The abstract should summarize the contents of the paper in short terms, i.e. 150–250 words. Identifying semantic phrases in text has become a key task in various applications, including information extraction and taxonomy building. The state-of-the-art approaches primarily rely on simple frequency-based statistical features in text sequence modeling for phrasal segmentation and require semantic training data created by human experts for phrase quality estimation, which limit the performance improvement of phrase discovery from massive text corpora. In this paper, we introduce a new framework named PDKEL (Phrase Discovery with Knowledge-enhanced Embedding Learning) to incorporate rich se mantic knowledge from knowledge bases into distributional embedding-based approach for automatic phrase discovery. PDKEL integrates phrasal segmentation with phrase quality estimation by exploring the embedding-based distributional phrase features and leveraging rich semantic information from available knowledge bases to generate high quality phrases. The learning of phrase embeddings is carried out by integrating semantic constraints from knowledge bases into a corpus-based approach. This allows the model to capture rich contextual in formation and generate semantically meaningful phrase embeddings, thereby improving phrasal segmentation and phrase quality estimation. Experimental results confirm that the PDKEL method outperforms existing cutting-edge approaches.