Large Language Models are Versatile Decomposers: Decompose Evidence and Questions for Table-based Reasoning

Benchmark Model Rank Results
semantic-parsing-on-wikitablequestionsDater#9Accuracy (Test): 65.9Accuracy (Dev): 64.8
table-based-fact-verification-on-tabfactDater#1Test: 93.0
table-to-text-generation-on-fetaqaCodex–BLEU: 27.96ROUGE-1: 62ROUGE-2: 40ROUGE-L: 52
table-to-text-generation-on-fetaqaCodex w/ Dater–BLEU: 30.92ROUGE-1: 66ROUGE-2: 45ROUGE-L: 56