LiFi: Lightweight Controlled Text Generation with Fine-Grained Control Codes
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917587111116800 |
|---|---|
| author | Shi, Chufan Cai, Deng Yang, Yujiu |
| author_facet | Shi, Chufan Cai, Deng Yang, Yujiu |
| contents | In the rapidly evolving field of text generation, the demand for more precise control mechanisms has become increasingly apparent. To address this need, we present a novel methodology, LIFI, which offers a lightweight approach with fine-grained control for controlled text generation. Unlike previous studies that train pre-trained language models to follow discrete, categorical, and exclusive control codes, LIFI learns controlled text generation under the guidance of continuous, relative, and nonexclusive control codes. These fine-grained codes are automatically derived from an attribute classifier, initially trained with a small amount of labeled data and subsequently employed to label abundant unlabeled data, thus garnering more extensive supervision signals. Moreover, to achieve efficient control, we incorporate the fine-grained control codes with adapters, a parameter- and compute-efficient way to steer a pre-trained language model. We evaluate LIFI on two conventional tasks -- sentiment control and topic control -- and one newly proposed task -- stylistic novel writing. Comprehensive experimental results validate the effectiveness of our proposed methods, demonstrating substantial performance improvements over existing baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_06930 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | LiFi: Lightweight Controlled Text Generation with Fine-Grained Control Codes Shi, Chufan Cai, Deng Yang, Yujiu Computation and Language In the rapidly evolving field of text generation, the demand for more precise control mechanisms has become increasingly apparent. To address this need, we present a novel methodology, LIFI, which offers a lightweight approach with fine-grained control for controlled text generation. Unlike previous studies that train pre-trained language models to follow discrete, categorical, and exclusive control codes, LIFI learns controlled text generation under the guidance of continuous, relative, and nonexclusive control codes. These fine-grained codes are automatically derived from an attribute classifier, initially trained with a small amount of labeled data and subsequently employed to label abundant unlabeled data, thus garnering more extensive supervision signals. Moreover, to achieve efficient control, we incorporate the fine-grained control codes with adapters, a parameter- and compute-efficient way to steer a pre-trained language model. We evaluate LIFI on two conventional tasks -- sentiment control and topic control -- and one newly proposed task -- stylistic novel writing. Comprehensive experimental results validate the effectiveness of our proposed methods, demonstrating substantial performance improvements over existing baselines. |
| title | LiFi: Lightweight Controlled Text Generation with Fine-Grained Control Codes |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2402.06930 |