I would like to ask for clarification regarding the distinction between statistical models (e.g., HMMs, CRFs) and neural models (e.g., RNNs, Transformers) as covered in our...
I would like to ask for clarification regarding the distinction between statistical models (e.g., HMMs, CRFs) and neural models (e.g., RNNs, Transformers) as covered in our recent lecture on sequence modeling in NLP. While I understand that both approaches aim to handle tasks like part-of-speech tagging and named entity recognition, I am still unclear about the key differences in terms of architecture, data requirements, and interpretability.
Could someone elaborate on when and why we might still use statistical models in contemporary applications, especially given the dominance of deep learning methods? Any recommended readings or examples that clearly contrast these approaches would be greatly appreciated.