Speech Act Classifier

Speech act classifier based on speech-act-analysis dataset.

How to train a speech act classifier

Step 1: Download dataset and create dataset splits

File: src/create_splits.py

• Create empty data folder in src/
• Download the following script: https://github.com/MelinaPl/speech-act-analysis/blob/main/data/version_1-1.json

$ python create_splits.py coarse version_1-1.json
$ python create_splits.py fine version_1-1.json
$ python create_splits.py merged version_1-1.json

Step 2: Train classifier

File: src/train.py

• Create empty model directory in src/
• Choose from the following models: "dbmdz/bert-base-german-uncased", "dbmdz/bert-base-german-cased", "deepset/gbert-base", "deepset/gelectra-base"

$ python train.py dbmdz/bert-base-german-uncased  fine

Step 3: Evaluate

File: src/evaluate.py

$ python evaluate.py MODELNAME DATAVERSION PATH_TO_CHECKPOINT