This universal computing language, called Interlingua, cannot be used by humans. The translation method applied by Google developers is called Zero-Shot translation. This technology is more sophisticated than the previous one and relies on an artificial intermediate language. This type of research is gaining momentum, and these systems are on their way to becoming the number one method of machine translation. The system's self-learning feature allows the neural network to accurately translate slang, jargon and neologisms, which do not exist in traditional dictionaries. The neural network can also exploit the letters with which words are constructed.
This is very useful for transliterating proper names congo-brazzaville business email list Language combinations The GSTAN system has significantly improved the translation of the two most widely used language combinations: Spanish-English and French-English. As a result, the percentage of translation accuracy increased to 85%. In 2017, Google asked regular Google Translate users to complete a large-scale opinion survey. They were asked to rate three translation options: statistical, neural, and human. The results are impressive: translation based on neural networks is in fact almost perfect for several language combinations. The following results table highlights a 6-point evaluation system to rate the quality of the translation.
The maximum score is 6, the minimum is 0. Statistical model Neural network Translation made by a human Spanish-English combination 4,885 5,428 5,550 English-French combination 4,932 5,295 5,496 English-Chinese combination 4,035 4,594 4,987 Spanish-English combination 4,872 5,187 5,372 French-English combination 5,046 5,343 5,404 Chinese-English combination 3,694 4,263 4,636 As you can see, the quality of the translation in the English-Spanish and French-English combinations is not far from that of a human translation. READ ALSO German-French technical translation: at the service of your business development This is not surprising since these combinations have been used for deep learning of GT algorithms.
Websites, apps, social media and blogs
-
- Posts: 151
- Joined: Mon Dec 23, 2024 3:59 am