Recent work has shown how to learn bettervisual-semantic embeddings by leveraging imagedescriptions in more than one language.Here, we investigate in detail which conditionsaffect the performance of this type of grounded language learning model. We show that multilingual training improves over bilingual training, and that low-resource languages benefit from training with higher-resource languages. We demonstrate that a multilingual model can be trained equally well on either translations or comparable sentence pairs, and that annotating the same set of images in multiple language enables further improvements via an additional caption-caption ranking objective