Applying AI into Translation, Can We Cross Language Gap?

In 1954 the first public demonstration of the Russian - English machine translation facilitated the research and stimulated founding in this area. However, due to the complicity of human language, for the next half century, there are few research could be put into a practical application.

The major developments of machine translation have only been achieved in near 10 years. In September 2016, Google released the Google Neural Machine Translation System (GNMTS), which claimed that the system was able to apply neural network technology to mimic the thinking model of the human brain, providing high-quality translations comparable to manual translations, reducing errors by 55% to 85%. Google has applied the technology to web page translation and mobile applications, and the quality of translations has improved significantly.

In this article I will use Itranslate as an example. So far this APP has the largest amount of languages in all the translation APPs I have seen. Comparing to the translation quality 3 years ago, I have witnessed a drastically improvement.

The instant translation does have a surprising development. There is no doubt that you can use this APP to conduct daily communication with English - speaking foreigners.

After testing several sentences, although there are some flaws, but the overall translation are still in place. However, the limitation is that this APP seems only be able to correctly translate English to other languages and vice verse. When tested by Mandarin to the Nepali language and the Nepali language to Mandarin, the structure of sentences will fall apart.

For the future development, the machine translation

may also need work for the following aspects.

1. Expend of colloquial corpus

The flaws or mistakes in colloquial translation are mostly due to the lack of colloquial material. The problem should be solved by the expansion of the colloquial corpus, to increase the oral text corpus, including common colloquial vocabulary, expressions and life languages.

2, Improve the ability to understand context

Machine translation is not based on understanding, but based on statistics and corpus, can only do roughly accurate translation but ineffective in translating sentences expressing emotions.