SandboxAQ's submission to MRL 2024 Shared Task on Multi-lingual Multi-task Information Retrieval

TL;DR

This paper evaluates the performance of five large language models on multilingual question answering and named entity recognition, highlighting the variability across languages and tasks, and emphasizing the importance of task-specific approaches.

Contribution

It provides a comprehensive comparison of prompting techniques across multiple languages and tasks, revealing the nuanced effectiveness of models and prompting strategies in multilingual NLP.

Findings

Advanced prompting improves QA performance

Model effectiveness varies by language and task

Language difficulty impacts model performance

Abstract

This paper explores the problems of Question Answering (QA) and Named Entity Recognition (NER) in five diverse languages. We tested five Large Language Models with various prompting methods, including zero-shot, chain-of-thought reasoning, and translation techniques. Our results show that while some models consistently outperform others, their effectiveness varies significantly across tasks and languages. We saw that advanced prompting techniques generally improved QA performance but had mixed results for NER; and we observed that language difficulty patterns differed between tasks. Our findings highlight the need for task-specific approaches in multilingual NLP and suggest that current models may develop different linguistic competencies for different tasks.