Ellipsis and Coreference Resolution as Question Answering

Rahul Aralikatte, Matthew Lamm, Daniel Hardt, Anders Søgaard

Research output: Working paperResearch

Abstract

Coreference (‘he eats potatoes’) and many forms of ellipsis (e.g., ‘so does Mary’) are similar to reading comprehension questions (‘who eats potatoes’ or ‘what does Mary do with potatoes’), in that in order to resolve these, we need to identify an appropriate text span in the previous discourse. This paper exploits this analogy and proposes to use an architecture developed for machine comprehension for ellipsis and coreference resolution. We present both single-task and joint models and evaluate them across standard benchmarks, outperforming the current state of the art for ellipsis by up to 48.5% error reduction – and for coreference by 37.5% error reduction.
Original languageEnglish
Place of PublicationIthaca, NY
PublisherArXiv
Number of pages10
Publication statusPublished - 2019

Cite this

Aralikatte, R., Lamm, M., Hardt, D., & Søgaard, A. (2019). Ellipsis and Coreference Resolution as Question Answering. Ithaca, NY: ArXiv.
Aralikatte, Rahul ; Lamm, Matthew ; Hardt, Daniel ; Søgaard, Anders. / Ellipsis and Coreference Resolution as Question Answering. Ithaca, NY : ArXiv, 2019.
@techreport{4b9303a9f47b40c6a3a67f548928f782,
title = "Ellipsis and Coreference Resolution as Question Answering",
abstract = "Coreference (‘he eats potatoes’) and many forms of ellipsis (e.g., ‘so does Mary’) are similar to reading comprehension questions (‘who eats potatoes’ or ‘what does Mary do with potatoes’), in that in order to resolve these, we need to identify an appropriate text span in the previous discourse. This paper exploits this analogy and proposes to use an architecture developed for machine comprehension for ellipsis and coreference resolution. We present both single-task and joint models and evaluate them across standard benchmarks, outperforming the current state of the art for ellipsis by up to 48.5{\%} error reduction – and for coreference by 37.5{\%} error reduction.",
author = "Rahul Aralikatte and Matthew Lamm and Daniel Hardt and Anders S{\o}gaard",
year = "2019",
language = "English",
publisher = "ArXiv",
address = "United States",
type = "WorkingPaper",
institution = "ArXiv",

}

Aralikatte, R, Lamm, M, Hardt, D & Søgaard, A 2019 'Ellipsis and Coreference Resolution as Question Answering' ArXiv, Ithaca, NY.

Ellipsis and Coreference Resolution as Question Answering. / Aralikatte, Rahul; Lamm, Matthew; Hardt, Daniel; Søgaard, Anders.

Ithaca, NY : ArXiv, 2019.

Research output: Working paperResearch

TY - UNPB

T1 - Ellipsis and Coreference Resolution as Question Answering

AU - Aralikatte, Rahul

AU - Lamm, Matthew

AU - Hardt, Daniel

AU - Søgaard, Anders

PY - 2019

Y1 - 2019

N2 - Coreference (‘he eats potatoes’) and many forms of ellipsis (e.g., ‘so does Mary’) are similar to reading comprehension questions (‘who eats potatoes’ or ‘what does Mary do with potatoes’), in that in order to resolve these, we need to identify an appropriate text span in the previous discourse. This paper exploits this analogy and proposes to use an architecture developed for machine comprehension for ellipsis and coreference resolution. We present both single-task and joint models and evaluate them across standard benchmarks, outperforming the current state of the art for ellipsis by up to 48.5% error reduction – and for coreference by 37.5% error reduction.

AB - Coreference (‘he eats potatoes’) and many forms of ellipsis (e.g., ‘so does Mary’) are similar to reading comprehension questions (‘who eats potatoes’ or ‘what does Mary do with potatoes’), in that in order to resolve these, we need to identify an appropriate text span in the previous discourse. This paper exploits this analogy and proposes to use an architecture developed for machine comprehension for ellipsis and coreference resolution. We present both single-task and joint models and evaluate them across standard benchmarks, outperforming the current state of the art for ellipsis by up to 48.5% error reduction – and for coreference by 37.5% error reduction.

M3 - Working paper

BT - Ellipsis and Coreference Resolution as Question Answering

PB - ArXiv

CY - Ithaca, NY

ER -

Aralikatte R, Lamm M, Hardt D, Søgaard A. Ellipsis and Coreference Resolution as Question Answering. Ithaca, NY: ArXiv. 2019.