Google Translate No Longer Needs Humans to Improve Translations

Google Translate Shuts Down Contribute

Google Translate has shelved its human feedback tool called Contribute. The Contribute feature allowed users to click a button to “improve this translation”, and write an alternative translation.

When the feature launched in 2014, Google said that Contribute would help to “incorporate your corrections and over time learn your language a little better.”

Google acknowledged the importance of Contribute to the development of Translate, saying “with your help, we’ve been able to add new languages and improvements over the years. Since then, our systems have significantly evolved, allowing us to phase out Contribute.”

However, users can still send feedback to Google, by rating the translation “good” or “poor”, and selecting why a translation may be poor from a select list of options.

Adam Bittlingmayer, CEO of ModelFront and former Google Translate engineer, told Slator that while “search, ads, videos or social feeds [have] successfully used human feedback loops at scale since the 2000s, […] so far, in translation, we’ve failed to use human feedback despite leading on model architecture.”

“The best feedback is incidental and just built into how every user uses the product”, he added.

The news has already generated some concern with Google Translate contributors. One volunteer, who contributed to improving Fulfulde — a Senegambian language spoken by approximately 36.8 million people — questioned if their efforts had gone to waste alongside other under-represented or long-tail languages in the tool: 

“As of 23rd March 2024, the language has reached a total of 52,600 contributions. This shows that the language has many dedicated people who are willing to promote and revive it; they don’t want their language to instinct”, added the user.

Google has recently touted its commitment to low-resource language translation in a different setting. Google’s AI Chief highlighted Gemini 1.5 Pro’s success in “learning” Kalamang, described as having “fewer than 200 speakers and therefore virtually no presence on the web.”

Hat tip to the Search Engine Journal for first covering the story.