Translating Literature with Large Language Models: Integrating External Knowledge and Maintaining Consistency in Named Entities
- Presenter
- Jiarui Liu
- Campus
- UMass Amherst
- Sponsor
- Mohit Iyyer, Department of Computer Science, UMass Amherst
- Schedule
- Session 2, 11:30 AM - 12:15 PM [Schedule by Time][Poster Grid for Time/Location]
- Location
- Poster Board A34, Campus Center Auditorium, Row 2 (A21-A40) [Poster Location Map]
- Abstract
As the world grows increasingly interconnected, the demand for accessible long-form documents across languages escalates. This research delves into the capabilities of Large Language Models (LLMs) in translating literature works, with a focus on enhancing consistency in named entity translations and integrating external knowledge sources.
Central to literary translations is the accurate and consistent translation of named entities such as main character names and significant proper nouns. Recent Language Models, often restricted by their context window or budget, face challenges in maintaining this consistency in long-form text translations. This study proposes a novel approach that merges the sophisticated capabilities of LLMs with traditional human translation practices such as Termbase, translation memory, and leveraging external knowledge bases like Wikipedia & linguistic corpus.
The methodology involves developing an auxiliary retrieval system and integrating translation models with retrieval-augmented generation (RAG) and in-context learning (ICL) approaches. The study aims to ensure the coherent translation of named entities and to enhance the accuracy of these translations in contexts.
The preliminary results have shown notable improvements in both the consistency and accuracy of named entity translations. This study increases the accessibility of literary works across languages, fostering a more inclusive global cultural exchange. It also encourages further exploration into the ways Large Language Models can be utilized and improved for complex tasks, bridging the gap between artificial intelligence and the nuanced art of human work.
- Keywords
- Natural language processing, Retrieval-Augmented Generation, Machine translation, Named Entity Translation
- Research Area
- Computer Science
SIMILAR ABSTRACTS (BY KEYWORD)
Research Area |
Presenter |
Title |
Keywords |
Computer Science |
Jung, Hayun |
|
Natural Language Processing
|
Computer Science |
Kakkar, Ishita |
|
Natural Language Processing
|
Computer Science |
Petlin, Algis O. |
|
Natural Language Processing (NLP)
|