Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ TDX (Tesis Doctorals...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
image/svg+xml Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao Closed Access logo, derived from PLoS Open Access logo. This version with transparent background. http://commons.wikimedia.org/wiki/File:Closed_Access_logo_transparent.svg Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao
versions View all 2 versions
addClaim

Post-editing effort and linguistically motivated evaluation of machine translation

Authors: Álvarez Vidal, Sergi;

Post-editing effort and linguistically motivated evaluation of machine translation

Abstract

Les recents millores en traducció automàtica neuronal (TAN) han provocat un canvi de la traducció automàtica estadística (TAS) a la TAN, que ha incrementat l’ús de la postedició en els fluxos de treball industrials. Tanmateix, molts traductors professionals afirmen que si la qualitat de la TA no és prou bona, eliminen tot el segment i el tradueixen de cap i de nou. El problema és que les mesures automàtiques habituals no sempre indiquen la qualitat de la TA, especialment quan aquesta és bona, i no hi ha una correlació directa entre l’esforç de postedició i les mesures de productivitat. Combinem mètodes quantitatius i qualitatius per estudiar algunes de les mesures automàtiques més habituals utilitzades per evaluar la qualitat de la TA, i les comparem amb les mesures de l’esforç de postedició. A continuació, estudiem en detall diferent mesures directes i indirectes d’esforç per establir-hi una correlació. Complementem aquest estudi amb l’anàlisi de les percepcions dels traductors que duen a terme aquesta tasca. Finalment, fem una anàlisis detallada dels errors de TA a partir de les correccions fetes en la postedició i proposem un enfocament basat en errors per evaluar la TA que inclou l’ús d’un conjunt de frases de prova.

The recent improvements in neural MT (NMT) have driven a shift from statistical MT (SMT) to NMT, which has propelled the use of post-editing (PE) in translation workflows. However, many professional translators state that if the quality of the MT output is not good enough, they delete the remaining segments and translate everything from scratch. The problem is that usual automatic measurements do not always indicate the quality of theMT output, especially with high quality outputs, and there is still no clear correlation between PE effort and productivity scores. We combine quantitative and qualitative methods to study some of the usual automatic metrics used to evaluate the quality of MT output, and compare them to measures of post-editing effort. Then, we study in detail different direct and indirect measures of effort in order to establish a correlation among them. We complement this study with the analysis of translators’ perceptions of the task. Finally, we conduct a fine-grained analysis of MT errors based on postediting corrections and suggest an error-based approach to evaluate raw MT output which includes the use of challenge sets.

Programa de doctorat en Traducció i Ciències del Llenguatge

Country
Spain
Keywords

Traducció automàtica neuronal (TAN), 81, Neural MT (NMT)

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green