A journalist has complained that Google Translate is translating the more neutral Spanish term ‘indocumentados’ – which means undocumented – into ‘illegal’. As a result, translated headlines and phrases where a writer has attempted to achieve political distance by referring to immigrants as ‘undocumented’ rather than ‘illegal’ do not have the tone the author intended.

Two days ago (August 12th), Jorge Rivas of ABC Univision’s Fusion TV channel wrote an open letter to Google about the issue, which was published by ABC News. In it, he called on the search machine to ensure it was “honest and accurate” in the translations it uses. However, he revealed yesterday that Google had responded saying it would not update the failed translation as it was up to users to suggest better translations to it. It is these recommendations that help the system to update itself.

In his original letter, Mr Rivas noted that in many countries, journalists are dropping “illegal” from their articles on immigrants and instead referring to them as “undocumented immigrants”. He pointed out this is because using the word ‘illegal’ could help encourage “anti-immigrant sentiment”. Such terms can also have an impact on peoples’ attitudes to those who speak a different language, he noted.

While there are currently campaigns underway to get news agencies and publishers to stop using the term ‘illegal immigrants’ altogether, in the cases where Google has translated the articles incorrectly the word ‘illegal’ did not appear in it to start with. As a result, not only does the translated article have a different tone and political leaning to the original, but it is also inaccurate, Mr Rivas explained. “When people use your system, you are promoting outdated and very harmful language even if that is not your intention,” he added.

To refer to immigrants as ‘illegal immigrants’ damages attempts at reform and stands in the way of the development of humane solutions to immigration reform, in addition to the language being “dehumanising”, Mr Rivas claimed. While he stated that he was not requesting that Google be politically correct with its translations, he would ask that it was accurate and that the translation was an honest reflection of the original.

Mr Rivas noted that his research had led him to discover that eight out of the ten stories featuring the word ‘indocumentados’ that he translated with the free software featured the English term “illegal immigrant” afterwards. However, in a statement delivered to Fusion by Google product communications specialist Ricardo Blanco, the search engine said its translation tool produced automatic translations based on material already present on the internet. Indeed, most machine translators work by analysing countless translated documents online, recognising repeating patterns in different language pairs and memorising these to assist in future translations. “Since the translations are generated by machine, they’re not always perfect, but we’re constantly working to improve the quality of our algorithms, and we appreciate this feedback,” Mr Blanco concluded.

Following the statement, Mr Rivas contacted Mr Blanco directly by telephone to urge him of the importance of correcting the issue. He urged that Google should regard it as a “failed translation”, but claimed that Mr Blanco did not appear to see the problem with the same urgency as him.

This is not the first time that a machine translation has been accused of being politically incorrect. A recent case study carried out as part of the Gendered Innovations project at Stanford University revealed that gender bias has also been noted in automatically-generated translations. It is likely this error is occurring for the same reason as the undocumented/illegal issue. Because Google and other machine translations work on statistics, they base their answers on the most commonly-occurring phrases. For instance, the case study found that English to German translation of the word ‘defendant’ would return the masculine form of the word, while the same action for the word ‘nurse’ would return the feminine form. This happened regardless of content.

While there is no denying machine translations have come on leaps and bounds, they remain a long way off from matching the skills and ability of a human translator. Until a computer can recognise the tone, context and political content of an article, a human translator will continue to produce the best results.