Citations as Queries: Source Attribution Using Language Models as Rerankers

TL;DR

This paper investigates using fine-tuned language models as rerankers to improve source attribution in texts, demonstrating that semi-supervised methods can match fully supervised performance without extensive annotations.

Contribution

It introduces a novel approach of using language models as rerankers for source attribution and compares supervised and semi-supervised methods across different datasets.

Findings

Semi-supervised reranking approaches perform nearly as well as fully supervised ones.

Language models effectively improve source attribution accuracy.

The method works across diverse languages and text types.

Abstract

This paper explores new methods for locating the sources used to write a text, by fine-tuning a variety of language models to rerank candidate sources. After retrieving candidates sources using a baseline BM25 retrieval model, a variety of reranking methods are tested to see how effective they are at the task of source attribution. We conduct experiments on two datasets, English Wikipedia and medieval Arabic historical writing, and employ a variety of retrieval and generation based reranking models. In particular, we seek to understand how the degree of supervision required affects the performance of various reranking models. We find that semisupervised methods can be nearly as effective as fully supervised methods while avoiding potentially costly span-level annotation of the target and source documents.