People primarily consult tables to conduct data analysis or answer specificquestions. Text generation systems that can provide accurate table summariestailored to users' information needs can facilitate more efficient access torelevant data insights. Motivated by this, we define a new query-focused tablesummarization task, where text generation models have to perform human-likereasoning and analysis over the given table to generate a tailored summary. Weintroduce a new benchmark named QTSumm for this task, which contains 7,111human-annotated query-summary pairs over 2,934 tables covering diverse topics.We investigate a set of strong baselines on QTSumm, including text generation,table-to-text generation, and large language models. Experimental results andmanual analysis reveal that the new task presents significant challenges intable-to-text generation for future research. Moreover, we propose a newapproach named ReFactor, to retrieve and reason over query-relevant informationfrom tabular data to generate several natural language facts. Experimentalresults demonstrate that ReFactor can bring improvements to baselines byconcatenating the generated facts to the model input. Our data and code arepublicly available at https://github.com/yale-nlp/QTSumm.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)