Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I don't think a conspiracy theory is needed for this case. It's just an unfortunate glitch of the machine learning. On the other hand, I acknowledge problems like these have serious consequences.

After Google Translate's 2016 algorithm update (https://spectrum.ieee.org/tech-talk/computing/software/googl...), anecdotally, many people started seeing it having problems on distinguish between a positive sentence and a negative sentence (e.g. has/has not, with/without), even simple ones.

Also, when Google cannot "understand" a sentence, it simply gives an output that happens to have the best match of its training patterns, often the opposite meaning. For short sentences without context, it's especially problematic.

In 2018, Google Translate briefly became a Twitter meme when it made the following translation. I cannot find the original thread, but I quickly Googled some keywords and found related tweets.

* 谷歌手机很卡顿 (Google's mobile phone is very laggish) => Google mobile phone is very fast https://twitter.com/abcdsxg/status/980985925894197248

I tried to find a random Wikipedia article, and I was still able to reproduce a similar problem today.

* 考虑到庭审涉及信息的敏感性,法庭分别进行了两次独立的听证会,一次公开进行,另一次不公开 (Because of the sensitive nature of information at stake in the trial, two separate hearings were conducted, one in public, and the other in private) => Taking into account the sensitivity of the information involved in the trial, the court conducted two separate hearings, one in public and one in the open.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: