Experiences from Creating a Benchmark for Sentiment Classification for Varieties of English