The document discusses three approaches for handling grammatical agreement in statistical machine translation (SMT): morphological generation, agreement constraints, and a class-based agreement model. Morphological generation predicts inflections for word stems to comply with agreement rules. Agreement constraints add constraints to an SCFG target grammar. A class-based model segments, tags with morphological classes, and scores based on class sequences to improve inflected translations. The approaches are evaluated on English-German and English-Arabic data, showing improvements over baselines.