Retrieval Augmented Generation (RAG) improves correctness of Question Answering (QA) and addresses hallucinations in Large Language Models (LLMs), yet greatly increase computational costs. Besides, RAG is not always needed as may introduce irrelevant information. Recent adaptive retrieval methods integrate LLMs’ intrinsic knowledge with external information appealing to LLM self-knowledge, but they often neglect efficiency evaluations and comparisons with uncertainty estimation techniques. We bridge this gap by conducting a comprehensive analysis of 35 adaptive retrieval methods, including 8 recent approaches and 27 uncertainty estimation techniques, across 6 datasets using 10 metrics for QA performance, self-knowledge, and efficiency. Our findings show that uncertainty estimation techniques often outperform complex pipelines in terms of efficiency and self-knowledge, while maintaining comparable QA performance.

The task of persuasion techniques detection is limited by several challenges, such as insufficient training data and ambiguity in labels. In this paper, we describe a solution for the Slavic NLP 2025 Shared Task. It utilizes multilingual XLM-RoBERTa, that was trained on 100 various languages, and Slavic BERT, a model fine-tuned on four languages of the Slavic group. We suggest to augment the training dataset with related data from previous shared tasks, as well as some automatic translations from English and German. The resulting solutions are ranked among the top 3 for Russian in the Subtask 1 and for all languages in the Subtask 2. We release the code for our solution - https://github.com/ssenichev/ACL_SlavicNLP2025.

Large Language Models (LLMs) often hallucinate in question answering (QA) tasks. A key yet underexplored factor contributing to this is the temporality of questions – whether they are evergreen (answers remain stable over time) or mutable (answers change). In this work, we introduce EverGreenQA, the first multilingual QA dataset with evergreen labels, supporting both evaluation and training. Using EverGreenQA, we benchmark 12 modern LLMs to assess whether they encode question temporality explicitly (via verbalized judgments) or implicitly (via uncertainty signals). We also train EG-E5, a lightweight multilingual classifier that achieves SoTA performance on this task. Finally, we demonstrate the practical utility of evergreen classification across three applications: improving self-knowledge estimation, filtering QA datasets, and explaining GPT-4o’s retrieval behavior.

Large Language Models (LLMs) are prone to hallucinations, and Retrieval-Augmented Generation (RAG) helps mitigate this, but at a high computational cost while risking misinformation. Adaptive retrieval aims to retrieve only when necessary, but existing approaches rely on LLM-based uncertainty estimation, which remains inefficient and impractical.In this study, we introduce lightweight LLM-independent adaptive retrieval methods based on external information. We investigated 27 features, organized into 7 groups, and their hybrid combinations. We evaluated these methods on 6 QA datasets, assessing the QA performance and efficiency. The results show that our approach matches the performance of complex LLM-based methods while achieving significant efficiency gains, demonstrating the potential of external information for adaptive retrieval.

We present our submission to SciHal Subtask 1: coarse-grained hallucination detection for scientific question answering. We frame hallucination detection as an NLI-style three-way classification (entailment, contradiction, unverifiable) and show that simple fine-tuning of NLI-adapted encoder models on task data outperforms more elaborate feature-based pipelines and large language model prompting. In particular, DeBERTa-V3-large, a model pretrained on five diverse NLI corpora, achieves the highest weighted F1 on the public leaderboard. We additionally explore a pipeline combining joint claim–reference embeddings and NLI softmax probabilities fed into a classifier, but find its performance consistently below direct encoder fine-tuning. Our findings demonstrate that, for reference-grounded hallucination detection, targeted encoder fine-tuning remains the most accurate and efficient approach.

In this work, we propose an efficient answer retrieval system **EARS**: a production-ready, factual question answering (QA) system that combines local knowledge base search with generative, context-based QA. To assess the quality of the generated content, we devise comprehensive metrics for both manual and automatic evaluation of the answers to questions. A distinctive feature of our system is the Ranker component, which ranks answer candidates based on their relevance. This feature enhances the effectiveness of local knowledge base retrieval by 23%. Another crucial aspect of our system is the LLM, which utilizes contextual information from a web search API to generate responses. This results in substantial 92.8% boost in the usefulness of voice-based responses. **EARS** is language-agnostic and can be applied to any data domain.

One of the main obstacles for deploying Active Learning (AL) in practical NLP tasks is high computational cost of modern deep learning models. This issue can be partially mitigated by applying lightweight models as an acquisition model, but it can lead to the acquisition-successor mismatch (ASM) problem. Previous works show that the ASM problem can be partially alleviated by using distilled versions of a successor models as acquisition ones. However, distilled versions of pretrained models are not always available. Also, the exact pipeline of model distillation that does not lead to the ASM problem is not clear. To address these issues, we propose to use adapters as an alternative to full fine-tuning for acquisition model training. Since adapters are lightweight, this approach reduces the training cost of the model. We provide empirical evidence that it does not cause the ASM problem and can help to deploy active learning in practical NLP tasks.

The task of generating long narratives using Large Language Models (LLMs) is a largely unexplored area within natural language processing (NLP). Although modern LLMs can handle up to 1 million tokens, ensuring coherence and control over long story generation is still a significant challenge. This paper investigates the use of summarization techniques to create extended narratives, specifically targeting long stories. We propose a special prompting scheme that segments the narrative into several parts and chapters, each generated iteratively with contextual information. Our approach is evaluated with GAPELMAPER, a sophisticated text coherence metric, for automatic evaluation to maintain the structural integrity of the generated stories. We also rely on human evaluation to assess the quality of the generated text. This research advances the development of tools for long story generation in NLP, highlighting both the potential and current limitations of LLMs in this field.