Fine-tuned KB BERT for identifying compounded introductions in the Riksdagen corpus