Papers by Ava Spataru
Know When To Stop: A Study of Semantic Drift in Text Generation (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for text generation make predictions token-by-token, without pre-established text structure. |
| Approach: | They develop a semantic drift score that measures the degree of separation between correct and incorrect facts in generated texts. |
| Outcome: | The proposed methods improve factuality by a large margin compared to baseline and early stopping methods. |