Fine-grained Controllable Text Generation through In-context Learning with Feedback
This addresses the challenge of fine-grained text control for NLP applications in data-sparse scenarios, though it appears incremental as it builds on existing in-context learning methods.
The paper tackles the problem of rewriting sentences to match specific linguistic features like dependency depth using in-context learning instead of finetuning, achieving accurate rewrites and matching state-of-the-art performance on school grade level rewriting.
We present a method for rewriting an input sentence to match specific values of nontrivial linguistic features, such as dependency depth. In contrast to earlier work, our method uses in-context learning rather than finetuning, making it applicable in use cases where data is sparse. We show that our model performs accurate rewrites and matches the state of the art on rewriting sentences to a specified school grade level.