Robert Hawkins
@rdhawkins
asst prof @Stanford linguistics | director of social interaction lab 🌱 | bluskies about computational cognitive science & language
congratulations to @judithfan.bsky.social on a tour de force talk for the 2026 lila gleitman prize!
good morning!! it's day 2 of #cogsci2026, and we're so excited to hear about new work from Seo-young Lee, Jess Mankewitz, Polina Tvilodub, Jinyi Kuang, and Karla Perez!
excited for the first #cogsci2026 poster session this afternoon! come by and chat with @ymacci.bsky.social @amirzur.bsky.social @clairebergey.bsky.social
got the bummer news this morning that my NSF proposal is being returned without review bc the program has been “paused”… 🥺
big nerd post: finally found hard bound copies of these absolute classics! (still shocking to me that they’re out of print!!)
huge thanks to napoleon katsos and the incredible team of organizers for making this year’s xprag such a wonderful and winning conference. long live xprag!
Flying over the Bay Area with @glupyan.bsky.social at the controls! Thanks to Jan Engelmann for capturing the moment and to Gary for an unforgettable flight.
Just got home from the loveliest visit to IU CogSci, my alma mater, where I got my start. Thanks to Rob Goldstone and Katy Börner for hosting me, forever grateful for your mentorship!
📄👀: Consonant lengthening marks the beginning of words across a diverse sample of languages www.nature.com/articles/s41...
just finished this exceptional book by @bnerlich.bsky.social, and it was a revelation to learn about these characters and the origins of the ideas we're now trying to formalize in computational models. Like how great is this from William Dwight Whitney in 1880: "men are always making language".
5/7: but aren't really sensitive to context when it comes to production (Exp 3). when describing familiar objects like dogs or tables, kids often use the same label regardless of context - even when they need to be more specific to avoid confusion.
2/7: ashley paired up 4-8yo children + parents to play a reference game with tangram shapes and found that parents naturally scaffold their kids' communication in clever ways - asking clarifying questions, proposing helpful labels, and prompting kids to expand when their descriptions are unclear.
new paper out with @ashleycleung.bsky.social and dan yurovsky in child development! 📄🔓: doi.org/10.1111/cdev... brief 🧵 below on our findings!
hyped to see @kushinm.bsky.social 's magnum opus out with @judithfan.bsky.social et al in NeurIPS (he's probably going to do a big thread on it at some point, so i'll save the punchline, but it's a huge new sketching benchmark spanning different levels of abstraction)
🧵4: Jang et al., (2023). Personalized Soups: Personalized large language model alignment via post-hoc parameter merging. "decomposing preferences into multiple dimensions" (there's an interesting connection to our mixed-effects transformer, arxiv.org/abs/2205.01749 )
🧵3: Liu et al. (2023). CRYSTAL: Introspective Reasoners Reinforced with Self-Feedback. [thinking about how to elicit structured situation models from LLMs... this approach is a relatively unstructured baseline, "using introspection to generate relevant knowledge statements."]
🧵1: Dingemanse & Enfield (2023). Interactive repair and the foundations of language. "As an always-available fallback option and an infrastructure for negotiating social commitments, interactive repair is foundational to the resilience, complexity, and flexibility of human language."
coolest paper i’ve seen today “Tits (Paridae sp.) use social information when locating and choosing nest lining material” link.springer.com/article/10.1...
just found this gem from a previous resident of my abode. it must have fallen behind the drawer.
CMU Philosophy and the Institute for Complex Social Dynamics is hiring -- due date Nov 1! apply.interfolio.com/131863
We proved a little theorem that real-valued decision problems are strictly more expressive than partition-based QUDs. So (1) if you like working with QUDs, any such model can be formally embedded into our decision-theoretic framework as-is, and ...
🧵 9/10 We find that *neither* maxim dominates: instead, participants make a graded tradeoff between the two. In other words, they’re willing to say things that violate epistemic norms of truthfulness *if and only if* the resulting belief helps the listener choose a good action!