Papers by Ivan Srba
MULTITuDE: Large-Scale Multilingual Machine-Generated Text Detection Benchmark (2023.emnlp-main)
Copied to clipboard
Dominik Macko, Robert Moro, Adaku Uchendu, Jason Lucas, Michiharu Yamashita, Matúš Pikuliak, Ivan Srba, Thai Le, Dongwon Lee, Jakub Simko, Maria Bielikova
| Challenge: | MULTITuDE benchmarks lack authentic and machine-generated text in languages other than English . defining characteristic of new generation of LLMs is increased quality of text . |
| Approach: | They propose a benchmarking dataset for multilingual machine-generated text detection that compares detectors with authentic and machine-generated texts in 11 languages. |
| Outcome: | The proposed dataset compares detectors with zero-shot and fine-tuned detectors in 11 languages. |
Better as Generators Than Classifiers: Leveraging LLMs and Synthetic Data for Low-Resource Multilingual Classification (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable multilingual capabilities, making them promising tools in both high- and low-resource languages. |
| Approach: | They use a multilingual LLM to generate synthetic datasets covering 11 languages and 4 classification tasks and use them to train smaller models. |
| Outcome: | The proposed model outperforms the large generator in low-resource languages and tasks. |
Comparing Specialised Small and General Large Language Models on Text Classification: 100 Labelled Samples to Achieve Break-Even Performance (2025.emnlp-main)
Copied to clipboard
| Challenge: | specialised small models outperform general large models with few labelled samples (on average 100) performance variance is taken into account when comparing the number of labelled sample required to tune a specialised model with a larger number of samples. |
| Approach: | They find that specialised small models need only few labelled samples to outperform general large models with limited labelled data. |
| Outcome: | The proposed models outperform general large models with few labelled samples and take performance variance into account. |
Multilingual Previously Fact-Checked Claim Retrieval (2023.emnlp-main)
Copied to clipboard
Matúš Pikuliak, Ivan Srba, Robert Moro, Timo Hromadka, Timotej Smoleň, Martin Melišek, Ivan Vykopal, Jakub Simko, Juraj Podroužek, Maria Bielikova
| Challenge: | Fact-checkers are often hampered by the sheer amount of online content that needs to be fact-checked. |
| Approach: | They propose a multilingual dataset for previously fact-checked claim retrieval using social media posts and 206k fact- checks in 39 languages written by professional fact- checkers. |
| Outcome: | The proposed method improves on the previously unsupervised method and shows that a multilingual dataset has its complexities and needs to be carefully interpreted. |
MultiSocial: Multilingual Benchmark of Machine-Generated Text Detection of Social-Media Texts (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for detecting social-media texts are limited to the English language and longer texts are not easily recognisable by humans. |
| Approach: | They propose to use a multilingual and multi-platform dataset to compare machine-generated text detection methods in the social-media domain to compare them to human-written texts. |
| Outcome: | The proposed dataset contains 472,097 texts, of which about 58k are human-written and approximately the same amount is generated by each of 7 multilingual LLMs. |
Fighting Randomness with Randomness: Mitigating Optimisation Instability of Fine-Tuning using Delayed Ensemble and Noisy Interpolation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Despite the success of fine-tuning, it still displays model performance instability, especially with limited data. |
| Approach: | They propose a new mitigation strategy that leverages the strengths of ensembling, noise regularisation and model interpolation while retaining computational efficiency. |
| Outcome: | The proposed mitigation strategy outperforms the best performing mitigation strategy (Ensemble) while using only a fraction of its cost. |
Effects of diversity incentives on sample diversity and downstream model performance in LLM-based text augmentation (2024.acl-long)
Copied to clipboard
| Challenge: | generative large language models (LLMs) have found their application in data augmentation tasks, where small numbers of text samples are paraphrased and then used to fine-tune downstream models. |
| Approach: | They propose to use taboo words, hints by previous outlier solutions, and chaining on previous outliest solutions to augment text datasets as part of instructions to LLMs augmenting text dataset. |
| Outcome: | The proposed methods increase diversity of generated texts, but performance is highest with hints. |
PEFT-Factory: Unified Parameter-Efficient Fine-Tuning of Autoregressive Large Language Models (2026.eacl-demo)
Copied to clipboard
| Challenge: | Parameter-Efficient Fine-Tuning (PEFT) methods address the increasing size of Large Language Models (LLMs). |
| Approach: | They propose a framework for efficient fine-tuning Large Language Models (LLMs) they aim to train only a small percentage of the full model's parameters . |
| Outcome: | Xu et al., 2023; Ding e t al, 2024; Lialin e al. 2023) show that using PEFT methods can improve performance. |
MultiCW: A Large-Scale Balanced Benchmark Dataset for Training Robust Check-Worthiness Detection Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are beginning to reshape how media professionals verify information, but support for detecting check-worthy claims remains limited. |
| Approach: | They propose a multilingual benchmark for check-worthy claim detection spanning 16 languages, six topical domains, and two writing styles. |
| Outcome: | The proposed model outperforms zero-shot LLMs on claim classification and strong generalization across languages, domains, and styles. |
Use Random Selection for Now: Investigation of Few-Shot Selection Strategies in LLM-based Text Augmentation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | generative large language models are increasingly used for data augmentation tasks . text samples are mostly selected randomly and a comprehensive overview of other sample selection strategies is lacking. |
| Approach: | They compare random sample selection strategies and random sample sampling strategies to evaluate their effects in a low-resource setting. |
| Outcome: | The proposed model performance improvements are compared with other sample selection strategies. |
Evaluation of LLM Vulnerabilities to Being Misused for Personalized Disinformation Generation (2025.acl-long)
Copied to clipboard
Aneta Zugecova, Dominik Macko, Ivan Srba, Robert Moro, Jakub Kopál, Katarína Marcinčinová, Matúš Mesarčík
| Challenge: | Recent large language models generate disinformation news articles following predefined narratives . personalization and disinformation abilities of LLMs have not been studied . |
| Approach: | They evaluate the personalization and disinformation abilities of large language models . they find personalization reduces the safety-filter activations, thus effectively functioning as a jailbreak . |
| Outcome: | The proposed model generates disinformation news articles in english with the lowest quality of personalization. |
Automatic Combination of Sample Selection Strategies for Few-Shot Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on small language models are characterised by a labelled data scarcity due to data collection/annotation costs or privacy considerations, making the training of typical deep learning models unfeasible. |
| Approach: | They propose a method for Automatic Combination of SamplE Selection Strategies to leverage the strengths and complementarity of various well-established selection objectives. |
| Outcome: | The proposed method outperforms all in-context learning strategies and performs on par or exceeds the in-constinction learning specific baselines. |
A Ship of Theseus: Curious Cases of Paraphrasing in LLM-Generated Texts (2024.acl-long)
Copied to clipboard
Nafis Irtiza Tripto, Saranya Venkatraman, Dominik Macko, Robert Moro, Ivan Srba, Adaku Uchendu, Thai Le, Dongwon Lee
| Challenge: | Using a computational approach, we discover that diminishing performance in text classification models is closely associated with the extent of deviation from the original author’s style. |
| Approach: | They propose to use large language models to determine whether a text retains original authorship when it undergoes numerous paraphrasing iterations. |
| Outcome: | The results suggest that authorship should be task-dependent . |
Authorship Attribution in Multilingual Machine-Generated Texts (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have reached human-like fluency and coherence, but distinguishing machine-generated text from human-written content becomes increasingly difficult. |
| Approach: | They propose a problem of multilingual authorship attribution (AA) that involves attributing texts to human or multiple LLM generators across diverse languages. |
| Outcome: | The proposed method can be adapted to multilingual settings, but still has significant limitations and challenges. |
RoSE: Round-robin Synthetic Data Evaluation for Selecting LLM Generators without Human Test Sets (2026.eacl-long)
Copied to clipboard
| Challenge: | Current large language models (LLMs) are powerful generators of synthetic data, which are used for training smaller, more efficient models. |
| Approach: | They propose a proxy metric for selecting the best LLM generator without human annotations and a metric that measures the performance of a model. |
| Outcome: | The proposed proxy metric outperforms intrinsic heuristics and comes within 0.76 percentage points of the optimal generator baseline. |
On Sensitivity of Learning with Limited Labelled Data to the Effects of Randomness: Impact of Interactions and Systematic Choices (2024.emnlp-main)
Copied to clipboard
| Challenge: | Using limited labelled data, learning with limited labels is sensitive to the effects of uncontrolled randomness. |
| Approach: | They propose to investigate the effects of individual randomness factors while taking the interactions between them into consideration. |
| Outcome: | The proposed method mitigates the effects of other factors while observing how the performance varies across multiple runs. |
Authorship Obfuscation in Multilingual Machine-Generated Text Detection (2024.findings-emnlp)
Copied to clipboard
Dominik Macko, Robert Moro, Adaku Uchendu, Ivan Srba, Jason Lucas, Michiharu Yamashita, Nafis Irtiza Tripto, Dongwon Lee, Jakub Simko, Maria Bielikova
| Challenge: | Recent advances in Language Modeling have birthed Large Language Models (LLMs), which exhibit significant improvements, including the ability to generate texts easily misconstrued as humanwritten. |
| Approach: | They compare authorship obfuscation methods against machine-generated text (MGT) in 11 languages and analyze their performance against 37 well-known AO methods. |
| Outcome: | The proposed methods can cause evasion of detection in all languages, with homoglyph attacks particularly successful. |
PEFT-Bench: A Parameter-Efficient Fine-Tuning Methods Benchmark (2026.eacl-long)
Copied to clipboard
| Challenge: | Parameter-Efficient Fine-Tuning (PEFT) methods reduce the number of trainable parameters while maintaining strong downstream performance. |
| Approach: | They propose a unified benchmark for evaluating diverse PEFT methods on autoregressive LLMs. |
| Outcome: | The proposed methods reduce trainable parameters while maintaining strong downstream performance. |