Examples of success and failure pairs from Turk and StackOverflow. On pairs (a) and (b), our DeepSketch is robust to the issues existing in natural language descriptions. On pairs (c) and (d), our approach fails due to the unrealistic semantics of the desired regexes. GrammarSketch succeeds in solving some complex pairs in StackOverflow, including (e) and (f). However, (g) and (h) fail because of insufficient examples or overly concise descriptions.