Can LMs Learn New Entities from Descriptions? Challenges in Propagating Injected Knowledge

TL;DR

This paper investigates whether pre-trained language models can effectively propagate injected knowledge about entities through inference, revealing limitations of current update methods and potential improvements via context prepending.

Contribution

The study introduces a new benchmark for testing knowledge propagation in LMs and compares different injection methods, highlighting the effectiveness of context prepending over traditional fine-tuning.

Findings

Existing update methods show limited knowledge propagation.

Performance improves with lexical overlap between facts and inferences.

Context prepending significantly enhances knowledge inference across settings.

Abstract

Pre-trained language models (LMs) are used for knowledge intensive tasks like question answering, but their knowledge gets continuously outdated as the world changes. Prior work has studied targeted updates to LMs, injecting individual facts and evaluating whether the model learns these facts while not changing predictions on other contexts. We take a step forward and study LMs' abilities to make inferences based on injected facts (or propagate those facts): for example, after learning that something is a TV show, does an LM predict that you can watch it? We study this with two cloze-style tasks: an existing dataset of real-world sentences about novel entities (ECBD) as well as a new controlled benchmark with manually designed templates requiring varying levels of inference about injected knowledge. Surprisingly, we find that existing methods for updating knowledge (gradient-based fine-tuning and modifications of this approach) show little propagation of injected knowledge. These methods improve performance on cloze instances only when there is lexical overlap between injected facts and target inferences. Yet, prepending entity definitions in an LM's context improves performance across all settings, suggesting that there is substantial headroom for parameter-updating approaches for knowledge injection.