Abstract
Word analogies is a relationship of the form "x is to y as w is to z", it can hold over an arbitrary number of ordered pairs. Automatically extracting analogies from text is a task that is not well explored for Arabic. Thus, in this paper we investigate the use of Zero-Shot Learning with word embeddings as an automatic approach for Arabic analogies extraction. We applied this approach on three corpora: King Saud University Corpus of Classical Arabic (KSUCCA), Gigaword, and Wikipedia embeddings. The best obtained precision result was 0.667 with Gigaword corpus. We further analyzed the extracted analogies relations manually to understand their types.