TY - JOUR
AU - 
AB - Daniel Khashabi Amos Ng 1 1 1;2 3 Tushar Khot Ashish Sabharwal Hannaneh Hajishirzi Chris Callison-Burch Allen Institute for AI, Seattle, WA, USA University of Washington, Seattle, WA, USA University of Pennsylvania, Philadelphia, PA, USA Abstract bases (Roberts et al., 2020; Lewis et al., 2021). Ex- isting effort, however, involves isolated studies on While day-to-day questions come with a vari- niche answer types, mainly short responses and, in ety of answer types, the current open question- a few cases, long responses (Joshi et al., 2017; Lee answering (QA) literature represents isolated et al., 2019; Bhakthavatsalam et al., 2021). efforts on niche response types, with a heavy In contrast, many of the everyday questions that focus on specific kinds of short responses (people, places, etc.). To address this gap, humans deal with and pose to search engines have we present G OOAQ, a large-scale dataset col- a more diverse set of response types, as illustrated lected from Google questions and answers, in Fig. 1. Their answer can be a multi-sentence containing 3 million questions with diverse an- description (a snippet) (e.g., ‘what is’ or ‘can you’ swer types ranging from factual short answers questions), a collection of items such as 
TI - GooAQ: Open Question Answering with Diverse Answer Types
JF - Findings of the Association for Computational Linguistics: EMNLP 2021
DO - 10.18653/v1/2021.findings-emnlp.38
DA - 2021-01-01
UR - https://www.deepdyve.com/lp/unpaywall/gooaq-open-question-answering-with-diverse-answer-types-V0BN009jUt
DP - DeepDyve
ER -