New preprint: NPI and depth charge illusions in Pythia LLMs of different sizes and at different training checkpoints. NPI illusion goes away but depth charge illusion stays or even gets stronger. Amazingly for me, I come out in favor of... construction grammar? arxiv.org/abs/2603.27855
Deadline 15 April: Professor (W3), Phonetics and Speech Processing, LMU Munich job-portal.lmu.de/jobposting/2...
Is there some open source spreadsheet that enforces and supports all these recommendations? I'd love to read in a methods section: "We entered and managed all data using OpenSheet with Broman-Woo guardrails."
New preprint with Sebastian Padó: Transformers do not reliably explain human morphosyntactic processing. Testing 11 models on agreement attraction shows they match human data for prepositional modifiers but fail on object-extracted relative clauses.
arxiv.org/abs/2603.16574
Sie war zuletzt eine der lautesten Kritikerinnen des deutschen Wissenschaftssystems. Jetzt gab Amrei Bahr, seit 2021 als Mitinitiatorin der Kampagne #IchbinHanna bekannt, die wissenschaftliche Forschung selbst aufzugeben.
Via @amreibahr.bsky.social
📢 CFA: 4th Workshop on Eye Movements and the Assessment of Reading Comprehension
Held in Koblenz on June 18/19/20.
Abstracts of up to 350 words can be submitted until April 1st.
Please share.
uni-ko.de/UTq3s
Think about where we could be if all that money was used to make open source / libre software better.
This study required a great deal of effort, but we didn't give up. Very grateful to my co-authors @benediktehinger.bsky.social, Anna Prysłopska, and Candy Adusei 🙏
Schematic representation of the sequence of tasks in the benchmark.
We compared the TRACKPixx3 eye-tracker against the EyeLink 1000 Plus. Across 8 tasks, accuracy was similar. One key difference: TPx3 aggressively smooths sample data, which may affect sub-fixation phenomena and fixation onsets. Check the paper for the full results.
osf.io/preprints/ps...
Exactly, for some reason we collectively decided that this is simply a given, but the fix is trivially easy.
This paper started as a student project in our eye-tracking course. It's amazing to see what cool projects students will put together if you let them.
New preprint on spatial bias in visual attention and how to counter it. In a two-alternative forced choice task, we show a 300% increase in statistical power when left-bias is neutralized. More is possible. Implications for visual world paradigm and perhaps the maze task.
osf.io/preprints/ps...
Followed. Thanks for the pointer! I ended up creating my own adapter PCB. Great opportunity to learn Kicad.
How do you know I'm not doing this already?
Obviously I'm aware of this. Not sure what you're trying to say.
HSP leadership, please move HSP to Canada so that we (= rest of the world) can attend again / can attend safely in the future.
Can we please stop holding important conference in the US?One of my students is affected by the full travel ban and for others it just doesn't seem sufficiently safe. I love my US colleagues and would have enjoyed attending HSP and ACL/CoNNL but it's simply not an option at this point.
Update: We’re working on a solution and have some initial results. Happy to share once we‘ve verified that it’s working correctly.
Does anyone have or know of a version of Veronica Boyce's software for A-Maze stimuli that works with German?
Among them is Gary Marcus. Too busy being a star on social media to check his own paper.
3+1-year fully funded PhD position in Experimental and/or Computational Psycholinguistics. Official application deadline today, but applications submitted by August 25 will receive full consideration.
tmalsburg.github.io/job_ad_2025_...
Wow, you're so virtuous!
Fully-funded PhD position in experimental and/or computational psycholinguistics:
tmalsburg.github.io/job_ad_2025_...
Application deadline is August 15.
Less obvious uses of LLMs that actually work.
That's the type of mundane stuff that actually works. Getting these things to write non-trivial code usually gives me results that are almost comically bad.
I'm offering a 3-year PhD position with benefits (1-year extension possible). Research topic open but broadly in incremental sentence comprehension. If you're into eye-tracking, even better! No teaching until Summer 2027, light teaching after that (English). Official ad soon. Please share 🙏
Nice, looking forward to seeing it!
Poster at #AMLaP2025 in Prag. Sebastian Padó and I report a larger-scale evaluation of transformers as cognitive models of morphosyntax processing.
Who else is going?
Weiterhin beschreibt Dieter #Lenzen in einem Gastbeitrag für die FAZ heute, was man als Hochschulleitung tun muss, um in der #ExStra Erfolg zu haben - nämlich: "negative Einflüsse von Gremien zu neutralisieren".
Ich denke mir das nicht aus. #ExitExzellenz
@amreibahr.bsky.social @kubon.bsky.social
