Double Trouble: Bilingual Pretraining Leaves Language-Conditioned Effects in Shared-Language Representations
Anjishnu Mukherjee, Ziwei Zhu, Antonios Anastasopoulos. EMNLP 2026 (Main).
Adding a second language to pretraining changes how a model represents English concepts, even after we align the English word embeddings of the two models. In all 32 controlled comparisons, bilingual and English-only models disagreed on held-out English concepts more than English-only models trained with different random seeds.
- Project page: https://iamshnoo.github.io/double_trouble/
- Paper: https://aclanthology.org/2026.emnlp-main.2062/
- arXiv: https://arxiv.org/abs/2608.26576
- Code: https://github.com/iamshnoo/bli
This repository holds the project page only. It is generated from iamshnoo.github.io by scripts/build-paper-sites.mjs. You can also edit index.html here directly: once the file is changed by hand, the generator leaves it as you wrote it.
@inproceedings{mukherjee-etal-2026-double,
title = "Double Trouble: Bilingual Pretraining Leaves Language-Conditioned Effects in Shared-Language Representations",
author = "Mukherjee, Anjishnu and
Zhu, Ziwei and
Anastasopoulos, Antonios",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Junyi Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2062/",
doi = "10.18653/2026.emnlp-main.2062",
pages = "46953--46977",
ISBN = "979-8-89176-459-0"
}