Machine translation technology has advanced significantly in the past several years. However, it still cannot replace the discerning judgment of a human translator. Due to the infinite variations involved in almost all languages, machine translations cannot (and probably will never) be able to detect and implement the correct translation in every instance. Hence, to have a highly reliable translation output when utilizing machine translations, post-editing will be necessary in most instances.

This study looks at the advisability of having students majoring in translation perform exercises in post-editing (PE) on outputs from machine translation (MT). It is a good way in which students could be introduced to MT and conclude that introducing PE at a pre-professional level can help raise students’ awareness of challenges facing both human and machine translation.

How is PE different from the traditional revising of human translation (RHT)? According to some linguists, PE is typically concerned with recurring, predictable errors, while RHT deals with inadvertent misunderstandings and omissions. PE focuses on words/expressions while RHT often involves correcting overall sentence- and discourse-level interpretations. Quality expectation is invariably high with RHT, but variable with PE, so the post-editors must be able to differentiate between style- and content-correction.

PE has proven its economic worth: scientists found that on average 2/3 of MT output required changes, even though the texts were technical manuals. They claim that MT + PE is 2-3 times faster than 100% human translation.

A final argument for PE is that Krings’ study found nothing to justify translators’ fear that PE makes them mere proofreaders (correctors). He found PE to be very similar to translation from scratch in terms of cognitive processes needed and the frequency distribution of these processes.

The English-to-Ukrainian PE experiment.

The goals of this experiment were to compare performance between “from-scratch” translation and revised MT output (for convenience referred to here as PE), and to determine student reaction to this exercise. Power Translator Pro (PTP) was the selected MT program. The first stage consisted of an error count of a passage translated by students without any MT intervention. The next year’s class was asked to revise MT output of the same passage, and another error count was done, both of the output and the PE results, using the same error classification. Finally, students filled out a questionnaire on the usefulness of the exercise.

Before giving the results of the error count, I will explain the categories underlying it:

Agreement: incorrect morphological concord between subject and verb, subject and reflexive pronoun, or noun and adjective. Anaphora: lack of object pronoun, incorrect person and/or number of possessive or relative pronoun, personal pronoun in place of demonstrative pronoun. Article: use of definite article where none is required. Generic: use of indefinite instead of definite article, or vice-versa. Literal: translation is word-for-word. Mistranslation: carelessness (e.g. Zimbabwe for Zambesi, plural for singular, unnecessary word, invented word). Omission: missing content word (noun, verb, adjective, adverb). Preposition: incorrect or missing preposition. Punctuation: mainly lack of a comma. Spelling: a typo rather than a grammatical error. Structure: syntactic error, e.g. using an intransitive verb transitively, incorrect pronoun object placement, gerund in place of infinitive or vice-versa. Tense: incorrect verb tense or mood. Word choice: error in polysemy or homonymy. Word order: e.g. failure to invert subject and communication verb in interpolated clause, placing an adverb after a heavy object.

Student comments. Students were asked to make a detailed analysis of the MT output and give a general assessment of the exercise both as a practical way to improve their translations and as a means of gaining insight into the nature and difficulties of translation.

Here is a sampling of comments: “This exercise may help me understand the logic or functioning of computers but it doesn’t help me improve my translations. Its mistakes are mechanical errors (e.g. word order, over-use of the article) which we have already learned to avoid”. “The exercise demonstrates basic problems underlying any translation, e.g. computer does largely a word-by-word replacement, ignoring context, general ideas (background knowledge, pragmatic conditions), and the culture of the source language”. “The machine is incapable of choosing among several possibilities the sense that fits best with the overall message”. “The exercise shows importance of being aware of concepts, rather than just searching for word or phrase equivalents. It made me aware of how much translation is an art rather than something mechanical”. “This exercise was extremely useful. My grammar isn’t very strong, so this exercise helped me check grammar points I’d forgotten”. “The computer can’t detect fixed expressions; it translates them word-for-word, which results in nonsense”. Finally, a number of students focused on specific limitations of MT, namely, that it can’t handle part of speech distinctions, word order shifts, pronoun reference, changes in register, context-dependent senses, polysemy, particle verbs, proper syntactic grouping, tone (e.g. humorous vs. serious), and agreement of discontinuous elements.

The students’ comments generally implied that this exercise made them aware of how important such factors are in any translation task.

Conclusion: pros and cons of teaching pe at a pre-professional level. After reviewing the literature on professional PE training and the results of my in-class PE experiments, I have concluded that having language students perform PE is a doubleedged sword.

On one hand, we have the disdainful (пренебрежительный) reaction of some good students whom the exercise exasperated (раздражать) because of the low-level, repetitive MT errors. Besides, the experiment did not allow time for real training to be comfortable with PE. The result is that students don’t really end up with an accurate view of what PE truly entails.

On the other hand, the experience turned out to be a confidence-building exercise for average and even poor students, who until then had always done translations from scratch. The PE training brought about increases in learners’ motivation, participation and, consequently, higher confidence in expressing themselves. In particular, post-editing drove impressed upon our students the importance of a holistic approach to interpreting the source text and translating by phrase rather than word. The activity also provided them with a taste of what to expect if they undertake a career in translation. From an intellectual standpoint, it gave students insight into the huge challenges which have confronted MT, especially the questions of how to deal with syntactic and lexical ambiguity, non-literal language, and interferencing.