Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

thedial.world

Post reply on HN
Print view View original

21–30 of 38 posts

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#21

Post by crote »

tgv wrote: post #16

> According to forensic linguists, we all use language in a uniquely identifiable way that can be as incriminating as a fingerprint. That's a bold and unproven statement, made worse because we can't really see that fingerprint.

It sounds like a fairly accurate statement to me, considering that there isn't a solid scientifically-based foundation behind fingerprint matching either. They aren't quite as unique as we've often been led to believe, and matching them is highly subjective with the same expert often interpreting the same comparison differently when provided with a different story for context.

Fingerprint matching of course isn't completely useless, but it's not as solid as you'd hope either.

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#23

Post by Suppafly »

TruffleLabs wrote: post #18

Reminds me of "Eats, shoots, and leaves" Or is it "Eats shoots and leaves"? ;) Book "Eats, Shoots & Leaves: The Zero Tolerance Approach to Punctuation" https://en.wikipedia.org/wiki/Eats,_Shoots_&_Leaves

or the hilarious "help your uncle jack off a horse" vs "help your uncle, Jack, off a horse" example. although it does also involve caps.

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#24

Post by seanhunter »

xandrius wrote: post #20
lqet wrote: post #2

> Spellings such as “wilfully” for “willfully” and “clew” for “clue” pointed to someone from the Chicago area, for example. Eventually, the linguistic evidence was strong enough to issue a search warrant for the home of a reclusive mathematician named Theodore Kaczynski, raised in Chicago but living in rural Montana One thing Kaczynski's brother noticed as particularly idiosyncratic was the consistent use of the phra…

It does make more sense to me though: of course I can have my cake and then eat it. But I can't eat it and still have it afterwards. So the order implies the temporality of the actions.

^ unabomber posts on hackernews.

Joking aside, the key word which is sometimes implied rather than included is “too”. The order isn’t important. The saying is both things can’t simultaneously be true.

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#26

Post by tgv »

crote wrote: post #21
tgv wrote: post #16

> According to forensic linguists, we all use language in a uniquely identifiable way that can be as incriminating as a fingerprint. That's a bold and unproven statement, made worse because we can't really see that fingerprint.

It sounds like a fairly accurate statement to me, considering that there isn't a solid scientifically-based foundation behind fingerprint matching either. They aren't quite as unique as we've often been led to believe, and matching them is highly subjective with the same expert often interpreting the same comparison differently when provided with a different story for context. Fingerprint matching of course isn't com…

But when two sets of fingerprints, are different, you can be fairly sure they're from different people. But when the percentage of some features is 20% in one text, and 30% in another, you still can't conclude anything. I write in different registers in contexts such as personal emails, professional emails to a large group, professional emails to a direct colleague, a quick post on the internet, an 'app' to a friend in another country, a text message on a phone, etc. I even write them in different languages. It's hard to imagine there's a well-defined, properly grounded model that can unite those yet distinguish them from written output by other people.

And now LLMs are going to add more noise to these features...

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#27

Post by lcnPylGDnU4H9OF »

yunruse wrote: post #11
wizzwizz4 wrote: post #7

Earlier quoted context omitted.

> Indeed, this used to be the most common form of the expression until the 1930s–1940s, when it was overtaken by the have-eat variant. — https://en.wikipedia.org/w/index.php?title=You_can%27t_have_...

I wonder if this has some sort of preference for ablaut reduplication [0]? I don't have the vowel phonics off by hand, but "have your cake and eat it" seems to flow a little more smoothly than "eat your cake and have it". [0] https://en.wikipedia.org/wiki/Reduplication#English

It sounds like reduplication is about individual words being repeated rather than a phrase.

> In linguistics, reduplication is a morphological process in which the root or stem of a word, part of that or the whole word is repeated exactly or with a slight change.

The ablaut section also seems to suggest that "eat" should come before "have" anyway.

> In ablaut reduplications, the first vowel is almost always a high vowel or front vowel (typically ɪ as in hit) and the reduplicated vowel is a low vowel or back vowel (typically æ as in cat or ɒ as in top).

I suspect you feel that it flows more smoothly because it's more familiar. You have to stop a brain process that's become a bit automatic to say that phrase and instead say it slightly differently.

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#29

Post by alganet »

ChiMan wrote: post #9

I think we can expect sophisticated criminals to start using AI to rewrite their correspondence.

That is by itself a linguistic choice that can be analyzed.

"Hey, that guy only communicates using cutouts from magazines, what a strange choice"

LLMs introduce all kinds of linguistic choices, and you can focus on those choices.

Re: Forensic linguists use grammar, syntax and vocabulary to help crack cold cases

#30

Post by seanhunter »

sgarland wrote: post #3

I wonder if distros like Tails [0] are going to start shipping lightweight LLMs specifically to reword messages. Though I’m also not sure how low of a spec you can go and still run an LLM decent enough to not be excruciatingly slow. [0]: https://tails.net/

You don’t need a special distro to run a local llm and something like ollama running llama3 7b would be just fine to reword a message on a normal laptop. Inference (ie actually using a model) is much much less compute intensive than training or finetuning.

I strongly suggest people try ollama - it takes a few minutes to set up, download a local model and you’re up and running. https://ollama.com/

Post reply on HN