Can multilinguality improve Biomedical Word Sense Disambiguation?

Area of Science:

• Biomedical Natural Language Processing (NLP)
• Computational Linguistics

Background:

• Ambiguity in biomedical text hinders accurate NLP tasks.
• Word Sense Disambiguation (WSD) is crucial for reliable information extraction, summarization, and classification.
• Existing monolingual approaches face limitations in addressing complex biomedical terminology.

Purpose of the Study:

• To investigate the effectiveness of multilingual resources in resolving biomedical word sense ambiguity.
• To identify conditions for improving WSD systems using multilinguality.
• To explore optimal methods for generating and utilizing multilingual resources.

Main Methods:

• Development of a WSD system based on co-occurrence graphs.
• Incorporation of biomedical concepts and textual information into graphs.
• Evaluation using a standard biomedical test dataset.
• Analysis of different languages and knowledge sources for resource generation.

Main Results:

• Multilingual resources demonstrated a significant improvement of over 7% compared to monolingual approaches.
• Enhanced performance was observed particularly with graphs constructed from a limited number of documents.
• Automatically translated resources proved valuable for biomedical WSD.

Conclusions:

• Multilingual strategies offer a tangible advantage for biomedical NLP tasks.
• The use of translated resources is a viable and effective method for enhancing WSD.
• Further research into multilingual resource generation can lead to more robust biomedical NLP systems.