A foolish headline. Maybe that’s the goal; this is just a baby step.
Twitter today had messages warning about an apparent subversion of Google Translate.
If you write China breaks promise the Chinese translation is China keeps promise. 中国信守诺言 If you were to write that the US breaks promise, the translation comes out correct 美国违约 — actually something like the US broke the treaty.
A few tries on people and animals found that Xi Jinping and pandas might break their promises in English but they do keep them in Chinese machine translation 熊猫信守诺言 as you can see below.
Not so Donald Trump and Bald Eagles 白头鹰违背了诺言 which break promises in both English and in Chinese translation.
Somehow there seems to be a reality distortion advantage for the Chinese home language team here.
Has Google Translate been brainwashed? It is a statistical translation system that operates by comparing large numbers of source and target language texts. The technical question of how this is done is intriguing. Could it be analogous to Google bombing — an algorithm subversion method — in some way? A few years ago some people subverted Google Search to make insults come up on a search on a US President
The problem seems to be in only English-Chinese machine translation and not the other way around.
Another Machine Translation Engine Passes the Test
I compared some of these machine to translation to a different machine translator in China at ichacha.net. Ichacha.net got them all right! Is there some malevolent attack on Google Translate to confuse Chinese people seeking to read online foreign language scriptures?That would be a Propaganda Department dream. Layers and layers of subtle obstacles is the trick for them. 南无阿弥陀佛！
Could there be some dark shadow out there? Many Chinese (well, many not as large a proportion as you might think but even a small proportion adds up to quite a few people) would like to get more news without the Party filter but the language barrier is high. Perhaps they will choose to use Google Translate, which can translate a webpage just by inserting the URL, to help them get forbidden fruit full of ideological poison. The Chinese Communist Party set up the Great Firewall of China to strives to protect people within the Garden of China from this poisons and foreign snakes who would lead them astray.
One wild theory I might suggest, without any evidence to support, is that this could be work of the Chinese Communist Party or some of its acolytes. The Party goes to great lengths to protect its ideological security which it sees as a core part of China’s national security since the Party towers high over the state in China.
Subverting Google Translate so that it speaks well of Chinese Communist Party General Secretary Xi Jinping might, in the minds of PRC propaganda people (propagating the Party line is the responsibility of all Party members but could be someone in the United Front Work Department or the Party Propaganda Department), help reduce the harm to people exposed to false foreign news reports and pernicious writings.
Filtering out and suppressing “bad ideas” is done layer-by-layer through persuasion at first gentle and only have repeated offenses gradually escalated. The Party seems to have found through experience (the Anti-Rightist Campaign? The Cultural Revolution?) that gentle warnings and gradual escalation works better than a sharp attack at first offense that would anger family and friends and attract sympathy. Intimidation not technical means is the Party’s most important method of control. See He Qinglian’s books and articles for more on media control in China.
Could this be confusion as a strategy! Reminds me of the story of President Eisenhower’s press secretary worried about a difficult question. Eisenhower told him, “Don’t worry, Jim; if that question comes up I’ll just confuse them.”
This is just my foolish hypothesis. Could be some patriotic hackers hongke 红客 ( 红黑客) working on contract or just raising hell (提升地狱 in an overly literal Google Translate mistranslation) for the fun of it too. Just because they can.
Or Just a Technical Problem Affecting Google Translate but Not Icha.net?
Or perhaps there is some technical subtlety that escape my poor understanding. I found that Google Translate mistranslate US breaks promise but translates US breaks its promise correctly. Reminds me of the old Shadow radio show opening “Who knows what evil lurks in the hearts of men?” Maybe it’s machines in this case.
This is just for your some situational awareness if you rely on Google Translate. Machines can be subverted. In machine translation studies, machine “errors” are called the poison cookie problem. Almost all delicious, but one poison. Can’t tell which one it is though!
Google Translate Helps Translators But Major Pitfalls, Shortcomings Remain
I find Google Translate a great help in doing translations these days. Although I do have to check line by line it does save time by helping get the basic structures and by being very accurate on things that have a narrow one-to-one relation in the source and target languages. Place names as well as scientific and technical terms for example. Fifteen years ago, I could translate faster than the time it would have taken me to untangle the machine-translated syntax and wording. Now Google Translate is accurate enough that I actually gain time by using it — fewer errors to correct.
But that it has gotten to that point only in the last ten years. How well Google Translate works depends also on the language pair and I imagine how many source and target language documents were input to the machine translation system. If you don’t know the target language, Google Translate should be used with caution — it works best on texts written in a simple declarative style and less on texts that are more abstract or have a more complex sentence structure.
For more on the weaknesses and strengths of Google Translate, see the article by master translator-master snformation scientist Douglas Hofstadter’s 2018 The Atlantic article The Shallowness of Google Translate.
See also the discussion of this in Professor Victor Mair’s article on Language Log.