https://serhii.net/tags/conference/ Recent content in Conference on serhii.net Hugo en-us Serhii Hamotskyi / serhii.net Wed, 13 May 2026 00:00:00 +0200 https://serhii.net/dtb/260513-1145-lrec-2026-notes/ Wed, 13 May 2026 00:00:00 +0200 https://serhii.net/dtb/260513-1145-lrec-2026-notes/ <h2 id="day-1">Day 1</h2> <h3 id="o4">O4</h3> <h4 id="assessing-the-political-fairness-of-multilingual-llms-a-case-study-based-on-a-21-way-multiparallel-europarl-dataset">Assessing the Political Fairness of Multilingual LLMs: A Case Study Based on a 21-Way Multiparallel EuroParl Dataset</h4> <ul> <li>Links: * <a href="https://arxiv.org/abs/2510.20508">Arxiv</a> - <a href="https://client.conflux.events/app/events/77/schedule/13841/presentations/95020?tab=about">LREC'26 | Schedule</a> - <a href="https://github.com/PaulLerner/21-EuroParl">Github</a></li> <li>Takeaways: <ul> <li><strong>BLEU</strong> variants not comparable across language pairs</li> <li><a href="https://en.wikipedia.org/wiki/Borda_count">Borda count - Wikipedia</a> is a way to fairly rank, rating = number of candidates below them in ranking <ul> <li>Originally for voting</li> <li>Used in IR to aggregate ratings and benchmarking models on tasks w/ different metrics</li> </ul> </li> </ul> </li> </ul> <h3 id="appeal-align-divide-stance-detection-for-group-directed-messages-in-german-parliamentary-debates">Appeal, Align, Divide? Stance Detection for Group-Directed Messages in German Parliamentary Debates</h3> <ul> <li>Links: <a href="https://lrec.elra.info/lrec2026-main-020">LREC</a></li> <li>Group appeals: <ul> <li>Negative: liberals want to X -&gt; bad</li> <li>+: we stand w/ millions of commuters</li> </ul> </li> <li>Benchmark + LLMs evaluation</li> <li>LLL &ldquo;our work is similar in spirit to&rdquo;</li> <li>QQQ/PPP &ldquo;Our work comes with limitations, the most obvious one being the high computational and, for closed- source models, financial cost of LLMs,&rdquo;</li> </ul> <h2 id="day-2">Day 2</h2> <h3 id="once-upon-a-kernel-extracting-important-events-from-narratives">Once upon a Kernel: Extracting Important Events from Narratives</h3> <ul> <li>Links <ul> <li><a href="https://client.conflux.events/app/events/77/schedule/13887/presentations/95479?tab=about">LREC</a></li> </ul> </li> <li>Kernel events in stories are key ones that survive across retellings</li> <li>Idea history: Narrative functions (1928), cardinal functions, kernel events (chatman 78)</li> <li>Fairy tales b/c simple</li> <li>Operational definition (=operationalization): three criteria</li> <li>Gold standard dataset of 50 fairytales 3.5k-ish kernels</li> </ul> <h2 id="temporal-expression-recognition-in-legal-transcripts">Temporal Expression Recognition in Legal Transcripts</h2> <ul> <li>Taxonomy <ul> <li>calendar date, duration, vague period (&lsquo;sometime in july&rsquo;) etc.</li> </ul> </li> </ul> <h3 id="multilingual-multimodal-pipeline-for-creating-authentic-and-structured-fact-checked-claim-dataset">Multilingual, Multimodal Pipeline for Creating Authentic and Structured Fact-Checked Claim Dataset</h3> <p><strong>TODO really really cool, need to look deeper</strong></p> <ul> <li><a href="https://arxiv.org/abs/2601.07985">[2601.07985] Multilingual, Multimodal Pipeline for Creating Authentic and Structured Fact-Checked Claim Dataset</a></li> </ul>