Joint dialogue act segmentation and recognition

Acknowledgement

• Most of this code is based on Tianyu Zhao's dialog processing repo; I copied relevant parts from src/tasks/joint_da_seg_recog
• Some SWDA preprocessing code is based on Hao Cheng's dynamic speaker repo
• Learning/optimizer common values, general tips, and good coding practices (though can't say I've successfully adopted them all) are based on Nikita Kitaev's self-attentive parser repo, as well as Nils Reimers' sentence transformers repo. These are beautiful codebases and I learned so much from just studying the code.

Steps

• Preprocessing scripts ...
• RNN models
• Transformer models

TODO items/issues

Priority:

• Implement sequence tagging model
• Change LR scheduling schemes
• Add ASR experiments

Longer-term:

• Fix run_test and overhaul config ... to not have to pass all the params again during testing
• Consolidate all models into one module; though needs some more thought as the tokenization is different