Towards Robustness of Text-to-Visualization Translation against Lexical and Phrasal Variability