A AI etymology deconstructor – can guess fake words | Hacker News Reader

A AI etymology deconstructor – can guess fake words

deconstructor.ayush.digital

27 pointshyusap52 comments

I think that by "can guess fake words" in the title OP means that the tool can come up with a plausible guess for the etymology, even for fake words. Unfortunately, the more common reading of that phrase is that it can tell fake words from real ones

Agree. It does a good job at coming up with a plausible meaning of novel words like "multiarborality" → "the state of relating to many trees" but it doesn't indicate that this is a "fake" word even though I just made it up.

>> even though I just made it up

One person inventing a word that they have never heard before doesn't negate the possibility of that word being in common use somewhere.

https://books.google.ca/books/about/Arboreality.html?id=S95Y...

"An expansion of the 2020 Theodore Sturgeon Memorial Award winning story. Arboreality is a finalist for the Philip K. Dick Award and the winner of the 2023 Ursula K. Le Guin Prize for Fiction."

Doesn't look fake to me. English is not a closed-world language, as far as I know.

Between things like "verbing" and "nouning", and the cultural acceptance of doing them in casual speech, I'd say English is a great language because you get to "invent" new words on the fly, and your interlocutors know what you mean.

In this sense, even if no one before ever said or wrote "multiarborality", it's pretty clear what it means (as long as you don't misread it), and IMHO it's perfectly fine to derive its etymology by deconstructing it back to "common" words and pulling etymology on those, recursively.

> I'd say English is a great language because you get to "invent" new words on the fly, and your interlocutors know what you mean.

I bet you could do that with most languages. I don’t see why English would be especially great at it.

I bet you could, but in the other languages I know, I believe it would be frowned upon. The cultural acceptance of this feature is just as important as the feature itself.

Conversely, in the other languages I know I see it happening all the time. I don’t think frowning on it has to do with the language, but with the person. There are sticklers and SNOOTs¹ everywhere, even in English.

¹ http://www.jamesgerity.com/biblio/authority.pdf

I can think of three reasons:

1. English doesn’t really have an official regulation body, like French does.

2. The lack of cases and complex grammar. Any language with a case system is going to have more complexity when it comes to adding new words, because otherwise you end up with awkward looking constructions.

3. English itself is something of a hybrid between Latin and Germanic languages, which to my knowledge gives it a more diverse ancestry than the typical language. Ergo having a new word of dubious origins is more natural.

I won’t comment on the second reason, since that seems like something a linguist should address, but I don’t really buy the first and last reasons:

1. Doesn’t seem relevant, as we’re discussing making up a word in conversation and not putting it into dictionaries.

3. Especially in this globalised world, English loanwords are everywhere. No one bats an eye at it and plenty of languages distort those words to fit their own language. For example: when referring to an internet post you say you’re “posting”; another language would keep the “post” but replace the “ing” with the modifier appropriate for them.

1. Well I think perhaps then we could reverse it and just see the lack of a regulatory body as a symptom of a culture that cares less about following strict linguistic rules. Compared to French, which also has a ton of slang and experimentation, but notably the power structures underlying the language care enough about maintaining a standard.

3. Loanwords are everywhere but I think they are easier to incorporate into everyday speech in English than in some other languages, especially ones with case endings. A word like taco, for example, has become indistinguishable from other “native” English words. Taco in say, Polish, requires more thinking about how it fits into the case system and what endings should be used. It’s a more complicated process than in English.

I still disagree but I think this would be too hard to discuss over asynchronous text. You have some arguments I’d like to explore and spend some back and forth trying to understand your point and explaining mine better, but unfortunately have stuff to still do today.

Still, I want to thank you for the polite and reasoned replies. I wish we were having this conversation in person, I’m confident it would’ve been interesting.

It's not more complicated if you know the language.

Language is a tool. People will use their language for what they need it for. If there isn't a word they'll make one up or steal it. This is absolutely universal. I know you can dig up any number of texts that say English is special, but they're all wrong.

"Languages differ not in what they can express, but in what they must express" as Roman Jakobson phrased it. (The "must express" refers to grammar - in some languages you need a subject, or to know the time something happened, or in which direction[0] it happened. Other languages don't care, but you can add that information if you need it.)

[0] E.g. Mam, spoken in Guatemala, marks all verbs for direction, even if they're abstract, then you add one based on convention or metaphor or maybe taste.

Loanwords are a different thing, IMHO - even those where some language would take the root and localize the verbing part. In loanwords, you're still actually lending a "real" word from English, one you'd find in an English dictionary. That's a distinct thing from ad-hoc inventing new words that are not in the dictionary and not intended to end up in one - words meant to exist for the duration of a conversation or some engagement, or within the scope of some work.

It is my impression that introducing such ad-hoc words in English is something people wouldn't bat an eye on, while in other languages/cultures I'm familiar with, it'd be something Serious that you probably shouldn't do unless absolutely necessary.

I agree with both comments, I just wanted to find a word that an AI wouldn't have heard before, not a word that is "fake".

I don't even know what a fake word is.

https://deconstructor.ayush.digital/w/flonkers

"Flonkers: A made-up word, likely humorous". Aren't all words made up? Edit: This one, unlike my previous example, is actually in use - flonkers: an animal that looks fat but is actually just fluffy.

turboencabulation, hydrocoptic. I bet you could look up a bunch of sci-fi for other examples of completely made up words

If you had used that word in a sentence, eg "what I like about this park is its multiarborality", I would have immediately understood what you meant.

Which it should since it knows all the words.

Yeah I just filled in "inbreathiate", which should be a fake / made up word but this tool generates a meaning anyway... which is also neat, but the "can guess fake words" description isn't accurate.

At least it did say "slartibartfast" was a fictional character.

I think OP means that it can make a guess at what a nonexistent word means, something Wiktionary and urban dic doesn't do as well.

I asked it about "ocelot". It said it was "tlatlauelolotl" which comes from the Nahautl for "tlatl" (field), "lauel" (unknown), and "olotl" (jaguar). Except according to [0], "tlatllauelolotl" doesn't exist (and neither does "lauel" - the search results for "tlatl" and "olotl" aren't in English which means I can't judge their validity.)

There's "tlacoocelotl"[1] which seems to be the actual word for "ocelot" ("semi-jaguar").

0/10 for the AI there.

[0] https://nahuatl.wired-humanities.org/

[1] https://nahuatl.wired-humanities.org/content/tlacoocelotl

Fyllipig is derived from "fyllan" and "pigge" and means "descriptive of something that is both full/abundant and pig-like."

Terryjambled means "mixed up in a confused or disorderly state, and covered with or resembling terry cloth."

Refugglemander means to "to manipulate electoral district boundaries in a way that impacts refugees."

I'm sorry, OP. This just isn't very good.

What etymologies would you have expected instead?

It's funny; I'm looking at your examples and coming to the opposite conclusion. I feel like it is very good because it provides explanations for unusual or novel words that are similar to what a human might conclude.

Etymology is a science. This is random guesswork, and it's not even very precise (deriving refuggle from refugee is definitely objectively wrong).

Maybe I'm coloured by having spent half a decade of my life on a linguistics degree, though.

Would you expect that it would point out that "refuggle" is not a word with any documented usage rather than drawing a strenuous connection to a slightly similar word?

I find the connections it draws amusing. Since I'm not the inventor of the word "refuggle," I can't say that I know its etymology or how it relates to "refugee." But I guess this is one weakness of LLMs: they're bad at admitting they don't have an answer.

We know alot about how English has evolved. Word forms develop along certain paths, sound changes follow laws that linguists have described, and the "fuuuug" in refugee does not become "fugg".

Etymology is probably the subfield of the humanities that provides the closest thing we get to testable hypotheses and laws. This is just mashing shit together.

It also seems to have a tendency to just mash semantics together. As though having something resembling "pig" and something resembling "full" means "full AND piggish". But that's not how plain juxtaposition works in Germanic: it's specification, not conjunction. So "pigful" would mean a type of full, full of pigs. (This is language specific, by the way; as I recall in Vietnamese "mother-father" is the normal word for parents - but in Danish it means grandfather, your mother's father.)

I'm sure I could get WhateverGPT to hallucinate something that looks like those drawings chemists make of molecules to most laypeople. That would be about as interesting as this is.

Cool, on my end it said "refugglemander" means "One who repeatedly flees and wanders" (re- fugere -ler mandros)

When I asked it to deconstruct "Babbage"[1] I got "Derived from Babba's place", Some others:

- phonenose: The ability to detect sounds or voices through the nose

- legpc: Acronym for Laptop Easy Personal Computer

- gitls: A command in Git to list files

- housefreezing: The action of hardening a house with cold

- uncleftish beholding[2]: The act of viewing something that is whole

In any case it's fun to play with and the UI is nice too.

Note, the title looks editorialized, it's currently "A AI etymology deconstructor – can guess fake words", but the website says just "deconstructor.".

[1]: https://en.wikipedia.org/wiki/Charles_Babbage

[2]: https://en.wikipedia.org/wiki/Uncleftish_Beholding

It doesn't really detect stringing together latinate nonsense, so all of my coinages got perfectly coherent definitions. On the flip side, I guess if you're erudite and willing to butcher other languages for sport, nothing has ever stopped you from making up perfectly quasiaryiumaryesque words.

I tried a nonsensical:

https://deconstructor.ayush.digital/w/Deundehydrate

"to reverse the process of reversing the removal of water"

There is clearly no such word, but a human would probably infer such a meaning.

I'm amazed at how unamazed we all seem to have become at such feats. We need more deunnamazing

Something I find interesting on the internet is the number of times an LLM can decode a human’s intent better than other humans. Claude 3.7 Sonnet also makes the common misinterpretation here that “can guess fake words” means “can detect words that don’t have an established meaning” rather than “can guess an etymology for words that don’t have an established meaning” but I gave it the chance to try a word on the tool and interpret the title and it successfully did.

This is interesting because it may eventually become the case that we should interpret other human texts through an LLM because it is better at understanding than many of us are.

My prompt, in case you are curious:

> I came across a HN story titled “AI etymology deconstructor - can guess fake words”. I want you to interpret the title and describe what this does. Then I’ll give you the chance to provide a test word and you will reinterpret the title.

I wish you could pick a language. Bahasa has a ton of dialects + txt, which make it difficult to guess sometimes.

Like "gua" in the dictionary means cave but in some dialects it's an informal "I". Sometimes it gets shortened to the phonetic "gue", sometimes "gu" which is similar to "aku" shortened to "ku", which is another form of "I". Really we have like 7 different ways to say informal "I". I think the tool guesses it as Chinese.

Malaysian tends to remove affixes when shortened, while Indonesian removes vowels but keeps the affixes, so something like "come here quickly" might be shortened to "come quick" or "here2 qkly". Less etymology, more about encoding.

For real words, better check Wikipedia. It's much more informative (compare for "dragon" : https://en.m.wiktionary.org/wiki/dragon )

This is fun, but what I think would be even more interesting and potentially actually useful would be to generate a word from multiple ideas — to help express complex ideas in one word. Like Frühstücksbrotübersättigung

You can just prompt for it:

"KI-Philologenstaunverunsicherung"

Breaking it down:

KI – (Künstliche Intelligenz, Artificial Intelligence) Philologen – (Philologists, scholars of language and linguistics) Staun – (from staunen, to marvel or be astonished) Verunsicherung – (a sense of uncertainty or unease)

https://chatgpt.com/share/67d80661-4bb4-8012-9328-77d56af52b...

The title seems very misleading: It seems to create false and real etymologies for both neologisms and commonly used words.

In other words, a false etymology constructor, not an etymology deconstructor :)

It actually knows about "Lisa the Iconoclast". Cool.

It's a perfectly cromulent tool

I was hoping it wouldn't. It does seem to give a reasonable etymology for "embiggen", though.

I love this kind of thing. It’s weird how many words we use without having any idea what they mean.

Words always end up getting broken with speech in a different way to the meaning.

Eg I put in pandemonium which you say as pande-moniun but I just learned is pan-demon-ium which makes a lot more sense!

Very cool. Well done.

Try “helicopter” for a real mind-blow.

I hate how everyone always breaks it up as "heli" and "copter" when it's "helico" and "pter".

> You've deconstructed 5 words! Enter your email to continue your language journey and get notified about new features.

Yeah, no. I tried the same word five times to test for inconsistencies. I don’t appreciate you weren’t straightforward with the limits and am not going to give you my email to “continue my language journey”. The fact you’re already using euphemisms like that makes me strongly distrust what you’d do with an email address.

I'd often casually wondered where the word scaffolding comes from. Its explanation seems sort of plausible.

"Can guess fake words"? It told me "shitbutt" was a real word.

It really struggled with Antidisestablishmentarianism

It did not like supercalifragilisticexpialidocious

I had the same idea, and it concluded that it meant "fantastic," which seems correct.

what is a "fake word"? New terminology is coined all the time. It's one of the fundamentals of how language works.

I think they mean never before seen words aka made up words like "antecarburetorism"

https://deconstructor.ayush.digital/w/antecarburetorism

I'll no longer misunderestimate AI

Very cool to use initially, but then completely locks you out with a modal in your face demanding an email address. That is despicable and you should be ashamed.

Reply on news.ycombinator.com