Elisabetta Fersini

Angel Felipe Magnossão de Paula, Giulia Rizzi, Elisabetta Fersini et al.

With the increasing influence of social media platforms, it has become crucial to develop automated systems capable of detecting instances of sexism and other disrespectful and hateful behaviors to promote a more inclusive and respectful online environment. Nevertheless, these tasks are considerably challenging considering different hate categories and the author's intentions, especially under the learning with disagreements regime. This paper describes AI-UPV team's participation in the EXIST (sEXism Identification in Social neTworks) Lab at CLEF 2023. The proposed approach aims at addressing the task of sexism identification and characterization under the learning with disagreements paradigm by training directly from the data with disagreements, without using any aggregated label. Yet, performances considering both soft and hard evaluations are reported. The proposed system uses large language models (i.e., mBERT and XLM-RoBERTa) and ensemble strategies for sexism identification and classification in English and Spanish. In particular, our system is articulated in three different pipelines. The ensemble approach outperformed the individual large language models obtaining the best performances both adopting a soft and a hard label evaluation. This work describes the participation in all the three EXIST tasks, considering a soft evaluation, it obtained fourth place in Task 2 at EXIST and first place in Task 3, with the highest ICM-Soft of -2.32 and a normalized ICM-Soft of 0.79. The source code of our approaches is publicly available at https://github.com/AngelFelipeMP/Sexism-LLM-Learning-With-Disagreement.

Daniel Scalena, Leonidas Zotos, Elisabetta Fersini et al.

With the rise of reasoning language models and test-time scaling methods as a paradigm for improving model performance, substantial computation is often required to generate multiple candidate sequences from the same prompt. This enables exploration of different reasoning paths toward the correct solution, however, allocates the same compute budget for each prompt. Grounded on the assumption that different prompts carry different degrees of complexity, and thus different computation needs, we propose EAGer, a training-free generation method that leverages model uncertainty through token-wise entropy distribution to reduce redundant computation and concurrently improve overall performance. EAGer allows branching to multiple reasoning paths only in the presence of high-entropy tokens, and then reallocates the saved compute budget to the instances where exploration of alternative paths is most needed. We find that across multiple open-source models on complex reasoning benchmarks such as AIME 2025, EAGer can reallocate the budget without accessing target labels, achieving the best efficiency-performance trade-off in terms of reasoning length and Pass@k. When target labels are accessible, EAGer generates up to 65% fewer tokens (hence saving compute) and achieves up to 37% improvement in Pass@k compared to the Full Parallel Sampling.

Elisa Leonardelli, Silvia Casola, Siyao Peng et al.

Many researchers have reached the conclusion that AI models should be trained to be aware of the possibility of variation and disagreement in human judgments, and evaluated as per their ability to recognize such variation. The LEWIDI series of shared tasks on Learning With Disagreements was established to promote this approach to training and evaluating AI models, by making suitable datasets more accessible and by developing evaluation methods. The third edition of the task builds on this goal by extending the LEWIDI benchmark to four datasets spanning paraphrase identification, irony detection, sarcasm detection, and natural language inference, with labeling schemes that include not only categorical judgments as in previous editions, but ordinal judgments as well. Another novelty is that we adopt two complementary paradigms to evaluate disagreement-aware systems: the soft-label approach, in which models predict population-level distributions of judgments, and the perspectivist approach, in which models predict the interpretations of individual annotators. Crucially, we moved beyond standard metrics such as cross-entropy, and tested new evaluation metrics for the two paradigms. The task attracted diverse participation, and the results provide insights into the strengths and limitations of methods to modeling variation. Together, these contributions strengthen LEWIDI as a framework and provide new resources, benchmarks, and findings to support the development of disagreement-aware technologies.

Daniel Scalena, Gabriele Sarti, Arianna Bisazza et al.

Large language models have simplified the production of personalized translations reflecting predefined stylistic constraints. However, these systems still struggle when stylistic requirements are implicitly represented by a set of examples, such as texts produced by a specific human translator. In this work, we explore various strategies for personalizing automatically generated translations when few examples are available, with a focus on the challenging domain of literary translation. We begin by determining the feasibility of the task and how style information is encoded within model representations. Then, we evaluate various prompting strategies and inference-time interventions for steering model generations towards a personalized style, with a particular focus on contrastive steering with sparse autoencoder (SAE) latents to identify salient personalization properties. We demonstrate that contrastive SAE steering yields robust style conditioning and translation quality, resulting in higher inference-time computational efficiency than prompting approaches. We further examine the impact of steering on model activations, finding that layers encoding personalization properties are impacted similarly by prompting and SAE steering, suggesting a similar mechanism at play.

Daniel Scalena, Elisabetta Fersini, Malvina Nissim

Adapting models to a language that was only partially present in the pre-training data requires fine-tuning, which is expensive in terms of both data and computational resources. As an alternative to fine-tuning, we explore the potential of activation steering-based techniques to enhance model performance on Italian tasks. Through our experiments we show that Italian steering (i) can be successfully applied to different models, (ii) achieves performances comparable to, or even better than, fine-tuned models for Italian, and (iii) yields higher quality and consistency in Italian generations. We also discuss the utility of steering and fine-tuning in the contemporary LLM landscape where models are anyway getting high Italian performances even if not explicitly trained in this language.

Daniel Scalena, Gabriele Sarti, Malvina Nissim et al.

Due to language models' propensity to generate toxic or hateful responses, several techniques were developed to align model generations with users' preferences. Despite the effectiveness of such methods in improving the safety of model interactions, their impact on models' internal processes is still poorly understood. In this work, we apply popular detoxification approaches to several language models and quantify their impact on the resulting models' prompt dependence using feature attribution methods. We evaluate the effectiveness of counter-narrative fine-tuning and compare it with reinforcement learning-driven detoxification, observing differences in prompt reliance between the two methods despite their similar detoxification performances.

Silvia Terragni, Ismail Harrando, Pasquale Lisena et al.

Topic models are statistical methods that extract underlying topics from document collections. When performing topic modeling, a user usually desires topics that are coherent, diverse between each other, and that constitute good document representations for downstream tasks (e.g. document classification). In this paper, we conduct a multi-objective hyperparameter optimization of three well-known topic models. The obtained results reveal the conflicting nature of different objectives and that the training corpus characteristics are crucial for the hyperparameter selection, suggesting that it is possible to transfer the optimal hyperparameter configurations between datasets.

Francesca Gasparini, Giulia Rizzi, Aurora Saibene et al.

In this paper we present a benchmark dataset generated as part of a project for automatic identification of misogyny within online content, which focuses in particular on memes. The benchmark here described is composed of 800 memes collected from the most popular social media platforms, such as Facebook, Twitter, Instagram and Reddit, and consulting websites dedicated to collection and creation of memes. To gather misogynistic memes, specific keywords that refer to misogynistic content have been considered as search criterion, considering different manifestations of hatred against women, such as body shaming, stereotyping, objectification and violence. In parallel, memes with no misogynist content have been manually downloaded from the same web sources. Among all the collected memes, three domain experts have selected a dataset of 800 memes equally balanced between misogynistic and non-misogynistic ones. This dataset has been validated through a crowdsourcing platform, involving 60 subjects for the labelling process, in order to collect three evaluations for each instance. Two further binary labels have been collected from both the experts and the crowdsourcing platform, for memes evaluated as misogynistic, concerning aggressiveness and irony. Finally for each meme, the text has been manually transcribed. The dataset provided is thus composed of the 800 memes, the labels given by the experts and those obtained by the crowdsourcing validation, and the transcribed texts. This data can be used to approach the problem of automatic detection of misogynistic content on the Web relying on both textual and visual cues, facing phenomenons that are growing every day such as cybersexism and technology-facilitated violence.

Federico Bianchi, Silvia Terragni, Dirk Hovy et al.

Many data sets (e.g., reviews, forums, news, etc.) exist parallelly in multiple languages. They all cover the same content, but the linguistic differences make it impossible to use traditional, bag-of-word-based topic models. Models have to be either single-language or suffer from a huge, but extremely sparse vocabulary. Both issues can be addressed by transfer learning. In this paper, we introduce a zero-shot cross-lingual topic model. Our model learns topics on one language (here, English), and predicts them for unseen documents in different languages (here, Italian, French, German, and Portuguese). We evaluate the quality of the topic predictions for the same document in different languages. Our results show that the transferred topics are coherent and stable across languages, which suggests exciting future research directions.

Flavio Massimiliano Cecchini, Elisabetta Fersini

We begin by introducing the Computer Science branch of Natural Language Processing, then narrowing the attention on its subbranch of Information Extraction and particularly on Named Entity Recognition, discussing briefly its main methodological approaches. It follows an introduction to state-of-the-art Conditional Random Fields under the form of linear chains. Subsequently, the idea of constrained inference as a way to model long-distance relationships in a text is presented, based on an Integer Linear Programming representation of the problem. Adding such relationships to the problem as automatically inferred logical formulas, translatable into linear conditions, we propose to solve the resulting more complex problem with the aid of Lagrangian relaxation, of which some technical details are explained. Lastly, we give some experimental results.