Writing content with machine translation in mind

Language IO’s mission is to help monolingual customer support teams seamlessly communicate with their customers in over 150 languages over any channel, including chat, email, knowledge base articles, and social messaging. To achieve this, Language IO provides unique Machine Translation (MT) technology that gets you and your teams up and running quickly with accurate and secure translations.

One of the key factors to ensure the quality of Machine Translation is the quality of the source content. The purpose of this article is to walk you through some basic concepts to remember when as an agent, you write to your end customers via chat or mail, so that you can make the most out of Language IO. Indeed, as part of a customer support team that uses a multilingual platform, you must always keep in mind that you are writing content for an end customer that does not speak the same language as you.

In this article

• Tip 1 - Watch out for Spelling, Grammar and Punctuation Mistakes
  • Spell check
  • Capitalization
• Tip 2 - Keep it simple
  • One sentence per concept or idea
  • Do not sacrifice simplicity for politeness
  • Use active voice over passive voice
  • Limit the use of negations and double negatives
  • Do not mix different languages in communications
• Tip 3 - Avoid ambiguity
  • Avoid using words that can have multiple meanings in different contexts
  • Use complete sentences
  • Repeat nouns instead of using pronouns when it is needed to avoid ambiguity 
  • Use articles (a, an, the) to help MT engines differentiate between verbs and nouns
  • Use terminology consistently
  • Avoid slang
  • Avoid unnecessary abbreviations or acronyms
  • Avoid idioms
  • Write dates in full
  • Keep numbers and measurements clear and consistent
• Tip 4 - Recycle whenever possible
• Tip 5 - Be culturally aware
  • Be aware of the appropriate greeting for the culture
  • Be mindful of your audience
• Tip 6 - Be mindful of how you use emojis
• Tip 7 - Make sure that key elements remain untranslated

Tip 1 - Watch out for Spelling, Grammar and Punctuation Mistakes

MT engines occasionally struggle with spelling mistakes, typos, grammatical and punctuation mistakes. Often, engines can also be very sensitive to capitalization.

Spell check

In the example below, missing punctuation led to the translation “Take care of Régis”,  which was not the intent.

Capitalization

Make sure that you use the right capitalization, especially if you deal with names of services or packages that can also be regular words.

Suppose that you work with a package called "Now" or "NOW". This makes it tricky for MT engines to distinguish between "Now" the subscription package and "now" the adverb. In this case, you must always make sure the package name is capitalized correctly.

Tip 2 - Keep it simple

Be mindful of sentence length and structure

Sentences that are overly long or complex increase the risk of mistranslation. Use short, simple sentences connected by conjunctions when needed. If necessary, split a sentence into smaller sentences instead of using complex subordinate clauses.

Do not sacrifice simplicity for politeness

If you write in English, limit yourself to expressing politeness by using "please" and modal verb constructions, and not much else. Long, polite circumlocutions tend to not translate well.

Both examples below illustrate cases of literal and unidiomatic translation:

Use active voice over passive voice

End users and MT engines understand active voice better:

Limit the use of negations and double negatives

Sentences with multiple negatives are harder to process for both humans and machines. Use positive constructions whenever possible.

Do not mix different languages in communications

Mixing languages can result in unexpected outputs from the MT engine.

For example, in the scenario below, the English-speaking agent used the Italian word "Ciao" in a Spanish chat. Unfortunately, the MT engine mistranslated it as "Cow" in Spanish:

Tip 3 - Avoid ambiguity

Avoid using words that can have multiple meanings in different contexts

This may be tricky at first, but this approach can go a long way in making a message more unequivocal. 

For example, "opening" can have very different meanings depending on the context. If you use it within an HR domain, use the more explicit expression "job opening".

In the example below "base" was translated as "military base" instead of "database". Always write full sentences to avoid these types of mistakes.

Use complete sentences

Always write full sentences to avoid potential mistranslations.

For example, for MT engines, "A year free" can mean "an available year" in many languages, and not "a year for free":

Repeat nouns instead of using pronouns when it is needed to avoid ambiguity 

Any pronoun (such as I, me, he, she, it, they, us, we, our, etc.) can be ambiguous depending on how the sentence is constructed. To clear up any ambiguity, just replace the pronoun with a noun.

For example, in the scenario below, the pronoun "he" is ambiguous because you do not know who is getting back to you: this could refer to either Mark or Andrew. Replacing "he" with the name of the person who carries out the action removes the ambiguity:

Use articles (a, an, the) to help MT engines differentiate between verbs and nouns

In the example blow, the engine mistranslated "skip" as "to jump", but it translated "the skip" correctly as "the container", thanks to the use of the definite article "the":

Use terminology consistently

For example, do not refer to "file types" and "file formats" in the same communication.

Avoid slang

Do not use slang words or expressions such as (for example):

• “gotcha”
• “no worries”
• “hang tight”
• “you’re all set”
• “on it”
• “my bad”

Avoid unnecessary abbreviations or acronyms

Do not use abbreviations or acronyms, unless you added a list of these expressions to the glossaries.

Avoid idioms

Most idioms do not translate well into other languages and cultures, and it is best to avoid their use as much as possible. Instead, use easy-to-understand sentences.

Examples of common idioms in English may include:

• be on the same page
• be under the weather
• ring a bell
• compare apples to apples
• compare apples to oranges
• call a spade a spade
• call it a day

Write dates in full

Avoid using abbreviations that may not be universal and may not be used in other languages. For example, use "May 10, 2025" instead of "05/10/2025" or "10/05/2025".

Keep numbers and measurements clear and consistent

Always write numbers clearly and use standard units. Avoid mixing formats (for example, using “1,000” and “1.000” in the same conversation) to prevent confusion in different languages.

Tip 4 - Recycle whenever possible

• Use templates for your content whenever possible.
• Make sure that your content:
  • takes into account the multilingual support environment, and
  • follows the guidelines presented in this document.

Tip 5 - Be culturally aware

Be aware of the appropriate greeting for the culture

In some cultures, it is more appropriate to greet the customer using their surname rather than first name.

For example:

• In Japan, customers are usually greeted using the surname followed by the suffix "-san" (meaning "dear" or "honorable"). For example, "Yamamoto-san" instead of "Hi, Etsuko".
• In Hungary, customers usually prefer to be greeted more formally, using their surname rather than first name.
• As a rule, avoid using an end user's first name too much during the conversation: in many cultures, it may come across as overly familiar.

Be mindful of your audience

Avoid cultural or local references that may be alien to your end users. For example, wishing an Egyptian customer "Happy Thanksgiving!" is not appropriate.

Tip 6 - Be mindful of how you use emojis

Currently Language IO only support emojis when the message is in plain text. There is no emoji support if the text is HTML. Do not use or leave emojis in HTML source texts as this could lead to corrupted characters or formatting errors.
As a rule of thumb, environments that do support emojis are Chats & Messaging (no HTML). Environments that do not support emojis are Case/Tickets and Guide/Articles (HTML).

Tip 7 - Make sure that key elements remain untranslated

In some cases, you may need key elements of a message to be left untranslated. When this happens, Language IO can configure this option, by adding untranslatable elements to your glossary.

One way to mark elements as non translatable on the fly is to enclose the relevant text segments in double square brackets like this: [[untranslated text]]. 

In the first example below, you can see that when the postal address is not protected, the word ‘House’ from the address gets automatically translated into French. In the second example, ‘Be’ from the Norwegian name of the menu option “Be om kontroll” is interpreted to be an English word and is translated to Norwegian again. 

Commonly protected elements can include:

• Names and surnames of people
  • This is especially relevant when the name is also a common noun, such as Heather, Rose, Willow, etc.
• Names of companies, products, features or services
  • The best way to manage this is to add the list of relevant terms to the Glossary.
• Email addresses
• Postal addresses
• URLs