Study: Back-to-basics approach can match or outperform AI in language analysis

ggiuliomagnifico 1 day ago 26 commentsRead Article on manchester.ac.uk

DE version is available. Content is displayed in original English for accuracy.

⚡ Community Insights

Discussion Sentiment

60% Positive

Analyzed from 773 words in the discussion.

Discussion (26 Comments)Read Original on HackerNews

spindump8930•about 20 hours ago

The article seems quite editorialized, shifting between describing "large-scale AI models" and "neural network-based approaches".

The underlying paper itself is more precise, comparing against LUAR, a 2021 method based on bert-style embeddings (i.e. a model with 82M parameters, which is 0.2% the size of e.g. the recent OS Gemma models). I don't fault the authors of the paper at all for this, their method is interesting and more interpretable! But you can check the publication history, their paper was uploaded originally in 2024: https://arxiv.org/abs/2403.08462

A good example of why some folks are bearish on journals.

"AI bad" seems to sell in some circles, and while there are many level-headed criticisms to be made of current AI fads, I don't think this qualifies.

adi_kurian•about 13 hours ago

I don't see it. Seems even-keeled for the most part. Not a polemic.

"Researchers found that a relatively simple, linguistically grounded method can perform as well as - and in some cases better than - complex artificial intelligence systems in identifying authorship.

The study suggests that increasingly sophisticated AI is not always necessary for high-performing writing analysis, particularly when methods are designed around established principles of how language works."

throwanem•about 19 hours ago

Are you prepared to demonstrate a superior result with models newer than those available when the research was done? Can you suggest a candidate experiment design to test your hypothesis?

E-Reverance•about 22 hours ago

I might be misinterpreting but the LUAR model (which is a transformer) seems to do decently well

https://www.nature.com/articles/s41599-025-06340-3/figures/2

spindump8930•about 20 hours ago

Yes, the paper itself tells a different story than the bullet points in this article.

glitchc•about 22 hours ago

If there's one problem that LLMs have solved, it's language. While an LLM may hallucinate, it does so in grammatically correct English sentences. Additionally, even the local version of gemma-4-26B can seamlessly switch between languages in the midst of a conversation while maintaining context. That's perhaps the most exciting part for me: We have a bonafide universal translator (that's Star Trek territory) and people seem more focused on its factual accuracy.

isjajciwifiwhdi•about 13 hours ago

Language is not about just grammatically correct sentences, it’s about expression, intent, and communication that goes beyond the spoken, written or even motioned word—not one of these things is in the realm of possibility for current (and dare I say even future) AI.

Your Star Trek comparison is also incorrect. Following your logic, we’ve had a “bonafide universal translator” for a while now with websites like Google Translate (and so on). But none of these websites or AI tools are capable of learning languages on the fly purely from context and with minimal input data (that’s the magic of Trek’s UT, what they call linguacode).

No, AIs have not “solved” language in any way.

tuvix•about 21 hours ago

Kind of a tangent I guess, but the coolest thing about Star Trek’s universal translator to me was that it could translate new languages mid-conversation with an extremely small amount of data. Makes me wonder how close we might be able to get to that eventually

pixl97•about 19 hours ago

https://en.wikipedia.org/wiki/Darmok

TL;DR, probably never.

kreyenborgi•about 10 hours ago

For the most widespread languages. There are thousands it still fails badly on.

porridgeraisin•about 21 hours ago

Tbh. The accuracy of translation is, while much better than prior methods, not that great yet. For tamil atleast.

adi_kurian•about 13 hours ago

I'd be curious to see them try and find Satoshi Yakamoto with back-to-basics and see if they do beter than the guy in the nyt last week https://www.nytimes.com/2026/04/08/business/bitcoin-satoshi-...

fideli0•about 20 hours ago

I wonder if this approach can be used to determine whether a text was generated by a specific LLM.

simianwords•about 22 hours ago

It should be obvious that LLMs would be able to beat this with ease. Not sure why this paper deliberately skipped comparing to current LLMs

Example of LLMs doing well in similar tasks: https://arxiv.org/abs/2602.16800

jeanettesherman•about 23 hours ago

Using LLMs for everything is going to be seen as a big fad in a few years. First we try them for everything, then we find what use cases actually make sense, then we scale back. Woe betide our 401(k)s when it happens, though.

computably•about 21 hours ago

> Woe betide our 401(k)s when it happens, though.

The stock market crashes once in a while. Shit happens. The long-term outlook is unlikely to change nearly as much, unless you think there will be systemic macroeconomic changes.

dnw•about 20 hours ago

Long-term relative to lifespan of the 401K holder. Outcome changes a lot for those who are ready to retire.

drBonkers•about 23 hours ago

This is a concise statement of what I've tried to articulate by analogizing it to railroad infra buildout.

What applications do you think make the most sense so far?

qsera•about 19 hours ago

Search, code review, some form of translation...

simianwords•about 22 hours ago

The paper did not compare against LLMs though.

isjajciwifiwhdi•about 13 hours ago

I just cannot wait for the “bubble bursting” moment, so to speak. It’s tiresome to be force fed this AI bullshit all the damn time, knowing full well it is not going to last.

gnabgib•about 13 hours ago

Agreed .. your two comments cancel each other out? https://news.ycombinator.com/item?id=47787688

z3c0•about 23 hours ago

Ha! To think that we're finally back to asking ourselves why we are using generative models for categorization and extraction. I wonder how much money has collectively been wasted by companies wittling away at square pegs.

otabdeveloper4•about 21 hours ago

> why we are using generative models for categorization and extraction

Because LLM models have already amortized the man-years cost of collecting, curating and training on text corpuses?

danielbln•about 19 hours ago

Yeah, LLMs are a solution to the cold start problem plus they are easy to integrate and if you know what you're doing in terms of evals, post processing and so on you can get excellent performance out of them, plus they can do semantic classification and reasoning that you won't get out of some bespoke traditional DS/ML model.

z3c0•about 16 hours ago

They amortized the creation of corpuses with trainable features, not the myriad of methods that can categorize text with a success rate in the levels required by high-stakes industries.