Hans Ole Hatzel

We are announcing our shared task on narrative similarity and narrative representations. SemEval 2026 Task 4: narrative-similarity-task.github.io. We invite you to benchmark LLMs, embedding models, or even test your favorite narrative formalism. Development data is now available!

Search/acc⏩Probably the hottest BoF at #EMNLP2024! Nearly 100 researchers packed the room for 12 back-to-back talks on search foundation models - covering everything from code embeddings, distilled rerankers, ColPali, ColBERT, late chunking & smaller LMs. Killer lineup featuring @HansHatzel @gangi_official @Robro612 @ManuelFaysse @zhichaoxu_ir @memray0 @SFResearch and many others! Big thanks to everyone who came out and showed such enthusiasm for search foundation models! 🙌 We are looking forward to seeing you next @emnlpmeeting!

Let’s discuss: - How could story embeddings be useful in your work? - Why don’t our embeddings work well for retellings? - How can story embeddings be improved upon? More details in the paper! Paper: inf.uni-hamburg.de/en/inst/ab/lt/… Github: github.com/uhh-lt/story-e… 🧵 (6/6)

Our paper has plenty of experiments on measuring embedding capabilities for retrieval and exploring which aspects of story summaries our embeddings focus on. Big surprise: the embeddings perform well on ROCStories, a commonsense reasoning task. 🧵 (5/6)

Excited to tell you all about our EMNLP paper on story embeddings! We present a novel approach for representing story summaries as embeddings. This is my first proper paper thread, so please be kind 😅 See you in Miami! 😎 🧵 (1/6)

What should the ACL peer review process be like in the future? Please cast your views in this survey: aclweb.org/portal/content… by 4th Nov 2024 #NLProc @ReviewAcl

@baohao_liao Does this refer to the "EMNLP 2024 Main" in the headline between visibility and publication date?

Not sure whether it is a bug, the #EMNLP result is out. You can check it in your submissions

"Tell me again! A Large-Scale Dataset of Multiple Summaries for the Same Story" a dataset of 96k summaries across 29l stories, harvested from five language versions of Wikipedia, and annotated with metadata from Wikidata. (Hatzel and Blemann, 2024) inf.uni-hamburg.de/en/inst/ab/lt/…

It’s cool how all of YouTube is funded by lying about the security of public WiFi

Wie wurde der vierte Tag der @DHdKonferenz grad so schön eingeläutet? CLS I und II = „die wichtigsten Sessions der Konferenz“ 🙌🏻 ich kann nicht widersprechen! Lauter spannende Vorträge von @HansHatzel @EvelynGius @BenKrautter @fotisja @janinajacke und vielen mehr #CLS #DHd

As Gregor Samsa awoke one morning from uneasy dreams, he found himself rewritten in Rust for performance reasons

@PyTorch Some native dependencies didn't compile on the target system (either due to compiler versions or native libraries). I ended up building a docker image which had everything pre-installed (based on the torchserve one); worked well but it's a bit of work. Conda is typically easier.

@PyTorch When deploying models using torchserve you can (iirc) specify a requirements.txt but this is not feasible for large libraries with native dependencies. Are there easy ways to specify a complete environment (e.g. via conda)? I found this to be the main pain point in demo hosting.

@debayan Yes! Wordle 220 4/6 ⬛🟨⬛🟩⬛ ⬛⬛⬛🟩🟩 🟨⬛⬛⬛🟨 🟩🟩🟩🟩🟩

@RealAAAI Not sure if this is a good way to reach out but this link and any paper from the conference seem to be unavailable (HTTP 500): aaai.org/ocs/index.php/…

posting it here to curse your timeline