Have you ever seen one of those advertisements where an English speaking tourist reaches an Asian country and is left bewildered because he doesn’t understand the local language? And then another tourist comes along and shares with him a small device in which you speak your tongue and it automatically translates into the language of the country our tourist was in? Yes, while this may quite literally be an example of machine translation (it involves a machine that translates), Machine Translation actually is much more complex. But what is Machine Translation exactly? Let us not confuse Machine Translation with Computer-Aided Translation (CAT).
Machine Translation is the task of converting one sequence of words in the source into the other sequence of words in the target while maintaining the same level of contextual information using a computer program. In simple language, it is a branch of computational linguistics that works with software that translates text or speech from one language into the other. While Machine Translation has definitely made the process of translation easier and more efficient, the mere substitution of words from one language to the other rarely produces a good translation. Not only does it overlook the context, but not all words in a language also have equivalent words in other languages. And if that wasn’t a hurdle enough, many words in many languages have multiple meanings. Today this problem is being addressed with research and development in rapidly growing fields such as deep learning where within it tremendous work is being done on linguistics typology, translation of idioms, etc.
But before we dive deeper into what is happening in this field today, let us take a quick look at the history of Machine Translation and how it reached the stage where it is today. Quinn DuPont in her article on the origins of Machine Translation traces back the beginnings of this technology to a 9th-century Arabic cryptographer called Al-Kindi who developed the techniques of systemic language translation which are used in modern-day machine translation. The purpose of Al Kindi’s development of cryptographic analysis of the Arabic language was to break it down to character, based on the frequency histogram and different permutations and combinations. Isn’t it ironic though, that the first guy who developed the concept of Machine Translation was Arabic and yet Arabic Machine translation has one of the lowest accuracies when it comes to a majorly used language in the world! In fact, Arabic Machine Translation is one of the most difficult to get an automatic translation without any human intervention. We, at EZ Works, have a team of R&D engineers and scientists who are tirelessly working on improving the quality of Arabic to English and vice versa Machine Translation using our proprietary dataset and the latest algorithms in the space.
However, in the modern era, Machine Translation can be traced back to the 1930s when Georges Artsrouni applied for the first patents for an automatic bilingual dictionary. Since then, this field has seen quite a few interesting proposals and demonstrations, leading up to the launch of Google Translate in 2006.
Today, there are about 4 broad categories of approaches within Machine Translation. These are :
- Rule-based Machine Translation (RBMT): RBMT translates on the basis of grammatical rules. It analyses source and target languages grammatically to produce the translated sentence. This, however, requires thorough proofreading. Besides, its inescapable dependency on lexicons demands a long period of time before any efficiency is achieved.
- Statistical Machine Translation (SMT): It refers to statistical models based on the analysis of large volumes of bilingual text. It tries to establish the correspondence between a word from the source language and a word from the target language. Although SMT is great for basic translation, its greatest drawback is that it does not factor in context, thereby leaving the translations full of errors.
- Hybrid Machine Translation: RBMT and SMT work together to make, as the name suggests, HMT. One of the outstanding features of HMT is its translation memory, making it rather popular today, despite its share of drawbacks. Since Machine Translation doesn’t have any translation memory system, the output of RBMT is fed into SMT to generate the final sentence. That is how the two work together to form HMT.
- Neural Machine Translation: NMT is a type of machine translation that is based on deep learning, which is neural network models, imitating the human central nervous system to develop probabilistic language models. The primary benefit of NMT is that it provides a single system that can be trained to decipher the source and target text. As a result, it does not depend on specialized systems that are common to other machine translation systems.
As machine translation applications are reaching significantly high accuracy levels, they are being increasingly employed in more areas of business, introducing new applications and improved machine-learning models. Some of these areas include Machine Translation in Industry for Business Use (for example in the government, software, and technology, military and defense, Healthcare, Finance among many others), Online / App Machine Translation for Consumer Use. These machine learning applications perform instant translation for textual, audio, and image files from a source language into a target language, among many others.
At EZ Works, although we are embracing, in fact contributing to all the developments in the field of Machine Translation, we are constantly aware of the fact that it is only to assist our brilliant service experts. We might not know what the future holds, but right now, at this moment, we believe technology is to assist human intelligence, not replace it entirely.
- Schubert, Lenhart, "Computational Linguistics", The Stanford Encyclopedia of Philosophy (Spring 2020 Edition), Edward N. Zalta (ed.)
- What Is Deep Learning?
- The Cryptological Origins of Machine Translation, Quinn DuPont, January 2018
- Uses and applications of machine translation. Presented at University of Westminster, February 2009