Shedding Light on Linguistic Shadows: A Cross-Lingual experiment on the Transparency of Nominal Compounds in Large Language Models