How Far Can We Go with Data Selection? A Case Study on Semantic Sequence Tagging Tasks