Dialogue act (DA) tagging is the task of labelling each utterance with its communicative function (e.g. statement, question, backchannel, agreement). It is usually solved with supervised learning on manually labelled data such as the Switchboard Dialog Act Corpus.
In this lab you work with the Switchboard data: loading and preprocessing the corpus, then training two different DA classification models. You will use PyTorch (or similar) for the models, and standard metrics (accuracy, per-class precision/recall) to compare them. The skills here — handling dialogue corpora, dealing with class imbalance, and interpreting confusion matrices — transfer directly to intent classification and slot tagging in task-oriented systems.
The lab notebook is available in the course materials (Lab2). Run the data download and preprocessing steps, then complete the model training and evaluation sections.