A timeline summarization (TLS) system extracts the key dates of some long lasting event, like a disaster or a political crisis, from a collection of news articles and outputs concise summaries of these dates. To the best of our knowledge, all previous work on TLS has focused on extractive summarization. We explore the feasibility of abstractive TLS using methods adapted from abstractive multi-document summarization. We show that our system performs on par with a state-of-the-art extractive system for TLS and can even outperform it in some cases. We also present an oracle summarization experiment that suggests that our abstractive approach has a higher upper bound on its performance than a purely extractive system when used on very large corpora that require strong compression. In this talk I will present this system and share some preliminary ideas on designing a protocol for manual evaluation of timeline summaries.