The parent post to mine was theorizing that the reason the English was so mangled was because it had been translated from English, then back to English. I was replying that, if the researchers didn't use English as a first language, it's ridiculously more likely that they were translating from their native tongue into English. You're misunderstanding where the notion of translating it twice came from.
And sure, there are reasons to translate a phrase from English, to another language, and back to English. This will be familiar to most people who've studied abroad, or done technical conferences on foreign soil, things of that nature. Let's say you're from Bolivia, attending a lecture at an English university, and are planning on referencing some of the content in a paper you're writing, in English. You speak passable English. The lecturer gets into the meat of the topic, and you realize you don't quite understand the context of what they're saying. Some of the conjugations are unfamiliar, so you just write it down as best you can and move on. Later, when writing the paper, you need a way to untangle the phrasing. A simple way is to put it into a translation application, translate to Bolivian, then try to parse it in native tongue. However, you know you have to explain and discuss this section in English; by translating it back, you'll get the English words, but some of the context and grammar structure will be from familiar Bolivian.
My wife was a professional translator and I did my master's thesis on the topic. With modern translation engines, there is no way you'll end up with "haze figuring" or "arbitrary timberland" translating one way from any source language into English. I also doubt you could get a very specific word like "timberland" from repeated translation of "forest", intentional synonym replacement is much more likely.
> there is no way you'll end up with "haze figuring" or "arbitrary timberland" translating one way from any source language into English.
Good to know; I've always wondered at the peculiarities of different translation engines, but never really dug into them, as most modern ones seem like neural network black boxes to me. I was pointing out that there are some realistic use cases for doing round trip translations. I've used this technique at a few conferences to help straighten out my hazy understanding of a complex idea in a language I spoke quite poorly. And I do agree it is bad form to use this directly in an academic paper.
> Also, they speak Spanish in Bolivia ;)
To be fair they speak Bolivian Spanish, along with many other native languages! I chose Bolivia as a random target without doing any research, so thanks for the pedantic push to go learn something new; things like this are why I do love HN!
I chose Bolivia because I know next to nothing about it and it seemed neutral; I should have used a notional country, like the Republic of United Swiss Emirates.
You've never had to do that before? Maybe I run into it a lot due to the nature of the conferences I attend. I know just enough of the language for functional conversation, but as soon as a complex idea is put forward, I need to be able to contextualize the familiar scientific portions of it quickly, and the round trip translation usually helps enough that I can parse it correctly.
Not really, but I came across some teammates in college who didn't seem to be able to follow a conversation yet they seemed to have quite good writing skills. This might be the reason why ;)
I find it unlikely a machine translator would spit out phrases like "counterfeit conscience" and "haze figuring" over AI and cloud computing with 1 pass. Plagiarism via multiple pass throughs seems much more likely.
Ironically this is exactly what I would have expected from machine translation 15 (edit: 20?) years ago. My friends and I used to get a kick out of running phrases through several rounds of machine translation in different languages and finally back into English, and then playing them with the Mac OS 9 text-to-speech system.
Maybe unlikely between two Indo-European languages with closely related sentence structures and vocabularies, but plausibly likely for others. DeepL just gave me "potter's screw" for "pan-head screw" in my language, for example.
It's also a common technique for people who don't speak English to translate it... In fact, quite a bit more common.