In Part 1, we examined the treacherous waters of relying on machine translation. But is there a way forward when it comes to using machine translation effectively? After all, many companies turn to machine translation for very understandable reasons, such as the desire to fit a tight budget or when faced with overwhelming amounts of text to translate in a short time.
Below, let’s take a look at several ways to engage with machine translation responsibly, while minimizing errors and without sacrificing accuracy.
When to use machine translation?
As discussed in the previous blog post, machine translation is not recommended for high-stakes environments such as hospitals, courtrooms, or anything involving tense international politics. Machine translation also tends to struggle with less widely-spoken languages.
Software’s inability to capture the nuances of tone, culture, humor, and colloquial language means machine translation is not ideal for industries such as advertising or marketing, which abound with slogans, slang, and quips. In contrast, MT can be best suited for translating large quantities of technical documents or when speed, rather than audience, is a priority. For example, machine translation might help with technical manuals, internal communications, or texts with a short life span and relatively minor impact, such as online product reviews.
Develop clear policies for using machine translation
It’s not just a question of when to use machine translation, but a question of how.
First and foremost, it’s essential to devise transparent policies for selecting and using machine translation software. Although translation apps proliferate on the market, there is little to no guidance for consumers when it comes to evaluating whether a certain software lives up to its claims of accurate translations. As a result, organizations that really shouldn’t depend on MT—such as law enforcement agencies or border patrol—find themselves wielding gadgets that haven’t been rigorously reviewed.
Industry experts have suggested that more education about machine translation is necessary, not only for companies but for regular users. Research groups such as the Machine Translation Literacy project at the University of Ottawa, for example, seek to inform the public about what MT software can—and cannot—realistically do for them. Translators have also suggested some kind of international evaluation board to help set metrics and assess the quality of machine translation software, similar to the ISO that certifies professional translation and interpretation providers.
Human input is a necessary complement
All said, the best practice for using machine translation comes back, in the end, to human translators—in particular, to human post-editing of machine translation.
In post-editing, human experts start with translations originally generated by software. The post-editor will compare each segment of the MT output to the source text and use their linguistic expertise to assess the output’s quality. Poor quality output may be deleted and re-translated from scratch by the post-editor; acceptable output may be allowed to remain, but further improved in terms of grammar, style, and accuracy. In general, human post-editors provide the grease, and polish, that can turn a clunky, erroneous machine translation into a text that flows smoothly and naturally.
Plus, another key element of human post-editing is that post-editors’ input helps improve the machine translation process for the future. Neural networks learn as they grow, so a post-editor’s input can set the stage for better translations going forward.
So, how should you avoid the perils of machine translation?
Be aware that machine translation isn’t the right fit for critical situations, and be willing to seek the expertise of human translators and interpreters instead. Then, if you do need to use machine translation, consult with a professional translation company like Trusted Translations for advice on which MT software will work best for you as well as a range of post-editing services.