Google Translate 1.0
Google translate has become ubiquitous in our world today, being used by students desperate to pass their language exams, to avid travelers in desperate need of some translation to be able to communicate abroad. However, translate has always missed that edge in not being able to extend its translation capabilities from a short sentence to a long corpus of text. For example, observe this translation of Hemingway’s “The Snows of Kilimanjaro” from the Japanese version to English:
Kilimanjaro is 19,710 feet of the mountain covered with snow, and it is said that the highest mountain in Africa. Top of the west, “Ngaje Ngai” in the Maasai language, has been referred to as the house of God. The top close to the west, there is a dry, frozen carcass of a leopard. Whether the leopard had what the demand at that altitude, there is no that nobody explained.
Clearly, the translation of this paragraph is sub-optimal. While the meaning is conveyed, the text loses all grammatical context and frankly is hopeless at creating a “human-like” translation. Google Translate just simply can’t understand the rules of language well enough. That is…until its most recent update.
Google Translate 2.0
On 15th November 2016, Google announced that they would be releasing a new version of Google translate, set to change the translation game for years to come. This new google translate is built from Neural Networks which, as opposed to the previous machine translation system, allows the machine to detect patterns in language and create features for translation by itself. The challenge in creating such a complex neural network is that it requires vast amounts of data, but it is safe to say that Google is the best positioned company to deal with this problem, especially with the recent spike in available data over the past decade.
So how good does this new system perform versus the old system? Let us compare the same extract from Hemingway’s book and see if there is any improvement:
Kilimanjaro is a mountain of 19,710 feet covered with snow and is said to be the highest mountain in Africa. The summit of the west is called “Ngaje Ngai” in Masai, the house of God. Near the top of the west there is a dry and frozen dead body of leopard. No one has ever explained what leopard wanted at that altitude.
The jump is simply astounding. It seems that the neural network has not only translated words correctly like the previous translation system did, but it now grasps the subtleties of grammar in each language and can adapt heavily different grammars (Japanese to English) incredibly well. More interestingly though, is comparing the newly translated extract to the original Hemingway (remember the above was from Japanese version of Hemingway to English):
Kilimanjaro is a snow-covered mountain 19,710 feet high, and is said to be the highest mountain in Africa. Its western summit is called the Masai “Ngaje Ngai,” the House of God. Close to the western summit there is the dried and frozen carcass of a leopard. No one has explained what the leopard was seeking at that altitude.
If it were not for the leopard reference near the end of the paragraph, it would be hard to tell which one was created by a human and which by a machine. Google has successfully created such a deep neural network, that it can pick up small intricacies in the languages it trains on, and replicate them almost to perfection.
Google’s Brain: A new language?
As seen above, Google’s Neural Machine Translation system has taken a massive jump in the world’s translation progress. A blog written on this matter by one of Google’s software engineers claims that Google Translate has progressed more overnight than it has in the past 10 years of tweaking.
The new translation system has also shown impressive results when translating between two languages it hasn’t seen before, for example, translating between Japanese and Korean when it has not seen direct translations between those two languages. Amazingly, the new system can figure out ways to translate such cases using a concept called an “interlingua”. An interlingua is a type of artificial language, the encoding of source sentences into almost a new language, allowing the system to then translate to target sentences. The AI uses this interlingua to understand how an unseen pair of languages could be translated. This brings us back to the debate of whether artificial intelligence systems are approaching human reasoning skills, and if we are accelerating towards singularity, the point in which us humans create systems far smarter than ourselves. It is a stretch to go from a simple translation machine to singularity, but the interlingua concept definitely makes me wonder really how far away we are to such a point in time.