The Differences Between AI and Human Translators

Computers can do many things better than humans – predictive analytics, driving cars, playing chess. But in language translation, humans reign supreme. Computers excel where rules and logic dominate, but language comes with far too many rules that are often broken or changing.

 

In 2018, iFlytek, one of China’s leading voice recognition companies, committed an AI translation scandal at the International Forum on Innovation and Emerging Industries Development. The company claimed translations were coming from its automated translation technology when, in fact, human translators were doing the work.

 

Even Google’s well-known Google Translate service isn’t trustworthy enough for use in professional settings. Studies have shown its unreliability in translating medical information for patients. This ineffectiveness extends to any profession where the need for accuracy is paramount.

 

Google itself warns that you should not use its translation technology in place of humans: “Reasonable efforts have been made to provide an accurate translation, however, no automated translation is perfect nor is it intended to replace human translators.”

 

When cognitive scientist and translator Douglas Hofstadter tested Google Translate for himself, he found that its translations often failed to relay the message’s correct meaning and intent.

 

Artificial intelligence has come a long way, but not quite as far as people tend to believe it has. Despite improvements, AI translation services will remain behind human translator services in accuracy for a long time – if not forever.

 

There are simply far too many components of communication that AI struggles to grasp.

 

The Subjective Nature of Language

Computers and AI work best in objective contexts, but language is subjective. Words have nuances in meaning; words have rules and grammar guidelines but with many unpredictable exceptions. On top of it all, words, their meaning, and their rules are constantly changing.

 

Slang changes in real-time, so computers can’t keep up. You can’t wait for new words to become official because people use them long before that, but you also can’t guarantee that the word won’t go out of use before becoming official.

 

To make it harder for AI, regional differences and dialects influence language as well.

 

Missing Out on Tone and Body Language

Communication is more than words. Tone informs intent and can be the deciding factor between genuineness and sarcasm or passive-aggressiveness and empathy. AI has a hard time differentiating between intent due to a lacking history with emotion.

 

You can train computers, but there’s a long way to go before they can interpret emotion with 100% accuracy. And then, they will have to learn how to convey it to the other party. Human interpreters account for tone and body language with ease.

 

Accounting for Culture

Language isn’t the only thing that changes between cultures. Cultures consist of diverse foods, customs, laws, and beliefs. They affect people’s opinions and ways of thinking and acting. Even common names, another problem area for AI, change among cultures.

 

A word or action that’s considered normal in one culture may be offensive in another. This is why professional interpreters must be as familiar with cultures as they are languages.

 

Machines aren’t capable of accounting for the broad range of cultural implications. It presents enough of a challenge when the same word holds different definitions across cultures and accents confuse speech recognition.

Accounting for Context

Machines often translate word by word. This is an unreliable method when words have multiple meanings depending on context. Some AI broadens the scope of consideration, but a word relies on the sentence for context, the sentence on the paragraph, the paragraph on the entire essay, and so on. It’s the case for oral translation as much as it is for document translation. It takes a human understanding to relay a fully contextualized understanding.

 

When AI is engineered to account for context to the best of its ability, it can still fail to pick up on
metaphors and underlying meanings, losing them in translation.

Robots Lack a Sense of Humor

Humor is another area where nuance matters. Many jokes, like puns and innuendo, are founded on wordplay, and this is easy to lose in translation. Direct jokes may also come across strangely in translation due to AI’s lack of context.

 

Humor is often used to express ourselves and connect. At best, you lose this opportunity to better connect with clients when using AI translation. At worst, you may deeply offend a client or customer without knowing it.

 

This problem goes both ways: imagine how uncomfortable it is for a client to open up with a joke and have you miss the point entirely.

The Danger of Learned Bias

On the topic of damaging relationships, learned bias can aggravate the situation. The problem with machine learning is that machines learn from their creators and the world around them. This makes it easy for AI to pick up its creators’ biases or the biases, racism, phobias, and judgments of others. This is a dangerous risk, especially when dealing with clients or patients.

Trouble With Oral Speech

While all the above applies to oral speech and text translation, translating oral speech comes with an additional challenge. Machines have to be able to hear and recognize the words people say out loud. Automatic speech recognition (ASR) can struggle to keep up with the speed of natural speech, especially with the other setbacks compounding the issue.

 

You may be familiar with everyday technology that relies on ASR like Siri and Alexa, but though impressive, they work in limited contexts with more predictable demands.

 

Consider a conference between government leaders or a healthcare conversation with an injured employee. What people say in these scenarios is way more varied and uncertain, but in these scenarios, translations must be exact.

 

In real-world circumstances, robots have yet to compare to humans’ speech recognition or translation ability.

 

Losing the Personal Touch

Human interpreters keep communication personal and foster an environment to connect and build trust. Clients and patients enjoy speaking to a person with a human face and voice. Machines are also unable to replicate a natural inflection and tone, making conversations more
robotic and awkward than the seamless experience you get with human interpretation.

 

Stick With Humans for Professional Translation

Artificial intelligence is capable of great things, but replacing human interpreters isn’t one of them. If you’re having fun with a friend or visiting a country where you know some of the language, Google Translate may be a reasonable option.

 

But if you’re working with others in a professional environment, neither Google Translate nor any other AI translation service will be able to guarantee reliable, accurate interpretations. They may even result in greater risk and misunderstanding.

 

If you’re looking for trusted translation and interpretation services that guarantee safe, effective results and improve client relationships, reach out to the human experts at iLingo2.