Anna Korhonen · page 2
Papers
- Multi 3 WOZ: A Multilingual, Multi-Domain, Multi-Parallel Dataset for Training and Evaluating Culturally Adapted Task-Oriented Dialog Systems (2023 · 17 citations)
Multilingual task-oriented dialog (TOD) systems need data that is hard to obtain. Existing multilingual TOD datasets are usually built by translating English data, which brings in "translationese" and inflates scores. Many are also small, ignore local culture, lack parallel dialogs across languages, or cover only… - Analyzing and Adapting Large Language Models for Few-Shot Multilingual NLU: Are We There Yet? (2025 · 15 citations)
Few-shot adaptation of NLP models to new languages can follow three paths: supervised fine-tuning (SFT) of compact multilingual models, supervised instruction tuning (SIT), or in-context learning (ICL) with large generative LLMs. - Explainability and Interpretability of Multilingual Large Language Models: A Survey (2025 · 11 citations)
Multilingual large language models (MLLMs) do well on cross-lingual tasks, but it remains unclear how they handle several languages internally, how knowledge moves between languages, and how they deal with scripts, word order or culture-specific bias. - A Rose by Any Other Name: LLM-Generated Explanations Are Good Proxies for Human Explanations to Collect Label Distributions on NLI (2025 · 9 citations)
Human annotators often disagree on natural language inference (NLI) items. That disagreement can be recorded as a human judgment distribution (HJD), but measuring it well takes many annotators per item. Earlier work showed that LLMs can estimate HJDs from a few human labels paired with human-written explanations.… - Can VLMs Predict Future States? Bootstrapping World Models from Inverse Dynamics (2026)
The paper asks whether unified vision–language models, which can interleave text and images during generation, can act as world models. - How Far Can Bias Go? Tracing Bias from Pre-Training Data to Alignment (2026)
This paper asks where gender-occupation stereotypes in LLMs come from, in particular whether imbalances in pre-training data carry over into what a model generates.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/anna-korhonen/page/2/