Let’s Talk about Machine Translation: The powering engine behind “Google Translate”

Published in

Becoming Human: Artificial Intelligence Magazine

6 min readApr 12, 2022

At some point, we’ve all used Google translate, Microsoft,DeepL or Bing translator to impress our friends/colleagues who speak a different language, or to understand a phrase, comment or post written in another language. Have you been curious about what power these engines? Well, Let’s talk about ‘Machine translation!!’

What is Machine Translation?

In simple terms, “Machine Translation” is an automatic translation between natural languages(bi-lingual, or multilingual).

Why Machine Translation?

Why we need translation

Typically, Without machine translation, a human translator who speaks both languages is required to facilitate communication between two dialect speakers. Is this human resource readily available at all times and in all circumstances?

The European Parliament, which speaks over 20 languages and has over 700 members, is an excellent example of this. Language translators are essential in bridging communication gaps for meetings, legal documents, and new laws, among other things.

Another interesting fact about translation is that it became more popular after World War II. The texts and messages that were passed around by enemies or collaborators piqued the interest and curiosity of warring factions.

How Machine Translation Started

The concept of machine translation arose from the need for faster translation in native dialects without the use of a human translator. Active Research in this area started since as far back as the 1950s c.a.

Imagine trying to read or respond to what’s app message/twitter message in another language, and everytime you require a language translator besides you.

Google Translate to the rescue!!

Amazing right!! How technology aids in bridging the gap in communication. We find this in several other applications such as;

Microsoft Translator
Yandex
Unbabel
Amazon Translate
IBM Watson Language Translator
Bing Translator, et cetera.

It’s all “ Machine Translation behind the scenes”. One its biggest motivation is — “Navigating through Language barriers”.

Interesting Fact: According to New York times, Google Translate alone serves more than 500 million monthly users and translates about 140 billion words a day”.

Now, Just like how food has different recipes and how we have different ways to solve a problem. Similarly, there are quite a range of approaches in Machine translation. Let’s take a brief look:

Approaches in Machine Translation

There is a concept called the “Vauquois Triangle”, which summarizes classical approaches of Machine Translation into 3 broad terms;

Direct Translation: The Lowest level which involves directly mapping words from one language to the other
Transfer-Based translation(syntactic or semantic); This approach involves either analysis first the structure of the source sentence, then do a mapping of the structure of the source and target languages, and finally generating an equivalent of the source language in the target language.
Interlingua Based translation: This takes into consideration the meaning of the source sentence first, then attempts a generation of the same meaning in the target language.

Professor Vauquois developed this concept in 1968, demonstrating visually how machine translation could be done at the time.

Several approaches have evolved over time, and researchers are constantly working to develop more useful machine translation systems. These are;

Dictionary Look-up
Rule-based Machine Translation
Example-based Machine Translation
Statistical Machine Translation(SMT)
Neural Machine Translation(NMT)

I have attached reference links below for further reading on these concepts. In my next series of articles, I will discuss Statistical MT and Neural MT with example implementations.

APPLICATIONS OF MACHINE TRANSLATION

Machine translation is available in a variety of formats, including text-to-text translation (as in Google Translate, Bing Translator, and others), speech-to-text translation (as in Voice Assistant), text-to-speech translation, and automatic video subtitles. Let’s take a look at some real-world examples from various industries.

Military use-case- Remember the origin of Machine translation?
Educational use-case: Coursera uses Machine translation to provide translation of a lecture in subtitle format for several languages. Another typical example is a conference or lecture where speech is delivered in one language and read or interpreted in textual format in another language. This is currently being applied by a university in Germany to assist foreign students who do not speak or understand German, to understand lectures been delivered in German.
Healthcare: There’s a company called “Canopy Innovations”, that developed an application called “Canopy speak” with the aim of bridging medical language barriers, using machine translator . The application “Canopy Speak” is a medical translator app, which runs on a phrase library. This was built with over 5000 medical phrases supporting over 15 languages including Spanish, Chinese, Arabic, Korean and several others. The company boasts of the app containing the largest corpus of pre-translated medical phrases ever developed. These phrases have been organized by frequently encountered procedures and medical specialties, and available in both text and audio.

E-commerce Use-case: Machine translation is also lifting the language barrier on global trade. One popular example is “Alibaba”, connecting global users to businesses with options of language translate embedded in the app.
Et cetera, there are many other use-cases applicable in our daily lives.

We’ve discussed what machine translation is, how it came to be, why we need it, and how it can be used. Is it possible that such a fascinating and useful technology has solved all of the problems associated with translation and communication barriers? The answer is-NO.

It is important to note that this is still a developing technology. There’s a long way to go still. Let’s look at some of the challenges of Machine translation:

Challenges of Machine Translation

Language is Ambiguous. One word could have different meanings in context, one word could mean different things in different languages. Ambiguity can be in form of part of speech(bank, as a verb and as a noun), pronouns(many languages use different forms of a word for different pronouns, example Arabic.)

Machine translation does not cover tonal differences such as Polite and Impolite speeches( I can’t speak with my boss/an official the same way I would speak with a peer/colleague), emotions, sounds and stress of words(a word’s meaning could change or have more depth depending on how stressed it the pronunciation is(i’m sure we all resonate with this)) et cetera
Some languages do not have a lot of resources that could be used to build a machine translation system. This is a very active area of research currently, called ‘ Low-resource Machine Translation” .For example, Languages such as English, Chinese, Spanish, French, Japanese and more of the European and Western languages are considered High-resource languages. High resource languages are languages for which many data resources exists, in form of texts/documents, videos, audio etc . On the other hand, low-resource languages have very little or no resources(either translated texts for religious purposes, historical texts etc).

Conclusion

In this article, I have shared a brief introduction on Machine translation, what machine translation is, why we need it and several approaches to it. The next series of articles will focus on Statistical and Neural MT approaches with implementation.

Thank you for reading!

References and Further Reading