DeepL vs Google Translate: who’s the best at translating MWEs from French into Polish? A multidisciplinary approach to corpora creation and quality translation of MWEs - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

DeepL vs Google Translate: who’s the best at translating MWEs from French into Polish? A multidisciplinary approach to corpora creation and quality translation of MWEs

Résumé

This article proposes a multidisciplinary approach to the creation of parallel French and Polish corpora annotated with multi-word expressions and the analysis of neural machine translation errors of annotated MWEs, from French into Polish. Our work first aims at building parallel FR-PL corpora from a French News corpus, taken from the WMT 2010 corpora of 40 000 words, by automatically translating them into Polish using the main commercial systems DeepL and Google Translate. The French source corpus has already been manually annotated with MWE, using the typology developed at LIDILEM (Tutin and Esperança-Rodier 2019). In a second step, the quality of the MWE translation of the FR-PL parallel corpora has been evaluated by annotating the translation errors creating a new error typology based on MQM-DQF (Lommel & al., 2018a) and the linguistic features of MWE translations using the ACCOLÉ platform (Brunet-Manquat and Esperança-Rodier, 2018). As a start, we have selected 154 sentences (4 332 French words) from the already MWE annotated French Document and translated them into Polish using DeepL (3 599 Polish words) and Google Translate (3 519 Polish words). The first general result shows that for the French to Polish language pair, DeepL translates MWE better than Google Translate even though it uses English as a pivot.
Fichier principal
Vignette du fichier
TC43DeeplvsGoogleFRPL_EER_FINAL_Publi_Relu.pdf (1.71 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03779450 , version 1 (26-09-2022)

Identifiants

  • HAL Id : hal-03779450 , version 1

Citer

Emmanuelle Esperança-Rodier, Damian Frankowski. DeepL vs Google Translate: who’s the best at translating MWEs from French into Polish? A multidisciplinary approach to corpora creation and quality translation of MWEs. Translating and the Computer 43, Asling, Nov 2021, Londres, United Kingdom. ⟨hal-03779450⟩
366 Consultations
224 Téléchargements

Partager

Gmail Facebook X LinkedIn More