We introduce POLAR - a framework that adds interpretability to pre-trained\nword embeddings via the adoption of semantic differentials. Semantic\ndifferentials are a psychometric construct for measuring the semantics of a\nword by analysing its position on a scale between two polar opposites (e.g.,\ncold -- hot, soft -- hard). The core idea of our approach is to transform\nexisting, pre-trained word embeddings via semantic differentials to a new\n"polar" space with interpretable dimensions defined by such polar opposites.\nOur framework also allows for selecting the most discriminative dimensions from\na set of polar dimensions provided by an oracle, i.e., an external source. We\ndemonstrate the effectiveness of our framework by deploying it to various\ndownstream tasks, in which our interpretable word embeddings achieve a\nperformance that is comparable to the original word embeddings. We also show\nthat the interpretable dimensions selected by our framework align with human\njudgement. Together, these results demonstrate that interpretability can be\nadded to word embeddings without compromising performance. Our work is relevant\nfor researchers and engineers interested in interpreting pre-trained word\nembeddings.\n