Token and Type Constraints for Cross-Lingual Part-of-Speech Tagging