Coreference resolution is an important part of natural language processing used in machine translation, semantic search, and various other information retrieval and understanding systems. One of the challenges in this field is an evaluation of resolution approaches. There are many different metrics proposed, but most of them rely on certain assumptions, like equivalence between different mentions of the same discourse-world entity, and do not account for overrepresentation of certain types of coreferences present in the evaluation data. In this paper, a new coreference evaluation strategy that focuses on linguistic and semantic information is presented that can address some of these shortcomings. Evaluation model was developed in the broader context of developing coreference resolution capabilities for Lithuanian language; therefore, the experiment was also carried out using Lithuanian language resources, but the proposed evaluation strategy is not language-dependent.

Online hate speech and abusive language pose a growing challenge for content moderation, especially in multilingual settings and for low-resource languages such as Lithuanian. This paper investigates to what extent modern multilingual sentence embedding models can support accurate hate speech detection in Lithuanian, Russian, and English, and how their performance depends on downstream modeling choices and feature dimensionality. We introduce LtHate, a new Lithuanian hate speech corpus derived from news portals and social networks, and benchmark six modern multilingual encoders (gemma, qwen, bge, snow, jina, and e5) on LtHate, RuToxic, and EnSuperset using a unified Python pipeline. For each embedding type, we train both a one-class histogram-based anomaly detector (HBOS) and a two-class gradient-boosted tree ensemble (CatBoost), with and without Principal Component Analysis (PCA) compression to 32-dimensional feature vectors. Across all datasets, two-class supervised models consistently and substantially outperform one-class anomaly detection, with the best configurations achieving up to 78.8% accuracy (Kappa 0.58, AUC ROC 0.87) in Lithuanian (jina), 92.2% accuracy (Kappa 0.77, AUC ROC 0.97) in Russian (e5), and 76.9% accuracy (Kappa 0.48, AUC ROC 0.86) in English (e5). PCA compression deteriorates the discriminative power of CatBoost only slightly, with much more negative impact for the HBOS model. These results demonstrate how modern multilingual sentence embeddings combined with gradient-boosted decision trees provides robust machine learning solutions for multilingual hate speech detection applications.

This paper examines whether incremental prompt engineering can enable reliable LLM based pre-annotation of corpus texts in a low resource language setting. Using Lithuanian as a case study, we systematically evaluate multiple LLM prompt designs and assess their suitability for generating emotional manipulation annotations for corpus development. We find that performance varies with task complexity, and systematic prompt refinement measurably reduces output instability. Cross-model evaluation of the best-performing prompting strategy shows consistent and similar trends over several modern LLMs. Our results demonstrate that while structured prompts substantially improve output consistency and LLM assisted annotation can roughly approximate human produced labels for well-defined categories, the quality of results produced by contemporary LLMs is unsatisfactory for automatic pre-annotation of emotional manipulation techniques in a low resource language.

This paper examines whether incremental prompt engineering can enable reliable large language model (LLM)-based pre-annotation of corpus texts in a low-resource language setting. Using Lithuanian as a case study, we systematically evaluate multiple LLM prompt designs and assess their suitability for generating emotional manipulation annotations for corpus development. We find that performance varies with task complexity, and systematic prompt refinement measurably reduces output instability. Cross-model evaluation of the best-performing prompting strategy shows consistent and similar trends over several modern LLMs. Our results demonstrate that while structured prompts substantially improve output consistency and LLM-assisted annotation can roughly approximate human-produced labels for well-defined categories, the quality of results produced by contemporary LLMs is unsatisfactory for automatic pre-annotation of emotional manipulation techniques in a low-resource language.

Anaphora resolution is an important part of natural language processing used in machine translation, semantic search and various other information retrieval and understanding systems. Anaphora resolution algorithms usually require linguistic pre-processing tools and various expensive resources for automatically identifying anaphoric expressions. Many smaller languages, like Lithuanian, lack such resources and tools. In this paper, an algorithm is proposed that requires only morphological annotations and recognized named entities. The paper presents experimental results showing the relevance of the solution for specific domains, and considers the further immediate ways towards dealing with the overall anaphora resolution problem for Lithuanian language.