Independent samples t-tests showed that there were significant differences between HT and PE students concerning accuracy errors, fluency errors and total error numbers. Clearly, differences in total error numbers can be largely accounted for by accuracy and fluency errors. In summary, post-editors’ target texts were of better overall quality, and this was because there were fewer accuracy and fluency errors in their texts.

These findings are in agreement with previous studies showing that PE texts are of better quality than HT texts (Arjani & Jamshidiha, 2024; Carl et al., 2011; Fiederer & O’Brien, 2009; Lee, Sun-Woo & Sang-Bin Lee, 2021; Plitt & Masselot, 2010).

Nevertheless, one final note should be made on the differences between HT and PE students. Terminology does not seem to fit into the overall pattern in which PE error numbers tend to be lower than HT error numbers: in this category, the mean value is slightly higher in the PE condition than in the HT condition, although the difference is not significant. Considering how unreliable NMT is with respect to terminology (Yamada, 2019), it makes sense to suggest that this problem should be further studied on larger samples. The lack of a significant difference can be due to a small sample size.

The next step in the analysis was to find out whether certain error types were significantly more common in students’ target texts than others. Repeated measures one-way ANOVA indicated that in the HT group, there were significantly more accuracy errors (M = 8.32) than terminology errors (M = 3.77), and significantly more stylistic errors (M = 5.77) than terminology errors. In the PE group, there were significantly more accuracy errors (M = 6.44) than terminology errors (M = 4.28) and fluency errors (M = 2.50). Fluency errors were also significantly lower in number than stylistic errors (M = 5.53) and terminology errors (M = 4.28).

The above findings suggest that accuracy errors still stand out as dominant error types in both groups at the end of translator training, although their predominance is less striking than in the first data collection wave. It is also worth highlighting that in the post-editor group, there are significantly fewer fluency errors than any other errors, suggesting that this aspect of target text production was the easiest for second-year post-editors.