Plausibility Vaccine: Injecting LLM Knowledge for Event Plausibility
This work addresses the challenge of discriminating plausible from implausible events in natural language processing, which is an incremental improvement for applications like text understanding and generation.
The paper tackles the problem of event plausibility prediction by injecting latent knowledge from large language models into pre-trained embeddings using parameter-efficient fine-tuning, resulting in improved performance on plausibility datasets.
Despite advances in language modelling, distributional methods that build semantic representations from co-occurrences fail to discriminate between plausible and implausible events. In this work, we investigate how plausibility prediction can be improved by injecting latent knowledge prompted from large language models using parameter-efficient fine-tuning. We train 12 task adapters to learn various physical properties and association measures and perform adapter fusion to compose latent semantic knowledge from each task on top of pre-trained AlBERT embeddings. We automate auxiliary task data generation, which enables us to scale our approach and fine-tune our learned representations across two plausibility datasets. Our code is available at https://github.com/Jacob-Chmura/plausibility-vaccine.