CALM: Unleashing the Cross-Lingual Self-Aligning Ability of Language Model Question Answering