Communication Technology: Google Translate

Introduction

There is a long list of things that function as barriers when people try to communicate with one another. One of these is language. When two people speak different languages it can seem impossible at times to get even the simplest ideas/thoughts across. Trying to ask where the bathroom is can turn into a frustrating, confusing event. These sorts of problems have been present throughout human history, and for hundreds of years, people have yearned for easier and more efficient ways to translate. 

History

The first record of an attempt to make translating all languages easier goes back to France during the year 1629. The philosopher Rene Descartes theorized that a universal alphabet, which could be applied to all languages, would make it easier to translate material. However, this was never accomplished. It wasn't until the mid 20th century that scientists and engineers started making serious attempts at "computer translation" (as it was dubbed in 1949 by Warren Weaver).

A replica of the Enigma

This increase in effort was fueled by nothing less than war. During WWII, top-secret messages and sensitive information were sent via ciphers. To make encoding and decoding messages faster, the Germans created the famous Enigma. This machine was extremely complex and solely mechanical (if you would like to learn more about how it worked click here), but it inspired scientists to pour more resources into researching machine translation. Scientists, like Warren Weaver, helped to spur others' interests in translation machines by writing about their potential and future goals for the field. By 1954, IBM created one of the first computers capable of basic translation.

How it Works

Fast forward 40 years and machine translation is still a challenge we have yet to tackle, but Google Translate (and other similar services) has come a long way. Now you're probably wondering, "How does Google Translate even work?" Back in 2006, Google Research announced the launch of the software. At this time, the main method used by the software was Phrase-Based Machine Translation (which breaks the entered phrase down into smaller segments before translating). Before this, word-based translation algorithms were used until they were proven to be less accurate by a team of researchers. 

An example of an older machine translators' process

Now, Google Translate uses its system called the Google Neural Machine Translation (GNMT) system (if you would like to learn more about its creation click here). GNMT is a form of AI (artificial intelligence) that takes the entire phrase entered into consideration when translating. The system relies on data sets of known translations between several languages. Unfortunately, it is impossible for me to explain in detail how this technology works because it is far too complex for me to understand fully in the time span of a week. I can, however, detail the steps it takes to reach its end goal. 

Let's say you're trying to translate something from Spanish to English. The first thing GNMT does is it uses an Encoder to transform the entered Spanish words into a list of vectors. Google's blog describes this action as "reading" as GNMT moves along the sentence(s) entered. Once all the words have been turned into values, a Decoder then "reads" the sentence(s) and gives the English translation of each Spanish word one at a time. As the Decoder moves along the sentence(s), it "pays attention" to the current word being translated and the ones following it. Then, through the power of advanced statistics and linguistics, it chooses the most relevant English translation. And there you have it, a basic explanation of machine translation. 

A representation of the GNMT Translation Process (found here)

The capabilities of GNMT are by no doubt amazing and a technological feat. Yet, this does not mean that the translations are perfect. Often Google Translate doesn't choose the right translations in context to the content, and sometimes original meaning can be lost. At this point, human translators still remain better than machines. 

Its Impact

Machine translators have forever changed the way we view language. It can be easy to forget that there are no perfect translations between languages because each one is unique. Each language has different grammar rules and ways of ordering subjects and verbs. The best way to learn all these complexities is to just submerge oneself into the world of the language they are trying to learn. They have to speak with native speakers, learn cultural references, and more on top of studying vocabulary words. 

Alexander Satola points out in their article that it is important for students, especially, to think through a translation before resorting to Google Translate. If a student immediately turns to use machine translators when presented with a question, they will become dependent on it. In the long run, students who are dependent on Google Translate (and other translators) won't retain the language in the same way that students who are not dependent will.

Although there are some negatives to Google Translate's success, overall, it has greatly increased people's ability to communicate with one another. Now, with just a few taps on your phone, you can easily "talk" to anyone no matter what language you speak. I believe this is very important because, in theory, the more we can talk to one another, the more we can understand each other.