Automatic sentence summarization produces a shorter version of a sentence,\nwhile preserving its most important information. A good summary is\ncharacterized by language fluency and high information overlap with the source\nsentence. We model these two aspects in an unsupervised objective function,\nconsisting of language modeling and semantic similarity metrics. We search for\na high-scoring summary by discrete optimization. Our proposed method achieves a\nnew state-of-the art for unsupervised sentence summarization according to ROUGE\nscores. Additionally, we demonstrate that the commonly reported ROUGE F1 metric\nis sensitive to summary length. Since this is unwillingly exploited in recent\nwork, we emphasize that future evaluation should explicitly group summarization\nsystems by output length brackets.\n