Flor Miriam Plaza-del-Arco

Flor Miriam Plaza-del-Arco, María-Teresa Martín-Valdivia, Roman Klinger

Within textual emotion classification, the set of relevant labels depends on the domain and application scenario and might not be known at the time of model development. This conflicts with the classical paradigm of supervised learning in which the labels need to be predefined. A solution to obtain a model with a flexible set of labels is to use the paradigm of zero-shot learning as a natural language inference task, which in addition adds the advantage of not needing any labeled training data. This raises the question how to prompt a natural language inference model for zero-shot learning emotion classification. Options for prompt formulations include the emotion name anger alone or the statement "This text expresses anger". With this paper, we analyze how sensitive a natural language inference-based zero-shot-learning classifier is to such changes to the prompt under consideration of the corpus: How carefully does the prompt need to be selected? We perform experiments on an established set of emotion datasets presenting different language registers according to different sources (tweets, events, blogs) with three natural language inference models and show that indeed the choice of a particular prompt formulation needs to fit to the corpus. We show that this challenge can be tackled with combinations of multiple prompts. Such ensemble is more robust across corpora than individual prompts and shows nearly the same performance as the individual best prompt for a particular corpus.

Giuseppe Attanasio, Flor Miriam Plaza-del-Arco, Debora Nozza et al.

Recent instruction fine-tuned models can solve multiple NLP tasks when prompted to do so, with machine translation (MT) being a prominent use case. However, current research often focuses on standard performance benchmarks, leaving compelling fairness and ethical considerations behind. In MT, this might lead to misgendered translations, resulting, among other harms, in the perpetuation of stereotypes and prejudices. In this work, we address this gap by investigating whether and to what extent such models exhibit gender bias in machine translation and how we can mitigate it. Concretely, we compute established gender bias metrics on the WinoMT corpus from English to German and Spanish. We discover that IFT models default to male-inflected translations, even disregarding female occupational stereotypes. Next, using interpretability methods, we unveil that models systematically overlook the pronoun indicating the gender of a target occupation in misgendered translations. Finally, based on this finding, we propose an easy-to-implement and effective bias mitigation solution based on few-shot learning that leads to significantly fairer translations.

Liliia Bogdanova, Shiran Sun, Lifeng Han et al.

This system paper describes our participation in the SemEval-2025 Task-7 ``Everyday Knowledge Across Diverse Languages and Cultures''. We attended two subtasks, i.e., Track 1: Short Answer Questions (SAQ), and Track 2: Multiple-Choice Questions (MCQ). The methods we used are retrieval augmented generation (RAGs) with open-sourced smaller LLMs (OS-sLLMs). To better adapt to this shared task, we created our own culturally aware knowledge base (CulKBs) by extracting Wikipedia content using keyword lists we prepared. We extracted both culturally-aware wiki-text and country-specific wiki-summary. In addition to the local CulKBs, we also have one system integrating live online search output via DuckDuckGo. Towards better privacy and sustainability, we aimed to deploy smaller LLMs (sLLMs) that are open-sourced on the Ollama platform. We share the prompts we developed using refinement techniques and report the learning curve of such prompts. The tested languages are English, Spanish, and Chinese for both tracks. Our resources and codes are shared via https://github.com/aaronlifenghan/FLANS-2026

Hiba Arnaout, Anmol Goel, H. Andrew Schwartz et al.

Although artificial intelligence (AI) shows growing promise for mental health care, current approaches to evaluating AI tools in this domain remain fragmented and poorly aligned with clinical practice, social context, and first-hand user experience. This paper argues for a rethinking of responsible evaluation -- what is measured, by whom, and for what purpose -- by introducing an interdisciplinary framework that integrates clinical soundness, social context, and equity, providing a structured basis for evaluation. Through an analysis of 135 recent *CL publications, we identify recurring limitations, including over-reliance on generic metrics that do not capture clinical validity, therapeutic appropriateness, or user experience, limited participation from mental health professionals, and insufficient attention to safety and equity. To address these gaps, we propose a taxonomy of AI mental health support types -- assessment-, intervention-, and information synthesis-oriented -- each with distinct risks and evaluative requirements, and illustrate its use through case studies.

Flor Miriam Plaza-del-Arco, Debora Nozza, Dirk Hovy

Large Language Models (LLMs) exhibit remarkable text classification capabilities, excelling in zero- and few-shot learning (ZSL and FSL) scenarios. However, since they are trained on different datasets, performance varies widely across tasks between those models. Recent studies emphasize the importance of considering human label variation in data annotation. However, how this human label variation also applies to LLMs remains unexplored. Given this likely model specialization, we ask: Do aggregate LLM labels improve over individual models (as for human annotators)? We evaluate four recent instruction-tuned LLMs as annotators on five subjective tasks across four languages. We use ZSL and FSL setups and label aggregation from human annotation. Aggregations are indeed substantially better than any individual model, benefiting from specialization in diverse tasks or languages. Surprisingly, FSL does not surpass ZSL, as it depends on the quality of the selected examples. However, there seems to be no good information-theoretical strategy to select those. We find that no LLM method rivals even simple supervised models. We also discuss the tradeoffs in accuracy, cost, and moral/ethical considerations between LLM and human annotation.

Flor Miriam Plaza-del-Arco, Amanda Cercas Curry, Susanna Paoli et al.

Emotions play important epistemological and cognitive roles in our lives, revealing our values and guiding our actions. Previous work has shown that LLMs display biases in emotion attribution along gender lines. However, unlike gender, which says little about our values, religion, as a socio-cultural system, prescribes a set of beliefs and values for its followers. Religions, therefore, cultivate certain emotions. Moreover, these rules are explicitly laid out and interpreted by religious leaders. Using emotion attribution, we explore how different religions are represented in LLMs. We find that: Major religions in the US and European countries are represented with more nuance, displaying a more shaded model of their beliefs. Eastern religions like Hinduism and Buddhism are strongly stereotyped. Judaism and Islam are stigmatized -- the models' refusal skyrocket. We ascribe these to cultural bias in LLMs and the scarcity of NLP literature on religion. In the rare instances where religion is discussed, it is often in the context of toxic language, perpetuating the perception of these religions as inherently toxic. This finding underscores the urgent need to address and rectify these biases. Our research underscores the crucial role emotions play in our lives and how our values influence them.

María Grandury, Javier Aula-Blasco, Júlia Falcão et al.

Leaderboards showcase the current capabilities and limitations of Large Language Models (LLMs). To motivate the development of LLMs that represent the linguistic and cultural diversity of the Spanish-speaking community, we present La Leaderboard, the first open-source leaderboard to evaluate generative LLMs in languages and language varieties of Spain and Latin America. La Leaderboard is a community-driven project that aims to establish an evaluation standard for everyone interested in developing LLMs for the Spanish-speaking community. This initial version combines 66 datasets in Basque, Catalan, Galician, and different Spanish varieties, showcasing the evaluation results of 50 models. To encourage community-driven development of leaderboards in other languages, we explain our methodology, including guidance on selecting the most suitable evaluation setup for each downstream task. In particular, we provide a rationale for using fewer few-shot examples than typically found in the literature, aiming to reduce environmental impact and facilitate access to reproducible results for a broader research community.

Flor Miriam Plaza-del-Arco, Amanda Cercas Curry, Alba Curry et al.

Large language models (LLMs) reflect societal norms and biases, especially about gender. While societal biases and stereotypes have been extensively researched in various NLP applications, there is a surprising gap for emotion analysis. However, emotion and gender are closely linked in societal discourse. E.g., women are often thought of as more empathetic, while men's anger is more socially accepted. To fill this gap, we present the first comprehensive study of gendered emotion attribution in five state-of-the-art LLMs (open- and closed-source). We investigate whether emotions are gendered, and whether these variations are based on societal stereotypes. We prompt the models to adopt a gendered persona and attribute emotions to an event like 'When I had a serious argument with a dear person'. We then analyze the emotions generated by the models in relation to the gender-event pairs. We find that all models consistently exhibit gendered emotions, influenced by gender stereotypes. These findings are in line with established research in psychology and gender studies. Our study sheds light on the complex societal interplay between language, gender, and emotion. The reproduction of emotion stereotypes in LLMs allows us to use those models to study the topic in detail, but raises questions about the predictive use of those same LLMs for emotion applications.

Flor Miriam Plaza-del-Arco, Alba Curry, Amanda Cercas Curry et al.

Emotions are a central aspect of communication. Consequently, emotion analysis (EA) is a rapidly growing field in natural language processing (NLP). However, there is no consensus on scope, direction, or methods. In this paper, we conduct a thorough review of 154 relevant NLP publications from the last decade. Based on this review, we address four different questions: (1) How are EA tasks defined in NLP? (2) What are the most prominent emotion frameworks and which emotions are modeled? (3) Is the subjectivity of emotions considered in terms of demographics and cultural factors? and (4) What are the primary NLP applications for EA? We take stock of trends in EA and tasks, emotion frameworks used, existing datasets, methods, and applications. We then discuss four lacunae: (1) the absence of demographic and cultural aspects does not account for the variation in how emotions are perceived, but instead assumes they are universally experienced in the same manner; (2) the poor fit of emotion categories from the two main emotion theories to the task; (3) the lack of standardized EA terminology hinders gap identification, comparison, and future goals; and (4) the absence of interdisciplinary research isolates EA from insights in other fields. Our work will enable more focused research into EA and a more holistic approach to modeling emotions in NLP.

Natalia Amat-Lefort, Mert Yazan, Amanda Cercas Curry et al.

Large Language Models (LLMs) are increasingly used not only for instrumental tasks, but as always-available and non-judgmental confidants for emotional support. Yet what drives adoption and how users perceive emotional support interactions across countries remains unknown. To address this gap, we present the first large-scale cross-cultural study of LLM use for emotional support, surveying 4,641 participants across seven countries (USA, UK, Germany, France, Spain, Italy, and The Netherlands). Our results show that adoption rates vary dramatically across countries (from 20% to 59%). Using mixed models that separate cultural effects from demographic composition, we find that: Being aged 25-44, religious, married, and of higher socioeconomic status are predictors of positive perceptions (trust, usage, perceived benefits), with socioeconomic status being the strongest. English-speaking countries consistently show more positive perceptions than Continental European countries. We further collect a corpus of 731 real multilingual prompts from user interactions, showing that users mainly seek help for loneliness, stress, relationship conflicts, and mental health struggles. Our findings reveal that LLM emotional support use is shaped by a complex sociotechnical landscape and call for a broader research agenda examining how these systems can be developed, deployed, and governed to ensure safe and informed access.

Alba Maria Marmol-Romero, Salud Maria Jimenez-Zafra, Flor Miriam Plaza-del-Arco et al.

This paper describes the participation of the SINAI team in the eRisk@CLEF lab. Specifically, two of the proposed tasks have been addressed: i) Task 1 on the early detection of signs of pathological gambling, and ii) Task 3 on measuring the severity of the signs of eating disorders. The approach presented in Task 1 is based on the use of sentence embeddings from Transformers with features related to volumetry, lexical diversity, complexity metrics, and emotion-related scores, while the approach for Task 3 is based on text similarity estimation using contextualized word embeddings from Transformers. In Task 1, our team has been ranked in second position, with an F1 score of 0.808, out of 41 participant submissions. In Task 3, our team also placed second out of a total of 3 participating teams.

Joachim Baumann, Paul Röttger, Aleksandra Urman et al.

Large language models are rapidly transforming social science research by enabling the automation of labor-intensive tasks like data annotation and text analysis. However, LLM outputs vary significantly depending on the implementation choices made by researchers (e.g., model selection or prompting strategy). Such variation can introduce systematic biases and random errors, which propagate to downstream analyses and cause Type I (false positive), Type II (false negative), Type S (wrong sign), or Type M (exaggerated effect) errors. We call this phenomenon where configuration choices lead to incorrect conclusions LLM hacking. We find that intentional LLM hacking is strikingly simple. By replicating 37 data annotation tasks from 21 published social science studies, we show that, with just a handful of prompt paraphrases, virtually anything can be presented as statistically significant. Beyond intentional manipulation, our analysis of 13 million labels from 18 different LLMs across 2361 realistic hypotheses shows that there is also a high risk of accidental LLM hacking, even when following standard research practices. We find incorrect conclusions in approximately 31% of hypotheses for state-of-the-art LLMs, and in half the hypotheses for smaller language models. While higher task performance and stronger general model capabilities reduce LLM hacking risk, even highly accurate models remain susceptible. The risk of LLM hacking decreases as effect sizes increase, indicating the need for more rigorous verification of LLM-based findings near significance thresholds. We analyze 21 mitigation techniques and find that human annotations provide crucial protection against false positives. Common regression estimator correction techniques can restore valid inference but trade off Type I vs. Type II errors. We publish a list of practical recommendations to prevent LLM hacking.

Paul Röttger, Giuseppe Attanasio, Felix Friedrich et al.

Vision-language models (VLMs), which process image and text inputs, are increasingly integrated into chat assistants and other consumer AI applications. Without proper safeguards, however, VLMs may give harmful advice (e.g. how to self-harm) or encourage unsafe behaviours (e.g. to consume drugs). Despite these clear hazards, little work so far has evaluated VLM safety and the novel risks created by multimodal inputs. To address this gap, we introduce MSTS, a Multimodal Safety Test Suite for VLMs. MSTS comprises 400 test prompts across 40 fine-grained hazard categories. Each test prompt consists of a text and an image that only in combination reveal their full unsafe meaning. With MSTS, we find clear safety issues in several open VLMs. We also find some VLMs to be safe by accident, meaning that they are safe because they fail to understand even simple test prompts. We translate MSTS into ten languages, showing non-English prompts to increase the rate of unsafe model responses. We also show models to be safer when tested with text only rather than multimodal prompts. Finally, we explore the automation of VLM safety assessments, finding even the best safety classifiers to be lacking.

Alba Maria Marmol-Romero, Flor Miriam Plaza-del-Arco, Arturo Montejo-Raez

This paper describes the participation of the SINAI team in the eRisk@CLEF lab. Specifically, one of the proposed tasks has been addressed: Task 2 on the early detection of signs of pathological gambling. The approach presented in Task 2 is based on pre-trained models from Transformers architecture with comprehensive preprocessing data and data balancing techniques. Moreover, we integrate Long-short Term Memory (LSTM) architecture with automodels from Transformers. In this Task, our team has been ranked in seventh position, with an F1 score of 0.126, out of 49 participant submissions and achieves the highest values in recall metrics and metrics related to early detection.

Flor Miriam Plaza-del-Arco, Paul Röttger, Nino Scherrer et al.

Large language models (LLMs) are increasingly integrated into our daily lives and personalized. However, LLM personalization might also increase unintended side effects. Recent work suggests that persona prompting can lead models to falsely refuse user requests. However, no work has fully quantified the extent of this issue. To address this gap, we measure the impact of 15 sociodemographic personas (based on gender, race, religion, and disability) on false refusal. To control for other factors, we also test 16 different models, 3 tasks (Natural Language Inference, politeness, and offensiveness classification), and nine prompt paraphrases. We propose a Monte Carlo-based method to quantify this issue in a sample-efficient manner. Our results show that as models become more capable, personas impact the refusal rate less and less. Certain sociodemographic personas increase false refusal in some models, which suggests underlying biases in the alignment strategies or safety mechanisms. However, we find that the model choice and task significantly influence false refusals, especially in sensitive content tasks. Our findings suggest that persona effects have been overestimated, and might be due to other factors.

Donya Rooein, Flor Miriam Plaza-del-Arco, Debora Nozza et al.

Given Farsi's speaker base of over 127 million people and the growing availability of digital text, including more than 1.3 million articles on Wikipedia, it is considered a middle-resource language. However, this label quickly crumbles when the situation is examined more closely. We focus on three subjective tasks (Sentiment Analysis, Emotion Analysis, and Toxicity Detection) and find significant challenges in data availability and quality, despite the overall increase in data availability. We review 110 publications on subjective tasks in Farsi and observe a lack of publicly available datasets. Furthermore, existing datasets often lack essential demographic factors, such as age and gender, that are crucial for accurately modeling subjectivity in language. When evaluating prediction models using the few available datasets, the results are highly unstable across both datasets and models. Our findings indicate that the volume of data is insufficient to significantly improve a language's prospects in NLP.

Jackson Trager, Francielle Vargas, Diego Alves et al.

Ensuring the moral reasoning capabilities of Large Language Models (LLMs) is a growing concern as these systems are used in socially sensitive tasks. Nevertheless, current evaluation benchmarks present two major shortcomings: a lack of annotations that justify moral classifications, which limits transparency and interpretability; and a predominant focus on English, which constrains the assessment of moral reasoning across diverse cultural settings. In this paper, we introduce MFTCXplain, a multilingual benchmark dataset for evaluating the moral reasoning of LLMs via multi-hop hate speech explanation using the Moral Foundations Theory. MFTCXplain comprises 3,000 tweets across Portuguese, Italian, Persian, and English, annotated with binary hate speech labels, moral categories, and text span-level rationales. Our results show a misalignment between LLM outputs and human annotations in moral reasoning tasks. While LLMs perform well in hate speech detection (F1 up to 0.836), their ability to predict moral sentiments is notably weak (F1 < 0.35). Furthermore, rationale alignment remains limited mainly in underrepresented languages. Our findings show the limited capacity of current LLMs to internalize and reflect human moral reasoning

Mikel K. Ngueajio, Flor Miriam Plaza-del-Arco, Yi-Ling Chung et al.

Automated counter-narratives (CN) offer a promising strategy for mitigating online hate speech, yet concerns about their affective tone, accessibility, and ethical risks remain. We propose a framework for evaluating Large Language Model (LLM)-generated CNs across four dimensions: persona framing, verbosity and readability, affective tone, and ethical robustness. Using GPT-4o-Mini, Cohere's CommandR-7B, and Meta's LLaMA 3.1-70B, we assess three prompting strategies on the MT-Conan and HatEval datasets. Our findings reveal that LLM-generated CNs are often verbose and adapted for people with college-level literacy, limiting their accessibility. While emotionally guided prompts yield more empathetic and readable responses, there remain concerns surrounding safety and effectiveness.

Justin Zhao, Flor Miriam Plaza-del-Arco, Benjamin Genchel et al.

As Large Language Models (LLMs) continue to evolve, evaluating them remains a persistent challenge. Many recent evaluations use LLMs as judges to score outputs from other LLMs, often relying on a single large model like GPT-4o. However, using a single LLM judge is prone to intra-model bias, and many tasks - such as those related to emotional intelligence, creative writing, and persuasiveness - may be too subjective for a single model to judge fairly. We introduce the Language Model Council (LMC), where a group of LLMs collaborate to create tests, respond to them, and evaluate each other's responses to produce a ranking in a democratic fashion. Unlike previous approaches that focus on reducing cost or bias by using a panel of smaller models, our work examines the benefits and nuances of a fully inclusive LLM evaluation system. In a detailed case study on emotional intelligence, we deploy a council of 20 recent LLMs to rank each other on open-ended responses to interpersonal conflicts. Our results show that the LMC produces rankings that are more separable and more robust, and through a user study, we show that they are more consistent with human evaluations than any individual LLM judge. Using all LLMs for judging can be costly, however, so we use Monte Carlo simulations and hand-curated sub-councils to study hypothetical council compositions and discuss the value of the incremental LLM judge.

Flor Miriam Plaza-del-Arco, Sercan Halat, Sebastian Padó et al.

The recognition of hate speech and offensive language (HOF) is commonly formulated as a classification task to decide if a text contains HOF. We investigate whether HOF detection can profit by taking into account the relationships between HOF and similar concepts: (a) HOF is related to sentiment analysis because hate speech is typically a negative statement and expresses a negative opinion; (b) it is related to emotion analysis, as expressed hate points to the author experiencing (or pretending to experience) anger while the addressees experience (or are intended to experience) fear. (c) Finally, one constituting element of HOF is the mention of a targeted person or group. On this basis, we hypothesize that HOF detection shows improvements when being modeled jointly with these concepts, in a multi-task learning setup. We base our experiments on existing data sets for each of these concepts (sentiment, emotion, target of HOF) and evaluate our models as a participant (as team IMS-SINAI) in the HASOC FIRE 2021 English Subtask 1A. Based on model-selection experiments in which we consider multiple available resources and submissions to the shared task, we find that the combination of the CrowdFlower emotion corpus, the SemEval 2016 Sentiment Corpus, and the OffensEval 2019 target detection data leads to an F1 =.79 in a multi-head multi-task learning model based on BERT, in comparison to .7895 of plain BERT. On the HASOC 2019 test data, this result is more substantial with an increase by 2pp in F1 and a considerable increase in recall. Across both data sets (2019, 2021), the recall is particularly increased for the class of HOF (6pp for the 2019 data and 3pp for the 2021 data), showing that MTL with emotion, sentiment, and target identification is an appropriate approach for early warning systems that might be deployed in social media platforms.