Engineering the Law-Machine Learning Translation Problem: Developing Legally Aligned Models