Can machine translation replace human translators?

How has machine translation improved accuracy and quality over time?
When will machine translation surpass the need for human translators?
CyraCom’s experts share the results of our research.

Organizations need to translate documents to better communicate with global audiences. But many hesitate to invest the time and budget required for human translation.

However, while machine translation has become more accessible over the last five years, is it wise for users to trust its accuracy and quality? Let’s review the latest machine translation developments to see how the technology compares to human translation.

Defining Machine Translation Types

The concept of machine translation, translating automatically without any human input, began in 1947, just one year after the completion of the first electronic numerical integrator and computer. Since then, humans have tried to teach technology how to convert one language to another using various processes and techniques. Let’s review the types of machine translation used over time:

  • Rule-based Machine Translation (RBMT)

    RBMT is the earliest form of machine translation. It relies on developers providing a full vocabulary and set of language rules for the source and target languages to function properly and relies upon significant post-editing from humans. The dynamic, ever-changing nature of languages makes this structured MT method difficult to use.

  • Statistical Machine Translation (SMT)

    Instead of moving forward word by word, SMT (also known as Phrase-Based Machine Translation (PBMT)) uses statistical models to string multiple words, phrases, and sentences together into the likeliest phrase by analyzing existing translations developed by humans. Like RBMT, SMTs can only translate a phrase if it exists in the reference texts.

  • Neural Machine Translation (NMT)

    NMT improves upon SMT’s ability to deconstruct and translate words and phrases individually by considering the entire framework and context of the source material. NMT software learns languages and constantly improves upon existing knowledge using artificial intelligence, much like the neural networks in the human brain. NMT has proven to be much faster and more accurate than other methods once the AI is trained in a language. Since Google adopted NMT as its primary model, this approach has rapidly become the industry standard.

  • NLLB-200

    Meta (formerly Facebook) claims its latest AI, called NLLB-200, can “translate 200 different languages and improve the quality of translations across […] a wider range of languages.” While most machine translation models handle only a handful of languages, Meta’s model is also interested in including “low-resource languages” in the model — languages with fewer than 1 million publicly-available translated sentence-pairs. These include many African and Indian languages not usually supported by commercial machine translation tools.

Benefits & Challenges of Machine Translation

Here are the top reasons many businesses have started using machine translation more often:

Benefits

Cost: Free machine translation platforms by big companies such as Google, Amazon, and Microsoft allow anyone access to translated content.
Time: Machine translation allows for fast delivery, which can be crucial for time-sensitive or high-volume projects.
Easy Use: Most machine translation platforms use a copy/paste structure and can integrate with other platforms.

Challenges

With these advantages, why isn’t human translation obsolete? Unfortunately, even with the breakthroughs machine translation offers, there are several challenges technology hasn’t overcome:

Liability:
Organizations using machine translation may be held responsible for errors or inaccuracies resulting in damages.

Literal Translation:
By default, machine translation chooses the most common vocabulary words when converting text. As a result, terms selected by automated translators can deviate from the original meaning.

Literacy Levels:
When you have a specific audience in mind, your content’s complexity should match their educational and literacy levels. Unlike human editors, current machine translation technology cannot adapt text complexity based on the intended reader.

Data Security:
Using public machine translation portals can lead to unintentional data leaks and privacy risks. Terms of use agreements may entitle the machine translation provider to store, ¬modify, reproduce, and distribute your submitted content.

Recent Machine Translation Mistranslation Examples:

  • In 2020, an officer used Google Translate to try to communicate with a Spanish-speaking driver during a traffic stop. When the case went to trial, the court stated:
       “The Court declines to infer that Google Translate accurately translated and communicated Conrad’s request to search Ramirez-Mendoza’s vehicle solely because it may work well generally. A record review shows that Google Translate is a useful tool with an alarming capacity for miscommunication and error. That the app can facilitate basic communication does not make it an adequate method for soliciting consent. It need only fail once to obviate a suspect’s consent.”
  • In 2017, Facebook mistranslated a post by a Palestinian man from “good morning” to “hurt them,” leading to his arrest by Israeli police.
       “Unfortunately, our translation systems made an error last week that misinterpreted what this individual posted.”

Varying Accuracy for Different Languages:

MT is more accurate for languages such as Spanish and Chinese. However, even the most accurate MT is still imperfect and can cause harm. For example, Slator’s article titled “Google Translate Not Ready for Use in Medical Emergencies But Improving Fast — Study” reviews data from a study of translated hospital discharge instructions.

Key Point
Quote

Examples of gross errors from Google Translate

English statements translated through Google Translation. "You need to take over-the-counter ibuprofen as needed for pain" translated to "You may take anti-tank missile as much as you need for oain" in Armenian. "Your Coumadin level was too high today. Do not take any more Coumadin until your doctor reviews the results" translated into "Your soybean level was too high today. Do not take any more soybean until your doctor reviews the results" in Chinese. "Do not blow your nose or put pressure on your facial fracture" translated into "Do not explode your nose because it could put pressure on the break in your face" in Farsi.

Citing the inconsistent performance between languages, the authors concluded:

The important implication of our study is that, despite recent reports of improvement in accuracy and the suggestion that GT has a role for use in the clinical setting, we found that GT accuracy varies substantially by language and is not yet a reliable tool in the clinical setting. Even for languages in which the accuracy is high, there is still the potential for important inaccuracies and the potential for patient harm. The best practice remains to use prewritten, professionally translated discharge instructions in the patient’s native language for general information about a diagnosis when such handouts are available in the electronic health record. For patient-specific instructions, clinicians should hand the patient a copy of their discharge instructions in English and use an interpreter to have the instructions verbally interpreted to the patient. While the interpreter is on the line, use a teach-back to be sure the patient understands the information.”

MT for Medical Professionals: Section 1557 proposed update on August 4

In this new proposed ruling, MT output must be reviewed by a “qualified human translator” for content that is “critical to the rights, benefits, or meaningful access of an LEP individual; when accuracy is essential; or when the source documents or materials contain complex, non-literal, or technical language.”

As noted in the above study, MT without human editing has the potential to cause harm to LEP patients. While Spanish machine translations are more accurate than other languages, is it worth the risk of even 2% of your patients being harmed?

Cultural Understanding:
Each language has distinct cultural norms, and violating them could send the wrong message to the reader.

Gender Bias:
MT often substitutes gender-neutral terms such as doctor or director with male vocabulary, while caregiver or cleaner end up female. Many companies strive to implement more inclusive, impartial, and equitable practices, yet simple MT bias can do the exact opposite.

Pronoun Confusion:
MT struggles with pronouns such as you or they. For example, Spanish has many versions of you, including tú (singular/informal), usted (singular/formal), vos (singular/formal and informal), and ustedes (plural/formal and informal) that MT programs can confuse, changing the meaning of the message.

Benefits & Challenges of Human Translation

Choosing human translation offers organizations these benefits:

Benefits

Accuracy:

When you need to communicate a precise message, human translation is the best method. Human linguists help deliver the correct meaning in the target language and can even incorporate the ideal tone and brand voice you’re looking for.

Quality Assurance Process:
Many language service providers offer a review process to confirm your content meets quality standards. For example, CyraCom has a second linguist proofread the original translator’s work before a project manager reviews the document for a final quality check. These extra steps help our clients feel confident about the integrity of our translations.

Localization & Cultural Awareness:
Beyond converting words from one language to another, translators can localize text to sound as though your message was originally written in the target language. Human linguists understand the culture behind the languages they speak, meaning it’s improbable your translation will be confusing or offensive to native speakers.

Subject Matter Experts:
A knowledgeable translator can prevent embarrassing or harmful mistakes if your content includes industry-specific terminology.

Data Security:
Choosing a provider with established data security protocols can help you avoid violating privacy or confidentiality laws or agreements. For example, CyraCom proved via a third-party audit that we preserve information privacy and security through a risk management process. In fact, we were the first US-based language services provider to obtain an ISO 27001 certification for information security management.

Challenges

There are only two ways human translation cannot compete with MT:

Cost:
Some organizations hesitate to invest in human translation to save their budget for other opportunities. However, is ‘free’ machine translation worth the risk?

Time:
Quality results take time to produce. Organizations may prioritize speed when precision and accuracy aren’t crucial.

Can machine translation replace humans? Not yet.

Although great strides have been taken in translation technology, machines cannot replace a human yet.

CyraCom has qualified human translators and editors to give you the reassurance you need for your translation projects. Partnering with CyraCom for your translation projects helps you access these advantages:

  • Human translators and editors with subject matter expertise in 100+ industries
  • 300 languages offered with literacy level adherence
  • Multiple projects can be completed simultaneously within our secure Translation Portal
  • Long-term savings and terminology consistency with our Translation Memory Database

Did you know CyraCom offers free quotes for translation projects?

Contact our team today to ask questions
or set up a zero-risk consultation.