Text this: Assessing interpretability in machine translation models for low-resource languages