In November 2021, we conducted an online survey of Hungarian translators on the acceptance and spread of machine translation (MT) and how and why machine translation post-editing (MTPE) is used in this profession. In the survey, we collected exactly 150 responses which were analyzed in detail in a separate publication which has been uploaded to our website.
However, we think that the main takeaways on this subject are worth a blog article as well. Let’s see what the main trends and opinions are on the current state of machine translation in Hungary.
Spread and acceptance
The first question in the survey investigated whether the respondents have already worked on machine translations.
69% of them already have, of which 41% do so rarely and 28% regularly. So, the majority of translators already have some MTPE experience. Those who responded ‘no’ were taken immediately to question 7. For those that answered ‘yes’, the next question was about the reasons they took such work and the type of client. 78% of them received the task from a translation company, but 14% of these decided to use MT voluntarily, not at the client’s request. Those that perform MTPE tasks for direct clients (20%) usually do so for their own reasons and not at the client’s request.
The question on whether most of the clients encourage, forbid or ignore machine translation showed very mixed results. Although many respondents (exactly 25%) of respondents say that most of clients encourage it, 21% answered that they forbid it, and the most popular answer was that for the time being, they ignore it. The fourth reply, that approximately half of them encourage it and half of them forbid it, also received 25%.
The first three questions show us that many more translators have experience with MT than do not, and while this tendency is due to translation companies, many of them forbid its use. Moreover, the proportion of translators who use it voluntarily is significant, at around 30%.
MT providers and quality
The question about which MT engine translators primarily use showed a 15% share of “I don’t know” responses. The rest of the answers were in a clear order: DeepL is the most popular engine at 41%, and Google Translate comes second (25%). Their combined share is around two thirds. The popularity of Amazon is very low, and the ‘other’ replies included some mentions of SDL NMT.
With regards to quality of MT, the majority of translators say it is “good enough”. This includes two different answers, namely “It is good enough to speed up my work” and “It is just good enough to be usable” (39% and 30% of all responses, respectively). The next runner-up was “It is very uneven”, with the remaining choices suggesting either that MT is almost perfect or completely useless for translators at around 6%.
Traits, advantages, and disadvantages of MT
For these questions, multiple choices could be selected by the respondents, so the percentages should be interpreted as a proportion of the total number of replies and not the share of respondents. First, we asked the opinion of the translators about post-editing compared to translation and revision. The most chosen option was “It is much like revising a bad translation” (35%), followed by “It is simpler than translation, because you only have to correct errors” and “I prefer translation” tied in second place (28% each). The rest of the answers underline that more translators prefer revision to post-editing than the other way around, and that very few of them like it better than translation. So, even if MTPE is accepted, overall it is less liked than both translation or revision.
The next question sheds light on the future: if only tasks of this type were available on the market, 41% of respondents would remain in the profession, but 13% would leave it, 27% would work part time, and the rest cannot decide.
In the question about the biggest advantages of MT the order is as follows: “It helps people who do not speak languages to understand a text written in a foreign language” (27%), “It does not speed up the work, but it sometimes helps in finding a solution” (21%), “It speeds up the work of translators” (18%), “It can help replace human translation in a situation when there is no time/possibility/budget for it” and finally “It is a quick aid in translating words or expressions”. So, unlike with translation memory, the upside of MT is mostly seen in uses other than speeding up the work.
The order of the biggest disadvantages is as follows: “It commits annoying or difficult-to-notice errors” (21% of all answers), “Its quality is not good (yet) in my working language(s)” (19%), “The suggested translation gives me a direction which is not always leading to a solution” (18%) and “It takes away the creativity from translation” (17%). The least chosen replies include “It does not pay as well as translation” (12%) and “It slows down or does not really speed up my work” (8%).
Finally, we asked the respondents about the most important change they would like to see in machine translation. Multiple choices were possible, and the order is as follows: “If it remained an option for less creative/important text types” (48%), “Better prices” (26%), “Almost perfect quality” (12%), “Better deadlines” (9%) and “Other” (5%). It is interesting to see these replies, especially if we compare them to the input from translators on how much they like post-editing versus translation and revision.
In this blog article, we wanted to give a general, overall picture about the subject – the detailed results and diagrams can be viewed in the PDF summary. Our survey has given some valuable insight on the topic to those who are interested. The spread of machine translation is wider than we expected, but its acceptance is very mixed among clients and translators alike. Its quality can be considered both acceptable and unsatisfactory, and the majority, even if they use it, treat it with caution and would like to have it as an option.
This survey should be treated as a specific snapshot at a given time, about a subject that is changing very quickly. We are sure that just a few years ago, or a few years from now, the results would be very different. We suggest all our readers download the full summary of the survey. Any feedback is welcome in any form.