One Trend in Machine Translation: case Facebook

Just last week Facebook published that it has stopped offering Microsoft’s machine translations (MT) in its service. Instead Facebook has developed its own MT system that currently does all the real-time translating in Facebook service.

This is one more confirmation of a trend among big internet giants: everybody is developing their own MT systems. This applies to e.g. companies like Google, Microsoft, Baidu, Yandex, Alibaba, Ebay, Amazon, etc.

The main reason for the trend is MT quality and ease of development. Currently developing own MT does not require that much research or technical understanding but instead requires a big collection of bilingual texts, i.e. texts in two languages. The MT ‘engines’ are publicly available and for developing own MT, only training the MT engine with the bilingual text collection is required. When the MT is trained with own texts, for that particular type of text its quality will be better than e.g. Microsoft’s or Google’s MT that has been trained with common, non-specialized texts.

The trend may also mean that MT is increasingly seen as a strategic capability that one is not willing to outsource from other companies.

Published by

Multilizer / Niko Papula

I am managing director of Multilizer, a Finnish software company specialising in software for enhancing translation quality, speed and cost.