Following these recommendations supports the creation of a full, multilingual post-editing tool. Good translations are defined through the systemic-functional model of House which aims at a contextual correct translation. Write your own post-editing guide and alter these recommendations to fit your own definitions. In most cases the browser-plugin of languageTool is useful as the proof-reading tool.
在 3.9 版本新加入.
可以在每個組件的菜單中找到的 :guilabel:`Community localization checklist`（社區本地化清單）可以給予指示，使您的本地化過程對於社區翻譯者更容易。
Post-editing of MT with terminology assignment influences each level of the translation process. The machine translation system can be adapted to the specific vocabulary and style with a continued training or neural fuzzy repair. Import your existing translation memory into weblate or create an initial scope with your basic terminology. In the end the lector should be instructed with additional terminology documents to guarantee a good knowledge and output in the field.
The quality of the automatic translation (often measured with the BLEU-score) correlates with editing time . Choose a machine backend which supports the needed languages and domains. Make clear how the translation backend functions and which quality the post-editor has to expect.
The translations should be reviewed by a second person after the post-editing. With an impartial and competent revisor, the two man rule reduces the errors and improves the quality and consistency of the content.
There are many 檢查和修復 in Weblate which provide structured feedback on quality of the translations.
In addition to the mentalistic and impact-based definitions which make a strong reduction, the text-based linguistic approach fits best with the implemented translation methods. A well-formulated theory for translation evaluation is House’s systemic-functional model, which focuses on the relation between original and translation. The model assumes that translation is an attempt to keep the semantic, pragmatic, and textual meaning of a text equivalent when crossing from one linguistic code to another.
The degree of quality of a translation is based on the degree of equivalence, the correspondence between the text profile and the text function. Because it cannot be calculated automatically, sufficient information should be collected to enable a uniform human evaluation. The two main parameters of agreement in a corresponding model are the macro-context – i.e. embedding in a larger social and literary context – and the micro-context consisting of field, tenor and mode.
Marina Sanchez-Torron and Philipp Koehn in Machine Translation Quality and Post-Editor Productivity, Figure 1: https://www.cs.jhu.edu/~phi/publications/machine-translation-quality.pdf
Joanna Best und Sylvia Kalina.Übersetzen und Dolmetschen: eine Orientierungs-hilfe. A. Francke Verlag Tübingen und Base, 2002. Möglichkeiten der Übersetzungskritik starting on page number 101
neural fuzzy repair, Bram Bulté and Arda Tezcan in Neural Fuzzy Repair: Integrating Fuzzy Matches into Neural MachineTranslation, 2019 https://aclanthology.org/P19-1175.pdf