Hypernymy plays a fundamental role in human cognition and various applications of natural language understanding. Supervised methods usually outperform unsupervised ones on the task of automatic hypernymy detection, but they require a sufficiently large number of annotated hypernymies as training examples, which are unavailable in low-resource languages. To deal with this problem, this paper proposes a neural multi-learning method to detect hypernymies in a low-resource target language. It works in a cross-lingually supervised manner, where the supervision information of annotated hypernymies comes from another resource-poor source language. By using English as the source language and Chinese as the target, experimental evaluation has shown that our method is more accurate than several state-of-the-art supervised and unsupervised methods on the task of hypernymy detection.