Leveraging Knowledge Graphs for Cross-Lingual Semantic Parsing: A Curriculum-Based Fine-Tuning Approach

Abstract

Semantic parsing involves converting natural language into a formal representation of its meaning, which is crucial for various applications such as question answering and information extraction. Cross-lingual semantic parsing aims to perform these tasks across multiple languages. This paper proposes a curriculum-based fine-tuning approach that leverages knowledge graphs to improve cross-lingual semantic parsing. By structuring training data in a curriculum fashion, the approach enhances the model's ability to generalize across languages and domains. The paper details the methodology, presents experimental results, and discusses the implications for multilingual NLP applications.