How to improve Translation Output (Prevent, Spot and Correct Machine Translation Errors)

correcting Machine Translation outputs

Machine Translation (MT) has made significant strides in the past decade, largely thanks to the advancements in neural networks and Artificial Intelligence (AI). Translations are fast, clear, and, for the most part, highly accurate. Yet, as with anything, perfection remains elusive. The quality of translation output is closely linked to both the quality of your input and the training material used.

So, are you providing the correct training data? Is your input easy to interpret for the MT system? Given the nature of AI, some errors may still emerge despite ticking all the boxes.

This article explores different types of machine translation errors and what you can do to find and fix them. We also look at various tips for preventing them in the first place! 

Types of machine translation errors

There are different types of errors associated with machine translation. We give some common ones below.

Morphological errors

Morphological errors occur when the structure of words is incorrectly translated, including issues with tense, plurality, gender, or poorly used verbs, adverbs, adjectives, or pronouns. For example, HSBC bank’s marketing tagline ‘Assume Nothing’ was mistakenly translated as ‘Do Nothing’ in non-English speaking countries, confusing the customers.

Semantic errors

Semantic errors happen when the translation is technically correct but misses meaning or context. For example, the global slogan for KFC(a chicken food brand) is ‘Finger lickin’ good.’ However, the literal translation of the words in Chinese, without context, resulted in a sentence the audience interpreted as ‘We’ll eat your fingers off.’ 

Lexical errors

Most languages have words that spell/sound the same but have different meanings. Sometimes, they may be mistranslated out of context. For example, when a massive earthquake hit Indonesia, many people shared their statuses on Facebook as Selamat, which means “to survive.” The term selamat alternatively also means “congratulations.” Facebook algorithm misinterpreted the word and added balloons and confetti to the post, creating a lot of public anger.

Orthographic errors

These errors pertain to the incorrect spelling or use of characters in the translated text, especially applicable to languages with different scripts. For example, some official papers handed by Britain to Croatia misspelled the name United Kingdom as Ujedinjeno Kraljevstvo instead of Ujedninjena Kralijevina, causing political embarrassment.

Mistranslation of Key Entities

This error occurs when a key entity, like a proper noun, technical term, or specific jargon, is incorrectly translated throughout the document. For example, on a Spanish government website, a department head’s name, “Dolores del Campo,” was replaced by a literal translation—It is pain of field, causing the department to become a subject of ridicule.

Negation and opposite meaning

Sometimes, a misplaced or missing negation like no, not, neither, nor can result in incorrect and even opposite translations. For example, a road sign : “No west, one way only.” was mistranslated into Welsh as “Un ffordd i’r gorllewin yn unig” or “One way west only.” The missed no resulted in confused drivers going in the opposite direction!

Why do machine translation errors happen

With so many different types of errors, it may seem easy to blame the machine translation system. Unfortunately, maximum errors happen due to the poor quality of the source text.  If your source text has grammatical errors, too long sentences, poor vocabulary, and inconsistent terms, the machine translation output quality drops too. 

Another major reason for machine translation errors is the quality of the training material. You won’t get good results if you train your MT engine on fictional text, but expect it to translate technical or official communication. MT system training requires parallel texts that directly compare the source and target languages. It is important to use parallel texts of the highest quality that are free of errors and cover the domains and topics you want to translate in.

Missing context is the third big reason why machine translation fails. In many languages, words and sentences change if the communication channel changes. A marketing ad and an official notice will use different language. Words with double meanings require context for correct interpretation. You must provide information on specific terminology, domains, style, and tone to improve translation quality.

Preventing machine translation errors

Preparing a simple, concise, and consistent source document is the best way to prevent machine translation errors. Having a professional translator involved in the process from the start is essential. The translator can do some light touch editing at the start, like:

  • Simplify sentences—Short, straightforward sentences are easier for a machine to translate accurately.
  • Standardize terminology—Consistent use of specific terms or jargon makes the translation more coherent.
  • Check spelling and grammar—Errors in the source text lead to significant mistranslations.
  • Provide context like comments or footnotes for phrases that rely heavily on context or cultural nuance.
  • Edit idioms and slang that translate poorly and can lead to misunderstandings.

It is also critical to choose a machine translation engine that best suits your needs, considering the language pair and the specific type of text you’re translating. For example, SYSTRAN developed its own NMT technology that yields high-quality automated translation in 50+ languages and includes pre-packaged industry-specific translation models for Finance, IT, Legal, Medical, and more.

With a well-prepared document and the right tools, you’re on your way to obtaining a high-quality machine translation!

Best practices to find and fix machine translation errors

Despite best efforts, minor errors may still creep in. As always, having a professional translator review and check the document is essential. 

Post-edit MT output

Machine translation post-editing involves human translators reviewing machine-generated text to correct errors and fine-tune nuances. Human expertise is necessary to capture idiomatic expressions, cultural references, and specific jargon that machines may not handle well.

Develop a style guide

Creating a translation style guide is essential for maintaining consistency and quality across different translations. A style guide outlines the preferred terminology, tone, and style, providing both human and machine translators with a consistent framework. 

Use quality assurance tools

Utilizing quality assurance tools specifically designed for translation can act as a safety net, flagging potential errors or inconsistencies before they become problematic. These tools can check for everything from basic spelling and grammar mistakes to complex issues like mistranslations or missed negations. Incorporating such tools into your translation workflow enhances accuracy and saves time in the post-editing stage, making the entire process more efficient and reliable.


Machine translation outputs can have different types of errors that can impact business reputation and cause embarrassment. The key to successful translation is using quality source material, quality translation tools, and post-translation editing. It is essential to involve translation experts from the start and to plan the process systematically for the best results!

Carolina - Linguist
5 Min Read
Newsletter Sign-Up
Find all the news and the latest technologies. A magazine designed by SYSTRAN