Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints

Craig Thomson; Clément  Rebuffel; Ehud Reiter; Laure  Soulier; Somayajulu Sripada; Patrick  Gallinari

Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints

Craig Thomson, Clément Rebuffel, Ehud Reiter, Laure Soulier, Somayajulu Sripada, Patrick Gallinari

Research output: Contribution to conference › Unpublished paper › peer-review

Abstract

Neural data-to-text systems lack the control and factual accuracy required to generate useful and insightful summaries of multidimensional data. We propose a solution in the form of data views, where each view describes an entity and its attributes along specific dimensions. A sequence of views can then be used as a high-level schema for document planning, with the neural model handling the complexities of micro-planning and surface realization. We show that our view-based system retains factual accuracy while offering high-level control of output that can be tailored based on user preference or other norms within the domain.

Original language	English
Number of pages	16
Publication status	Accepted/In press - 1 Sept 2023
Event	16th International Natural Language Generation Conference - OREA Hotel Pyramida, Prague., Prague, Czech Republic Duration: 11 Sept 2023 → 15 Sept 2023 https://inlg2023.github.io/index.html

Conference

Conference	16th International Natural Language Generation Conference
Country/Territory	Czech Republic
City	Prague
Period	11/09/23 → 15/09/23
Internet address	https://inlg2023.github.io/index.html

Access to Document

https://sigdialinlg2023.github.io/paper_inlg97.htmlLicence: Unspecified

Cite this

@conference{f29d13d6d9d8420f87890c4bd435a179,

title = "Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints",

abstract = "Neural data-to-text systems lack the control and factual accuracy required to generate useful and insightful summaries of multidimensional data. We propose a solution in the form of data views, where each view describes an entity and its attributes along specific dimensions. A sequence of views can then be used as a high-level schema for document planning, with the neural model handling the complexities of micro-planning and surface realization. We show that our view-based system retains factual accuracy while offering high-level control of output that can be tailored based on user preference or other norms within the domain.",

author = "Craig Thomson and Cl{\'e}ment Rebuffel and Ehud Reiter and Laure Soulier and Somayajulu Sripada and Patrick Gallinari",

year = "2023",

month = sep,

day = "1",

language = "English",

note = "16th International Natural Language Generation Conference ; Conference date: 11-09-2023 Through 15-09-2023",

url = "https://inlg2023.github.io/index.html",

}

TY - CONF

T1 - Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints

AU - Thomson, Craig

AU - Rebuffel, Clément

AU - Reiter, Ehud

AU - Soulier, Laure

AU - Sripada, Somayajulu

AU - Gallinari, Patrick

PY - 2023/9/1

Y1 - 2023/9/1

N2 - Neural data-to-text systems lack the control and factual accuracy required to generate useful and insightful summaries of multidimensional data. We propose a solution in the form of data views, where each view describes an entity and its attributes along specific dimensions. A sequence of views can then be used as a high-level schema for document planning, with the neural model handling the complexities of micro-planning and surface realization. We show that our view-based system retains factual accuracy while offering high-level control of output that can be tailored based on user preference or other norms within the domain.

AB - Neural data-to-text systems lack the control and factual accuracy required to generate useful and insightful summaries of multidimensional data. We propose a solution in the form of data views, where each view describes an entity and its attributes along specific dimensions. A sequence of views can then be used as a high-level schema for document planning, with the neural model handling the complexities of micro-planning and surface realization. We show that our view-based system retains factual accuracy while offering high-level control of output that can be tailored based on user preference or other norms within the domain.

UR - https://inlg2023.github.io/index.html

M3 - Unpublished paper

T2 - 16th International Natural Language Generation Conference

Y2 - 11 September 2023 through 15 September 2023

ER -

Enhancing Factualness and Controllability of Data-to-Text Generation via Data Views and Constraints

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Cite this