Human vs Machine Translation: How Do They Compare?

Tips for Localizing Content for Southeast Asia
August 19, 2020
Important Localization and Translation Terms to Know
September 2, 2020

Google Translate and other machine translation software has come a long way since their infancy. Many users’ first experiences using Google Translate resulted in humorous responses and these experiences have ingrained the belief for many that human translation is always better.

Is that still the case? In this blog we explore some of the recent trends in machine translation, how human translation compares, and what the possible future might hold for human and machine translation.

[Average read time: 4 minutes]

A MacBook with lines of code on its screen on a busy desk

photo by Christopher Gower

Machine Translation: Uses and Trends

Machine translation is the use of a computer program to translate text from one language into another. Research in the field is said to have begun around the late 1940s or early 1950s by the U.S. in order to track the Russian government and their operatives. Since then, the field of artificial intelligence (AI) has grown exponentially as well as its use in machine translation. Using AI, software has become more accurate and efficient, making machine translation comparatively much more cost-effective and faster than human translation.

However, machines still have issues with capturing the nuances of text and thus human translators are still needed to make sure that the translated text is accurate and conveys the original meaning. That said, companies are able to leverage machine translation to process low priority, high volume content such as user comments, social media posts, and other user-generated content.

A big trend in machine translation is post-editing machine translation (PEMT) in which a human translator revises or edits text that has been translated by a machine. Recent studies have shown that PEMT with neural machine translation resulted in 36% higher productivity compared to a human translator translating all the text from scratch.

Potential customers are hungry for video content, a reality that has become increasingly true with video sites like YouTube surpassing the 2 billion user mark. With this demand comes the need for video translation. Video translation (aka video localization) is the process of making a video suitable for a target audience that speaks another language. Machine translation tools are now available to help meet this demand, including automatic transcription and translation that listen to the speech of the video and transcribe it in the same language or translate it simultaneously into another language. This can help increase the speed for creating captions and subtitles, however, if the translations are not double checked by a human, it may lead to strange results.

man holding black book and pen

photo by Ilyass Seddoug

Human Translation is Here to Stay (for now)

Human translation is not going away anytime soon: humans are still able to understand the context and nuances of written and spoken language to a degree of accuracy that machines currently cannot. There are many cultural and idiomatic phrases that machine translation still does not know how to translate accurately. Phrases like, “I’m getting cabin fever” or “the cat’s out of the bag”  are still being translated literally by machine translation software instead of their intended meanings (feeling stuck inside or revealing a secret, respectively).

Communication is another big part of why human translators still win. You can’t get on the phone with a computer program and let it know the details and/or emotional tone that you’re going for with a project. With human translators, companies are able to discuss project objectives, brand style, special key terms, etc…with another human being and help make the translations richer and more accurate.

Style and tone of language is something that may keep humans at the forefront of translation for a while. Machine translation learns how to translate from data input and will generally use the most common terms and register when translating. However, a change of phrase or the use of a different word to mean the same thing can greatly alter the feeling of a sentence. For example, a common synonym of the word “happy” is “joy”. However, native English speakers will generally say that “joy” comes off as stronger, a rarer feeling–this is a nuance that is hard to train machine translation to pick up on.

Also, words with multiple meanings confuse machines. In the last paragraph, I used the word “register” to refer to the politeness level when speaking/writing to another person. However, register can also mean “to register to vote” or the “cash register”: the same word with completely different meanings. Because we humans understand context much better than machines, we can infer the correct meaning.

However, machine translation is getting more and more accurate in large part thanks to humans. What makes companies like Google so powerful is their vast user community with millions if not billions of Google Translate users helping edit or correct translations free of charge. Therefore, phrases like “it’s raining cats and dogs” when translated to Chinese, can be corrected thanks to community (aka human) input–Google Translate translates the phrase correctly here.

smiling man facing laptop

photo by Anastasila Kamil

Looking Forward

The latest development in machine translation is neural machine translation (NMT). NMT programs are designed in a way that mimic how the human brain processes information (thus the “neural” aspect). NMT analyzes entire sentences as opposed to one word at a time and studies the similarities between words, allowing for a more fluid translation.

What we’re seeing now are human translators using NMT and other machine translation models as tools to help increase their productivity. Also, machine translation is learning from humans: as we build more sophisticated translation programs, we need more human input and expertise to teach the programs idiomatic phrases, cultural nuances, and much more.

English is currently the top language on the internet, but Chinese Mandarin is a close second with content in Spanish and Arabic growing at fast rates. The demand to translate to and from these languages is growing and thus is pushing the need to translate at higher volumes, which machines can do, but also accurately, which humans are better at.

Thus for the foreseeable future, it’s likely we’ll see companies leverage machine translation with human translators providing post-editing and making sure it sounds human.

Want to make sure your content is translated and subtitled correctly? Contact us now for a free quote!