The brain behind Google Translate can actually take you for a ride

Google Translate from Google

Google has made our lives a lot easier than we realize. It has literally made it a lot better for us. There are certain tools that we use every single day which make our work a piece of cake for us. No, this piece is not written to praise Google for its extraordinary work in many fields, but just to put some light on one of the most underrated features developed by it; the Google Translate.

Google Translate might not feel like a big deal but there are people who have given their life towards it. There are people working 24/7 for a breakthrough so that we can have a better product and ease of use. The brain that has gone behind developing the Google Translate is immense and the concepts are aplenty. There are things beyond our imagination that have gone into making this technology. That said, I am going to try my level best to decode it and lay it plain and simple.

Google Translate, when it came into existence used to translate word for word. That was the reason for it being so inaccurate and out of context. Talking about context, Google Translate never understood the context of the sentence that was being translated, it always used to translate word to word. The people working on this realized the problem and then, in came Wittgenstein’s theory.

It was 1953 when Ludwig Wittgenstein’s theories were published posthumously, there were people studying it, and those studies have been used by Google to perfect Translate. His theories say that a specific word does not have a specific meaning, it depends on the context in which it was used. This theory fits Google Translate perfectly. The reason for it is that Google Translate, translated word to word without the context and if you apply Wittgenstein’s theory, the problem was solved. However, it is not as easy as dragging and dropping the theory to Translate. The developers had to help its machine learn something which was written hundreds of years ago and that too by a philosopher.

Now, here comes the interesting part, actual algebra can be applied to the Google Translate. There are vector representations of each word and by applying simple algebra we can produce a conceptually perfect meaning. So for starters, we will consider a vector representing the word ‘King’, now if we minus the vector representing the word ‘man’ and add a vector representing the word ‘woman’ then we will land in a vector space that represents ‘queen’.

For translations to actually work and bring out meaningful results, programmers have to write code which is thousands of lines long. The programmers have to create a ‘neural network’ which is a form of machine learning which learns or is trained to understand how every word relates to each other. There are words which have different meanings, like the word ‘left’ can mean the direction or ‘left’ as in the past participle of ‘leave’. Now for Google Translate to understand this, it needs to understand the context it is being used in. For this, Google Translate then goes through the neural network. This neural network then goes through millions of texts which focuses on the word that follows or precedes any one word. This is done in seconds.

When that is done, there is Artificial Intelligence which then does the calculations which is more like the permutation and combination between each word and this then forms a vector space which is multi-dimensional and impossible for us to understand. And that space is our translation.


That is how Google Translate actually works. Apart from this, there is a lot more that goes into it but this is the gist.

-Unmesh Phule

Comment here