CorpusBrain: Pre-train a Generative Retrieval Model for Knowledge-Intensive Language Tasks

TL;DR

CorpusBrain introduces a pre-trained generative retrieval model that simplifies knowledge-intensive task retrieval, replacing traditional pipelines with end-to-end training, leading to state-of-the-art results on KILT benchmarks.

Contribution

The paper presents a novel single-step generative retrieval model, CorpusBrain, pre-trained with specialized tasks, enabling end-to-end optimization and improved performance on knowledge-intensive tasks.

Findings

Outperforms strong baselines on KILT benchmark

Effective in zero- and low-resource settings

Encodes entire corpus information in model parameters

Abstract

Knowledge-intensive language tasks (KILT) usually require a large body of information to provide correct answers. A popular paradigm to solve this problem is to combine a search system with a machine reader, where the former retrieves supporting evidences and the latter examines them to produce answers. Recently, the reader component has witnessed significant advances with the help of large-scale pre-trained generative models. Meanwhile most existing solutions in the search component rely on the traditional ``index-retrieve-then-rank'' pipeline, which suffers from large memory footprint and difficulty in end-to-end optimization. Inspired by recent efforts in constructing model-based IR models, we propose to replace the traditional multi-step search pipeline with a novel single-step generative model, which can dramatically simplify the search process and be optimized in an end-to-end manner. We show that a strong generative retrieval model can be learned with a set of adequately designed pre-training tasks, and be adopted to improve a variety of downstream KILT tasks with further fine-tuning. We name the pre-trained generative retrieval model as CorpusBrain as all information about the corpus is encoded in its parameters without the need of constructing additional index. Empirical results show that CorpusBrain can significantly outperform strong baselines for the retrieval task on the KILT benchmark and establish new state-of-the-art downstream performances. We also show that CorpusBrain works well under zero- and low-resource settings.