Publication detail

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

FAJČÍK, M. DOČEKAL, M. ONDŘEJ, K. SMRŽ, P.

Original Title

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

Type

conference paper

Language

English

Original 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.

Keywords

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

Authors

FAJČÍK, M.; DOČEKAL, M.; ONDŘEJ, K.; SMRŽ, P.

Released

11. 11. 2021

Publisher

Association for Computational Linguistics

Location

Punta Cana

ISBN

978-1-955917-10-0

Book

Findings of the Association for Computational Linguistics: EMNLP 2021

Edition

Findings of the Association for Computational Linguistics

Pages from

854

Pages to

870

Pages count

17

URL

https://aclanthology.org/2021.findings-emnlp.73.pdf

BibTex

@inproceedings{BUT175855,
  author="Martin {Fajčík} and Martin {Dočekal} and Karel {Ondřej} and Pavel {Smrž}",
  title="R2-D2: A Modular Baseline for Open-Domain Question Answering",
  booktitle="Findings of the Association for Computational Linguistics: EMNLP 2021",
  year="2021",
  series="Findings of the Association for Computational Linguistics",
  pages="854--870",
  publisher="Association for Computational Linguistics",
  address="Punta Cana",
  isbn="978-1-955917-10-0",
  url="https://aclanthology.org/2021.findings-emnlp.73.pdf"
}