Conventional approaches to relation extraction usually require a fixed set ofpre-defined relations. Such requirement is hard to meet in many realapplications, especially when new data and relations are emerging incessantlyand it is computationally expensive to store all data and re-train the wholemodel every time new data and relations come in. We formulate such achallenging problem as lifelong relation extraction and investigatememory-efficient incremental learning methods without catastrophicallyforgetting knowledge learned from previous tasks. We first investigate amodified version of the stochastic gradient methods with a replay memory, whichsurprisingly outperforms recent state-of-the-art lifelong learning methods. Wefurther propose to improve this approach to alleviate the forgetting problem byanchoring the sentence embedding space. Specifically, we utilize an explicitalignment model to mitigate the sentence embedding distortion of the learnedmodel when training on new data and new relations. Experiment results onmultiple benchmarks show that our proposed method significantly outperforms thestate-of-the-art lifelong learning approaches.