Horizonal lines of varying thickness on a black background

Machine Translation and Catastrophic Errors

Protect your company from the consequences of catastrophic translation errors with automated quality checks

Translation matters. Accurate translations will enable you to effectively reach your target markets no matter what language people speak. As companies increasingly turn to Machine Translation (MT) to expand their reach to new markets and translate more content faster for all their audiences, they increasingly run the risk of disseminating catastrophic translation errors.

It is critical to be aware of MT’s shortcomings and prevent these types of errors from reaching your audiences because they can be highly detrimental to your business. Savvy MT providers, like Lionbridge, can administer automated quality checks during translation to detect critical errors while preserving MT speed and reducing the need for human intervention.

What Are the Two Main Types of Errors Associated With Machine Translation?

Although MT is becoming increasingly reliable due to the vast amounts of text that systems can learn from and neural networks that enable Neural Machine Translation (NMT), engines still have the propensity to make errors. There are two major types of errors associated with Machine Translation; each type has a different scale of severity.

Standard Errors: What Are They and Should You Care?

Standard MT errors are the more benign of the two types of errors. These mistakes in the target language pertain to the content’s linguistic features. Standard errors include grammar, spelling, or punctuation mistakes. While native speakers are likely to notice these slip-ups, these mistakes rarely cause disastrous consequences for a company.

In fact, extensive research indicates end users are willing to accept less-than-perfect quality.

A survey of 8,709 consumers in 29 countries found:

  • 65% of buyers prefer to purchase a product when content is in their local language, even if that content contains some errors
  • 66% of users turn to online Machine Translation when evaluating products available in other languages
  • 40% of consumers will not buy products that are presented in other languages

(Source: “Can’t Read, Won’t Buy – B2C,” CSA Research, June 2020)

Perfect translations are not always necessary. In these cases, companies need not be overly concerned about the occasional standard error.

Catastrophic errors transcend linguistics and occur when engine output dangerously deviates from the intended message. Resulting misinformation or misunderstandings have the potential to cause companies reputational, financial, or legal repercussions and lead to adverse public safety or health consequences.

Catastrophic Errors: What Are They and Why Should You Care?

Catastrophic MT errors — as the name suggests — involve serious mistakes. A company may face detrimental consequences if it disseminates translated content that contains catastrophic errors.

These errors transcend linguistic mistakes and occur when the engine output does not accurately convey the intention of the source text. The deviation from the intended message may spread misinformation, cause confusion and misunderstandings, and even possibly lead to conflict.

For instance, if law enforcement or health officials put out directives to the public that contain catastrophic errors because of MT, the mistakes may negatively impact the public’s well-being and cause the agencies to lose credibility. A company that accidentally disseminates content containing catastrophic errors may face reputational, financial, or legal repercussions.

What Causes Catastrophic Errors?

Think of a catastrophic error as an MT engine malfunction. It can occur if the engine doesn’t understand the context of the text, such as when one word has two meanings or if there is a typo in the source text. These errors can happen if the engine is not trained well or a flawed glossary is used, which then results in the same mistakes appearing repeatedly.

Catastrophic errors occur because engines are imperfect despite their sophistication. Machines cannot exercise judgment the way people can.

Skyscrapers illuminate city at night

When Are Catastrophic Errors Most Likely To Appear in Your Text?

Catastrophic errors show up in different contexts. There are three primary categories to be aware of when anticipating the potential for catastrophic translation errors — the mistranslation of key entities, negation and opposite meaning, and hallucinations.

Mistranslation of Key Entities

The mistranslation of key entities refers to translation errors of proper names (individuals or organizations), important numbers, or units of measurement.

In the case of proper names, mistakes may occur when the name is also a common word. The public witnessed a real-world example of a catastrophic error involving a proper name on a Spanish governmental agency website. In that instance, the name of the department head, Dolores del Campo, was omitted from the ministry’s official site. Instead, the literal translation — It is pain of field — appeared in place of the name.

During Machine Translation, the engine may translate the unit of currency — let’s say from yen to dollars — but fail to take the conversion rate into account when translating the corresponding numbers. This will result in a catastrophic error that can cause confusion and lead to adverse economic consequences.

And here’s how things can go wrong during the translation of a unit of measurement. If a medical document specifies a certain dosage in milligrams, but the engine erroneously translates the dosage into grams, the patient reading the translated copy has the potential to receive the wrong dosage and face harmful medical consequences. The company may be held liable for the error and possibly incur legal expenses and damages.

Negation and Opposite Meaning

Negation and opposite meaning catastrophic errors refer to mistakes in the target language that result in the opposite meaning of what was intended by the original text.

An example may be when a company memo for shareholders is translated from English to Spanish, and the resulting Spanish translation indicates share prices went down when, in fact, prices went up.

Hallucinations

On rare occasions, MT can introduce content that is simply not present in the source. These are called hallucinations. When this type of catastrophic error happens, there is a problem with the MT engine software itself. Engines may generate offensive, profane, aggressive, or highly sensitive words under certain circumstances.

Strings of beaded orange lights in a dark tunnel lead to a bright opening

How Can Automated Methods Prevent Catastrophic Errors During Machine Translation?

It is essential to prevent catastrophic errors from compromising your content. But it’s not that easy. Companies will be better protected from the threat of catastrophic errors when computer scientists improve existing MT technology to stop these errors from happening. Until such time, we can use automated technology to identify potential issues, revise problematic sentences, and promote accuracy during the translation process.

Lionbridge administers specific automated quality checks in translated texts via its Smart MT™ offering — and in conjunction with its cutting-edge Smart Content language AI — to detect errors while maintaining MT speed and minimizing the need for post-editing by human translators.

These automated methods detect:

  • Incorrect translations of proper names, including the names of individuals or organizations, by identifying entities in the source text that can be both a named entity and a common word
  • Offensive, profane, or highly sensitive words by combining a supervised Machine Learning (ML) algorithm and a list of offensive terms
  • Opposite meanings between original and translated texts by identifying negative particles (sentences that contain the word not or its shortened form n’t) in the original text or the translated text, but not in both
  • Hallucinations in the translation via dictionaries or a list of offensive words when the hallucinations are insulting 

Automated quality checks will not guarantee the elimination of catastrophic errors. Automated checks can miss errors, causing a false negative result. Nonetheless, they are highly effective at helping us find problems. By using this approach, we can focus professional translators on the sentences that are flagged and avoid reworking the whole document. When we can alert professional translators to where problems are most likely to be found, we enhance the efficiency of the localization process.

Get in Touch

If you’d like to learn more about how Lionbridge can help you implement a winning MT strategy that protects you from catastrophic errors, contact us today.

linkedin sharing button

Luis Javier Santiago with Janette Mandell
AUTHOR
Luis Javier Santiago with Janette Mandell
  • #blog_posts
  • #translation_localization