keyword

multi-hop questions

Multi-hop questions are queries that require retrieving, integrating, and reasoning across multiple discrete pieces of information or intermediary facts to determine the final answer, rather than relying on a single directly stated fact. In computational linguistics and automated question answering, resolving a multi-hop question involves traversing a sequential chain of logical steps, where the resolution of an initial relationship or sub-query provides the necessary entity or premise to solve the next. This structure distinguishes multi-hop reasoning from single-fact retrieval, serving as a standard framework for evaluating whether an information system or language model can perform compositional reasoning and synthesize connected knowledge.

1 item

MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions

Zexuan Zhong, Zhengxuan Wu, Christopher D. Manning, Christopher Potts, Danqi Chen

OrganizationsPrinceton UniversityStanford University

Why you should read this

Introduces the MQuAKE benchmark to evaluate whether knowledge-edited language models can propagate updates through multi-hop reasoning, revealing that existing weight-editing methods fail catastrophically and proposing a memory-based prompting alternative, MeLLo, that outperforms them by a large margin.

The information stored in large language models (LLMs) falls out of date quickly, and retraining from scratch is often not an option. This has recently given rise to a range of techniques for injecting new facts through updating model weights. Current evaluation paradigms are extremely limited, mainly validating the recall of edited facts, but changing one fact should cause rippling changes to the model's related beliefs. If we edit the UK Prime Minister to now be Rishi Sunak, then we should get a different answer to Who is married to the British Prime Minister? In this work, we present a benchmark, MQuAKE (Multi-hop Question Answering for Knowledge Editing), comprising multi-hop questions that assess whether edited models correctly answer questions where the answer should change as an entailed consequence of edited facts. While we find that current knowledge-editing approaches can recall edited facts accurately, they fail catastrophically on the constructed multi-hop questions. We thus propose a simple memory-based approach, MeLLo, which stores all edited facts externally while prompting the language model iteratively to generate answers that are consistent with the edited facts. While MQuAKE remains challenging, we show that MeLLo scales well with LLMs (up to 175B) and outperforms previous model editors by a large margin.¹

Source

https://aclanthology.org/2023.emnlp-main.971.pdf

Added

2026-09-26