Machine Learning Translation and the Google Translate Algorithm
Machine Learning Translation and the Google Translate Algorithm
Text translation systems use lots of complex algorithms to generate accurate results. Learn how to evaluate the complex engine of machine learning translations.
Join the DZone community and get the full member experience.Join For Free
Bias comes in a variety of forms, all of them potentially damaging to the efficacy of your ML algorithm. Read how Alegion's Chief Data Scientist discusses the source of most headlines about AI failures here.
Years ago, it was very time-consuming to translate text from an unknown language. Using simple vocabularies with word-for-word translation was hard for two reasons:
The reader had to know the grammar rules.
The reader needed to keep in mind all language versions while translating the whole sentence.
Now, we don’t need to struggle so much — we can translate phrases, sentences, and even large texts just by putting them in Google Translate. But most people don’t actually care how the engine of machine learning translation works. This post is for those who do care.
Deep Learning Translation Problems
If the Google Translate engine tried to keep the translations exact, even for short sentences, it wouldn’t work because of the huge number of possible variations. The best idea is to teach the computer sets of grammar rules and teach it to translate the sentences according to those rules.
Now if only it were as easy as it sounds.
If you have ever tried learning a foreign language, you know that there are always a lot of exceptions to rules. When we try to capture all these rules, exceptions, and exceptions to the exceptions, the quality of translation drops.
Modern machine translation systems use a different approach: They allocate the rules from a text by analyzing a huge set of documents.
Creating your own simple machine translator would be a great project for any data science resume.
Let’s try to investigate what hides in the “black boxes” that we call machine translators. Deep neural networks can achieve excellent results in very complicated tasks (speech/visual object recognition), but despite their flexibility, they can be applied only for tasks where the input and target have fixed dimensionality.
Recurrent Neural Networks
Here is where long short-term memory networks (LSTMs) come into play, helping us work with sequences whose length we can’t know a priori.
LSTMs are a special kind of recurrent neural network (RNN) capable of learning long-term dependencies. All RNNs look like a chain of repeating modules.
So the LSTM transmits data from module to module and, for example, for generating Ht, we use not only Xt but also all previous input values X. To learn more about the structure and mathematical models of LSTM, check out this article.
Our next step involves bidirectional recurrent neural networks (BRNNs). A BRNN splits the neurons of a regular RNN into two directions: One for positive time (or forward states) and another for negative time (or backward states). The output of these two states is not connected to inputs of the opposite direction states.
To understand why BRNNs can work better than a simple RNN, imagine that we have a sentence of nine words and we want to predict the fifth word. We can make it know either only the first four words or the first four words and last four words. Of course, the quality in the second case would be better.
Now we’re ready to move to sequence to sequence models (also called seq2seq). The basic seq2seq model consists of two RNNs: an encoder network that processes the input and a decoder network that generates the output.
Finally, we can make our first machine translator!
However, let’s think about one trick. Google Translate currently supports 103 languages, so we should have 103x102 different models for each pair of languages. Of course, the quality of these models varies according to the popularity of languages and a number of documents needed for training this network. The best that we can do is make one NN take any language as input and translate it into any language.
That very idea was realized by Google engineers at the end of 2016. The architecture of NNs was built on the seq2seq model.
The only exception is that between the encoder and decoder, there are eight layers of LSTM-RNN that have residual connections between layers with some tweaks for accuracy and speed. If you want to go deeper with that, take a look at this article.
The main thing about this approach is that now the Google Translate algorithm uses only one system instead of a huge set for every pair of languages.
The system requires a “token” at the beginning of the input sentence that specifies the language you’re trying to translate the phrase into.
This improves translation quality and enables translations, even between two languages which the system hasn’t seen yet — a method termed “zero-shot translation.”
What Characterizes a Better Translation?
When we’re talking about improvements and better results from Google Translate algorithms, how can we correctly evaluate that the first candidate for translation is better than the second?
It’s not a trivial problem because, for some commonly used sentences, we have sets of reference translations from the professional translators that have, of course, some differences.
There are a lot of approaches that partly solve this problem, but the most popular and effective metric is BLEU (bilingual evaluation understudy). Imagine that we have two candidates from machine translators:
Candidate 1: Statsbot makes it easy for companies to closely monitor data from various analytical platforms via natural language.
Candidate 2: Statsbot uses natural language to accurately analyze business metrics from different analytical platforms.
Although they have the same meaning, they differ in quality and have different structures.
Let’s look at two human translations:
Reference 1: Statsbot helps companies closely monitor their data from different analytical platforms via natural language.
Reference 2: Statsbot allows companies to carefully monitor data from various analytics platforms by using natural language.
Obviously, Candidate 1 is better, sharing more words and phrases compared to Candidate 2. This is a key idea of the simple BLEU approach. We can compare n-grams of the candidate with n-grams of the reference translation and count the number of matches (independent from their position). We use only n-gram precisions because calculating recall is difficult with multiple refs and the result is the geometric average of n-gram scores.
Now, you can evaluate the complex engine of machine learning translations. Next time when you translate something with Google Translate, imagine how many millions of documents it analyzed before giving you the best language version.
Published at DZone with permission of Daniil Korbut . See the original article here.
Opinions expressed by DZone contributors are their own.