Modeling Relevance in Statistical Machine Translation: Scoring Alignment, Context, and Annotations of Translation Instances