Research map / Graph learning / Knowledge and temporal graphs
Enhancing Machine Translation Experiences with Multilingual Knowledge Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents Wiki-MT, a demo that retrieves entities from Wikidata with a multilingual Contriever and appends their names and translations to queries fed into a fine-tuned NLLB-200 translator, improving entity-name translation and interpretability.
Problem
Machine translation often fails to correctly translate entity names, which may require transcreation rather than literal translation, and augmenting training data to memorize entity names is costly and quickly outdated.
Approach
Instead of memorizing entity translations in model weights, retrieve relevant entities from Wikidata via a multilingual dense retriever and append their source-language names and target-language aliases to the input query, so a fine-tuned MT model can use human-curated translations; the retrieved knowledge is also shown to users for interpretability.
Key results
- No quantitative results are reported; the paper presents a demo with a qualitative UI illustration of entity translation (e.g., The Catcher in the Rye to Il Giovane Holden).
Limitations
- No quantitative evaluation of translation quality is reported.
- Requires fine-tuning the MT model with new special tokens on a dataset.
- Supports only 11 languages in the demonstrated system.
Experimental setup
Datasets: Wikidata dump (May 2023, 11 languages) Mintaka
Baselines: Wiki-MT without knowledge retriever
Builds on: NLLB-200 Contriever Wikidata Mintaka
Closest papers by meaning
- DelTA: An Online Document-Level Translation Agent Based on Multi-Level Memory (ICLR 2025 · similarity 0.85)
- A Systematic Exploration of Knowledge Graph Alignment with Large Language Models in Retrieval Augmented Generation (AAAI 2025 · similarity 0.85)
- Translate Meanings, Not Just Words: IdiomKB’s Role in Optimizing Idiomatic Translation with Language Models (AAAI 2024 · similarity 0.85)
- MKGL: Mastery of a Three-Word Language (NeurIPS 2024 · similarity 0.85)
- K-ON: Stacking Knowledge on the Head Layer of Large Language Model (AAAI 2025 · similarity 0.84)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.