Hate speech spreads widely online and harms both individuals and communities, making automatic detection essential for large-scale moderation. However, accurately detecting hate speech remains a difficult task. Part of the challenge lies in subjectivity: what one person flags as hate speech, another may see as benign. Traditional annotation agreement metrics, such as Cohen’s k, oversimplify this disagreement, treating it as an error rather than meaningful diversity. Meanwhile, Large Language Models (LLMs) promise scalable annotation, but prior studies demonstrate that they cannot fully replace human judgement, especially in subjective tasks. In this work, we reexamine LLM reliability using a subjectivity-aware framework, cross-Replication Reliability (xRR), revealing that even under fairer lens, LLMs still diverge from humans. Yet this limitation opens an opportunity: we find that LLM-generated annotations can reliably reflect performance trends across classification models, correlating with human evaluations. We test this by examining whether LLM-generated annotations preserve the relative ordering of model performance derived from human evaluation (i.e. whether models ranked as more reliable by human annotators preserve the same order when evaluated with LLM-generated labels). Our results show that, although LLMs differ from humans at the instance level, they reproduce similar ranking and classification patterns, suggesting their potential as proxy evaluators. While not a substitute for human annotators, they might serve as a scalable proxy for evaluation in subjective NLP tasks.

We explore the use of discourse parsers for extracting a particular discourse structure in a real-world social media scenario. Specifically, we focus on enhancing parser performance through the integration of synthetic data generated by large language models (LLMs). We conduct experiments using a newly developed dataset of 1,170 local RST discourse structures, including 900 synthetic and 270 gold examples, covering three social media platforms: online news comments sections, a discussion forum (Reddit), and a social media messaging platform (Twitter). Our primary goal is to assess the impact of LLM-generated synthetic training data on parser performance in a raw text setting without pre-identified discourse units. While both top-down and bottom-up RST architectures greatly benefit from synthetic data, challenges remain in classifying evaluative discourse structures.

This paper presents an original approach for SemEval 2025 Task 11. Our study investigates various strategies to improve Text-Based Multi-label Emotion Detection task. Through experimental endeavors, we explore the benefits of contextualized vector representations by comparing multiple BERT models, including those specifically trained for emotion recognition. Additionally, we examine the impact of hyperparameters adjustments on model performance. For Subtask A, our approach achieved F1 scores of 0.71 on the English dataset and 0.84 on the Russian dataset. Our findings underscore that (1) monolingual BERT models demonstrate superior performance for English, whereas multilingual BERT models perform better for Russian; (2) pretrained emotion detection models proving less effective for this specific task compared to models with reduced vocabulary and embeddings focused on specific languages;(3) exclusive use of BERT-based models, without incorporating additional methods or optimization techniques, demonstrates promising results for multilabel emotion detection.