SynTQA: Synergistic Table-based Question Answering via Mixture of Text-to-SQL and E2E TQA

Text-to-SQL parsing and end-to-end question answering (E2E TQA) are two mainapproaches for Table-based Question Answering task. Despite success on multiplebenchmarks, they have yet to be compared and their synergy remains unexplored.In this paper, we identify different strengths and weaknesses throughevaluating state-of-the-art models on benchmark datasets: Text-to-SQLdemonstrates superiority in handling questions involving arithmetic operationsand long tables; E2E TQA excels in addressing ambiguous questions, non-standardtable schema, and complex table contents. To combine both strengths, we proposea Synergistic Table-based Question Answering approach that integrate differentmodels via answer selection, which is agnostic to any model types. Furtherexperiments validate that ensembling models by either feature-based orLLM-based answer selector significantly improves the performance overindividual models.