Publication Details

R2-D2: A Modular Baseline for Open-Domain Question Answering

FAJČÍK Martin, DOČEKAL Martin, ONDŘEJ Karel and SMRŽ Pavel et al. R2-D2: A Modular Baseline for Open-Domain Question Answering. In: Findings of the Association for Computational Linguistics: EMNLP 2021. Findings of the Association for Computational Linguistics. Punta Cana: Association for Computational Linguistics, 2021, pp. 854-870. ISBN 978-1-955917-10-0. Available from: https://aclanthology.org/2021.findings-emnlp.73.pdf
Czech title
R2-D2: Modulární systém pro odpovídání na otázky nad otevřenou doménou
Type
conference paper
Language
english
Authors
Fajčík Martin, Ing., Ph.D. (DCGM FIT BUT)
Dočekal Martin, Ing. (DCGM FIT BUT)
Ondřej Karel, Ing. (DCGM FIT BUT)
and others
URL
Keywords

question answering, QA, ODQA, ensemble modeling, retrieval corpora

Abstract

This work presents a novel four-stage open-domain QA pipeline R2-D2 (Rank twice, reaD twice). The pipeline is composed of a retriever, passage reranker, extractive reader, generative reader and a mechanism that aggregates the final prediction from all systems components. We demonstrate its strength across three open-domain QA datasets: NaturalQuestions, TriviaQA and EfficientQA, surpassing state-of-the-art on the first two. Our analysis demonstrates that: (i) combining extractive and generative reader yields absolute improvements up to 5 exact match and it is at least twice as effective as the posterior averaging ensemble of the same models with different parameters, (ii) the extractive reader with fewer parameters can match the performance of the generative reader on extractive QA datasets.

Published
2021
Pages
854-870
Proceedings
Findings of the Association for Computational Linguistics: EMNLP 2021
Series
Findings of the Association for Computational Linguistics
Conference
Conference on Empirical Methods in Natural Language Processing, Punta Cana, DO
ISBN
978-1-955917-10-0
Publisher
Association for Computational Linguistics
Place
Punta Cana, DO
EID Scopus
BibTeX
@INPROCEEDINGS{FITPUB12624,
   author = "Martin Faj\v{c}\'{i}k and Martin Do\v{c}ekal and Karel Ond\v{r}ej and Pavel Smr\v{z} and et al.",
   title = "R2-D2: A Modular Baseline for Open-Domain Question Answering",
   pages = "854--870",
   booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
   series = "Findings of the Association for Computational Linguistics",
   year = 2021,
   location = "Punta Cana, DO",
   publisher = "Association for Computational Linguistics",
   ISBN = "978-1-955917-10-0",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/12624"
}
Back to top