How to teach machine translation post-editing?
Experiences from a post-editing course
Maarit Koponen maarit.koponen@helsinki.ﬁ
Department of Modern Languages, University of Helsinki, Helsinki, Finland
Advances of machine translation technology have in recent years increased its use in various
contexts. In the translation industry, work processes involving the use of machine translated
texts as a raw translation to be post-edited by a translator are becoming increasingly com-
mon. The increasing use of post-editing processes also raises questions related to teaching and
training of translators and post-editors, and institutions offering translator training have started
to incorporate post-editing in their curricula. This paper describes a machine translation and
post-editing course arranged at the University of Helsinki in Fall 2014. From the teacher’s
perspective, we discuss experiences of planning and teaching of the post-editing course. The
development of the students’ experiences and perception of the course contents, machine trans-
lation technology and post-editing are also discussed based on reﬂective essays written by the
students after the course.
The development of machine translation (MT) quality has led to growing use of the technology
in recent years in many contexts. In the professional context, an increasingly common workﬂow
involves the use of machine translated text as a raw translation to be corrected or post-edited
by a translator. Tools and practices for such workﬂows are being developed in large multilin-
gual organizations, such as the European Commission (Bonet, 2013), and by language service
providers who increasingly offer their clients tailored MT systems and post-editing (PE) ser-
vices. Based on a recent survey of 438 stakeholders in the translation and localisation ﬁeld,
Gaspari et al. (2015) report that 30% currently use MT, and the majority (70%) of the MT users
combine that with PE at least some of the time. The use and usability of MT and PE varies
greatly in different countries and language pairs, however.
The increasing use of MT and PE workﬂows in the translation industry has also attracted
interest to how translators and post-editors are, or should be, trained. The need for speciﬁc
training for post-editing has been argued by O’Brien (2002), for example. More recently, based
on a review of various translation industry surveys in addition to their own survey, Gaspari et al.
(2015) argue that there is a growing demand for MT services as well as a growing demand for
expertise in PE skills, and emphasize the impact of “familiarity with translation technology” on
the employability of future translators.
To answer to this changing industry landscape, universities training translators have there-
fore started to incorporate MT and PE in their curricula. In this paper, we describe experiences
from a special course focusing on post-editing, arranged at the University of Helsinki, Finland,
during Fall term 2014. The aim of the course was to introduce translator students to the theory
and practice of MT and PE, covering topics ranging from the technical principles of MT to
PE practice, and research related to PE processes and MT evaluation. This paper presents an
overview of the course contents, learning outcomes and the lecturer’s experiences from plan-
ning and teaching the course. The students’ perspectives are also discussed based on reﬂective
essays written after the course.
The remainder of the paper is arranged as follows: Section 2 presents a brief overview of
the current status of MT and PE use in Finland to provide context for the course described. Sec-
tion 3 presents related work on MT and PE in translator training. Section 4 provides an overview
of the course itself and an evaluation of the practical organization and learning outcomes. Sec-
tion 5 discusses the themes arising from the students’ reﬂective essays as well as issues raised
during the course. Finally, Section 6 presents the conclusions and some recommendations for
2 MT and PE use in Finland
The adoption rate of MT and PE processes naturally varies in different countries and language
pairs. For some language pairs, such as English-Spanish, MT systems have already for some
time been able to provide sufﬁcient quality to make PE a viable alternative. For some other
language pairs, however, MT quality has lagged behind and PE has not been seen as feasible.
Finland is one of the countries where the use of MT and PE has not been particularly
widespread. In large part, the situation is connected to MT quality achievable with Finnish as
one part of the language pair. As a morphologically rich language with relatively free word
order, Finnish has proven difﬁcult for MT systems. For a more detailed discussion of the par-
ticularities of the Finnish language related to MT, see Koskenniemi et al. (2012). The quality
issues are reﬂected, for example, in the European Commission trials, where English-Finnish
translators considered MT at most sufﬁcient to suggest ideas for expressions or not usable at
all and better replaced by translation from scratch (Leal Fontes, 2013). In contrast, for more
successful language pairs like French-Spanish, French-Italian and French-Portuguese, most MT
segments were rated reusable.
On the other hand, also the small market area has attracted relatively little interest in devel-
oping MT systems for Finnish. In addition to free online statistical machine translation systems
like Google Translate, the survey by Koskenniemi et al. (2012) mentions only two rule-based
machine translation R&D projects, one of which did not reach product stage. A recent survey
of 238 Finnish translators found that MT-related skills were not considered important by the
translators themselves, which is likely connected to the unavailability and low quality of MT
systems involving Finnish (Mikhailov, 2015). The report of this survey does not provide exact
numbers on how many respondents used MT systems, but noted that most were only familiar
with free online systems, with only three having used systems other than Google (Mikhailov,
2015, p. 111, endnote 5). Development work is, however, reportedly being carried out by
Finnish language service providers.
3 MT and PE in translator training
As the use of MT and PE workﬂows has increased, research on, and teaching of, skills speciﬁc
to post-editing has become necessary. Part of the skill set is likely to be shared with “traditional”
human translation, such as source and target language proﬁciency, subject area knowledge, text
linguistic skills, cultural and intercultural competence, as well as general documentation and
research skills (see O’Brien, 2002; Rico and Torrej´
on, 2012; Austermuehl, 2013). However, PE
has been found to differ from human translation as well as revision of human translated texts
both in terms of the cognitive processes and the practical goals and processes (Krings, 2001;
O’Brien, 2002). This likely leads to there being skills that are speciﬁc to PE.
In her proposal for PE course content, O’Brien (2002) adds to the general skills mentioned
above also certain a list of speciﬁc PE skills, including general knowledge of MT technology,
terminology management skills, knowledge of pre-editing and controlled language, some pro-
gramming skills, and text linguistic skills. O’Brien (2002) also notes the importance of a pos-
itive attitude toward MT, which is brought up also by many other writers (Rico and Torrej´
2012; Doherty and Moorkens, 2013; Pym, 2013).
Rico and Torrej´
on (2012) discuss their view of necessary PE skills that are divided into
three groups: core competences, linguistic skills, and instrumental competence. Core com-
petences are “attitudinal or psycho-physiological competences”, which relate to dealing with
subjectivity in PE speciﬁcations, client expectations and uncertainty, as well “strategic com-
petence” for reaching informed decisions regarding PE alternatives. Linguistic skills relate to
source and target language, communicative and textual skills as well as cultural, intercultural
and subject area competence, and instrumental competence involves various technical skills
related to understanding MT technologies, terminology management, corpora and controlled
languages, as well as some basic programming skills.
Pym (2013) also discusses the general skills necessary for working with MT. He em-
phasizes the need to “learn to learn”, or learn how to pick up any new software quickly, as
speciﬁc tools and skills related to them soon become outdated due to technological develop-
ment. Related to this is the skill of evaluating the tools on offer. Another necessary skill set is
learning to evaluate the MT suggestions (as well as translation memories) and the usability of
segments, learning to make only necessary changes and to discard suggestions that require too
many changes. Pym (2013) also suggests some speciﬁc revision skills involving detecting and
correcting suprasentential errors (punctuation, cohesion) as well as stylistic revision, working
as a part of a review team, and revising to a speciﬁed quality level.
Post-editing and MT have been included in at least some translation courses, often in
the context of translation technology teaching (for example, Kliffer, 2005, 2008; Austermuehl,
2013; Fersoe et al., 2013; Kenny and Doherty, 2014; Doherty and Kenny, 2014). A commercial
PE training course with assignments in various languages is also offered by the TAUS resource
center 1. In translator training, MT and PE may also be used as part of larger projects: for
example, Shuttleworth (2002) discusses a course involving a large translation project, where
MT and PE were used together with translation memory and terminology tools. Torrej´
Rico (2002) describe a hands-on MT and PE exercise for a translation course. This exercise
takes the form of a full translation assignment, starting from the deﬁnition of the assignment
(client, text, schedule, purpose and style of the translation), followed by an error analysis of the
MT output and pre-editing the ST to produce a new, improved MT version to be post-edited in
accordance with PE guidelines.
Doherty et al. (2012) describe a course consisting of lectures and practical sessions fo-
cusing on statistical machine translation (see also Doherty and Moorkens, 2013). The topics
forming the course content include introduction to MT history and concepts underlying the
systems, MT evaluation, and the role of statistical MT and humans in the workﬂow (includ-
ing pre- and post-processing of MT). Doherty and Moorkens (2013) also report on the student
evaluations of the sessions, as well as themes arising from experiences during the course.
Kenny and Doherty (2014) also discuss some of the speciﬁc things translators should un-
derstand about statistical MT, where human translators ﬁt in the workﬂow and what interven-
tions that may be fruitful. Doherty and Kenny (2014) examine how these aims can be oper-
ationalized and included in translator training syllabus. The syllabus outlined in Doherty and
Kenny (2014) covers both translation memories (TM) and MT, and includes topics such as basic
concepts and implementation, evaluation of tools and MT output, pre- and post-processing, as
well as professional issues like ethics, payment and collaboration.
Flanagan and Christensen (2014) describe a MT and PE course module consisting of two
workshops introducing the translator students to MT and PE, and two PE assignments. The
assignments involved PE according to PE guidelines where two different PE quality levels are
deﬁned (TAUS, 2010), and Flanagan and Christensen (2014) discuss how the translator students
interpreted and followed these guidelines in their assignments.
To the best knowledge of the author, courses speciﬁcally focusing on PE have not been
previously offered at Finnish universities, although PE exercises have been incorporated as parts
of translation or translation technology courses. One example of such an exercise is discussed
by Salmi and Koponen (2014). The exercise involved a for the students to translate a passage of
English newspaper text into Finnish using two different MT systems, then compare the results
and post-edit the version they considered easier to edit.
4 Course description
The course on post-editing was arranged at the University of Helsinki in the fall term 2014. It
is an intermediate level course offered as part of the Translation Studies curriculum, and can
be taken either by Bachelor’s level students as part of their minor, or by Master’s level students
as a specialized course. The practical organization of the course consisted of seven two-hour
lectures, ﬁve homework assignments and some in-class exercises, and a reﬂective essay written
at the end of the course. The Moodle platform was used for distributing materials and collecting
The following subsections describe the students who took part in the course, the objectives
of the course and the topics covered, and the practical exercises.
The course was attended by 15 students, 13 of whom completed the ﬁnal reﬂective essay. Trans-
lation Studies courses offered at the University of Helsinki are not limited to translation in
any speciﬁc language pair or students of speciﬁc languages, but rather intended to support the
studies of various language and translation-related majors. Therefore, the students’ study back-
grounds included various language and translation subjects (English, German, Russian, Swedish
and Nordic languages, Baltic languages). One of the students was from a more technical back-
ground, majoring in Language Technology. The language of teaching in the course was Finnish.
The students were native speakers of Finnish, with the exception of one non-native speaker, also
ﬂuent in Finnish. All were ﬂuent in English.
4.2 Course objectives and topics
The course aimed to introduce the students to the use of MT and PE in the translation industry,
as well as research in the post-editing ﬁeld. The course combined both theoretical background
and practical exercises. The topics covered during the course include:
•Theory and history of MT and PE
•Practical use of MT and PE
•Controlled language and pre-editing for MT
•Post-editing without source text
•Post-editing process research
•Post-editing quality levels and guidelines
•MT quality evaluation and PE effort
In line with some of the recommendations and proposals discussed in Section 3, it was
considered important to provide the students with knowledge of the theoretical principles of
MT, as well as historical and practical context to the use of MT and PE. In particular, the outline
can be compared to Doherty and Kenny (2014). The second topic, practical use of MT and PE,
addressed also some of the professional issues brought up by Doherty and Kenny (2014), such
as ethical questions and compensation in relation to the use of MT and PE. No lectures or
practice on programming in general or building an MT system as such were included due to
the limited course time and resources. On the theoretical side, we also considered it useful to
provide the students with an overview of research concerning PE processes and competences,
particularly the commonalities and differences compared to translation.
The theoretical material covered during the course was supplemented with ﬁve homework
assignments involving post-editing machine translated texts and evaluating the quality of the
machine translations, as well as some practical in-class exercises. The languages used in the
exercises were mainly Finnish and English, with most post-editing done from English into
Finnish. For some exercises, the students were offered options for using some other language
pairs in the exercises.
4.3 Course assignments
The course assignments, as well as the texts, MT systems and other tools used are discussed be-
low. Some general observations about the assignments and students’ comments are discussed in
Section 4.5. As part of each assignment, the students were also asked to write a brief comment,
which they could then later use as a starting point for their reﬂective essays.
Comparison of MT versions: The objective of the ﬁrst assignment was to familiarize the
students with different types of MT systems: rule-based MT and statistical MT. The theoretical
background to these MT principles was discussed in the lecture preceding the assignment. To
observe the differences in action, the students were given two MT versions of the ﬂu treatment
text, provided as MS Word documents. One MT version was produced using the the a statistical
system (Google) and one by a rule-based system (Sunda). The students were asked to compare
the two versions, examine their differences, and then post-edit the version they considered easier
to edit. For this exercise, the students were instructed to produce a translation that could be
published on the Internet for any reader looking for information on the ﬂu, but no explicit PE
guidelines were given.
Pre-editing: The objective of the second assignment was to introduce the concept of pre-
editing. The students were given a brief overview of potential issues and a summary of com-
monly used English Controlled Language rules. They were asked to take at least ﬁve of the most
problematic sentences from the text used in the previous assignment, try rewriting the sentences
following the Controlled Language rules presented, and translate the rewritten versions with
the MT systems used in the ﬁrst assignment. The option of machine translating the sentences
to some other target language was also given. As a further in-class exercise, the students were
also able to use the terminology editor included in the Sunda tool in order to test the effect of
user-speciﬁed terminology on the MT output.
PE without source text: The third assignment involved a scenario where the post-editor
has no access to the source text. The students were introduced to research investigating the
potential for such a scenario and its possible practical uses. For the assignment, they were given
only the raw Finnish MT version of the dish washing liquid text, translated by the European
Commission system. They were instructed to edit it so that it was grammatical and conveyed
the meaning as they understood it; publication quality was not required. The text was again
provided in a MS Word document. The students were also asked to track and report their own
time use in this assignment.
Quality levels: The objective of the fourth assignment was to introduce two different PE
quality levels. The students were provided with the TAUS guidelines (TAUS, 2010), where two
quality levels are deﬁned. For the assignment, they were asked to post-edit two text passages,
one according to the guidelines for “good enough” quality (semantically correct, no informa-
tion added or omitted, no offensive or inappropriate content, correct spelling, corrections for
stylistic reasons or ﬂuency not required) and the other according to “human translation qual-
ity” (grammatically, syntactically and semantically correct, correct terminology, no information
added or omitted, no offensive or inappropriate material, correct spelling and punctuation). For
this assignment, the two press release passages were used. The students were divided into two
groups and the text passages were alternated so that one group edited passage A according to the
guidelines for “good enough quality” and passage B according to the guidelines for publication
quality, and the text passages were reversed for the other group. This assignment was carried
out using the Appraise tool, which also records PE time per sentence.
In class after the assignment, the students were able to compare their own PE time data
to others in the class. They were also introduced to the use of edit distance metrics like HTER
(Snover et al., 2006), which compare the number and type of changes made during post-editing.
The students then examined their own edits using the metrics in Asiya Online Toolkit.
MT quality evaluation: The objective of the ﬁfth assignment was to introduce the students
to the topic of MT quality and evaluation, and to have them consider the distinction of ﬂuency of
language and adequacy of meaning in translation. They were provided with the LDC guidelines
for evaluating ﬂuency and adequacy (LDC, 2005), and asked to rank different MT versions of
sentences either based on ﬂuency or adequacy. For this assignment, the tourist phrases were
used, and the evaluation was carried out using the Appraise tool. The students were offered
different possible evaluation tasks in Finnish, Swedish, French, or German.
4.4 Evaluation of the practical organization of the course
Some issues related to planning and teaching the class can be noted. The course was planned to
include many different types of practical exercises to provide the students with a wide view of
different topics. On the other hand, the limited course time made it difﬁcult to deal with many
issues in depth, and some of the topics remained at a rather superﬁcial level. The selection of
texts and tools was complicated by the fact that the translator students’ main working language
differed, and providing them all with materials speciﬁc to their own language pairs was not
feasible within the scope of this course. The language pair English-Finnish was settled on for
most exercises, as these were languages in which all had at least a good level of proﬁciency.
For texts to be edited during the course, we aimed to select relatively general interest texts
that would not require particular expertise. One text consisted of 17 sentences (263 words)
extracted from a longer informative text with instructions for treating the ﬂu, intended for the
general public. The second text consisted of 11 sentences (281 words) taken from a European
Commission text on requirements for dish washing liquids. The third and fourth texts were
extracts from English EU press releases. For the evaluation task, machine translated sentences
were taken from the tourist phrasebook dataset described in Rautio and Koponen (2013). The
set contains short, simple sentences involving asking directions, making purchases and similar,
translated using different MT systems.
The selection of MT systems was limited by the availability of systems for Finnish. In
the end, translations used in the PE exercises were produced with three systems: the statistical
system developed by the European Commission, a rule-based system by Sunda Systems2, and
Google Translate3. Texts translated with the European Commission system were included to
represent perhaps the most likely intended scenario for PE: using an in-house system to translate
specialized texts. The Sunda system is customizable through the use of user-speciﬁc terminol-
ogy, and is reportedly used by some Finnish freelance translators and companies. Google was
included mainly due to its familiarity and easy availability, although its usability for real-life PE
scenarios is uncertain.
In terms of tools used, two of the PE tasks were given to the students in MS Word document
format. One of the PE assignments and the evaluation task were carried out using the Appraise
evaluation tool (Federmann, 2012). Although this tool is mainly intended for MT evaluation
rather than PE, it provides basic functionality for post-editing sentences, and records PE time,
for example. As an additional tool, a class exercise utilized the Asiya Online Toolkit (Gonz`
et al., 2012), which is an online interface for using various automatic MT evaluation metrics.
The initial plan for the course had been to use a translation memory software for carrying
out the assignments during the course. However, due to unexpected technical issues combined
with the fact that surprisingly, most students indicated they had very little practical experience
with the use of TM tools made their use not feasible during this course. This situation reﬂects
previous observations (Doherty and Moorkens, 2013; Doherty and Kenny, 2014) about the im-
portance of technical resources for teaching, and the challenges related to them.
4.5 Evaluation of learning outcomes
The objective of the PE course module described was to introduce the students to the use of MT
and PE in the translation industry, as well as research in the post-editing ﬁeld. The course aimed
to provide the students with general knowledge of MT principles, the use of PE in practical
scenarios and related professional issues, and related issues such as controlled language and
pre-editing. Further, the goal of the course was to foster a positive attitude towards technology,
together with the ability to critically evaluate the tools and processes, both of which have been
found important (O’Brien, 2002; Pym, 2013; Kenny and Doherty, 2014). This section discusses
these objectives in light of the class discussion as well as the reﬂective essays written by the
students after the course. The reﬂective essays are discussed in more detail in Section 5.
Based on the ﬁrst assignment where the different MT versions (rule-based and statistical)
were compared, all students appeared to have grasped the basic principles of MT, and the dif-
ferences between these two approaches. They were able to analyze the types of errors found
in each version and reﬂect them against their understanding of the type of MT technology in
question. In general the students found both versions surprisingly understandable, although not
perfect. Some said there was “not much” to edit, that the MT provided a surprisingly good start-
ing point, and that they felt post-editing had saved time and was not particularly challenging.
Others, however, felt that they probably would have translated the text at least as fast.
The students were also able to assess the use of MT and PE critically. Most students
doubted particularly the feasibility of PE without source text, based on their experience. Al-
though they considered the text relatively understandable, they found it hard to trust the MT,
and many pointed out that it might have been faster to translate the text using the source text.
The students also did not consider pre-editing particularly beneﬁcial. Although the application
of some of the rules improved some sentences, or parts of sentences, many errors remained, and
many students commented that it seemed rather futile to spend time pre-editing if one still had
to do post-editing. They also felt that the introduction to controlled language in the preceding
class had not been detailed enough.
The students also appeared to gain awareness of their own editing processes, and evaluate
their own work. Most students commented that they found it somewhat difﬁcult to differentiate
between the PE quality levels, and to determine what changes were necessary for the “good
enough” quality. Examining their own corrections, many commented that they likely were
correcting “too much”. The in-class exercise examining the students’ own PE times and edits
was overall very well received. The students commented that it was interesting and informative
to compare the PE times and examine their own corrections after the fact. The students also
commented that the evaluation task, where they rated MT quality focusing on language only or
meaning only made them more aware of the issues.
Based on the course assignments and the students’ reﬂective essays, the course appears to
have succeeded in introducing them at least to the basic principles of MT and PE. The students
had a positive attitude towards the tools and workﬂows, but also evaluated them critically. As
the course described here was a relatively short (7 weeks), many of the topics could only be
discussed on a relatively superﬁcial level. The PE assignments were also relatively short, and
more extensive practice will naturally be required to make the students ready for PE work.
Nevertheless, the students’ comments also show an awareness of their own editing processes,
which will hopefully help them to further develop their skills.
5 Themes identiﬁed in the students’ reﬂective essays
At the end of the course, the students were asked to write a reﬂective essay on the central ideas
and experiences during the course. They were instructed to write a 1000–1500 word essay
discussing the theoretical background covered during the course and reﬂecting on their own
experiences and observations from the assignments. As a starting point for the essay, they were
asked to reﬂect on the following questions:
1. How did your understanding of the use of MT post-editing and related phenomena devel-
oped during the course?
2. What beneﬁts and opportunities do you see in the use of MT and PE from the perspective
of a translator, an organization requiring translation services, an individual MT user? What
about problems or limitations?
3. How has the ﬁeld developed and how do you believe it will develop in the future?
In addition, the students were instructed to raise other issues based on their readings and
observations. They were also asked to comment on the course content, practical organization
and potential improvements.
These reﬂective essays were analyzed for themes raised by the students. Similarly to the
qualitative evaluation discussed in Doherty and Kenny (2014), extracts of the topics and ob-
servations discussed by each student were compared to determine whether multiple students
brought up similar issues, and commonly occurring topics were grouped into larger themes.
The themes related to the students’ expectations and attitudes toward MT and PE, observations
on MT quality and usability, PE processes, PE levels and requirements, the students’ visions for
the future of translation and PE, and training. The following subsections discuss these themes
in more detail.
5.1 Expectations and attitudes
One clear theme commented on by the students related to their expectations and attitudes before
the course, and some of the ways they changed. About half (7 out of 13) of the students stated
that they had very little knowledge about MT and PE before the course. Most were at least
somewhat familiar with free online MT systems, but as one student stated, the experience was
mainly limited to laughing at examples of silly MT errors. One of the students had some prior
theoretical and practical knowledge of MT, and another mentioned that the basic principles had
been covered in a prior course taken at another university. One student had worked in a company
where MT was in use, although she had not used it in practice.
In terms of attitudes, most did not expect the quality of MT to be particularly high, and
had not assumed it would be used, or usable, in professional contexts. Three of the students
speciﬁcally described their own attitude toward MT as negative at the beginning of the course.
Only one student mentioned having a generally positive attitude toward MT and PE. During the
course, their perception of MT apparently changed, with most commenting that they saw more
potential in MT after the practical experiences of the course.
Student attitudes was also one of the central themes identiﬁed by Doherty and Moorkens
(2013), and in general attitude toward MT has been considered important in MT and PE related
training. In this sense, the course appears to have succeeded in fostering a positive attitude, as
most of the students commented on seeing MT in a more positive light after the course.
5.2 MT quality and trust
As noted in the previous subsection, most of the students did not have very high expectations of
MT quality. In fact, most of the students (10 out of 13) explicitly stated that the quality of the
machine translations handled during the course was better than they had expected. This could
already be seen in the ﬁrst assignment, where students commented that although neither version
was “publication-ready”, containing some unintelligible parts, both were more understandable
than they had expected. Their comments were even more positive concerning the texts translated
using the European Commission system.
On the other hand, although positively surprised by the readability, the students found it
hard to trust the MT. This point, which was raised in the essays by more than half of the students
(8 out of 13), had also come up in class in connection with the PE assignment without source
text. The students pondered whose responsibility the correctness (with regard to meaning)
would be, and stated they would not feel secure taking responsibility for the ﬁnal translation.
The trust issue was not limited to this task, and more than one student discussed the slowing
effect of constantly having to check the MT version against the source text, looking for potential
It should be noted that although the students appeared to be even somewhat impressed with
the quality of the MT used, all the texts still contained various issues, as MT quality into Finnish
remains far from ﬂuent. Similarly to the theme of quality thresholds discussed by the students
in Doherty and Kenny (2014), the student of this course also identiﬁed situations where the MT
quality was likely too poor for any intervention to be fruitful. The MT texts, however, appear to
be at the level deemed most usable for teaching purposes by Kliffer (2005): not bad enough to
contain such frequent low-level errors that PE would be completely pointless, but not such high
quality that there would be hardly anything to correct.
5.3 PE process and time
Half of the students (7 out of 13) discussed the unfamiliarity of the PE work compared to their
prior experience with translation, and stated that it took time to get used to the idea of working
with a “raw” translation. The difﬁculties mentioned involved, for example, getting used to
the MT differing so greatly from a raw version they might create themselves, and having to
deal with two sources, the source text and the MT. Some also commented that the sentence-
by-sentence way of working imposed by the Appraise tool was difﬁcult, while others found it
intuitive. One even stated he started to quite like using the tool.
As could be seen in the third assignment, where the students tracked their own time, their
speed varied greatly. Of the 13 who completed the assignment, 4 had taken less than 30 minutes,
5 had taken between 30 and 60 minutes, and 4 students had taken more than 60 minutes to edit
the relatively short passage. Some of the differences in PE speed may of course be related to
the students’ overall language skills, as the language pair English-Finnish did not represent the
main working languages for all of the students. A more detailed analysis of the relationship
between language skills, edits and PE times was not within the scope of this paper, but might
be informative in the future.
Based on their class comments and the reﬂective essays, they also differed in how they
viewed their time use. Some stated that once they started to feel more comfortable, they felt
they were saving time compared to translation from scratch. One student estimated that PE
probably took only about one third of the time that translating the same text passage would
have. On the other hand, not every student felt the PE was making them faster. One of the
students even expressed some frustration and feeling she was still working too slow and making
The students’ observations can be compared to the ﬁndings of the survey reported in Salmi
and Koponen (2014), where most stated they either would have translated the same text faster
or that there was no difference in time. Only 10 of the 49 students answering that survey felt PE
made them faster. In studies where speed in PE versus translation has been explicitly compared,
ıa (2010, 2011) have indeed found that for translation students, PE was only marginally
faster. These experiments, however, were conducted with students who had no training in PE,
so partly this may be due to unfamiliarity.
Interestingly, one student explicitly connected her PE speed to the amount of experience,
noting that the PE probably speeded up her work speciﬁcally because she is not yet a particularly
experienced translator. She wrote in her essay that if her translation processes had been more
established, she might have found it more difﬁcult to work with the MT. No detailed analysis
was performed on the students’ edits and it is therefore difﬁcult to say how proﬁcient this
particular student was, but her self-assessment reﬂects the observation by Garc´
ıa (2011) that
the “poor” students beneﬁted more from MT, and that one possible explanation was that they
had not yet formed effective translation skills. On the other hand, as Garc´
ıa (2011) points out,
these students could also have some other characteristics that make them particularly suited for
5.4 Adjusting to PE quality requirements
As with MT and PE in general, the students indicated that they had relatively little knowledge
of the different use scenarios and quality levels for PE. Particularly the idea of “good enough”
quality was new to them, and most (9 out of 13) explicitly commented that they had a hard time
adjusting to a task where such quality was required. Thinking of the translation as a text that
needed to be simply sufﬁcient to convey the meaning needed some adjustment. However, most
indicated that after the course, they now had a better understanding of differing situations where
a lesser quality of language might sufﬁce. Most also explicitly mentioned that they considered
it important to learn to adapt to different requirements.
Most of the students (7) who raised this theme also explicitly connected it to their prior
translator training, where the need for “perfect” language is generally emphasized. They found
it difﬁcult to accept a translation that was “less than perfect”. The students appeared to react to
this in two ways. Some wanted to work on the translation until they were happy with the ﬁnal
version, but recognized that they were probably “editing too much”, which was also connected
to their observations on time use. Referring speciﬁcally to the ﬁrst PE assignment, one student
stated her ﬁnal version should properly be called “inspired by the MT” rather than “post-edited”.
Other students decided to try to manage only with minimal corrections, and not try to make the
text their “own”, but expressed concerns about how the often unidiomatic MT affected their
The concerns about quality of the ﬁnal translation are reﬂected, for example, in the survey
reported by Salmi and Koponen (2014), where many of the students expressed similar issues.
Their apparent desire to edit the ﬁnal translation to be as perfect as possible, even if it meant
rewriting, however, contrasts with the ﬁndings of Depraetere (2010). Based on the analysis of
the corrections made by students in a PE project, Depraetere (2010) reports that the students
“did not feel the urge to rewrite it”, and did not rephrase the text if the meaning was clear.
The observations regarding difﬁculty of adjusting to the quality levels can also be compared to
Flanagan and Christensen (2014), whose students also discussed making potentially “unneces-
sary” stylistic changes, even when the guidelines called for “good enough” quality. Similarly
to those students, our students struggled with adhering to the guidelines related to style. As
noted by Flanagan and Christensen (2014), it may also be that the guidelines themselves are not
5.5 Future visions
Since the essay instructions had speciﬁcally asked the students to reﬂect on potential future
visions, all of the students did discuss views regarding MT and PE both as a general practice
and in their future careers. Of the 13 students, 11 students explicitly stated they believed the
use of MT and PE would increase. In contrast with their rather low initial expectations for MT,
and even self-professed negative opinions before the course, most seemed to have a much more
positive outlook for such future. One student who described her view as very negative before
the course even stated she had completely changed her mind on MT. Over half (8 out 13) of the
students explicitly stated that they viewed MT as a tool, not a threat for them as translators.
Nearly all, however, also discussed the limitations of PE, particularly of any scenario in-
volving PE without source text. The essays also reﬂected a clear understanding that MT and PE
are not feasible in every situations, which is an important point related to the need of helping
students become also critical users of technology, as argued by Kenny and Doherty (2014).
Some of the students also explicitly commented on how they saw PE as a potential career
prospect. Five of the students stated that they could well see themselves doing PE work in the
future. Two commented that they would not be interested in PE work; one speciﬁed that she
was mainly interested in literary translation, where MT is likely to be of little use. The other six
students did not comment on PE in terms of their own future plans.
5.6 PE training and potential improvements to the course
Nearly all of the students also explicitly commented on MT and PE related training. In general,
they considered the topics and competences covered important, even essential for translator
training. Even the students who stated that they did not consider PE an attractive career option
for them personally felt the course would overall beneﬁt translator trainees. One of them noted
that it was good to know what the work is like, so she can better decide whether to accept
potential jobs offered in the future.
As this course was the ﬁrst time a speciﬁc PE course was taught, it was very important to
hear the students’ views on how the course contents and practical organization worked. Most
stated they were overall very satisﬁed with the course and found the topics interesting. The
chance to examine one’s own PE time data and edits was mentioned as a particularly interesting
and useful exercise. Most commonly mentioned improvement was that future courses should
include options for more varied language pairs. Some students also hoped that more concrete
examples would have been used in introducing topics like controlled language and PE levels,
and that there would have been more detailed discussions of speciﬁc examples from the edited
texts. With regard to speciﬁc topics, many found pre-editing least interesting and useful. Some
students did not ﬁnd PE without source text particularly useful, whereas one hoped even more
class time had been spent on that topic.
6 Conclusion and future teaching
In this paper, we have described experiences from a translator training course focusing on MT
and post-editing. The course was taught for the ﬁrst time during fall term 2014 at the University
of Helsinki, Finland. After a brief overview of the context of MT and PE use in the country,
and an overview of related work on teaching MT and PE, we have presented a description of
the course contents, practical organization and assignments intended to introduce the students
to various theoretical topics and practical issues in the ﬁeld.
Common themes arising from reﬂective essays written by the students after the course
were also identiﬁed and discussed in combination with observations and comments from the
assignments and class discussions. The central themes identiﬁed involved students’ expecta-
tions and attitudes, MT quality and trust, the PE process and time use, adjusting to PE quality
requirements, future visions regarding PE, and PE training. A particularly interesting result was
seen in the students’ attitudes, in that their generally low expectations and even negative atti-
tudes toward MT before the course appeared to change to a much more positive view. However,
they also expressed rather realistic views on the potential and limitations of MT.
The materials and tools selected for the course appeared mostly suitable for the purpose.
Although MT quality into Finnish still lags behind many others, the systems used were able to
produce translations of the source texts selected with high enough quality to make PE feasible.
Two of the MT systems used were known to be used in practice by at least some translators
working in the ﬁeld, which was considered important for a realistic view of the situation. When
selecting the MT systems to be used, Google was included due to its familiarity and availability
to the students. Interestingly, the survey by Gaspari et al. (2015) has found that the vast majority
of their respondents who used MT, used free online MT systems like Google or Bing. Although
it is difﬁcult to say whether that includes translators working into Finnish, there is also anecdotal
evidence pointing to its use in this country.
Some practical and technical challenges were also observed. The most important chal-
lenges relate to the fact that the course is offered to students in all language and translation
subjects, which makes it difﬁcult to provide materials for all the language pairs they may be
working in. Another issue involved technical challenges, and the inability to use translation
memory tools for the assignments.
The course is being taught again in Fall term 2015. Based on the experiences of the course
in 2014, we have implemented PE exercises in more varied language pairs. To better reﬂect
the practical scenario, most of the assignments are also carried out using a TM system. The
exercise where the students were able to examine their own edits and PE times appeared par-
ticularly useful, and we plan to incorporate this aspect already in the earlier assignments. We
hope that this may help the students observe their own processes more closely over several as-
signments. Together with class discussions, they may be then able to try different approaches
in the following assignments. The ordering of the assignments will also be reconsidered. In
particular, the adequacy versus ﬂuency evaluation task might be more fruitful in the beginning
of the course, as it could assist the introduction of PE requirements and “good enough” quality,
which the students found particularly hard to accept.
The author wishes to thank all the students for participation in the course, and the European Commission,
Sunda Systems Oy, and professor Lauri Carlson for offering tools and materials for use in the course.
Austermuehl, F. (2013). Future (and not-so-future) trends in the teaching of translation technology. For-
o I technologies de la traducci´
o. Revista Tradum`
atica: tecnologies de la traducci ´
Bonet, J. (2013). No rage against the machine. Languages and Translation, 6:4–5.
Depraetere, I. (2010). What counts as useful advice in a university post-editing training context? Report on
a case study. In EAMT 2010: Proceedings of the 14th Annual conference of the European Association
for Machine Translation, St Raphael, France.
Doherty, S. and Kenny, D. (2014). The design and evaluation of a Statistical Machine Translation Syllabus
for translation students. The Interpreter and Translator Trainer, 8(2):276–294.
Doherty, S., Kenny, D., and Way, A. (2012). Taking statistical machine translation to the student translator.
In Proceedings of the Tenth Biennial Conference of the Association for Machine Translation in the
Americas, San Diego, California.
Doherty, S. and Moorkens, J. (2013). Investigating the experience of translation technology labs: peda-
gogical implications. Journal of Specialised Translation, 19:122–136.
Federmann, C. (2012). Appraise: An open-source toolkit for manual evaluation of machine translation
output. The Prague Bulletin of Mathematical Linguistics, 98:25–35.
Fersoe, H., Hansen, D. H., Offersgaard, L., Olsen, S., and Povlsen, C. (2013). Let’sMT! as a learning
platform for SMT. In Proceedings of the XIV Machine Translation Summit, pages 353–359, Nice,
Flanagan, M. and Christensen, T. P. (2014). Testing post-editing guidelines: how translation trainees
interpret them and how to tailor them for translator training purposes. The Interpreter and Translator
ıa, I. (2010). Is machine translation ready yet? Target, 22(1):7–21.
ıa, I. (2011). Translating by post-editing: Is it the way forward? Machine Translation, 25(3):217–237.
Gaspari, F., Almaghout, H., and Doherty, S. (2015). A survey of machine translation competences: in-
sights for translation technology educators and practitioners. Perspectives: Studies in Translatology,
alez, M., Gim´
enez, J., and M`
arquez, L. (2012). A Graphical Interface for MT Evaluation and Error
Analysis. In The 50th Annual Meeting of the Association for Computational Linguistics.
Kenny, D. and Doherty, S. (2014). Statistical machine translation in the translation curriculum: overcom-
ing obstacles and empowering translators. The Interpreter and Translator Trainer, 8(2):295–315.
Kliffer, M. (2005). An experiment in MT post-editing by a class of intermediate/advanced French majors.
In Proceedings EAMT 10th Annual Conference, pages 160–165, Budapest, Hungary.
Kliffer, M. D. (2008). Post-editing machine translation as an FSL exercise. Porta Linguarum, 9:53–68.
Koskenniemi, K., Lind´
en, K., Carlson, L., Vainio, M., Arppe, A., Lennes, M., Westerlund, H., Hyv¨
M., Bartis, I., Nuolij¨
arvi, P., and Piehl, A. (2012). Suomen kieli digitaalisella aikakaudella – The
Finnish Language in the Digital Age. META-NET White Paper Series. Georg Rehm and Hans Uszkoreit
(Series Editors). Springer. Available online at http://www.meta-net.eu/whitepapers.
Krings, H. P. (2001). Repairing texts: Empirical investigations of machine translation post-editing pro-
cess. The Kent State University Press, Kent, OH.
LDC (2005). Linguistic Data Annotation Speciﬁcation: Assessment of Adequacy and Fluency in Trans-
lations. Revision 1.5. Technical report, Linguistic Data Consortium. http://www.ldc.upenn.
Leal Fontes, H. (2013). Evaluating Machine Translation: preliminary ﬁndings from the ﬁrst DGT-wide
translators’ survey. Languages and Translation, 6:10–11.
Mikhailov, M. (2015). Minor language, major challenges: the results of a survey into the IT competences
of Finnish translators. Journal of Specialised Translation, 24:89–111.
O’Brien, S. (2002). Teaching Post-editing: A Proposal for Course Content. In Proceedings of 6th EAMT
Workshop Teaching Machine Translation, pages 99–106, Manchester, UK.
Pym, A. (2013). Translation skill-sets in a machine-translation age. Meta, 58(3):487–503.
Rautio, J. and Koponen, M. (2013). MOLTO evaluation and assessment report. Technical report, MOLTO
Rico, C. and Torrej´
on, E. (2012). Skills and Proﬁle or the New Role of the Translator as MT Post-editor.
o, canvi de paradigma? Revista Tradum`
atica: technologies de la traducci´
Salmi, L. and Koponen, M. (2014). Machine Translation, Post-Editing and Respeaking: Challenges for
Translator Training. In Man vs. Machine? The Future of Translators, Interpreters and Terminologists.
Proceedings of the XXth FIT World Congress Vol. I, pages 138–145, Berlin, Germany.
Shuttleworth, M. (2002). Combining MT and TM on a Technology-oriented Translation Masters: Aims
and Perspectives. In Proceedings of 6th EAMT Workshop on Teaching Machine Translation, pages
123–129, Manchester, UK.
Snover, M., Dorr, B., Schwartz, R., Micciulla, L., and Makhoul, J. (2006). A Study of Translation Edit
Rate with Targeted Human Annotation. In 7th Conference of the Association for Machine Translation
in the Americas, pages 223–231, Cambridge, Massachusetts.
TAUS (2010). MT Post-editing guidelines.
on, E. and Rico, C. (2002). Controlled translation: a new teaching scenario tailor-made for the
translation industry. In Proceedings of 6th EAMT Workshop Teaching Machine Translation, pages
107–116, Manchester, UK.