Sequence Shortening for Context-Aware Machine Translation