Abstract:
Machine Translation (MT) on its own is generally not good enough to produce high-quality texts, so to improve the MT output, it is common to have humans intervening in the translation process. One such typical intervention is post-editing (PE), the process whereby a human translator corrects the errors in the MT output. Yet another, more recent form of intervention is the so-called interactive translation prediction (ITP), which involves an MT system presenting a translator with translation suggestions they can accept or reject, actions that the MT system then uses to present them with new, corrected suggestions. In this thesis I present two empirical studies with professional English-to-Spanish translators investigating various translation productivity aspects of two such types of translation scenarios. Both studies are primarily quantitatively oriented, with qualitative data used, where possible, to interpret the quantitative findings. In the traditional PE setting, I found that decreases in MT quality were associated with increases in technical effort and processing time, suggesting that, in some cases, the BLEU score of an MT system can give a good indication of the expected temporal and technical effort a professional translator would have to exert when post-editing its output. In the ITP scenario, I found that ITP may be a viable alternative to PE based on a number of translation productivity indicators collected over time and translators’ qualitative feedback. Specifically, relative to PE, ITP resulted in shorter processing time and less technical effort. In terms of translator preferences, most translators preferred ITP over PE and would use it in real-life translation situations.
