Seeing the Abstract: Translating the Abstract Language for Vision Language Models