Advanced fuzzy matching in the translation of EU texts

Šoštarić, Margita. (2018). Advanced fuzzy matching in the translation of EU texts. Diploma Thesis. Filozofski fakultet u Zagrebu, Department of English Language and Literature. [mentor Pavlović, Nataša and Simeon, Ivana].

Preview

PDF (English) Download (921kB) | Preview

Abstract

In the translation industry today, CAT tool environments are an indispensable part of the translator’s workflow. Translation memory systems constitute one of the most important features contained in these tools and the question of how to best use them to make the translation process faster and more efficient legitimately arises. This research aims to examine whether there are more efficient methods of retrieving potentially useful translation suggestions than the ones currently used in TM systems. We are especially interested in investigating whether more sophisticated algorithms and the inclusion of linguistic features in the matching process lead to significant improvement in quality of the retrieved matches. The used dataset, the DGT-TM, is pre-processed and parsed, and a number of matching configurations are applied to the data structures contained in the produced parse trees. We also try to improve the matching by combining the individual metrics using a regression algorithm. The retrieved matches are then evaluated by means of automatic evaluation, based on correlations and mean scores, and human evaluation, based on correlations of the derived ranks and scores. Ultimately, the goal is to determine whether the implementation of some of these fuzzy matching metrics should be considered in the framework of the commercial CAT tools to improve the translation process.

Actions (login required)

View Item