- Table of Contents
1. History of AI Translation
AI translation (machine translation by computer) first appeared in the 1950s. The technology initially used was rule-based, but in the late 1980s, statistical machine translation and in the 2010s, neural machine translation were introduced.
In rule-based machine translation, rules for translation are manually created based on dictionaries and grammar. It was necessary to create a huge number of rules, which took a lot of time and effort for development and updating with new words. The accuracy was not high and it could only translate mainly standard phrases.
In statistical machine translation, computers learn rules instead of humans. They read a large number (such as 1 million sentences) of pairs of original and translated texts, and learn by statistically matching words and phrases from the data (called a corpus). If additional original and translated texts are collected, it is relatively easy to match them to new terms. However, translation between languages with significant grammatical differences, such as English and Japanese, is difficult and the translation accuracy is not yet practical.
In addition, there are also technologies such as hybrid machine translation that combines statistical machine translation and rule-based translation, as well as example-based machine translation that extracts similar parts from existing source and target pairs for translation. These machine translations have improved translation accuracy compared to rule-based and statistical machine translation.
In neural machine translation, similar to statistical machine translation, the computer is trained by inputting a large amount of pairs of source and target texts. However, by using neural networks and deep learning, a type of machine learning, it is able to extract much more information than statistical machine learning and use it for translation. Compared to traditional machine translation, the translation accuracy has greatly improved. The output is characterized by fluent translations that resemble those done by humans. With the emergence of neural machine translation, machine translation has gained attention and is now widely used in daily life and work.
2. Differences between Conventional and Latest Automatic Translation Technologies
The main difference between traditional rule-based machine translation and statistical machine translation, and the latest automatic translation technology, neural machine translation, is the fluency of the translated text. Traditional machine translation often produces unnatural text that is easily recognized as being machine translated. Additionally, since it translates sentence by sentence, the connection between sentences can be awkward. On the other hand, the translated text from the latest neural machine translation is natural. If it can translate by paragraph, like DeepL, the connection between sentences is natural and it can be difficult to determine if the text was translated by a human or a machine when reading through a paragraph.
On the other hand, there are also issues that have become more prominent with neural machine translation. These include translation duplication and omissions. In traditional machine translation, each word or phrase in the source text is reflected in the translation, so duplication and omissions rarely occur. However, with neural machine translation, there may be cases where words or phrases are duplicated or omitted. This is especially true because the translations produced by neural machine translation are often fluent, making it difficult to spot omissions when reading the translation alone. To identify omissions, it is necessary to compare the source text and the translation.
3. Features of Popular Machine Translation in the World
The representative machine translation services are DeepL, Google, and Microsoft, all of which use neural machine translation technology. Among these, DeepL is particularly noteworthy. The feature of DeepL is its fluency. And what makes this fluency possible is paragraph translation. With paragraph translation, the context and field can be accurately understood by translating the text in paragraph units, resulting in not only the use of appropriate terminology, but also a natural flow between sentences. As a result, the translated text becomes more fluent.
For the latest trends in machine translation and a comparison of DeepL, Google, Microsoft, and Amazon, please see "Latest Trends in Machine Translation and Comparison of "DeepL" and "Google Translate" (https://www.science.co.jp/nmt/blog/32334/).
4. Summary
AI translation has evolved from rule-based machine translation, statistical machine translation, to neural machine translation. Translations using neural machine translation are more fluent than before and may be indistinguishable from those translated by humans. However, it is still necessary to compare the original text with the translated text to check for any repetitions or omissions.
We offer a translation software "MTrans for Office" (MTrans for Office) that incorporates machine translation services from DeepL, Google, and Microsoft. With just one click, you can translate Microsoft Office products (Word, Excel, PowerPoint, Outlook), which can also lead to reduced workload. Please try our 14-day free trial to confirm the quality and usability.
Related Services
Translate Office with the easy translation software MTrans for Office
Return to the top of the page
+81-3-5321-3111[Tokyo]
Contact Us
Request for Information