Andreas Rücklé

Three papers accepted to ACL main with my current/former students! A proud supervisor moment to see all my current PhD students getting accepted to ACL main at the same time. 🎉🎉🎉 1️⃣ Map of Encoders: There are way too many sentence encoders (over 17K sentence-transformer backends alone!) out there and their landscape is not clear. Which encoders are similar to what others and in what ways? We show that Quantum Relative Entropy is a great metric to compare encoders and create a map of encoders. (arxiv.org/abs/2602.08740) w/ @GaifanZhang 2️⃣ Synthetic Data for Diversified Commonsense Generation: 8B people on Earth are talking to a handful of LLMs these days. LLMs should not be returning templated, fixed, boring responses. How can we make them return more diversified responses? We create the first-ever synthetic dataset for post-training LLMs for diversified response generation. Tianhui is in job market and wrapping up his thesis now with many xACL papers! Grab him while you can. (arxiv.org/abs/2603.18361) w/ Bei Peng and @ThuiZhanglsy 3️⃣ Multilingual Social Bias Benchmark: Thinking process, not just the final answer matters when evaluating social biases in LLMs. This benchmark reveals some of the hidden social biases in LLMs that were not easily surfaced by prior evaluation frameworks. If you care about social biases in LLMs please take a look. w/ @MasahiroKaneko_ and @eltimster #NLProc #ACL2026

In our new paper “What to Pre-Train on? Efficient Intermediate Task Selection” we investigate efficiently selecting the best intermediate training tasks for a target task in an adapter-based transfer setup. w. @PfeiffJo @arueckle tinyurl.com/WhatToPreTrain…

🚨New paper alert 🚨 🍻 BEIR: a heterogeneous benchmark for IR. 17 datasets, 9 tasks with diverse domains. 9 SOTA retrieval models evaluated in a zero-shot setup. w/ @Nils_Reimers @arueckle @abhesrivas, IG at @UKPLab pdf: arxiv.org/abs/2104.08663 More details, code 👇 #NLProc

In our new paper “What to Pre-Train on? Efficient Intermediate Task Selection” we investigate efficiently selecting the best intermediate training tasks for a target task in an adapter-based transfer setup. w. @PfeiffJo @arueckle tinyurl.com/WhatToPreTrain…

Check out our new publication: “TWEAC: Transformer with Extendable QA Agent Classifiers” We propose a meta-QA to find the most fitting QA systems within a pool for a given question. w/ @Nils_Reimers @arueckle IGurevych arxiv.org/abs/2104.07081 Code & data released as well

We are happy to introduce our new Python toolkit - Trankit, a light-weight transformer-based toolkit for multilingual natural language processing that can process raw text and support fundamental NLP tasks for 56 languages. github.com/nlp-uoregon/tr….