Leveraging annotator rationales for more interpretable and sample efficient classification.| machine learning musings