Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2406.11830 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916289877901312 |
|---|---|
| author | Li, Belinda Z. Liu, Emmy Ross, Alexis Zeitoun, Abbas Neubig, Graham Andreas, Jacob |
| author_facet | Li, Belinda Z. Liu, Emmy Ross, Alexis Zeitoun, Abbas Neubig, Graham Andreas, Jacob |
| contents | When the world changes, so does the text that humans write about it. How do we build language models that can be easily updated to reflect these changes? One popular approach is retrieval-augmented generation, in which new documents are inserted into a knowledge base and retrieved during prediction for downstream tasks. Most prior work on these systems have focused on improving behavior during prediction through better retrieval or reasoning. This paper introduces ERASE, which instead improves model behavior when new documents are acquired, by incrementally deleting or rewriting other entries in the knowledge base each time a document is added. In two new benchmark datasets evaluating models' ability to answer questions about a stream of news articles or conversations, ERASE improves accuracy relative to conventional retrieval-augmented generation by 7-13% (Mixtral-8x7B) and 6-10% (Llama-3-8B) absolute. Code and data are available at https://github.com/belindal/ERASE |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_11830 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Language Modeling with Editable External Knowledge Li, Belinda Z. Liu, Emmy Ross, Alexis Zeitoun, Abbas Neubig, Graham Andreas, Jacob Computation and Language Artificial Intelligence When the world changes, so does the text that humans write about it. How do we build language models that can be easily updated to reflect these changes? One popular approach is retrieval-augmented generation, in which new documents are inserted into a knowledge base and retrieved during prediction for downstream tasks. Most prior work on these systems have focused on improving behavior during prediction through better retrieval or reasoning. This paper introduces ERASE, which instead improves model behavior when new documents are acquired, by incrementally deleting or rewriting other entries in the knowledge base each time a document is added. In two new benchmark datasets evaluating models' ability to answer questions about a stream of news articles or conversations, ERASE improves accuracy relative to conventional retrieval-augmented generation by 7-13% (Mixtral-8x7B) and 6-10% (Llama-3-8B) absolute. Code and data are available at https://github.com/belindal/ERASE |
| title | Language Modeling with Editable External Knowledge |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2406.11830 |