Tortured phrases: A dubious writing style emerging in science • Hacker News Board

Re: Tortured phrases: A dubious writing style emerging in science

#71

Post by ksaj » Sun, Aug 08, 2021, 7:14 PM UTC

I noticed this happening in other areas a few years ago, but with faked blogs. The titles and subjects would sound interesting, but then when you tried to read them, you'd need a specialized decoder to get through the utterly baffling word replacements. But they already got their ad revenue by the time you notice the article is complete gibberish.

The first one I found was about dog illnesses. They kept referring to dogs with phrases like "Your domesticated canine," and it was quite a chore trying to figure out most of the symptoms that they were listing. "Heart worms" was translated to "love snakes," which I thought was delightful.

Re: Tortured phrases: A dubious writing style emerging in science

#72

Post by lostlogin » Sun, Aug 08, 2021, 7:15 PM UTC

gzer0 wrote: post #35

This is anecdotal evidence at best, but it is worth considering. I know of several individuals who were able to complete their entire Master's thesis utilizing a combination of AI generated content (GPT-3) and a paraphrasing tool. The generated text was well over 50 pages, completely bypassed all known content/plagiarism checks and was even included in the Universities "exemplary examples". To this day, it is still t…

> third party content generation and paraphrasing tools.

Presumably this is an arms race against things like https://www.turnitin.com/

Empower students ‘to do their best, original work’ and this is what you get. Though what the alternative is, I have no idea.

Re: Tortured phrases: A dubious writing style emerging in science

#73

Post by zozbot234 » Sun, Aug 08, 2021, 7:16 PM UTC

wmf wrote: post #61

zozbot234 wrote: post #22

No mentions of tortured phrases in the humanities and softer social sciences? For all their supposed appreciation of les belles-lettres (viz., "fine writing") those researchers sure seem to like their tortured phrasings.

That's a completely separate issue that shouldn't be conflated.

> That's a completely separate issue

How so? It seems quite related to me. Anecdotally, one would expect a pretty clear negative correlation between torturedness in the sense of this article and indicators of research quality.

Re: Tortured phrases: A dubious writing style emerging in science

#74

Post by ipsum2 » Sun, Aug 08, 2021, 7:19 PM UTC

A high profile case (on the internet) similar to the one described in the article is when Siraj Raval plagiarized a paper on quantum ML and made some amusing replacement phrases:

complex Hilbert space -> Complicated Hilbert space

Quantum gate -> Quantum door

https://www.theregister.com/2019/10/14/ravel_ai_youtube/

Re: Tortured phrases: A dubious writing style emerging in science

#75

Post by hdjjhhvvhga » Sun, Aug 08, 2021, 7:21 PM UTC

atrettel wrote: post #60

I have encountered something similar to this for a submission that I reviewed for a scientific journal. I will not list any names or give much detail past those generalities, but I pointed out that the authors were misusing a particular technical term. In my review I defined the term and explained it briefly. I asked the authors to revise their submission accordingly. The paper was not bad but the authors did not kno…

> The paper was not bad but the authors did not know English very well, so it was quite difficult to read. That was its main problem.

This seems to confirm my suspicion than these cases are not so much about AI-generated content but rather a result of machine translation.

Re: Tortured phrases: A dubious writing style emerging in science

#76

Post by Freak_NL » Sun, Aug 08, 2021, 7:22 PM UTC

aaron-santos wrote: post #67

FabHK wrote: post #62

Some of these tortured phrases are great. My favourites: "flag to clamor" for signal to noise "individual computerized collaborator" for PDA (personal digital assistant) "haze figuring" for cloud computing "information stockroom" for data warehouse "focal preparing unit" for CPU "discourse acknowledgement" for voice recognition "mean square blunder" for MSE (mean square error) "arbitrary right of passage" for random…

I enjoyed finding "counterfeit consciousness" for artificial intelligence. To me it evokes a kind of science fiction that's shown up occasionally on HN[1]. [1] https://qntm.org/mmacevedo

Also “haze figuring” for cloud computing.

It sounds like something you'd find in 30s, 40s, 50s sci-fi for sure! Like “visiplate” (E.E. “Doc” Smith, Heinlein) for a computer display screen. (Along with ticker tape printouts and tape reels in the far future of course.)

Re: Tortured phrases: A dubious writing style emerging in science

#77

Post by robertlagrant » Sun, Aug 08, 2021, 7:25 PM UTC

rusk wrote: post #68

In telecoms they call all the backend infrastructure “back haul” and have never read a satisfactory explanation. I’m convinced that somebody once coined “back hall” with the intention of invoking the image of service passages like what you see in the mall, it was misheard (as is often the case in Telecomms given it’s global nature) and the metaphor of the bulldozer tail stuck for ever after

IME I think backhaul is just sending data to the main internet, not all backend. I thought it just meant it hauled the data back into the core network.

Re: Tortured phrases: A dubious writing style emerging in science

#78

Post by hdjjhhvvhga » Sun, Aug 08, 2021, 7:27 PM UTC

bjourne wrote: post #64

gzer0 wrote: post #35

This is anecdotal evidence at best, but it is worth considering. I know of several individuals who were able to complete their entire Master's thesis utilizing a combination of AI generated content (GPT-3) and a paraphrasing tool. The generated text was well over 50 pages, completely bypassed all known content/plagiarism checks and was even included in the Universities "exemplary examples". To this day, it is still t…

I really doubt you can computer generate a Master's thesis. Completing a Master's thesis at an accredited institution is a heck of a lot of work and even a cursory reading of a thesis by an examiner, supervisor, opponent, or other interested party would give the generated content away. Maybe if you get your degree from a diploma mill you could get away with it, but then your degree wouldn't be worth toilet paper anyw…

I agree that in most cases it would be very difficult to do. But I can imagine some specific circumstances where it could be pulled off, possibly with some manual modifications: soft sciences like sociology (you can't imagine the amount of bs I've read during my college years), the subject matter being very different from the area your supervising prof specializes in, the topic that allows for arbitrary speculation, an underfunded university branch with profs having a more lax attitude.

Re: Tortured phrases: A dubious writing style emerging in science

#79

Post by robertlagrant » Sun, Aug 08, 2021, 7:28 PM UTC

tnzm wrote: post #37

waterhouse wrote: post #27

Earlier quoted context omitted.

Sometimes technological solutions work really well to solve social problems. For example, at one point, one person using the internet would tie up the phone line for everyone else in the house, and vice versa. Negotiating this shared resource could be considered a household social problem. But now there's no such interference, and most people have their own cell phones.

This is a social problem around the shared use of a technological resource. I'm reminded of the old saying, "computers can only solve problems that are created with computers". But then again you can view _all_ solutions to social problems as inherently technological in the broader sense; I adhere to that paradigm.

That saying seems silly. Computers (i.e. Zoom) help with the problem of needing socially distanced education during Covid lockdowns.

Re: Tortured phrases: A dubious writing style emerging in science

#80

Post by dimatura » Sun, Aug 08, 2021, 7:30 PM UTC

ksaj wrote: post #71

I noticed this happening in other areas a few years ago, but with faked blogs. The titles and subjects would sound interesting, but then when you tried to read them, you'd need a specialized decoder to get through the utterly baffling word replacements. But they already got their ad revenue by the time you notice the article is complete gibberish. The first one I found was about dog illnesses. They kept referring to…

Yes, this may be a specific example of a more widespread phenomenon. There's certain websites out there that republish articles from well-established publications (e.g., New York Times) almost word for word, except that they are rife with synonym swaps that may or may not make sense in context, presumably to escape some kind of automated copy detection. Results can be amusing. For example, the copied article said "“Drukqs” acquired a blended essential response..." where the original said "“Drukqs” received a mixed critical response...".