How Google’s Improvement to Translation, is helping Global Companies Provide an Improved Service to non-English Users.

We’ve all heard the buzz around AI, Machine Learning & Automation. In fact, it’s not so much of a buzz but a seemingly unstoppable trend.

The leading organisations are embracing the changes and looking to build these improvements into their “Digital Strategies”.

The technology that supports such change is constantly improving and providing more value. In this short article, the focus is on translation and the improving service available.

Translation Support

Translation Applications

I think most of us who have travelled to foreign lands have used ‘Google Translate’ to try and make sense of signs, menus, simple questions to locals. The application available on iOS and Android is a must for frequent flyers.

Businesses are also often using Google Translate to convert messages and requests from their foreign speaking users into their local language. Exide Technologies is a perfect example.

How the Business consumes the translation services can have a huge effect on both Agent and User Satisfaction. This is an approach favoured by a number of firms.

Regardless of the entry point to the translation service, the quality of the translation itself is going to be the deciding factor on overall satisfaction.

The Technology

Google’s translation service is a veteran of their portfolio, first becoming available in 2006. I don’t think it’s unfair to say that historically the translations weren’t perfect.

The technology used was Narrow AI. An AI that has been used for many tasks and operations. Narrow AI, sometimes referred to as Weak AI is a pre-programmed approach that completes a set of tasks in sequence.

With Google, this technology would break a sentence down into parts, known as phrase-based translation.

These would be translated into the new language and the process would be repeated along the sentence, translating each part in sequence. The final step was reassembling all the parts in the new language.

This is problematic as languages are structured differently. Certain words are only used in one particular context and never used in others. The translation was also not able to judge the context of the situation.

Due to this, the use of translation was really limited to translating words or very short phrases.

Introducing GNMT

In 2016, Google introduced a Neural Machine Translation System (GNMT). This was a huge step forward in the quality of the language translations.

The GNMT consists of a deep LSTM network with 8 encoder and 8 decoder layers using attention and residual connections.

If you’re keen on seeing the breakdown of this, take a look at this article.

If that makes sense to you – well done. If, like me, you’re a little lost, consider it as a similar way as a human brain starts to understand patterns.

Our mind brings together patterns and situations, starting to learn from experience. We store this knowledge in our memory which is the key to continued learning.

Artificial Intelligence used by translation

Improved Translation

For most people, their interest is that GNMT brought improvements.

You can see an example of the quality change with this translation from Chinese to English, a notoriously difficult translation pairing. Other examples can be seen here.

Chinese Translation by Google's GNMT

Image courtesy of Google. 

Google stated that the switch to GNMT, reduced errors in translation from 55% – 85%. This came from testing samples on Wikipedia and News Websites and was checked by Human Translators.

What does it mean for Businesses?

In a global world, where businesses operate in numerous countries, employing and servicing people who speak many languages – Google’s translation can offer many possibilities for improving service support.

This article provides more perspective on the topic.