Yulia Tsvetkov · page 2
Papers
- Native Language Cognate Effects on Second Language Lexical Choice (2018 · 52 citations)
Psycholinguistic research suggests that bilinguals favor second-language words that resemble equivalents in their mother tongue (cognate facilitation). - Resolving Knowledge Conflicts in Large Language Models (2023 · 48 citations)
When LLMs receive retrieved or prompted text, that text often contradicts what they absorbed during pretraining. - CulturalBench: A Robust, Diverse and Challenging Benchmark for Measuring LMs' Cultural Knowledge Through Human-AI Red-Teaming (2025 · 38 citations)
Measuring how well language models understand everyday cultural practice worldwide is difficult. Static hand-written test sets soon become too easy. Benchmarks built from web pages, value surveys or synthetic generation may overlap with training data or carry over model biases. - Knowledge Card: Filling LLMs' Knowledge Gaps with Plug-in Specialized Language Models (2023 · 35 citations)
Large language models are frozen after training and costly to update. As a result, they often produce outdated, hallucinated, or missing facts, especially for long-tail and fast-changing topics. Knowledge Card addresses this by attaching small, domain-specific language models ("knowledge cards") to a black-box LLM.… - Cross-Cultural Similarity Features for Cross-Lingual Transfer Learning of Pragmatically Motivated Tasks (2021 · 28 citations)
Choosing a source language for cross-lingual transfer is usually guided by typological or genealogical closeness. - Biased LLMs can Influence Political Decision-Making (2025 · 20 citations)
The paper asks whether partisan bias in chat-based LLMs can change people's political views and choices during short, free-form conversations. - MAGNET: Improving the Multilingual Fairness of Language Models with Adaptive Gradient-Based Tokenization (2024 · 19 citations)
Byte-level and subword language models usually split non-Latin-script text into many more pieces than English. - Among Us: Measuring and Mitigating Malicious Contributions in Model Collaboration Systems (2026 · 1 citation)
Multi-LLM systems increasingly combine models built by different parties, which raises the question of how a compromised participant affects the whole.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/yulia-tsvetkov/page/2/