Publicação
Quality in human post-editing of machine-translated texts : error annotation and linguistic specifications for tackling register errors
| Resumo: | During the last decade, machine translation has played an important role in the translation market and has become an essential tool for speeding up the translation process and for reducing the time and costs needed. Nevertheless, the quality of the results obtained is not completely satisfactory, as it is considerably variable, depending on numerous factors. Given this, it is necessary to combine MT with human intervention, by post-editing the machine-translated texts, in order to reach high-quality translations. This work aims at describing the MT process provided by Unbabel, a Portuguese start-up that combines MT with post-editing provided by online editors. The main objective of the study is to contribute to improving the quality of the translated text, by analyzing annotated translated texts, from English into Italian, to define linguistic specifications to improve the tools used at the start-up to aid human editors and annotators. The analysis of guidelines provided to the annotator to guide his/her editing process has also been developed, a task that contributed to improve the inter-annotator agreement, thus making the annotated data reliable. Accomplishing these goals allowed for the identification and the categorization of the most frequent errors in translated texts, namely errors whose resolution is bound to significantly improve the efficacy and quality of the translation. The data collected allowed us to identify register as the most frequent error category and also the one with the most impact on the quality of translations, and for these reasons this category is analyzed in more detail along the work. From the analysis of errors in this category, it was possible to define and implement a set of rules in the Smartcheck, a tool used at Unbabel to automatically detect errors in the target text produced by the MT system to guarantee a higher quality of the translated texts after post-edition. |
|---|---|
| Autores principais: | Testa, Ingrid |
| Assunto: | Língua inglesa - Tradução para italiano Língua inglesa - Tradução automática Língua italiana - Tradução automática Tradução - Metodologia Tradução automática Teses de mestrado - 2018 |
| Ano: | 2018 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | During the last decade, machine translation has played an important role in the translation market and has become an essential tool for speeding up the translation process and for reducing the time and costs needed. Nevertheless, the quality of the results obtained is not completely satisfactory, as it is considerably variable, depending on numerous factors. Given this, it is necessary to combine MT with human intervention, by post-editing the machine-translated texts, in order to reach high-quality translations. This work aims at describing the MT process provided by Unbabel, a Portuguese start-up that combines MT with post-editing provided by online editors. The main objective of the study is to contribute to improving the quality of the translated text, by analyzing annotated translated texts, from English into Italian, to define linguistic specifications to improve the tools used at the start-up to aid human editors and annotators. The analysis of guidelines provided to the annotator to guide his/her editing process has also been developed, a task that contributed to improve the inter-annotator agreement, thus making the annotated data reliable. Accomplishing these goals allowed for the identification and the categorization of the most frequent errors in translated texts, namely errors whose resolution is bound to significantly improve the efficacy and quality of the translation. The data collected allowed us to identify register as the most frequent error category and also the one with the most impact on the quality of translations, and for these reasons this category is analyzed in more detail along the work. From the analysis of errors in this category, it was possible to define and implement a set of rules in the Smartcheck, a tool used at Unbabel to automatically detect errors in the target text produced by the MT system to guarantee a higher quality of the translated texts after post-edition. |
|---|