The Gulf Between Japanese and English Google Image Search

By Anri Asagumo, Oxford Uehiro/St Cross Scholar, (with input from Dr Tom Douglas and Dr Carissa Veliz)

Trigger Warning: This article deals with sexual violence, which could be potentially upsetting for some people.

Although Google claims in its policy that it restricts promotion of adult-oriented content, there is a district in the online world where their policy implementation seems loose: Google image search in the Japanese language. If one looks up ‘reipu’, a Japanese word for rape on Google, the screen fills up with a heap of explicit thumbnails of porn movies, manga, and pictures of women being raped by men. The short descriptions of the thumbnails are repugnant: ‘Raping a girl at my first workplace’, ‘Raping a junior high-school girl’, ’Raping cute girls’, ‘Raping a female lawyer’, ‘Raping a girl in a toilet’. As if rape in itself were not repulsive enough, many descriptions go even further, implying child rape. Similar results show up with ‘reipu sareta; I was raped’. It is strikingly different from the world of English Google image search, in which the top images usually send strong messages of support for victims and zero-tolerance for sexual offenders. Another example of how the Japanese Google world is different from that of English is ‘Roshia-jin; Russian people’. Searching in Japanese yields 17 pictures of young, beautiful Russian women, while searching in English returns pictures of people of different age and sex.

What is the cause of these differences? There are several plausible explanations. One reason could be that there are far fewer websites and articles written in Japanese about offering help for rape victims or shedding light on rape crime, compared to those in English, so Google inevitably suggests the hideous images, lacking alternatives. Another reason could be that Japanese people prefer to click on these websites, and the algorithm learns and perpetuates their preferences. Yet another explanation could be that Google’s safe search filter is just far less effective in Japanese.

Either way, Google should be taking steps to ensure more ethical results. To do so, it could use search preferences from one language to improve their search result ranking for another. Google might say that they are giving each language group what, taken collectively, they want. Moreover, they might point to difficulties in finding equivalent search terms across languages, perhaps raising the examples of ‘false friends’, words in two languages that look or sound similar but differ significantly in meaning. ‘Mansion’ in English refers to a large, impressive house, while the exact translation of it into Japanese, ‘manshon’, means ‘apartment building’. In such a case, Google might claim that extrapolating preferences from one language group to the other would skew search results.

However, I still argue that Google should take a more paternalistic approach when the images concern harm to people. We should not let the victims of sexual offences hurt again when they need help and are looking for support. We should not send the wrong messages to potential sexual offenders that what they look for online is something normal, or socially acceptable, as the availability and easy access suggest.

If the Japanese image results for ‘reipu’ are due to a shortage of Japanese websites that offer supportive content for rape victims, Google might claim that it is not their fault that the extremely repellent images come on top. In such a case, I would go so far as to argue that they should try to show the images that would be found when searching in a different language. Google has an obligation to improve auto-translation so that these websites with right messages reach people who need them.

A somewhat related problem is the difficulty Japanese people experience when they look for trustworthy information on diseases and medical treatments. When searching in English on google.co.uk, the NHS’s official website and other trustworthy sources tend to show up first. However, in Japanese, the top results generally come from blogs or the websites of individual hospitals. To find information from more authoritative sources, one generally needs to specify a particular medical society in one’s search.

The divide between the Japanese online world and the English online world is enormous. It is so immense that it gives parents a reason to bring up children to be fluent in other languages just to help them navigate the digital world more safely. Parents nowadays try many methods to protect their children from the dangers of the Internet — limitations on screen time, monitoring social media accounts, filtering harmful websites and so on. Perhaps as part of this ‘digital education’, we should encourage children to learn new languages so that they are better equipped to find trustworthy information.

There is much that can and should be done to improve Google search in Japanese. Google could ensure more ethical results with its cutting-edge technologies. The Japanese government could create a trustworthy standard list of symptoms, diseases, and treatments. Parents could encourage their children to learn other languages to navigate the digital world better. All parties should work hand in hand to improve the online world.