Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignment
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866929539505979392 |
|---|---|
| author | Wu, Zhaofeng Balashankar, Ananth Kim, Yoon Eisenstein, Jacob Beirami, Ahmad |
| author_facet | Wu, Zhaofeng Balashankar, Ananth Kim, Yoon Eisenstein, Jacob Beirami, Ahmad |
| contents | Aligning language models (LMs) based on human-annotated preference data is a crucial step in obtaining practical and performant LM-based systems. However, multilingual human preference data are difficult to obtain at scale, making it challenging to extend this framework to diverse languages. In this work, we evaluate a simple approach for zero-shot cross-lingual alignment, where a reward model is trained on preference data in one source language and directly applied to other target languages. On summarization and open-ended dialog generation, we show that this method is consistently successful under comprehensive evaluation settings, including human evaluation: cross-lingually aligned models are preferred by humans over unaligned models on up to >70% of evaluation instances. We moreover find that a different-language reward model sometimes yields better aligned models than a same-language reward model. We also identify best practices when there is no language-specific data for even supervised finetuning, another component in alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_12318 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignment Wu, Zhaofeng Balashankar, Ananth Kim, Yoon Eisenstein, Jacob Beirami, Ahmad Computation and Language Aligning language models (LMs) based on human-annotated preference data is a crucial step in obtaining practical and performant LM-based systems. However, multilingual human preference data are difficult to obtain at scale, making it challenging to extend this framework to diverse languages. In this work, we evaluate a simple approach for zero-shot cross-lingual alignment, where a reward model is trained on preference data in one source language and directly applied to other target languages. On summarization and open-ended dialog generation, we show that this method is consistently successful under comprehensive evaluation settings, including human evaluation: cross-lingually aligned models are preferred by humans over unaligned models on up to >70% of evaluation instances. We moreover find that a different-language reward model sometimes yields better aligned models than a same-language reward model. We also identify best practices when there is no language-specific data for even supervised finetuning, another component in alignment. |
| title | Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignment |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2404.12318 |