7 Tips to Create Machine-Friendly, Translatable Content for Customer Service

24. November 2020y

Machine translation has been a huge boon for customer service teams who have adopted it. Being able to communicate with customers in their preferred language demonstrates respect and care, and can be a valuable competitive advantage. A full 70% of customers in one survey said they are more loyal to brands that offer native language support, and 29% of brands have lost customers due to not offering multilingual support. 

Of course, hiring a full team of agents who speak every language of every customer can be near impossible. This is where machine translation algorithms can be so powerful. But machine translation will only be accurate and successful if the algorithm is fed clean, translatable text. 

There are certain nuances in diction, grammar, and style that vary across languages, dialects, regions, and cultures. This can impact how messages are written and how they are translated by machines. It’s important to understand the best practices of writing messages that can be easily and clearly translated by an algorithm. 

Attention to detail in customer service especially can make a huge difference. Here’s what to think about when writing customer service communications for machine translation. 

Stick to short, simple sentences

Because different languages have different grammatical rules and structures, it’s best to express just one idea per sentence. This will increase the odds that the translation properly tracks the original meaning. Sentences should not be longer than 30 words, generally speaking, and shorter is better. Short sentences can be parsed most easily by machines and are likely to lead to clear, concise translations. The simplest, most straightforward way of expressing an idea is the best path. 

Avoid idioms and slang

Many idioms do not have an equivalent in every language. When they are translated literally, they may be confusing and muddy the intended meaning. This is true with informal idiomatic language like the English phrase, “I got you,” meaning, “I will help you.” It’s also true with idiomatic expressions like, “It’s raining cats and dogs,” which does not mean anything when translated literally into most other languages. Stick to more formal and direct language and try to avoid slang at all times. 

Use active voice

Active voice is a best practice in general, but when it comes to machine translation it has an extra benefit. Active voice reduces ambiguity by making the subject and verb crystal clear. This improves the likelihood of getting meaning across clearly. So rather than writing, “The ball was kicked,” write, “John kicked the ball.”

Don’t use abbreviations

Abbreviations are difficult for machine translation systems to handle, since they do not provide the full context on intended meaning or may not have a translation in the algorithm if they are informal or slangy. Abbreviations include any shortened form of a word or phrase, like “IOU” for “I owe you” or “thru” for “through.” Whenever possible, write the words out in full to reduce translation errors or confusion. 

Use correct native grammar

Whatever language you are writing in, use appropriate grammar for that language and allow the machine to translate it to appropriate grammar for the next language. Be sure not to skip any words or add any unnecessary ones in. Re-read messages and ensure they make sense and read naturally in your language before sending them for translation. Pay attention to small details like extra or missing whitespaces, incorrectly used punctuation, and typos, which can trip up algorithms. It’s also key to double-check that words agree in number, gender, and tense. 

Format messages correctly

Extra line breaks and unnecessary paragraph breaks can confuse algorithms. For most short messages, there is no need to add in line breaks. Additionally, use correctly formatted bullets, not asterisks. 

Err on the side of formality and politeness

In many other languages, such as German and French, it is considered impolite to address someone you do not have a close personal relationship with in an overly informal manner. For this reason, in customer service, it’s often best to stick to a formal register. For example, use titles and polite greetings and closings like, “Dear Mr. Smith,” and “With kind regards,” rather than informal language like “Hey,” or “See you!” This blog post on tone of voice also shares some helpful tips on how to ensure that your polite intentions are translated effectively. 

Bonus tip: Follow the golden rule

This last tip isn’t about translation per se, but it’s key to positive customer service interactions. Treat customers the way you would want to be treated if you were in their shoes. This means every interaction matters, and you should strive to offer quality and clarity without sacrificing speed. A little bit of respect goes a long way, and “please” and “thank you” apply across any cultural context or demographic. 

The post 7 Tips to Create Machine-Friendly, Translatable Content for Customer Service appeared first on Unbabel.

About the Author

Profile Photo of João Graça
João Graça

João Graça ist Mitbegründer, Chief Technology Officer, und das Rechengenie hinter Unbabel. Der gebürtige Portugiese João studierte Informatik auf Doktorniveau an einer der angesehensten technischen Universitäten von Lissabon, dem Instituto Superior Técnico de Lisboa. Während des Studiums veröffentlichte er mehrere angesehene Arbeiten über maschinelles Lernen, Computerforschung und Computerlinguistik, die heute die Grundlage für die maschinelle Übersetzung von Unbabel bilden. Nach seinem Abschluss arbeitete João bei INESC-ID, wo er Forschungsarbeiten im Bereich der Verarbeitung natürlicher Sprache (NLP) durchführte. Anschließend absolvierte er sein Postdoktorat im Bereich NLP an der University of Pennsylvania. João erhielt ein Marie-Curie-Welcome-II-Stipendium verliehen (2011), das er aber zugunsten seines Einstiegs in die Geschäftswelt abgelehnt hat. Er arbeitete zusammen mit dem heutigen CEO von Unbabel, Vasco Pedro, an der Entwicklung von Algorithmen für sprachliches Lernen und Tools für das maschinelle Lernen und war in verschiedenen Positionen in der Forschungswissenschaft tätig, bevor er Unbabel 2013 mitbegründete.

DeutschFrançaisNLdanskSvenskaEnglish