Machine translation is one the most important innovation that helps many people solve their problems in communication, i.e. the language barrier. This is actually not a new thing since the first machine translation was actually developed almost 100 hundred years ago. The purpose was the same as today which is to help people understand other languages without previously taking a long course or hiring a translator.
The History Machine Translation
The first machine translation was born during the World War 2 era for military purposes. It was actually a mechanical tool for translating medical documents that would be used to save more lives during wartime.
The designer of this very first mechanical bilingual tool was a French engineer Georges Artsrouni. This tool was designed to store information retrieved from printed documents and store them for further application. The designer claimed this as a mechanical brain that could function as a bilingual dictionary and other usages.
The development of mechanized translation tools had been continued by a Russian scholar named Peter Troyanskii who introduced a more complex bilingual dictionary that included grammatical rules in language transferring. His work had never resulted in a final remarkable product that was widely used, but it set a foundation for the development of computerized machine translation.
Several years after World War 2 over, the invention of the computer had given a significant technological leap for the development of machine translation. Many countries were in a race for pursuing this technology even though it was only dedicated to military usage. This early model of MT was still not comparable with the machine we used for free today. There were very limited language pairs and also a limited area of translation the machine could handle at a time.
Luckily, the ability of modern computers to quickly store and process information has been very significant in improving the quality of translation results. The ability also improves the numbers of language pairs compared to the previous version developed during the cold war era.
How can a machine translate a language?
The first generation of machine translation used a set of rules and statistics to translate a text into a selected target language. It might sound like stark and stiff translation during that time but the invention had attracted further development that resulted in technologies that are almost flawless and almost as natural as the human translation
There were actually several popular approaches of translation applied in machine translations during the early computer era. The first computer-based translation was using the ‘rule-based’ approach or also called a classical approach. This approach was to feed the computer program with data containing fundamental linguistic features taken from dictionaries and grammar rules. The program also used semantic, morphological, and syntactical rules from both languages to make the translation.
The second generation of machine translation no longer applied the first classical rule. The new approach was based on practical language usage or called “example-based machine translation”. This approach was not merely using grammatical rules and other linguistic features, but using the bilingual corpus of parallel sentences. In the process, the machine would break down the source language and analyze the part which would be transferred using analogy to the target language.
Although this approach was popular during 1980’s era, today CAT translation tools are still applying the system named translation memory to help translators quickly translate a language into a target language.
When computer device was more capable in the 1990’s and 2000’s, the schema for translation was also improved. The translation became a statistical process that converted the input into mathematical code based on word-level, phrase-level, and sentence-level. The machine would use the encoded data to create a new form of data from the target language. It could also learn from syntactic differences between language pairs that could result in a better translation compared with the previous version.
The machine translation we know today is actually a rooted and programmed system that is far more complex compared to its preceded generation, of the early computer translation tools until early 2000’s machine translation. They said that the current version of translation tools uses a newer framework that is able to mimic the human brain process to make an analogy and recognize the relationship between language pairs.
The neural approach is applied in this version which can learn mathematical functions and use machine learning algorithms amidst the process. The artificial neural network is a more capable system that deploys a deep learning process to create an equivalent translation by harmonizing gender and context. Different from the previous statistic approach, the neural process does not require taking the source separately instead of establishing a network of interconnected components as it will always learn to tune up itself while being continuously used.
How Machine Translation is Used Today?
Since the first time it was born, machine translation is a designated platform that can help humans understand language faster and more accurately. With globalization has come to its peak, many parties need MT to help them to communicate with other parties who speak other languages, quickly and reliably.
On a larger scale, many giant companies use the service of MT to help them localize their product to the target customers. It is far cheaper compared with the human translation that can cost hundreds of times higher but takes a longer time to complete. It is a vast market size that continuously grows with an estimated global value of 3 billion USD in 2027.
Comparing the MT and Human Translation will trigger a long debate that each has its own strength which cannot be leveled by the other. Yet, it is still a sensitive debatable topic if one day human translation will be replaced completely by artificial intelligence.