A dubious writing style emerging in science | Hacker News Reader

A dubious writing style emerging in science

arxiv.org

110 pointsdoener40 comments

I'm dying. Table 1 is incredible. It's machine learning jargon from a parallel universe. "counterfeit consciousness" "fog computing" "information stockroom" "profound neural organization" "credulous Bayes" "mean supreme blunder"

I came here to hahaha at the same thing

> machine learning

Did you mean apparatus scholarship?

From a coextending cosmos.

The thing is - I really like counterfeit consciousness in some contexts - where the AI is being used deceptively.

The phrase is perhaps apt in some cases, but the term consciousness has no universal scientific definition. I would therefore not expect to see it used in a serious manner in a scientific paper.

Does the term "Intelligence" have a universal scientific definition, then?

"designs preparing unit" = graphics processing unit (GPU)

"focal preparing unit" = central processing unit (CPU)

"individual computerized collaborator" = personal digital assistant (PDA)

It really is quite hilarious.

And, branching out to biology: "underground creepy crawly province" -- among others -- for "ant colony".

What all of these phrases have in common is that their words have been individually replaced with synonyms, with no regard to context. The author correctly identifies this behavior as characteristic of a text spinning service -- I've seen the same sort of thing frequently in spam.

They also sound like mistranslations from round-tripping through another language.

Followup:

The phrase "underground creepy crawly", in isolation, appears to be unique to whatever text spinning tool is being used to generate these papers. A Google search for that phrase turns up dozens of suspicious papers full of other tortured phrases.

Notably, this phrase is not used by Spinbot. It only uses "insect" or "subterranean insect" as substitutes for "ant".

"language input implement" = keyboard

"muscle directed navigational utensil" = mouse

Keyboard = piano Mouse = rat

"fog computing" is an actual thing though, was discussed in one of my undergrad courses.

Oh god, it's already spreading! /s

+1 for haze figuring (cloud computing)

Now I want to read some papers about reflected-sound-of-underground-spirits or in-sewer-ants-policy.

It’s almost like reading software patents.

> counterfeit consciousness

That’s the proof right there that the AI field is full of closeted Marxists.

Fog computing is real. It occurs during the period between waking up and the first instance of caffeination.

Motion:commotion = signal:noise? Gullible bayes? hahaha

See https://news.ycombinator.com/item?id=28107614 for recent discussion

Until we stop using the number of papers written as a measure of success, we will continue to have things such as this and low quality papers from predatory journals polluting the literature.

I won't be too surprised if authors of all those papers come out to reveal it was their novel research, to test pat.. paper review and acceptance process in academic community. /s

It's like in that episode of Friends where Joey discovered Thesaurus and turned his sentence "They're warm, nice people with big hearts" into "They're humid, prepossessing Homo sapiens with full-sized aortic pumps."

These content spinning techniques were once widely used in SEO search engine spam. Back then, it worked because Google couldn't spot that they were the same text with synonyms. Presumably whatever anti-plagiarism software journals are using today is similarly limited.

The motivations appear similar too. If you make academic success dependent on an artificial measure like publishing volume (or search ranking for SEO), people are going to try to game the system.

> Presumably whatever anti-plagiarism software journals are using today is similarly limited.

Sadly, I am guessing the problem here is the anti-plagiarism software was never run, nor was the article ever even read. Especially given that the issues were found disproportionately in one specific journal, this seems to be outright corruption on both sides of the equation.

I just checked on one of the common journals in my field, which is reasonable impact, but generally has a reasonable reputation. I could not find any matches for the phrases that makes sense (SNR, MSE).

I think this is really because no review happened. The problem is really that Elsevier forces libraries to also subscribe to all this crap they generate when you want to subscribe to their high-impact journals. They seem to be a highly unethical organisation.

>“ant colony” -> ”underground creepy crawly settlement”

Wow, and it passed peer review.

It passed squint survey.

IMO, 'A dubious writing style emerging in science' is a misleading title. It's not a style emerging in science, it's a discovery of a few more dumpster-journals, where they accept anything submitted including GPT generated text.

Let's be real, I have trouble trusting people who use the words ‘tortured phrases’ instead of ‘bogus sentences’.

Initially I thought this article was going to be about the needless use of esoteric jargon when sufficiently specific language exists. Or badly named jargon.

Naming in programming is hard but if it was done like science journal writing, everything would be someone’s last name or a single invented word that sort of makes sense if one is good at Greek and Latin roots.

The wording of the abstract indicates that this article itself has been auto-generated, right? :)

[dupe]

A few days ago, lots of discussion:

https://news.ycombinator.com/item?id=28107614

In this real world GAN, the next fake paper can just run replacements on their bogus terms from the table and keep spewing bogus.

Turns out they don't even have to do that as demonstrated by the paper.

The review process itself is broken and the world of academia is drowning in bogus publications and journals taking advantage of the "publish or perish"-mentality of contemporary academic activities.

Stats have become meaningless - whenever publications try to "rate" countries by their output (be that scientific publications or patent filings), I die a little inside, knowing that most patents are trivial BS and most papers are plagiarism or generated word salat published in pay-per-submission online "journals".

I wouldn't be surprised if some of this came from translation tools too. Those use very similar systems.

The researchers did the comparison in their control datasets.

There is a significant and detectable difference between generated garbage and failed translation.

Speaking of which, just out of curiosity I tried and translated one of their examples to simplified Chinese and back into English using Google translate. The results were shockingly bad, but until I discovered that it was the virtual line breaks introduced by the UI (i.e. the textbox) that caused it.

Turns out it's a horrible idea to introduce line breaks in the middle of Chinese character sequences as the meaning of individual symbols varies depending on adjacent symbols. I'd be willing to bet that a lot of translation mistakes stem from that horrible design decision.

If we can get some AI bots to review these papers, we've come full circle.

This is already standard practice to a point: many universities use AI to detect plagiarism for example.

Reply on news.ycombinator.com