Academic Papers

Retrieval-augmented Multilingual Knowledge Editing

Large language models (LLMs) are becoming ever more pervasive, but there’s a catch: they often struggle to keep their knowledge up-to-date, especially when it comes to different languages. Imagine having a helpful resource that can answer your questions, but only in English!


To update knowledge in LLMs, methods for knowledge editing (KE) have been developed. But current KE research focuses on editing knowledge in one language, typically English whereas there’s a need for editing knowledge in multiple languages for companies serving a multilingual customer base.


​To address this, the researchers  propose Retrieval-augmented Multilingual Knowledge Editor (ReMaKE), which combines multilingual retrieval from a knowledge base with in-context learning. This allows for editing knowledge in one language and querying it in multiple languages.


​ReMaKE consists of two stages: multilingual knowledge retrieval and multilingual in-context editing. The retrieval stage finds the most relevant fact in the multilingual knowledge base for a given query, while the editing stage combines the retrieved knowledge with the user’s input to generate accurate responses. The knowledge base can be updated in any language; its contents can be accessed in all languages.


Experimental results show that ReMaKE outperforms other knowledge editing methods in a multilingual setting.  This means that ReMaKE can be used to update knowledge in LLMs in multiple languages, making it a valuable tool for companies with a diverse customer base.


The research team, including Dr Barry Haddow, Aveni’s Head of Natural Language Processing and Dr Alexandra Birch, Head of Aveni Labs, created a dataset in 12 languages to help others develop even better ways to update knowledge in these language models.


In conclusion, “Retrieval-augmented Multilingual Knowledge Editing” presents ReMaKE as a groundbreaking solution for multilingual knowledge editing in LLMs. By leveraging retrieval-augmented techniques, ReMaKE offers a more efficient and scalable approach to updating knowledge across diverse languages. The experimental results highlight the effectiveness of ReMaKE in improving the performance of LLMs in multilingual settings, paving the way for enhanced knowledge editing capabilities in a global context.


Key takeaways from the paper: 


  • ReMaKE introduces a novel approach to multilingual knowledge editing in Large Language Models (LLMs).


  • It leverages retrieval-augmented techniques to update knowledge in one language and propagate it to answers in multiple languages.


  • Experimental results demonstrate that ReMaKE outperforms traditional methods in multilingual settings.


  • The MzsRE dataset, provided in 12 languages, facilitates further research in multilingual knowledge editing.


  • ReMaKE offers a scalable and efficient solution for updating knowledge across diverse languages in LLMs.


Download the research paper

Other resources

Vulnerable Customers

Transformational efficiency direct from the customer voice

Adoption of Machine Learning in human-centred advice

Consumer Duty: Board Reports Generating the right information for the right reports

Understanding Aveni’s customer engagement process

Assessing the Reliability of Large Language Model Knowledge

Hallucinations in Large Multilingual Translation Models

Age Partnership: Driving quality assurance efficiency

Improve Quality Assurance processes

Aveni Assist explainer video

Aveni’s platform uses the latest in NLP to transform productivity and risk oversight.

Scale compliance at a fraction of the cost

Cut financial advice admin from hours to minutes with Aveni’s AI assisitant

Aveni Assist

Get up and running with Aveni Assist and how it can help transform productivity and compliance. 

Aveni Detect

Get up and running with Aveni Detect and how it can help transform productivity and compliance. 

Read the latest articles from Aveni

Access our latest whitepapers, webinars, brochures and more

Jargon-bust your way to a better understanding of all things AI