https://orcid.org/0000-0003-1761-8918
![Sample our Language & Literature journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5941/TOC-Subject-Sample-2021-Language-Literature.jpg"})
ABSTRACT
This study combines product- and process-oriented research methods and tools to observe whether and how the presence of pre-translated text affects translation quality and influences the translator’s research patterns. It is part of the LeMaTTT project, a simulated longitudinal empirical study exploring the impact of MT on info-mining and thematic competences in legal translation. Data were elicited through a translation task completed by a cohort of 110 final-year MA trainees with training in legal translation and basic MT literacy, and a cohort of 54 first-year MA trainees with no to very limited experience in specialised translation and post-editing. This paper provides a first analysis of selected process-related data concerning the allocation of time and the use of external resources throughout the translation or post-editing processes of 40 participants, 20 from each cohort. Preliminary results highlight some correlations between the use of time and external resources and both (a) the development of thematic and info-mining competences and (b) the specific type of task, i.e. whether from-scratch translation or post-editing.
Acknowledgments
The author wishes to acknowledge the dedication and assistance of students Lucrezia Lemma, Alice Mazzatenta, Matilde Moretti, and Virginia Rossato in the development of support and research material, the conduct of the empirical phase, and/or the extraction of data within the LeMaTTT project.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. https://lex-machina.ch/ (accessed January 2023).
2. Whether translator training should mirror or mould the increasingly technocratic translation market, and the socio-professional implications of MT in both training and the profession are issues that go beyond the scope of this investigation and cannot be thoroughly addressed here for reasons of space. Some insights into these controversial questions are offered, among others, by Canfora and Ottman (Citation2020); Kenny, Moorkens and do Carmo (Citation2020); Sakamoto and Yamada (Citation2020).
3. Since Citation2009, the EMT’s framework has been updated twice, i.e. in Citation2017 and Citation2022. In its most recent versions, both thematic and info-mining sub-competences are subsumed under the ‘translation’ sub-competence. For the purpose of this study, it is vital to keep the two components set apart from other translation-related abilities, as the project focuses precisely on how MT can influence the development of domain knowledge and info-mining skills. For this reason, the labels proposed in the old version of EMT’s model are used.
4. The language combination and nature of the source text (e.g. general vs specialised text) were found to affect this result and some studies also provided conflicting evidence indicating reduced productivity gains in PE (cf. Koponen Citation2016 for an overview).
5. The author does not specify which type of MT engine was enabled, but given that GTT was released the same year (i.e. 2009) and relied on Google Translator to produce machine-translated matches, it can be presumed that it ran a statistical engine.
6. The MT engine used for the study is Google Translate, but no details about the type of MT used at the time of the study are provided.
7. Since both Daems et al. (Citation2016) and Daems et al. (Citation2017) used Google Translate, which only shifted from statistical to neural MT in 2016, it can be presumed that the outputs used in both studies were produced by its old statistical system.
8. Simulated longitudinal studies are those considering comparable samples of participants at different stages in a developmental process, e.g. trainees with different years of experience or training, rather than repeating the same measurements on the same participants at regular intervals over a long period (cf. Göpferich and Jääskeläinen Citation2009, 183).
9. https://www.flashbackrecorder.com/ (accessed January 2023).
10. Retrievable at: https://www.sec.gov/Archives/edgar/data/1141107/000095014407001746/g05741exv24.htm (accessed January 2023).
11. Cf. https://site.matecat.com/ (accessed April 2024).
12. The MT engine enabled for the task was the one provided by Matecat, i.e. the neural engine ModernMT (cf. https://www.modernmt.com/, accessed January 2023).
13. The paper analyses the datasets of the trainees with the following IDs: N1, N2, N3, N4, N5, N7, N8, N10, N12, and N13 (G1MT); N28, N29, N30, N31, N32, N33, N34, N35, N36, and N37 (G1FS); I1, I2, I3, I4, I5, I7, I8, I9, I10, and I11(G2MT); I56, I57, I58, I59, I60, I61, I62, I63, I64, and I65 (G2FS).
14. G1MT: M = 25’58’’, Mdn = 23’08’’; G1FS: M = 43’15’’, Mdn = 39’29’’; G2MT: M = 1h6’41’’, Mdn = 1h8’; G2FS: M = 1h9’40’’, Mdn = 1h13’22’’.
15. The ratio was calculated by dividing the number of searches made by the group using a specific resource (e.g. bilingual general dictionaries) by the total number of searches made by the same group. This allows to compare the results of two groups despite any differences in their absolute frequencies.