Machine Translation

Machine Translation Overview This article provides an overview of Machine Translation (MT). Since the 1950's, many research groups have been investigating the original idea of ​​MT, and many MT systems have been created and developed around the world. Three ways of MT system: direct translation, transfer, and language-to-language method is a common system. The main idea of ​​the direct translation method is to replace words with words before converting the structure from source language (SL) to target language (TL).

Machine translation (MT) automatically translates text from one natural language (source language) to another language (target language) using machine capabilities. The idea of ​​using a machine for translation was first proposed by Warren Weaver in 1949. Machine translation has been done for a long time (from the 1950's to the 1980's) by examining the language information of the source language and the translation language, and generating a translation based on the dictionary. And the grammar is called Rule Based Machine Translation (RBMT). Along with the development of statistics, statistical models became applied to machine translation, and translation was generated based on an analysis of bilingual text corpus. This approach, known as Statistical Machine Translation (SMT), achieved superior performance than RBMT and dominated over the 1980s and early 2000s. Their research laid the foundation for the future application of neural networks in machine translation.

Machine translation usually uses phrase-based statistical methods, but neural networks are rapidly becoming dominant. My favorite architecture is trained in a multilingual environment where a single model has been translated from any source language to any target language and has been set to be weakly supervised (or not fully supervised). Game Although the clearly written Go program has been developed for a long time, AlphaGo Zero, ConvNet, which examines the original state of the board, is currently the most powerful player in the game. I hope that you will get very similar results in other fields such as DOTA 2 and StarCraft.

Machine translation between languages ​​is an example of a rule-based machine translation method. In this approach, the source language, ie the text to be translated, is converted to an inter-language language, which is an independent "language neutral" representation between languages. Next, we generate a target language from an intermediate language. The main advantage of this system is that as the number of languages ​​that can be converted to a target increases, the value of the intermediate language increases. However, the only inter-language machine translation system operating at the commercial level is the KANT system (Nyberg and Mitamura, 1992), which aims to translate Caterpillar Technical English (CTE) into other languages.