Understanding the Effect of Textual Adversaries in Multimodal Machine Translation

Koel Dutta Chowdhury; Desmond Elliott

doi:10.18653/v1/D19-6406

Understanding the Effect of Textual Adversaries in Multimodal Machine Translation

Abstract

It is assumed that multimodal machine translation systems are better than text-only systems at translating phrases that have a direct correspondence in the image. This assumption has been challenged in experiments demonstrating that state-of-the-art multimodal systems perform equally well in the presence of randomly selected images, but, more recently, it has been shown that masking entities from the source language sentence during training can help to overcome this problem. In this paper, we conduct experiments with both visual and textual adversaries in order to understand the role of incorrect textual inputs to such systems. Our results show that when the source language sentence contains mistakes, multimodal translation systems do not leverage the additional visual signal to produce the correct translation. We also find that the degradation of translation performance caused by textual adversaries is significantly higher than by visual adversaries.

Originalsprog	Udefineret/Ukendt
Titel	Proceedings of the Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN)
Antal sider	6
Udgivelsessted	Hong Kong, China
Forlag	Association for Computational Linguistics (ACL)
Publikationsdato	1 nov. 2019
Sider	35-40
DOI	https://doi.org/10.18653/v1/D19-6406
Status	Udgivet - 1 nov. 2019

Adgang til dokumentet

10.18653/v1/D19-6406

https://www.aclweb.org/anthology/D19-6406

Citationsformater

Understanding the Effect of Textual Adversaries in Multimodal Machine Translation. / Dutta Chowdhury, Koel; Elliott, Desmond.
Proceedings of the Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN). Hong Kong, China: Association for Computational Linguistics (ACL), 2019. s. 35-40.

Publikation: Bidrag til bog/antologi/rapport › Konferencebidrag i proceedings › Forskning › peer review

@inproceedings{7b0367d9fa1b4d16a44bef614b637523,

title = "Understanding the Effect of Textual Adversaries in Multimodal Machine Translation",

abstract = "It is assumed that multimodal machine translation systems are better than text-only systems at translating phrases that have a direct correspondence in the image. This assumption has been challenged in experiments demonstrating that state-of-the-art multimodal systems perform equally well in the presence of randomly selected images, but, more recently, it has been shown that masking entities from the source language sentence during training can help to overcome this problem. In this paper, we conduct experiments with both visual and textual adversaries in order to understand the role of incorrect textual inputs to such systems. Our results show that when the source language sentence contains mistakes, multimodal translation systems do not leverage the additional visual signal to produce the correct translation. We also find that the degradation of translation performance caused by textual adversaries is significantly higher than by visual adversaries.",

author = "{Dutta Chowdhury}, Koel and Desmond Elliott",

year = "2019",

month = nov,

day = "1",

doi = "10.18653/v1/D19-6406",

language = "Udefineret/Ukendt",

pages = "35--40",

booktitle = "Proceedings of the Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN)",

publisher = "Association for Computational Linguistics (ACL)",

address = "USA",

}

TY - GEN

T1 - Understanding the Effect of Textual Adversaries in Multimodal Machine Translation

AU - Dutta Chowdhury, Koel

AU - Elliott, Desmond

PY - 2019/11/1

Y1 - 2019/11/1

N2 - It is assumed that multimodal machine translation systems are better than text-only systems at translating phrases that have a direct correspondence in the image. This assumption has been challenged in experiments demonstrating that state-of-the-art multimodal systems perform equally well in the presence of randomly selected images, but, more recently, it has been shown that masking entities from the source language sentence during training can help to overcome this problem. In this paper, we conduct experiments with both visual and textual adversaries in order to understand the role of incorrect textual inputs to such systems. Our results show that when the source language sentence contains mistakes, multimodal translation systems do not leverage the additional visual signal to produce the correct translation. We also find that the degradation of translation performance caused by textual adversaries is significantly higher than by visual adversaries.

AB - It is assumed that multimodal machine translation systems are better than text-only systems at translating phrases that have a direct correspondence in the image. This assumption has been challenged in experiments demonstrating that state-of-the-art multimodal systems perform equally well in the presence of randomly selected images, but, more recently, it has been shown that masking entities from the source language sentence during training can help to overcome this problem. In this paper, we conduct experiments with both visual and textual adversaries in order to understand the role of incorrect textual inputs to such systems. Our results show that when the source language sentence contains mistakes, multimodal translation systems do not leverage the additional visual signal to produce the correct translation. We also find that the degradation of translation performance caused by textual adversaries is significantly higher than by visual adversaries.

U2 - 10.18653/v1/D19-6406

DO - 10.18653/v1/D19-6406

M3 - Konferencebidrag i proceedings

SP - 35

EP - 40

BT - Proceedings of the Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN)

PB - Association for Computational Linguistics (ACL)

CY - Hong Kong, China

ER -