Multilingual Pretraining and Instruction Tuning Improve Cross-Lingual Knowledge Alignment, But Only Shallowly
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914743435919360 |
|---|---|
| author | Gao, Changjiang Hu, Hongda Hu, Peng Chen, Jiajun Li, Jixing Huang, Shujian |
| author_facet | Gao, Changjiang Hu, Hongda Hu, Peng Chen, Jiajun Li, Jixing Huang, Shujian |
| contents | Despite their strong ability to retrieve knowledge in English, current large language models show imbalance abilities in different languages. Two approaches are proposed to address this, i.e., multilingual pretraining and multilingual instruction tuning. However, whether and how do such methods contribute to the cross-lingual knowledge alignment inside the models is unknown. In this paper, we propose CLiKA, a systematic framework to assess the cross-lingual knowledge alignment of LLMs in the Performance, Consistency and Conductivity levels, and explored the effect of multilingual pretraining and instruction tuning on the degree of alignment. Results show that: while both multilingual pretraining and instruction tuning are beneficial for cross-lingual knowledge alignment, the training strategy needs to be carefully designed. Namely, continued pretraining improves the alignment of the target language at the cost of other languages, while mixed pretraining affect other languages less. Also, the overall cross-lingual knowledge alignment, especially in the conductivity level, is unsatisfactory for all tested LLMs, and neither multilingual pretraining nor instruction tuning can substantially improve the cross-lingual knowledge conductivity. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_04659 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Multilingual Pretraining and Instruction Tuning Improve Cross-Lingual Knowledge Alignment, But Only Shallowly Gao, Changjiang Hu, Hongda Hu, Peng Chen, Jiajun Li, Jixing Huang, Shujian Computation and Language Despite their strong ability to retrieve knowledge in English, current large language models show imbalance abilities in different languages. Two approaches are proposed to address this, i.e., multilingual pretraining and multilingual instruction tuning. However, whether and how do such methods contribute to the cross-lingual knowledge alignment inside the models is unknown. In this paper, we propose CLiKA, a systematic framework to assess the cross-lingual knowledge alignment of LLMs in the Performance, Consistency and Conductivity levels, and explored the effect of multilingual pretraining and instruction tuning on the degree of alignment. Results show that: while both multilingual pretraining and instruction tuning are beneficial for cross-lingual knowledge alignment, the training strategy needs to be carefully designed. Namely, continued pretraining improves the alignment of the target language at the cost of other languages, while mixed pretraining affect other languages less. Also, the overall cross-lingual knowledge alignment, especially in the conductivity level, is unsatisfactory for all tested LLMs, and neither multilingual pretraining nor instruction tuning can substantially improve the cross-lingual knowledge conductivity. |
| title | Multilingual Pretraining and Instruction Tuning Improve Cross-Lingual Knowledge Alignment, But Only Shallowly |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2404.04659 |