Advances in monolingual and crosslingual automatic disability annotation in Spanish

Area of Science:

• Natural Language Processing (NLP)
• Medical Informatics
• Computational Linguistics

Background:

• Automatic recognition of disabilities lags behind disease recognition in medical NLP.
• Lack of annotated corpora hinders progress in disability annotation.
• Neural networks can translate between representation types for sequence tagging.

Purpose of the Study:

• To present advances in monolingual (Spanish) and crosslingual (English-Spanish) automatic disability annotation.
• To identify disability mentions in Spanish biomedical abstracts.
• To improve the accuracy and coverage of disability annotation systems.

Main Methods:

• Combined deep learning models with varying embedding granularities for sequence-to-sequence tagging.
• Integrated an acronym and abbreviation detection module to enhance coverage.
• Utilized neural architectures for automatic annotation tasks.

Main Results:

• Monolingual experiments combining different word embedding representations outperformed single representations.
• Achieved state-of-the-art results in Spanish disability annotation.
• Crosslingual transfer (zero-shot) between English and Spanish showed promising results.

Conclusions:

• Combining diverse word embeddings significantly improves Spanish disability annotation accuracy.
• Crosslingual transfer learning offers a viable solution to data scarcity in disability annotation.
• The developed methods advance the field of medical NLP for disability recognition.