README.md

The MAD-X² setup consists of 3 adapters:

1. MAD-X² language adapters
2. MAD-X² domain adapters
3. QA head adapter

First, activate the conda environment: conda activate m2qa_experiments
Then train the adapters by starting the bash scripts:

1. Domain & Language Adapter:
   • Part of the german creative writing is the Corpus of German-Language Fiction (Fischer et al. 2015), load it from here: https://figshare.com/articles/dataset/Corpus_of_German-Language_Fiction_txt_/4524680 and put it in the directory Corpora/corpus-of-german-fiction-txt/
   • Execute the train_domain_language_adapter.py through the start_training.sh bash file
2. The QA adapter head is trained by executing QA_Head/start_train_squad_head_adapter.sh
   • Note: In this train_squad_head.py script, you must edit the paths so that they point to the previously trained English language and Wikipedia domain adapters.