Do you want to understand how language models work, and how they can change language science? I'm recruiting PhD students at UBC Linguistics! The research will be fun, and Vancouver is lovely. So much cool NLP happening at UBC across both Ling and CS!

RT @jurafsky: Now that school is starting for lots of folks, it's time for a new release of Speech and Language Processing! Jim and I adde….

RT @aclmeeting: 📅 25-Year ToT Award (2000). Daniel Gildea & Daniel Jurafsky:. “Automatic Labeling of Semantic Roles”. ACL 2000. 🔗 https://t….

@antara_raaghavi is applying for PhDs this fall! She’s super impressive and awesome to work with, and conceived of this project independently and carried it out very successfully! Keep an eye out 🙂.

More in the preprint! This project was led by Antara, with @elmelis and Kate Davidson.

So is it really this implicit operators thing that’s tripping them up? We try many other ablations, looking at the effect of giving extra context in the prompt, using numbers vs words, left-to-right ordering, and subtractive systems, and none of them seem to have that much effect.

Our experiments are based on Linguistics Olympiad problems that deal with number systems, like the one here. We created additional hand-standardized versions of each puzzle in order to be able to do all of the operator ablations.

This shows the types of reasoning and variable binding jumps that are hard for LMs. It’s hard to go one level up, and bind a variable to have the meaning of an operator, or to understand that an operator is implicit.

If we alter the problems to make the operators explicit, the models can solve these problems pretty easily. But it’s still harder to bind a random symbol or word to mean an operator like +. It’s much easier when we use the familiar symbols for the operators, like + and x.

Our main finding: LMs find it hard when *operators* are implicit. We don’t say “5 times 100 plus 20 plus 3”, we say “five hundred and twenty-three”. The Linguistics Olympiad puzzles are pretty simple systems of equations that an LM should solve – but the operators aren’t explicit

Why can’t LMs solve puzzles about the number systems of languages, when they can solve really complex math problems? Our paper led by @antara_raaghavi looks at why this intersection of language and math is difficult, and what this means for LM reasoning!

Check out our ACL paper! We use shapley interactions to see which words (and phones) interact non-linearly -- what we lose when we assume linear relationships between features. Chat to Diganta in Vienna!.

RT @SashaBoguraev: A key hypothesis in the history of linguistics is that different constructions share underlying structure. We take advan….

Man, those sparse autoencoder concepts sure can be unstable: you see one thing if you train it once and something totally different if you train it again. unless you use Archetypal SAEs!.

RT @iseeaswell: 😼SMOL DATA ALERT! 😼Anouncing SMOL, a professionally-translated dataset for 115 very low-resource languages! Paper: https://….

RT @QuantaMagazine: Computational linguists Julie Kallini (left) and Christopher Potts think that large language models might have somethin….

I will be at NeurIPS starting tomorrow! Would love to chat about interpretability, linguistics, language structure, meaning in LLMs. Reach out!. Aaaand if you love Vancouver, apply to do a PhD at UBC and work with me!

RT @jcrwhittington: 🔊Panel Announcement! Very excited to ask “Is this the end of Scale?” at the Global Summit in Boston on Dec 4: https://t….

I'm fascinated by the lack of determiner on (the?) Labour Conference, and why this sounds so strange to me! We do name-ize things like "at ICLR" with no determiner, but all the "at Labor Conference" kind of breaks my brain. Would someone who has nothing to do with Labour use it?.

Isaac does some of the most impactful NLP work that I know! This is not just 'Google LM magic', it's the result of extremely hard-nosed and outside-the-box data work and linguistic work, as well as working with speakers. And the whole combo that makes Isaac Isaac!.