Measuring content preservation in textual style transfer

Stuart Fitzpatrick; Laurence Park; Oliver Obst

doi:10.1007/978-981-19-8746-5_1

Measuring content preservation in textual style transfer

Stuart Fitzpatrick, Laurence Park, Oliver Obst

Research output: Chapter in Book / Conference Paper › Chapter

Abstract

Style transfer in text, changing text that is written in a particular style such as the works of Shakespeare to be written in another style, currently relies on taking the cosine similarity of the sentence embeddings of the original and transferred sentence to determine if the content of the sentence, its meaning, hasn't changed. This assumes however that such sentence embeddings are style invariant, which can result in inaccurate measurements of content preservation. To investigate this we compared the average similarity of multiple styles of text from the Corpus of Diverse Styles using a variety of sentence embedding methods and find that those embeddings which are created from aggregated word embeddings are style invariant, but those created by sentence embeddings are not.

Original language	English
Title of host publication	Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings
Editors	Laurence A. F. Park, Heitor Murilo Gomes, Maryam Doborjeh, Yee Ling Boo, Yun Sing Koh, Yanchang Zhao, Graham Williams, Simeon Simoff
Place of Publication	Singapore
Publisher	Springer
Pages	3-14
Number of pages	12
ISBN (Electronic)	9789811987465
ISBN (Print)	9789811987458
DOIs	https://doi.org/10.1007/978-981-19-8746-5_1
Publication status	Published - 2022

Bibliographical note

Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Access to Document

10.1007/978-981-19-8746-5_1

Cite this

Fitzpatrick, S., Park, L., & Obst, O. (2022). Measuring content preservation in textual style transfer. In L. A. F. Park, H. M. Gomes, M. Doborjeh, Y. L. Boo, Y. S. Koh, Y. Zhao, G. Williams, & S. Simoff (Eds.), Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings (pp. 3-14). Springer. https://doi.org/10.1007/978-981-19-8746-5_1

Fitzpatrick, Stuart ; Park, Laurence ; Obst, Oliver. / Measuring content preservation in textual style transfer. Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings. editor / Laurence A. F. Park ; Heitor Murilo Gomes ; Maryam Doborjeh ; Yee Ling Boo ; Yun Sing Koh ; Yanchang Zhao ; Graham Williams ; Simeon Simoff. Singapore : Springer, 2022. pp. 3-14

@inbook{69431465f880411c829120bf2be2f988,

title = "Measuring content preservation in textual style transfer",

abstract = "Style transfer in text, changing text that is written in a particular style such as the works of Shakespeare to be written in another style, currently relies on taking the cosine similarity of the sentence embeddings of the original and transferred sentence to determine if the content of the sentence, its meaning, hasn't changed. This assumes however that such sentence embeddings are style invariant, which can result in inaccurate measurements of content preservation. To investigate this we compared the average similarity of multiple styles of text from the Corpus of Diverse Styles using a variety of sentence embedding methods and find that those embeddings which are created from aggregated word embeddings are style invariant, but those created by sentence embeddings are not.",

author = "Stuart Fitzpatrick and Laurence Park and Oliver Obst",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.",

year = "2022",

doi = "10.1007/978-981-19-8746-5_1",

language = "English",

isbn = "9789811987458",

pages = "3--14",

editor = "Park, {Laurence A. F.} and Gomes, {Heitor Murilo} and Maryam Doborjeh and Boo, {Yee Ling} and Koh, {Yun Sing} and Yanchang Zhao and Graham Williams and Simeon Simoff",

booktitle = "Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings",

publisher = "Springer",

}

Fitzpatrick, S, Park, L & Obst, O 2022, Measuring content preservation in textual style transfer. in LAF Park, HM Gomes, M Doborjeh, YL Boo, YS Koh, Y Zhao, G Williams & S Simoff (eds), Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings. Springer, Singapore, pp. 3-14. https://doi.org/10.1007/978-981-19-8746-5_1

Measuring content preservation in textual style transfer. / Fitzpatrick, Stuart; Park, Laurence ; Obst, Oliver.
Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings. ed. / Laurence A. F. Park; Heitor Murilo Gomes; Maryam Doborjeh; Yee Ling Boo; Yun Sing Koh; Yanchang Zhao; Graham Williams; Simeon Simoff. Singapore: Springer, 2022. p. 3-14.

Research output: Chapter in Book / Conference Paper › Chapter

TY - CHAP

T1 - Measuring content preservation in textual style transfer

AU - Fitzpatrick, Stuart

AU - Park, Laurence

AU - Obst, Oliver

PY - 2022

Y1 - 2022

N2 - Style transfer in text, changing text that is written in a particular style such as the works of Shakespeare to be written in another style, currently relies on taking the cosine similarity of the sentence embeddings of the original and transferred sentence to determine if the content of the sentence, its meaning, hasn't changed. This assumes however that such sentence embeddings are style invariant, which can result in inaccurate measurements of content preservation. To investigate this we compared the average similarity of multiple styles of text from the Corpus of Diverse Styles using a variety of sentence embedding methods and find that those embeddings which are created from aggregated word embeddings are style invariant, but those created by sentence embeddings are not.

AB - Style transfer in text, changing text that is written in a particular style such as the works of Shakespeare to be written in another style, currently relies on taking the cosine similarity of the sentence embeddings of the original and transferred sentence to determine if the content of the sentence, its meaning, hasn't changed. This assumes however that such sentence embeddings are style invariant, which can result in inaccurate measurements of content preservation. To investigate this we compared the average similarity of multiple styles of text from the Corpus of Diverse Styles using a variety of sentence embedding methods and find that those embeddings which are created from aggregated word embeddings are style invariant, but those created by sentence embeddings are not.

UR - https://hdl.handle.net/1959.7/uws:78065

U2 - 10.1007/978-981-19-8746-5_1

DO - 10.1007/978-981-19-8746-5_1

M3 - Chapter

SN - 9789811987458

SP - 3

EP - 14

BT - Data Mining: 20th Australasian Conference, AusDM 2022, Western Sydney, Australia, December 12-15, 2022, Proceedings

A2 - Park, Laurence A. F.

A2 - Gomes, Heitor Murilo

A2 - Doborjeh, Maryam

A2 - Boo, Yee Ling

A2 - Koh, Yun Sing

A2 - Zhao, Yanchang

A2 - Williams, Graham

A2 - Simoff, Simeon

PB - Springer

CY - Singapore

ER -

Measuring content preservation in textual style transfer

Abstract

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this