Getting Started with Machine Translation

Tailoring MT to work with your content

Machine Translation (MT) has come a long way since the 1950s when it was first introduced. With the incorporation of artificial intelligence technologies, automated translations are now better than ever. However, that doesn’t mean that MT is suitable for every kind of material. In most cases, machine-translated content still requires post-editing by a human linguist in order to ensure its consistency and quality.

When to use Machine Translation Post-Editing (MTPE)

Machine translation is best suited to technical fields, especially legal, medical and scientific subject matter. These types of content tend to have predetermined structures, limited semantic domains¹ and repeated language patterns, all factors that contribute to efficient and accurate MT conversions. On the other hand, creative writing will result in literal translations as MT doesn’t recognize context, shades of meaning or nuance.

Not
Recommended
Recommended

Marketing material

News articles

Subtitles

Slogans / Taglines

Product descriptions

Recipes

Addresses

Instruction manuals

Medical records

Contracts

Weather reports

Of course, language pairs should also be taken into account, as common language pairs work better – especially when coupled with a translation memory. When working with pairs that use a different syntax or grammatical structure, you can often find unusual turns of phrase in the MT output.

Swedish
Portuguese
German
Source text
A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend. A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend. A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend.
Translation*
En stor asteroid som har sin egen lilla kompanionsmåne är inställd på att whiz över jorden i helgen. Um grande asteróide que tem sua própria lua minúscula está pronto para passar pela Terra neste final de semana. Ein großer Asteroid mit einem eigenen winzigen Begleitmond wird an diesem Wochenende über die Erde sausen.
Back translation*
A large asteroid that has its own little companion moon is set to whiz over the earth this weekend. A large asteroid with its own tiny moon is ready to pass Earth this weekend. A big asteroid with its own tiny moon will whiz across the earth this weekend.

* Translated using Google Translate API

Korean
Thai
Vietnamese
Source text
A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend. A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend. A large asteroid that has its own tiny companion moon is set to whiz past Earth this weekend.
Translation*
자신의 작은 동반자 달이있는 큰 소행성은 이번 주말 지구를지나 소변으로 설정됩니다. ดาวเคราะห์น้อยขนาดใหญ่ที่มีของตัวเองดวงจันทร์ของมันเล็ก ๆ ถูกตั้งค่าให้หวือที่ผ่านมาโลกสุดสัปดาห์นี้ Một tiểu hành tinh lớn có riêng trăng đồng hành nhỏ bé của nó được thiết lập để rít trên Trái đất vào cuối tuần này.
Back translation*
Large asteroid with his small companion moon is set in the urine over the weekend district. A large asteroid with its own moon, a tiny set to whiz past Earth this weekend. A large asteroid has its own small companion moon set to hiss on Earth this weekend.

* Translated using Google Translate API

In the case of Asian languages, machine translations often end up being very literal. This is because many of them are high-context, meaning that they require some “reading between the lines.” There are also variations in formality and tone that needs to be taken into account when translating certain Asian languages. While an experienced linguist would be able to respond flexibly to such needs, an MT program would simply replace the source text with what it considers to be the most suitable target equivalent. Furthermore, differences in syntax, punctuation and lack of clarity (for the software) can also complicate things.

The following is an example of a long, complicated sentence resulting in an unclear translation.

SOURCE TEXT
Summer in Tokyo tends to be hot and humid but it’s a great time to visit and experience the city like a local and enjoy traditional festivals and fireworks, try shaved ice, hit the beach, join friends for drinks at a rooftop beer garden, stroll the shady parks or chill out in the city’s excellent museums, indoor experiences and department stores.
JAPANESE TRANSLATION*
東京の夏は暑く湿気が多いですが、地元のように街を訪れたり、伝統的なお祭りや花火を楽しんだり、かき氷を試したり、ビーチに出かけたり、屋上のビアガーデンで飲み物を飲みに友人と過ごしたりしましょう。日陰の公園や市内の優れた美術館、屋内での経験、デパートでのんびり過ごしましょう。
BACK TRANSLATION*
Summer in Tokyo is hot and humid, but you can visit the city like locals, enjoy traditional festivals and fireworks, try shaved ice, go out to the beach, and drink with friends in the rooftop beer garden with friends Let’s spend it. Let’s spend a relaxing time in the shady parks, the city’s excellent museums, indoor experiences and department stores.

Compare the above to this next example. The source text has been edited for clarity, and the target has been reviewed by a linguist to ensure readability. The final translation sounds much more natural (e.g. “you can unwind” rather than “let’s spend a relaxing time”) due to these additional steps.

SOURCE TEXT
Summer in Tokyo tends to be hot and humid but it’s a great time to visit and Summer in Tokyo tends to be hot and humid but it’s a great time to visit and experience the city like a local. Enjoy traditional festivals and fireworks, try shaved ice, hit the beach, join friends for drinks at a rooftop beer garden, stroll the shady parks or chill out in the city’s excellent museums, indoor experiences and department stores.
JAPANESE TRANSLATION*
東京の夏は暑くて湿気がある傾向がありますが、地元の人のように街を訪問したり体験したりするには絶好の機会です。伝統的な祭りや花火を楽しんだり、かき氷を試したり、ビーチに出かけたり、屋上のビアガーデンで飲み物を飲みながら友達と過ごしたり、日陰のある公園を散策したり、市内の優れた美術館、屋内体験、デパートでくつろいでください。
MTPE
東京の夏は時に蒸し暑いですが、地元の人のように街を訪問したり体験したりするには絶好の機会です。伝統的な祭りや花火を楽しんだり、かき氷を食べてみたり、ビーチに出かけたり、屋上のビアガーデンで友達と過ごしたり、日陰のある公園を散策するのはいかがでしょうかまた、市内の優れた美術館、屋内体験やデパートでくつろぐこともできます
BACK TRANSLATION*
Tokyo’s summer is sometimes hot and humid, but it is a great opportunity to visit and experience the city like a local. Enjoy traditional festivals and fireworks, eat shaved ice, go to the beach, spend time with friends in the rooftop beer garden, or take a walk in the shady park. You can also unwind at the city’s excellent museums, indoor experiences and department stores.

* Translated using Google Translate API

What to expect

Machine translations are rarely perfect, let alone “ready for use”. Like any software that relies on data, it needs to be tailored and trained to work with your content. As discussed in the previous sections, it’s important to understand that the quality of the machine translation will largely depend on a) how suited the content is for MT but also b) how well the content is prepared. Make sure to also leverage tools such as glossaries and translation memories to streamline your MT strategy.

Pre-MTPE Checklist

Post-editing can help fine-tune machine translations, but depending on the quality of the translation, this may take as long as it would to actually translate the content from scratch! Rest assured, there are steps that you can take to get better MT results and save time in this step. Take a look at the following checklist to see what you can do to help your next MT project run smoothly.

  • Review the content
    Gauge how technical the subject matter is. Make sure the writing is clear and concise.
  • Assess the language pair
    Common language pairs with similar syntax and vocabulary (such as English – French) will have better results.
  • Build a glossary
    Identify proper nouns and industry jargon. Add variations of words/phrases to your glossary whenever possible (e.g. function, functional, main function, standard function).
  • Check for consistency
    Ensure that key terms match your glossary, and are used consistently through the file (e.g. “international flights” should be used throughout a file, instead of synonyms such as “overseas flights” or “non-domestic flights”)
  • Rewrite, remove, rearrange
    Don’t be afraid to shorten sentences or remove unnecessary expressions if needed.
  • Work with experienced linguists
    Make sure that the linguists handling post-editing are familiar with the MT process.

¹ Semantic domains are areas of meaning and the words used to discuss them. For example, the domain vehicle can include a range of words including car, truck, tank, road, motor, gasoline and transportation.