Lea Hirlimann

Ahmad Dawar Hakimi, Lea Hirlimann, Isabelle Augenstein et al.

Instruction-tuned LLMs can annotate thousands of instances from a short prompt at negligible cost. This raises two questions for active learning (AL): can LLM labels replace human labels within the AL loop, and does AL remain necessary when entire corpora can be labelled at once? We investigate both questions on a new dataset of 277,902 German political TikTok comments (25,974 LLM-labelled, 5,000 human-annotated), comparing seven annotation strategies across four encoders to detect anti-immigrant hostility. A classifier trained on 25,974 GPT-5.2 labels (\$43) achieves comparable F1-Macro to one trained on 3,800 human annotations (\$316). Active learning offers little advantage over random sampling in our pre-enriched pool and delivers lower F1 than full LLM annotation at the same cost. However, comparable aggregate F1 masks a systematic difference in error structure: LLM-trained classifiers over-predict the positive class relative to the human gold standard. This divergence concentrates in topically ambiguous discussions where the distinction between anti-immigrant hostility and policy critique is most subtle, suggesting that annotation strategy should be guided not by aggregate F1 alone but by the error profile acceptable for the target application.

Yihong Liu, Runsheng Chen, Lea Hirlimann et al.

In large language models (LLMs), certain \emph{neurons} can store distinct pieces of knowledge learned during pretraining. While factual knowledge typically appears as a combination of \emph{relations} and \emph{entities}, it remains unclear whether some neurons focus on a relation itself -- independent of any entity. We hypothesize such neurons \emph{detect} a relation in the input text and \emph{guide} generation involving such a relation. To investigate this, we study the LLama-2 family on a chosen set of relations, with a \textit{statistics}-based method. Our experiments demonstrate the existence of relation-specific neurons. We measure the effect of selectively deactivating candidate neurons specific to relation $r$ on the LLM's ability to handle (1) facts involving relation $r$ and (2) facts involving a different relation $r' \neq r$. With respect to their capacity for encoding relation information, we give evidence for the following three properties of relation-specific neurons. \textbf{(i) Neuron cumulativity.} Multiple neurons jointly contribute to processing facts involving relation $r$, with no single neuron fully encoding a fact in $r$ on its own. \textbf{(ii) Neuron versatility.} Neurons can be shared across multiple closely related as well as less related relations. In addition, some relation neurons transfer across languages. \textbf{(iii) Neuron interference.} Deactivating neurons specific to one relation can improve LLMs' factual recall performance for facts of other relations. We make our code and data publicly available at https://github.com/cisnlp/relation-specific-neurons.

Ivo Bueno, Lea Hirlimann, Enkelejda Kasneci

Detecting hate speech in memes is challenging due to their multimodal nature and subtle, culturally grounded cues such as sarcasm and context. While recent vision-language models (VLMs) enable joint reasoning over text and images, end-to-end prompting can be brittle, as a single prediction must resolve target, stance, implicitness, and irony. These challenges are amplified in multilingual settings. We propose a prompted weak supervision (PWS) approach that decomposes meme understanding into targeted, question-based labeling functions with constrained answer options for homophobia and transphobia detection in the LT-EDI 2026 shared task. Using a quantized Qwen3-VLM to extract features by answering targeted questions, our method outperforms direct VLM classification, with substantial gains for Chinese and Hindi, ranking 1st in English, 2nd in Chinese, and 3rd in Hindi. Iterative refinement via error-driven LF expansion and feature pruning reduces redundancy and improves generalization. Our results highlight the effectiveness of prompted weak supervision for multilingual multimodal hate speech detection.

Leonor Veloso, Lea Hirlimann, Philipp Wicke et al.

Knowledge of slang is a desirable feature of LLMs in the context of user interaction, as slang often reflects an individual's social identity. Several works on informal language processing have defined and curated benchmarks for tasks such as detection and identification of slang. In this paper, we focus on queer slang. Queer slang can be mistakenly flagged as hate speech or can evoke negative responses from LLMs during user interaction. Research efforts so far have not focused explicitly on queer slang. In particular, detection and processing of queer slang have not been thoroughly evaluated due to the lack of a high-quality annotated benchmark. To address this gap, we curate SLAyiNG, the first dataset containing annotated queer slang derived from subtitles, social media posts, and podcasts, reflecting real-world usage. We describe our data curation process, including the collection of slang terms and definitions, scraping sources for examples that reflect usage of these terms, and our ongoing annotation process. As preliminary results, we calculate inter-annotator agreement for human annotators and OpenAI's model o3-mini, evaluating performance on the task of sense disambiguation. Reaching an average Krippendorff's alpha of 0.746, we argue that state-of-the-art reasoning models can serve as tools for pre-filtering, but the complex and often sensitive nature of queer language data requires expert and community-driven annotation efforts.