Prompting Vision-Language Models For Aspect-Controlled Generation of Referring Expressions F i n d i n g s - N A A C L