Large Language Models are Limited in Out-of-Context Knowledge Reasoning

TL;DR

This paper systematically evaluates the out-of-context knowledge reasoning abilities of large language models using a synthetic dataset, revealing their limitations in combining multiple knowledge sources and transferring knowledge across languages.

Contribution

The paper introduces a synthetic dataset with seven OCKR tasks to assess LLMs' out-of-context reasoning, highlighting their limited capabilities and challenges in knowledge retrieval and cross-lingual transfer.

Findings

LLMs show limited out-of-context knowledge reasoning ability.

Training with reasoning examples does not significantly improve OCKR.

Explicit knowledge retrieval training aids attribute knowledge retrieval but not relation knowledge.

Abstract

Large Language Models (LLMs) possess extensive knowledge and strong capabilities in performing in-context reasoning. However, previous work challenges their out-of-context reasoning ability, i.e., the ability to infer information from their training data, instead of from the context or prompt. This paper focuses on a significant aspect of out-of-context reasoning: Out-of-Context Knowledge Reasoning (OCKR), which is to combine multiple knowledge to infer new knowledge. We designed a synthetic dataset with seven representative OCKR tasks to systematically assess the OCKR capabilities of LLMs. Using this dataset, we evaluated several LLMs and discovered that their proficiency in this aspect is limited, regardless of whether the knowledge is trained in a separate or adjacent training settings. Moreover, training the model to reason with reasoning examples does not result in significant improvement, while training the model to perform explicit knowledge retrieval helps for retrieving attribute knowledge but not the relation knowledge, indicating that the model's limited OCKR capabilities are due to difficulties in knowledge retrieval. Furthermore, we treat cross-lingual knowledge transfer as a distinct form of OCKR, and evaluate this ability. Our results show that the evaluated model also exhibits limited ability in transferring knowledge across languages.