Translating between human languages is something which artificial intelligence – specifically machine learning – has proven to be very competent at.
So much so that the CEO of one of the world’s largest employers of human translators has warned that many of them should be facing up to the stark reality of losing their job to a machine.
One Hour Translation CEO Ofer Shoshan told me that within one to three years, neural machine technology (NMT) translators will carry out more than 50% of the work handled by the $40 billion market.
His words stand in stark contrast to the often-repeated maxim that, in the near future at least, artificial intelligence will primarily augment, rather than replace, human professionals.
Shoshan told me that the quality of machine translation has improved by leaps and bounds in recent years, to the point where half a million human translators and 21,000 agencies could soon find themselves out of work.
He says, "The analogy that we can use is Kodak and digital photography - Kodak didn't see it coming … and before that, Corona typewriters and word processors
"Two years ago, translation technology would produce something that at best would let you get a general understanding of what the text was about – but in most cases, a professional translator would tell you they would rather just translate from scratch because they couldn't understand a lot of the output.
“Today with neural machines, for a growing amount of material and categories, they only need to make a very small number of changes to what a machine outputs, in order to get a human-quality translation.”
Quantifying this, Shoshan tells me that today on average 10% of a machine-translated document needs to be fine-tuned by humans to meet the standards expected by his company's Fortune 500 clients. Just two years ago, that figure was around 80%.
This has been made possible by the switch to neural machine translation – sometimes known as deep learning – adopted by the most advanced machine translation tools. Previously these relied on a method known as statistical translation. Google, Bing, and Amazon now all use NMT in their translation engines.
Training a neural machine to translate between languages requires nothing more than feeding a large quantity of material, in whichever languages you want to translate between, into the neural net algorithms.
To adapt to this rapid transformation, One Hour Translation has developed tools and services designed to distinguish between the different translation services available, and pick the best one for any particular translation task.
"For example, for travel and tourism, one service could be great at translating from German to English, but not so good at Japanese. Another could be great at French but poor at German. So we built an index to help the industry and our customers. We can say, in real time, which is the best engine to use, for any type of material, in any language."
This work – comparing the quality of the output of NMT generated translation, gives a clue as to how human translators could see their jobs transforming in coming years. Humans rate the output of each engine and compile the index. In the case of One Hour Translation's index, this is done once per quarter, to reflect the speed at which NMT is evolving, and new players are emerging onto the market.
If that sounds like a silver lining, however, then things may not be quite that straightforward. The level of training and expertise required to rate machine translations, or to translate while “augmented” by a machine, is far lower than for traditional, “from scratch” translation.
“You need someone smart, with good language skills – but they don’t need to be a professional, traditionally-trained translator, because fixing one word here or there is much easier,” says Shoshan.
So, as was the case during the first industrial revolution, are we likely to see gangs of translators rioting in the streets and smashing up the intelligent machines which are threatening their livelihoods?
"I hope not," Shoshan says. "But actually, it is an issue, just like autonomous trucking will be an issue for the four million or so truck drivers employed in the US.
"And importantly, we're not talking about five to ten years; we're talking one to three years.
“It’s obvious that if machines can do what you can do, then you have a problem. A lot of translators and agencies will tell you that there are certain highly specialized translation services which will require a human touch for the foreseeable future – and that may be true.
“But for the bulk – I would estimate 80% - of the material that corporate customers pay to have translated on the market today, it will be machine translatable in the next one to three years.”
Some advice for translators wanting to keep their heads above water could include specializing in languages which are less widely spoken. NMT services rely on huge bodies of literature being available, to train the algorithms – and for languages with a smaller user base, that quantity of material may not be readily available, particularly in specialist, technical or scientific subjects.
Another, as is the case with One Hour Translation, could be to get used to working alongside machines. While they will do the bulk of the work, there will be a need for people able to assess different translation technology and apply the correct tools for specific jobs.
So much so that the CEO of one of the world’s largest employers of human translators has warned that many of them should be facing up to the stark reality of losing their job to a machine.
One Hour Translation CEO Ofer Shoshan told me that within one to three years, neural machine technology (NMT) translators will carry out more than 50% of the work handled by the $40 billion market.
His words stand in stark contrast to the often-repeated maxim that, in the near future at least, artificial intelligence will primarily augment, rather than replace, human professionals.
Shoshan told me that the quality of machine translation has improved by leaps and bounds in recent years, to the point where half a million human translators and 21,000 agencies could soon find themselves out of work.
He says, "The analogy that we can use is Kodak and digital photography - Kodak didn't see it coming … and before that, Corona typewriters and word processors
"Two years ago, translation technology would produce something that at best would let you get a general understanding of what the text was about – but in most cases, a professional translator would tell you they would rather just translate from scratch because they couldn't understand a lot of the output.
“Today with neural machines, for a growing amount of material and categories, they only need to make a very small number of changes to what a machine outputs, in order to get a human-quality translation.”
Quantifying this, Shoshan tells me that today on average 10% of a machine-translated document needs to be fine-tuned by humans to meet the standards expected by his company's Fortune 500 clients. Just two years ago, that figure was around 80%.
This has been made possible by the switch to neural machine translation – sometimes known as deep learning – adopted by the most advanced machine translation tools. Previously these relied on a method known as statistical translation. Google, Bing, and Amazon now all use NMT in their translation engines.
Training a neural machine to translate between languages requires nothing more than feeding a large quantity of material, in whichever languages you want to translate between, into the neural net algorithms.
To adapt to this rapid transformation, One Hour Translation has developed tools and services designed to distinguish between the different translation services available, and pick the best one for any particular translation task.
"For example, for travel and tourism, one service could be great at translating from German to English, but not so good at Japanese. Another could be great at French but poor at German. So we built an index to help the industry and our customers. We can say, in real time, which is the best engine to use, for any type of material, in any language."
This work – comparing the quality of the output of NMT generated translation, gives a clue as to how human translators could see their jobs transforming in coming years. Humans rate the output of each engine and compile the index. In the case of One Hour Translation's index, this is done once per quarter, to reflect the speed at which NMT is evolving, and new players are emerging onto the market.
If that sounds like a silver lining, however, then things may not be quite that straightforward. The level of training and expertise required to rate machine translations, or to translate while “augmented” by a machine, is far lower than for traditional, “from scratch” translation.
“You need someone smart, with good language skills – but they don’t need to be a professional, traditionally-trained translator, because fixing one word here or there is much easier,” says Shoshan.
So, as was the case during the first industrial revolution, are we likely to see gangs of translators rioting in the streets and smashing up the intelligent machines which are threatening their livelihoods?
"I hope not," Shoshan says. "But actually, it is an issue, just like autonomous trucking will be an issue for the four million or so truck drivers employed in the US.
"And importantly, we're not talking about five to ten years; we're talking one to three years.
“It’s obvious that if machines can do what you can do, then you have a problem. A lot of translators and agencies will tell you that there are certain highly specialized translation services which will require a human touch for the foreseeable future – and that may be true.
“But for the bulk – I would estimate 80% - of the material that corporate customers pay to have translated on the market today, it will be machine translatable in the next one to three years.”
Some advice for translators wanting to keep their heads above water could include specializing in languages which are less widely spoken. NMT services rely on huge bodies of literature being available, to train the algorithms – and for languages with a smaller user base, that quantity of material may not be readily available, particularly in specialist, technical or scientific subjects.
Another, as is the case with One Hour Translation, could be to get used to working alongside machines. While they will do the bulk of the work, there will be a need for people able to assess different translation technology and apply the correct tools for specific jobs.
Comments
Post a Comment