Probabilistic Assumptions Matter: Improved Models for Distantly-Supervised Document-Level Question Answering

By Hao Cheng, Ming-Wei Chang, Kenton C. T. Lee, Kristina Toutanova. 2020

22 citations recorded by the source corpus.

We address the problem of extractive question answering using document-level distant super-vision, pairing questions and relevant documents with answer strings. We compare previously used probability space and distant supervision assumptions (assumptions on the correspondence between the weak answer string labels and possible answer mention spans). We show that these assumptions interact, and that different configurations provide complementary benefits. We demonstrate that a multi-objective model can efficiently combine the advantages of multiple assumptions and outperform the best individual formulation. Our approach outperforms previous state-of-the-art models by 4.3 points in F1 on TriviaQA-Wiki and 1.7 points in Rouge-L on NarrativeQA summaries.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jianfeng Gao
2. Weizhu Chen
3. Xiaodong Liu
4. Pengcheng He
5. Yelong Shen
6. Hao Cheng
7. Samuel R. Bowman
8. Mohit Bansal

In Computer Vision and Pattern Recognition

1. Jianfeng Gao
2. Weizhu Chen
3. Xiaodong Liu
4. Pengcheng He
5. Yelong Shen
6. Hao Cheng
7. Zhe Gan
8. Luc Van Gool

In Natural Language Processing

1. Jianfeng Gao
2. Xiaodong Liu
3. Weizhu Chen
4. Pengcheng He
5. Hao Cheng
6. Yelong Shen
7. Caiming Xiong
8. Richard Socher

Citations in the corpus

• R2-D2: A Modular Baseline for Open-Domain Question Answering (2021, methodology): “Cheng et al. (2020a) experimented with different assumptions for MML, showing improvement when marginalizing over components of span probability independently.”
• UnitedQA: A Hybrid Approach for Open Domain Question Answering (2021, methodology): “In our work, we further extend the multi-objective formulation proposed in (Cheng et al., 2020) with the hard EM learning (Min et al.”
• ReadTwice: Reading Very Large Documents with Memories (2021, unknown): “For a similar setup, see multi-segment based QA tasks (Clark and Gardner, 2018; Cheng et al., 2020).”
• Rethinking the Objectives of Extractive Question Answering (2020, background): “…a s , a e (cid:105) – has stayed the same over the course of years in many influential works (Xiong et al., 2017; Seo et al., 2017; Chen et al., 2017; Yu et al., 2018; Devlin et al., 2019; Cheng et al., 2020) since the earliest work on this dataset – the submission of Wang and Jiang (2017).”