Self-Adaptive Paraphrasing and Preference Learning for Improved Claim Verifiability