Introducing MTPE pricing in translator training: a concrete proposal for MT instructors

ABSTRACT

In recent years, machine translation post-editing (MTPE or PE for short) has been steadily gaining ground in the language industry. However, studies that examine translators’ perceptions of, and attitudes towards, MTPE paint a somewhat negative picture, with PE pricing methods and rates being a major source of dissatisfaction. While the European Master’s in Translation Competence Framework stresses the importance of preparing translation graduates for market challenges, to date there have been no concrete suggestions for practical activities designed to introduce MTPE-pricing-related topics into the translation classroom. The present article aims to address this gap by describing a teaching unit developed for master’s students. The activity includes comparing three MTPE pricing methods commonly used in the industry: word-based, time-based and effort-based rates. Using authentic performance data from individual PE tasks carried out in MateCat, students were able to discover the different levels of remuneration they would receive for the same task, depending on the pricing method applied. The results, which show wide variation across both methods and students, proved useful in raising students’ awareness of the thorny issue of setting PE rates and sparking reflection on the financial implications of accepting PE assignments.

KEYWORDS:

• translator training
• machine translation post-editing (MTPE)
• PE pricing
• PE rates
• PE module

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. https://search.proz.com/employers/rates (accessed 27 December 2023).

2. See for instance https://wiki.proz.com/wiki/index.php/Determining_your_rates_and_fees_as_a_translator (accessed 27 December 2023).

3. https://www.sft.fr/fr/fiche-metier-post-edition (accessed 3 November 2022).

4. https://community.rws.com/product-groups/linguistic-ai/b/weblog/posts/isn-t-it-time-to-embrace-machine-translation-post-editing-the-localization-use-case-for-mt (accessed 27 December 2023).

5. TER was designed to ‘[measure] the amount of editing that a human would have to perform to change a system output so it exactly matches a reference translation’ (Snover et al. 2006, 223).

6. See for instance GALA https://www.gala-global.org/events/events-calendar/gala-connected-2021-pricing-impact-machine-translation-fairness-and (accessed 3 November 2022)

7. In their article, Scansani and Mhedhbi (2022, 397) report: ‘[The final coefficient] will be then subtracted from the per-word rate for no-match segments when MT is used’. It is unclear, however, whether this rate corresponds to a full translation rate.

8. See for instance Hunnect https://hunnect.com/mtpe_fair_pricing/ (accessed 27 December 2023).

9. https://www.matecat.com/.

10. In order to provide all students with the same DeepL translation, we created a translation memory with the MT segments and deactivated the suggestions coming from other sources in MateCat.

11. https://www.wooclap.com/.

12. It is worth noting that not all the students enrolled in the module took part in the weekly activities, which were not compulsory. Furthermore, although 20 students performed the PE tasks described, not all of them participated in the MTPE-pricing session, nor answered all the live quizzes. For this reason, in this section we will include the number of participants (n) at each stage of the teaching unit.

13. https://site.matecat.com/deliverables/d5–3/key-performance-indicators/ (accessed 15 November 2022).

14. In MateCat, payable words do not correspond to actual source words: they are the outcome of intricate calculations based, inter alia, on TM (translation memory) fuzzy matches and previously edited MT segments. See:

https://guides.matecat.com/how-matecat-calculates-payable-words (accessed 27 December 2023).

15. https://copeco.unige.ch/.