← Search

Josef Kucha{\v{r}}

2 accepted papers

2025

Can Out-of-Distribution Evaluations Uncover Reliance on Prediction Shortcuts? A Case Study in Question Answering

EMNLP 2025

A large body of recent work assesses models’ generalization capabilities through the lens of performance on out-of-distribution (OOD) datasets. Despite their practicality, such evaluations build upon a strong assumption: that OOD evaluations can capture and reflect upon possible failures in a real-w

2025

Pre-trained Language Models Learn Remarkably Accurate Representations of Numbers

EMNLP 2025

Pretrained language models (LMs) are prone to arithmetic errors. Existing work showed limited success in probing numeric values from models’ representations, indicating that these errors can be attributed to the inherent unreliability of distributionally learned embeddings in representing exact quan