In the biblical story of the Tower of Babel, humans initially all spoke one language. They wanted to make a name for themselves and began construction of a huge tower to reach the heavens. As the tower grew, God saw the hubris of the humans and came down to confuse their language (“babel” comes from the Hebrew word “to confuse”). The architects and workers, now all speaking different languages, were no longer able to communicate and abandoned construction. They then scattered across the world…and the human translation business has been booming ever since.
Since the birth of spoken and written language, human translation was essential for communication between different regions. Whether it was translating Egyptian hieroglyphics to ancient Greek text in 196 BC or from Cantonese to Portuguese at an 18th century Canton seafood market, humans, with our unique ability to understand context and cultural differences, were the only ones able to translate between different languages.
In the present day though, artificial intelligence (AI) is challenging that idea. How is AI changing the translation landscape? Will human translators be out of a job in the near future? Let’s dive in and explore the present and future of translation.
[Average read time: 4 minutes]
To understand AI, let us first look into its origins. Ancient myths around the world have described inanimate objects coming to life; one such tale is the Jewish story of the golem, a clay figure which is brought to life through magic. Philosophers have interpreted these tales as symbolic representations of human intelligence and it’s limitations.
Modern AI took shape first as a thought experiment by British polymath, Alan Turing. He developed the “Turing Test” in 1950: a human questioner would ask two unseen respondents a series of questions through a terminal. One of the respondents would be a computer, the other a human. After a series of tests with multiple participants, if 50% or more of the questioners guessed that the computer was human, the computer was considered to have passed the test and exhibited human-like intelligence.
However, the vocabulary and technology for what Turing was describing had not been developed yet. The term “artificial intelligence”, AI, was actually first coined by researcher John McCarthy at the Dartmouth Summer Research Project on Artificial Intelligence conference in 1956. This event was monumental and sparked an interest in machine learning and machine translation that has led to the advancements of today, such as Neural Machine Translation.
Neural Machine Translation: How AI Translates Human Language
Children learn language by listening to others and detecting patterns in the language. Similar to children, pattern recognition is used in the form of AI called Neural Machine Translation (NMT). NMT uses an electronic, neural network trained to recognize patterns in the input data set (e.g. a Mandarin sentence) and translate it into a desired output data set (e.g. an English sentence). For example, for Mandarin to English translation, the network would be trained by receiving input of millions of Chinese Mandarin and English language pairs.
The computer would receive a Mandarin sentence and then will guess what the English sentence would be. It will then be told how accurate its translation was relative to the correct English translation pair. Repeated millions of times, the computer will learn how to be more and more accurate. Human engineers would then test the system with a new sentence not used during training to see if the system had learned how to generalize.
Google Translate uses NMT and it’s not bad for widely spoken and written languages like Mandarin and English. However, if you expand to less widely used languages like Samoan, you can get inaccuracies like the one below:
“Honeymoon”, a vacation taken by newly wed couples, is translated to “Samoa”, a Pacific island nation. With less language pairs to train it, the NMT system will generate more translation errors.
Other limitations: certain languages do not have gender specific pronouns (e.g. Malayalam, Uzbek), but if you’re translating to a gender specific language (e.g. French, English), NMT can create gender errors, even if the gender is mentioned in a previous sentence. Also, cultural idioms like “bull in a china shop” often get translated literally (e.g. a bull in a shop in China) to much confusion.
As we can see, AI translation is still not perfect and humans still have the upper hand, but compared to ten years ago, AI translation has improved incredibly. With the help of present-day translation devices like the CM translator, Japanese store owners can do business in multiple languages instead of hiring employees to help translate.
Prior to AI, translators would do all the translations themselves with the assistance of a hardbound translation dictionary. Now, some translators will use an NMT engine like Google Translate to do the first round of translations and then edit those translations.
It’s estimated that by 2022, most business translations will be done by NMT with human editors to come in after to clean up the text. This is not hard to believe as NMTs can translate materials at very low to no cost as well as go through a much higher volume of text at a faster rate than humans.
Even so, we’re still in a transition period where NMTs rely heavily on human translators to correct issues. Last year, South Korea held a translation contest between a team of professional translators and a NMT to translate text from Korean to English. 90% of the the NMT text was deemed “awkward” and definitely not from a native speaker.
Aside from text to text translation, there have also been incredible advancements in live speech translation, also known as interpretation. Items like Google’s Pixel Buds, Waverly Labs’ Pilot, and Bragi are hearables (electronic ear buds) that can translate speech of one language live and play the audio back to you in your native language. Though not perfect–hearables have issues isolating human voice in a noisy environment and are still not close to the level of a professional interpreter–they’re still good enough to assist millions of travelers and businesses each year.
Humans may soon get to a point, through the help of AI in the form of an NMT system, a hearable, or some other technology, where every language is understood. It would be as if we returned to the mythical time before the Tower of Babel. As AI gets more and more advanced, it will supersede human intelligence exponentially and change our lives in ways we have yet to fathom. The Hebrew word for babel (to confuse) comes from the Akkadian word bab-ilu, which means “Gate of God”. We may soon no longer need to create a tower to reach the heavens, since we would have created god here on earth.
Want to make sure your captions and subtitles are translated and delivered properly? Download our free E-Book below!