Generative vs. Encoder Models for Multilingual NER: A Comprehensive Empirical Study on Naamapadam
The episode discusses a study comparing generative and encoder models for Named Entity Recognition (NER) across eleven languages using the Naamapadam benchmark. The hosts highlight practical recommendations for fixing model shortcomings, such as using entity-aware hybrid sampling and tailoring strategies based on language clusters to build equitable multilingual AI systems.