Despite large language models (LLMs) being known to exhibit bias against non-mainstream varieties, there are no known labeled datasets for sentiment analysis of English. To address this gap, we introduce BESSTIE, a benchmark for sentiment and sarcasm …