional.

Of course, this option is only realistic if you have peers who are fluent in the English language and have the time to review your paper.

Closing Thoughts

It’s unfair when a researcher’s language skills prevent them from sharing their work. Though MT tools might seem like a cheap solution to this problem, they can’t replace human translators and editors.

For ESL authors on a budget, the best solution may be a combination of MT tools and professional or peer support.