Katerina Margatina

selected publications

1. ACL

   CONFETTI: Conversational Function-Calling Evaluation Through Turn-Level Interactions

   Tamer Alkhouli, Katerina Margatina, James Gung, Raphael Shu, Claudia Zaghi, Monica Sunkara, and Yi Zhang

   In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. 2025

   Abstract arXiv PDF Blog

   We introduce Conversational Function-Calling Evaluation Through Turn-Level Interactions (CONFETTI), a conversational benchmark1 designed to evaluate the function-calling capabilities and response quality of large language models (LLMs). Current benchmarks lack comprehensive assessment of LLMs in complex conversational scenarios. CONFETTI addresses this gap through 109 human-simulated conversations, comprising 313 user turns and covering 86 APIs. These conversations explicitly target various conversational complexities, such as follow-ups, goal correction and switching, ambiguous and implicit goals. We perform off-policy turn-level evaluation using this benchmark targeting function-calling. Our benchmark also incorporates dialog act annotations to assess agent responses. We evaluate a series of state-of-the-art LLMs and analyze their performance with respect to the number of available APIs, conversation lengths, and chained function calling. Our results reveal that while some models are able to handle long conversations, and leverage more than 20+ APIs successfully, other models struggle with longer context or when increasing the number of APIs. We also report that the performance on chained function-calls is severely limited across the models. Overall, the top performing models on CONFETTI are Nova Pro (40.01%), Claude Sonnet v3.5 (35.46%) and Llama 3.1 405B (33.19%) followed by command-r-plus (31.18%) and Mistral-Large-2407 (30.07%).

2. REALM

   A Study on Leveraging Search and Self-Feedback for Agent Reasoning

   Karthikeyan K, Michelle Yuan, Elman Mansimov, Katerina Margatina, Anurag Pratik, Daniele Bonadiman, Monica Sunkara, Yi Zhang, and Yassine Benajiba

   In Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025). 2025

   Abstract arXiv PDF

   Recent works have demonstrated that incorporating search during inference can significantly improve reasoning capabilities of language agents. Some approaches may make use of the ground truth or rely on model’s own generated feedback. The search algorithm uses this feedback to then produce values that will update its criterion for exploring and exploiting various reasoning paths. In this study, we investigate how search and model’s self-feedback can be leveraged for reasoning tasks. First, we explore differences in ground-truth feedback and self-feedback during search for math reasoning. Second, we observe limitations in applying search techniques to more complex tasks like tool-calling and design domain-specific approaches to address these gaps. Our experiments reveal challenges related to generalization when solely relying on self-feedback during search. For search to work effectively, either access to the ground-truth is needed or feedback mechanisms need to be carefully designed for the specific task.

3. Thesis

   Exploring Active Learning Algorithms for Data Efficient Language Models

   Katerina Margatina

   2024

   Abstract HTML

   Supervised learning is based in the premise that models can effectively solve tasks by learning from numerous examples, mapping inputs to outputs through iterative learning. However, contemporary deep learning models often require vast amounts of labeled data, termed training examples, for optimal performance. Unfortunately, not all training examples contribute equally to the learning process, leading to inefficiencies and resource wastage. Active Learning (AL) has emerged as a powerful paradigm for training language models in a data-efficient manner. By iteratively selecting informative unlabeled data points, which are then annotated by humans to form the training set, AL intelligently guides the training process, optimizing data selection for model improvement over random sampling. This thesis investigates various aspects of active learning algorithms for language mod- els, focusing on model training, data selection, in-context learning and simulation. The thesis is structured along four key publications that tackle these topics respectively. The first publication addresses the effective adaptation of pretrained language models for AL, highlighting the importance of task-specific fine-tuning. The second publication introduces a novel acquisition function, Contrastive Active Learning (CAL), which selects contrastive examples to improve AL performance. The third publication explores active learning principles for in-context learning with large language models, emphasizing the selection of informative demonstrations for few-shot learning. Lastly, the fourth publication critically examines the limitations of simulating AL experiments and pro- poses guidelines for future research. Through these contributions, this thesis aims to advance our understanding of AL algorithms for data-efficient language model training.

4. NeurIPS

   🏆 Best Paper

   The PRISM Alignment Project: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models

   Hannah Rose Kirk, Alexander Whitefield, Paul Röttger, Andrew Bean, Katerina Margatina, Juan Ciro, Rafael Mosquera, Max Bartolo, Adina Williams, Bertie Vidgen He He, and Scott A. Hale

   In Proceedings of the Conference on Neural Information Processing Systems (NeurIPS) Track on Datasets and Benchmarks. 2024

   Abstract arXiv HTML Blog TL;DR

   Human feedback plays a central role in the alignment of Large Language Models (LLMs). However, open questions remain about the methods (how), domains (where), people (who) and objectives (to what end) of human feedback collection. To navigate these questions, we introduce PRISM, a new dataset which maps the sociodemographics and stated preferences of 1,500 diverse participants from 75 countries, to their contextual preferences and fine-grained feedback in 8,011 live conversations with 21 LLMs. PRISM contributes (i) wide geographic and demographic participation in human feedback data; (ii) two census-representative samples for understanding collective welfare (UK and US); and (iii) individualised feedback where every rating is linked to a detailed participant profile, thus permitting exploration of personalisation and attribution of sample artefacts. We focus on collecting conversations that centre subjective and multicultural perspectives on value-laden and controversial topics, where we expect the most interpersonal and cross-cultural disagreement. We demonstrate the usefulness of PRISM via three case studies of dialogue diversity, preference diversity, and welfare outcomes, showing that it matters which humans set alignment norms. As well as offering a rich community resource, we advocate for broader participation in AI development and a more inclusive approach to technology design.

5. EMNLP-Findings

   Active Learning Principles for In-Context Learning with Large Language Models

   Katerina Margatina, Timo Schick, Nikolaos Aletras, and Jane Dwivedi-Yu

   In Findings of the Conference on Empirical Methods in Natural Language Processing (EMNLP) 2023 2023

   Abstract arXiv TL;DR

   The remarkable advancements in large language models (LLMs) have significantly enhanced the performance in few-shot learning settings. By using only a small number of labeled examples, referred to as demonstrations, LLMs can effectively grasp the task at hand through in-context learning. However, the process of selecting appropriate demonstrations has received limited attention in prior work. This paper addresses the issue of identifying the most informative demonstrations for few-shot learning by approaching it as a pool-based Active Learning (AL) problem over a single iteration. Our objective is to investigate how AL algorithms can serve as effective demonstration selection methods for in-context learning. We compare various standard AL algorithms based on uncertainty, diversity, and similarity, and consistently observe that the latter outperforms all other methods, including random sampling. Notably, uncertainty sampling, despite its success in conventional supervised learning scenarios, performs poorly in this context. Our extensive experimentation involving a diverse range of GPT and OPT models across 24 classification and multi-choice tasks, coupled with thorough analysis, unambiguously demonstrates that in-context example selection through AL prioritizes high-quality examples that exhibit low uncertainty and bear similarity to the test examples.

6. EMNLP

   ✨ Oral ✨

   Active Learning by Acquiring Contrastive Examples

   Katerina Margatina, Giorgos Vernikos, Loïc Barrault, and Nikolaos Aletras

   In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP) 2021

   Abstract arXiv HTML PDF TL;DR Code Poster Slides

   Common acquisition functions for active learning use either uncertainty or diversity sampling, aiming to select difficult and diverse data points from the pool of unlabeled data, respectively. In this work, leveraging the best of both worlds, we propose an acquisition function that opts for selecting \textitcontrastive examples, i.e. data points that are similar in the model feature space and yet the model outputs maximally different predictive likelihoods. We compare our approach, CAL (Contrastive Active Learning), with a diverse set of acquisition functions in four natural language understanding tasks and seven datasets. Our experiments show that CAL performs consistently better or equal than the best performing baseline across all tasks, on both in-domain and out-of-domain data. We also conduct an extensive ablation study of our method and we further analyze all actively acquired datasets showing that CAL achieves a better trade-off between uncertainty and diversity compared to other strategies.