On-The-Fly Information Retrieval Augmentation for Language Models
This addresses the challenge of enhancing language model capabilities efficiently for NLP applications, though it appears incremental as it builds on existing retrieval-augmented methods.
The authors tackled the problem of improving language model performance without retraining by augmenting GPT 2.0 with information retrieval, achieving a 15% relative reduction in perplexity on the Gigaword corpus in a zero-shot setting and validating it on an event co-reference task.
Here we experiment with the use of information retrieval as an augmentation for pre-trained language models. The text corpus used in information retrieval can be viewed as form of episodic memory which grows over time. By augmenting GPT 2.0 with information retrieval we achieve a zero shot 15% relative reduction in perplexity on Gigaword corpus without any re-training. We also validate our IR augmentation on an event co-reference task.