The Impact of Machine Translation Postediting on Translation Industry: A

Qualitative Research Synthesis

Wagdi Rashad Ali Bin-Hady, Department of English, Hadhramout

University, Yemen
wagdyrashad@hu.edu.ye
https://orcid.org/0000-0002-4116-113X

Abstract

Artificial intelligent applications (AIA) flock into our daily life including education. Such
widely used applications are Google translate and ChatGPT. However, their wide influx
into language learning has made many EFL practitioners look at such applications with
inferior sights due to their inability to perceive the benefits that such applications provide
to the users. This qualitative research synthesis aims to gauge the features of postediting
and the criteria used in evaluating machine translate quality. It also examines the
characteristics of postediting machine translation outputs. The study explored the findings
of 20 articles published between 2015 and 2023. Findings reported that postediting
encompasses four features: translation approaches, profession, skills and tasks.
Postediting quality was evaluated according to the translation acceptability, postediting
time and postediting efforts. Finally, postediting outputs includes comparison between
neuro machine translation and machine translation segment and machine translation and
human translation. The study confirmed the importance of using MT. It recommended
training EFL students with postediting skills to enable them to use the advantages of
machine translation in the translation industry.

Keywords: Evolution, neuro machine translation (NMT), postediting quality, translation

outputs.

Introduction

However, the countless advantages that technology supports our era in various fields,
some researchers believed that technology does not make everything better or viable
(Pym, 2011). Regarding the use of artificial intelligence, (AI) in translation or what is
called machine translation (MT), it is believed that humans are still far away from the
ability of programming software that would be able to produce intelligible translation in
various language pairs and genres (Eszenyi & Dóczi, 2020). However, MT has got a
considerable recognition lately on translation industry (Kasperė et al., 2023). This
positive recognition is proved by the huge number of use which reached 140 billion
words daily translated by Google translate (Hu et al., 2020). MT also used in different
fields and by different users in work, search and housework (Kasperė et al., 2021).
Therefore, it is irrational and unwise to prevent using such advantages (Povilaitienė &
Kasperė, 2022). Many critics dislike machine translation due to the repetitive errors it
produces which no human translators may commit (O’Brien, 2002). Translators who have
not been trained on machine translation post-editing (henceforth MTPE) could not
professionally do the post-editing tasks due to the many skills and competencies required
to be both practically and theoretically a translator be acquainted with (Aranberri, 2017).

Statement of the problem

MTPE is perceived to take high portion of the translator's work in the future (Koponen,
2016). This study explored the state-of-the-art on status of MT when accompanied with
postediting on end users or readers. Therefore, no matter who produces the translation,
whether human or machine, translation is evaluated according to its quality and
acceptability. Previous research on MTPE was shaped by experiment design, (e.g.,
Aranberri, 2017; Kasper˙e et al., 2019), or quantitative design (Yamada, 2019). There is a
lack of research which focuses on qualitative synthesis. Therefore, this research aims to
bridge a methodological gap in state-of-the-art research on MTPE to verify these
questions:

Research questions

1. What are the features of postediting?

2. How is the machine translate quality evaluated?
3. What are the characteristics of postediting machine translation outputs?

Literature review

Machine translation

A few studies have been given to translators on what to do with MT raw products and
how to improve such translation obtained by machines (Aranberri, 2017). Furthermore,
measures were established to evaluate the quality of MT, (Koponen, 2010; Rossi & Carré,
2022). Other studies compared the acceptability of MT outputs to human translation
(Eszenyi & Dóczi, 2020). However, the fundamental role of post-editing in MT,
translation trainees had not been taught such skills (O’Brien, 2002). Stoyanova-Georgieva
(2021) highlighted the viable role of postediting training course for developing neural
machine translation (NMT) outputs in business. MT has passed by many phases
beginning by statistical machine translation (SMS) and reached to NMT.

SMT, as the word 'statistical' donates, refers to the use of statistical methods to translation
form one language into another using corpus. SMT, as Lopez (2008) perceived depends
on "algorithms automatically learn how to translate" (p.1). SMT dominated the field of
MT studies for two centuries (Hearne & Way, 2011). Plethora of research studies
reported that NMT is more efficient than phrase machine translation (PMT) or using
transition memory when referring to the post editing less efforts and translation outputs
efficiency (Sánchez-Gijón et al., 2019; Yamada, 2019). Zouhar et al. (2021) stated that
NMT has the privileged over SMT regarding translation quality. NMT has been obtained
by majority of translation service providers because it becomes the state-of-the-art
approach. NMT depends on 'artificial neural network' to predict the context of word. This
approach began to be used by Google translate in 2016. However, the great development
in the field of NMT outputs, it does not reach human parity (Eszenyi & Dóczi, 2020),
therefore, for ensuring the acceptability of NMT, the translator should edit the MT
outputs (Almaaytah, 2022).

Postediting approaches and skills

Post-editing is perceived as a strategy to freeing the machine translating form errors and
make it human sounding (Almaaytah, 2022). Postediting implies the modification of
translation outputs including format, style, terminology and the like mistranslation
(Eszenyi & Dóczi, 2020). There are various strategies and methods used in postediting.
Mrinalini et al. (2016) suggested language identification system along with postediting n-
best translation approach. Such approach enhanced translation by 3-5%. Dunne, (2022)
proposed crowd-source post-edits to MT. This approach is based on full document
context. It identifies the place of errors. The study concluded that crowd-sourcing
annotations is a creditable application for evaluating MT. Various authors suggested
interrelated post-editing strategies which should be applied to ensure the creditability of
the translation products. As postediting is not easy to master, O’Brien (2002)
recommended the importance of teaching postediting skills in two studying semesters to
translation students. He suggested the following skills to be trained on "specialized
translation skills; basic linguistics; basic terminology management; IT skills; an
introduction to language technology (focusing on translation memory tools" (p. 103).
Vardaro et al. (2019) studied how European Commission’s Directorate-General for
Translation identifies and corrects errors in NMT and PE. The study found that they
focused on mistranslation terms or stylistics.

Machine translation quality evaluation

Translation quality is determined according to the reader's satisfactory with the MT

outputs. Translation quality is viewed as "the degree to which the target or output text
meets the needs and expectations of its reader(s) or user(s)" (Castilho et al., 2018, p. 8). It
is also perceived as the "accuracy, suitability, and well-formedness," (AbuSa’aleek, 2016,
p. 185) of the translation outputs. This leads us to translation quality assessment (TQA)
which indicates to "the usability and satisfaction results by end-users insofar as the
implementation of light PE both increased the usability and acceptability of the PE
instructions and led to satisfaction being reported," (Castilho & O’Brien, 2017, p. 120).
Popović (2018) reported that MT may be assessed by a bilingual translator who
understand both source and target language. A scale is produced which ranged translation
quality from one for poor into five for perfect translation. Evaluators focus on meaning
preservation and grammatically fluency.

Pragmatic approach is pertained to measure the quality of MT due to its dependency of

user's usefulness indicators i.e., satisfaction and productivity (Rossi & Carré, 2022).
Koponen (2010) adopted an error analysis for identifying MT errors. The study reported
that texts translated by humans have different errors than texts translated by machines.
Sanchez-Torron and Koehn (2016) studied the impact of bilingual evaluation understudy
(BLEU) algorithm in MT on postediting speed. The study reported that postediting is
linearly impacted by BLUE score. That is to say, MT weak outputs lead to worse
transition quality even after postediting. To make this clearer, translation quality is the
degree of accepting MT outputs by service users. This needs to define the term
acceptability.

Acceptability stands for the level of matchness between the MT product and user'
expectation (Castilho et al., 2018). Castilho (2016) affirmed that measuring MT
acceptability is important to get the impact of translation product on the readers.
AbuSa’aleek (2016) measured the acceptability of four MT systems on translating Islamic
texts according to accuracy, well-formedness and suitability. Findings showed that
Google translate is the best acceptable translation over Babylon translation, World lingo
and Bing translation.

Previous studies

Previous studies on MT can be classified on MT acceptability and comparison between

NMT and MT segments. Alshehri and Alowedi (2023) explored the status of PEMT in
the Saudi context in translation center. 18 professional translators were surveyed.
Findings showed that the majority of participants postedit the MT product. They are also
aware about the importance of PE certified program which enables post editors. To begin
by the acceptability of MT, Kasperė et al. (2023) previewed the acceptability of MT
amongst professional translators and ordinary users. Findings showed that professional
translators read MT product deeply and critically whereas nonprofessional users had low
awareness about MT production which mean they MT quality. Castilho and O’Brien
(2017) studied the acceptability of MT among group of users in which texts were
translated form English into Japanese, Chinses and German. Using TQA, end users
expressed their satisfaction of the MT production. Findings also reported that applying
PE makes the text usable and acceptable.

Studies also disseminated the difference between MT and NML (Moorkens et al., 2018;
Sánchez-Gijón et al., 2019; Yamada, 2019; Zouhar et al., 2021). Yamada (2019)
compared between statistical machine translation (SMT) of Google translate before 2014
and now a days Google NMT of the same source texts. Findings resulted from postediting
evaluation revealed no cognitive efforts was found between SMT and NMT. However,
significant difference is reached on editing amount. The output of NMT followed with PE
is more efficient with SMT with PE in case of number of errors. Sánchez-Gijón et al.
(2019) explored the postediting difference between NMT and TM between English and
Spanish. Results revealed that NMT postediting required less editing effort than using
TM segments. Moorkens et al. (2018) probed the difference between translation output
resulted from SMT and NMT in translation literature between English and Catalan. The
comparison was conducted under three dimensions, to translate from scratches, NMT
postediting and SMT postediting. The professional translators revealed through an
interview and questionnaire their preference to translate from scratches than to be
restricted on translation segment. Zouhar et al. (2021) investigated via an experimental
study the postediting quality and time via using NMT and MT system between English
and Czech. Findings indicated that good MT system produces output with less errors. The
study found not relationship between MT system quality and time spent in PE. SMT
correlates with PE time though. Jia et al. (2019 a) studied the translation of 30 fresh
postgraduate Chinse students on Google neural machine translation (GNMT) and scratch
translation in dominate and general text between English and Chinese. Findings reported
that postediting in GNMT is faster than scratch translation in both text types. GNMT
lacks the cognition of the student translators. Interestingly, PE parity with the quality of
scratch translation. Jia et al. (2019 b) conducted a study in which 9 evaluators assess the
translation output quality of SMT and NMT in term of fluency and accuracy. They found
NMT produced better translation quality than SMT. Findings also showed that postediting
which accompanied NMT is not necessary faster than SMT, however, it reduced
cognitive and technical efforts.

Methods

Research design

This study uses a qualitative research synthesis design. This kind of research involves
investigating research studies on certain criteria (Chong & Reinders, 2020). In this study,
postediting in MT research was investigated.

Identify keywords

In this study the researcher set keywords as "postediting, machine translation quality, and
machine translation acceptability". These keywords were searched using Google scholar
which gave 2,620 results. The search was narrowed down by searching within a period
from 2015-2023 in which 1,500 results was reached. The researcher selected this phase
because the topic develops whereas technology develops. So, old findings do not provide
reliable information for this research.

Conduct literature search

The researcher works from the beginning to avoid bias in selecting the research papers in
a way which ensures representativeness. The research searched on the three keywords
identified above on Google scholar. The database includes high indexed journals, books
chapters, theses and conference papers. In this literature research, 40 studies were reached
and listed. Not all of these studies were used in the qualitative synthesis, because some of
them do not sit in the inclusion criteria.

Evaluate literature using inclusion criteria

The 40 research studies were appraisal using some criteria were taken into account while
including paper in the analysis.
 1. The paper should be published within the year 2015-2023,
2. The article should exhibit empirical study, i.e., has conducted on certain sample,
therefore, reviewed articles were excluded,
3. The paper should be published in high indexed data base, including Scopus, web
of science and the like.
4. The article should adopt qualitative or experimental design, therefore, purely
quantitative articles were excluded.
5. The findings of these articles should show answer to one of more of the research
questions listed in this study.

Out of the 40 studies, 20 studies were listed for the final analysis, Table 1. Articles
which are based on purely quantitative design, or reviews of previous research were
excluded from this qualitative synthesis.

Table 1
Studies included in the synthesis
Research title No. of
Study Journal Category
themes
1. Aranberri HERMES-Journal of Skills What do professional translators do when 2
(2017) Language and post-editing for the first time? First insight
Communication in into the Spanish-Basque language pair
Business
2. Azer & Advances in Language and An evaluation of output quality of machine 1
Aghayi Literary Studies Acceptability translation (Padideh Software vs. Google
(2015) Translate)
3. Béchara et Informatics Postediting The role of machine translation quality 1
al. (2021) efforts estimation in the post-editing workflow
4. Castilho & Linguistica Antverpiensia, Acceptability Acceptability of machine-translated content: 1
O’Brien New Series–Themes in A multi-language evaluation by translators
(2017) Translation Studies and end-users
5. Daems et Frontiers in psychology Postediting Post-editing effort of a novel with statistical 1
al. efforts and neural machine translation.
(2017)
6. Machine Translation NMT/MTS Post-editing neural machine translation 1
Jia et al.
versus phrase-based machine translation for
(2019)
English–Chinese.
7. Macken et Informatics Postediting Quantifying the effect of machine translation 1
al. (2020) efforts in a high-quality human translation
production process
8. Moorkens The Interpreter and MT vs HT What to expect from neural machine 1
(2018) Translator Trainer translation: a practical in-class translation
evaluation exercise
9. Sánchez- Machine Translation NMT/MTS Post-editing neural machine translation 1
Gijón et al. versus translation memory segments.
(2019)
10. Temizöz Perspectives Profession Postediting machine translation output: 1
(2016) Subject-matter experts versus professional
translators.
11. Tezcan et Computer Speech & Postediting time Estimating post-editing time using a 1
al. Language gold-standard set of machine
(2019)
translation errors
12. Informatics Approaches Translation quality and error recognition in 1
Vardaro et
professional neural machine translation post-
al. (2019)
editing
13. Yang et al. Translation and MT vs HT Measuring the usability of machine 2
(2021) Interpreting Studies translation in the classroom context
14. Yang & Computers & Education Approaches Modeling the intention to use machine 1
Wang translation for student translators: An
(2019) extension of technology acceptance model
15. Yang & Human Behavior and MT vs HT On postediting of machine translation and 1
Mustafa Emerging Technologies workflow for undergraduate translation
(2022) program in China
Book chapters
16. Eszenyi & Fit-for-market translator Tasks Rage against the machine–will post-editing 1
Dóczi, and interpreter training in assignments outnumber translations in the
2020 a digital age future
Thesis/dissertation
17. Dunne, MA thesis, The University Approaches A post-editing approach to machine 1
2022, of Dublin translation evaluation at the document-level,
18. Mitchell Doctoral dissertation, Postediting time Community post-editing of machine- 1
(2015) Dublin City University translated user-generated content
Conference/proceeding
19. Conference on Empirical Postediting Neural machine translation quality and post- 1
Zouhar et
Methods in Natural efforts editing performance
al., 2021
Language Processing.
20. Conferences of the Postediting time Machine translation quality and post-editor 1
Sanchez-
Association for Machine productivity
Torron &
Translation in the
Koehn
Americas: MT
(2016)
Researchers' Track

Findings and discussion

This qualitative synthesis includes themes from 15 journals, 2 theses/dissertation, 2

conference proceedings and a book chapter, Table 1. The results of the synthesis generate
22 initial codes, subdivided into 9 categories. These 9 categories are affiliated to the 3
research questions investigated. The postediting features are divided into four categories
which are postediting profession, approaches, skills and tasks. These categories
encompass the initial codes that provide an answer to the first research question. The
second element in this qualitative synthesis is the evaluation of postediting quality. It
includes three categories which are acceptability, postediting time and efforts. These
categories include themes which answer the second research question. Finally, the last
element in this study is the characteristics of postediting outputs. It has two categories
which are human translation (HT) vs MT and NMT vs MTS. They contain themes which
answer the third research question, Table 1.

Postediting features

While investigating the previous publications, four codes were reached about the features
of postediting. They are postediting profession, approaches, skills, and tasks. They will be
displayed accordingly.

Postediting profession. Temizöz, (2016) finds “a degree in translation does not directly
correlate with postediting quality, unless it is combined with
subject-matter knowledge and professional experience in translation" (p. 646). Posteditors
should be prepared well to do the task of posteding properly. It is traditionally believed
that posteditors should own a degree in translation while others (contemporary view)
claimed that knowledge in the field to be edited is necessary. The second perspective is
confirmed by this finding. It means that a posteditor may be one who has knowledge in
the field and masters the language pairs. This finding is confirmed by Ramos (2020) who
reported that during the development of MT, the postediting work will be shifted from
translators to linguistic experts.
Postediting approaches. The study revealed three major approaches for postediting. The
first approach is crowd-sourcing document-level post-edits. In this regard, Dunne (2022)
states that crowd-source editing “gives full document context, removes subjective
numerical judgements and it can specify where MT systems go wrong” (p. ii). While the
second approach, i.e., linear mixed-effect, enables translator to "predict what kind of
behavior…[the translator] experts is associated with the correction of different error types
during the post-editing process" (Vardaro et al., 2019). The third approach is quasi-
circular model. The quasi-circular model developed by Yang and Wang (2019) "not only
reveals significant factors for MT adoption, but suggests positive effects of using MT"
(p.101). Crowd-sourcing document-level approach has some advantages, the most of
which is that it guides the posteditor to the place of mistranslation. Herbig et al. (2019)
recommended the combination of touch, pen and speech in supporting the tasks of
posteditors.
Postediting skills. Postediting implies the modification of certain faults in MT outputs.
These modifications whether be in terminologies, syntax, cultural-specific or cohesion
and coherence. It has been reported that a posteditor "shifts from the task of identifying
and fixing errors, to that of “patchwork” where post-editors identify the machine
translated elements to reuse and connect them using their own contributions"
(Aranberri, 2017, p. 89). Other focused that a posteditor should also have experience in
the topic being edited as well in translation profession. On the contrary, posteditors are
warned from editing what is already right. It was found that posteditor “primarily focus
on correcting machine translation errors but often fail to restrain themselves from editing
correct structures" (Aranberri, 2017, p. 89). This finding is consistent with Popović
(2018) who reported that to ensure the best MT outputs, editor should also have
experience in the topic being edited as well in translation profession.
Postediting tasks. Postediting is an extensive task in which posteditors should focus on
style, format, terms and so forth. This is affirmed by Eszenyi and Dóczi (2020) who
found that "postediting implies the modification of translation outputs including format,
style, terminology and the like mistranslation" (p. 119). The context is also more
important in doing the posteding. It was stated that professional translators while postedit
MT outputs, they "should pay attention to context, logical relationships, four word
phrases, and so on"(Chu & Liu, 2021, p. 129). To conclude, the tasks of posteding may
produce us to the traditional view that posteditors should have a major in translation and
the counterparty view which shift the work of postediting from translators to experts in
linguistics. This can be confirmed by Kasperė et al. (2023) who found that professional
translators read the MT outputs deeply and critically whereas nonprofessional users
accept the translation outputs due to their low awareness about MT quality.
Machine translation quality evaluation

Findings reported that MTPE is evaluated according the end users' acceptability to the
translation outputs, the postediting time and efforts required to produce a final copy of the
translation. These three criteria will be discussed underneath:

Acceptability. Findings showed that MTPE is acceptable by users. The MT outputs can be
relied on in providing translation service. Azer and Aghayi (2015) state, "the machine-
generated translations are intelligible and acceptable in translating certain text- types, for
end-users and Google Translate is more acceptable from end-users point of view" (p.
226). Furthermore, Castilho and O’Brien (2017) pinpoint “…the usability and satisfaction
results by end-users insofar as the implementation of light PE both increased the
usability and acceptability of the PE instructions and led to satisfaction being
reported" (p. 120). These findings indicated that pinpoint for MT quality is the end-user's
acceptability to the translation outputs. These findings are aligned with the findings of
AbuSa’aleek (2016), who established three criteria for accepting MT to Islamic texts:
accuracy, coherence, and appropriateness. Like, Azer and Aghayi's (2015) findings,
AbuSa’aleek (2016) found that Google translate produces the best acceptable translation
over other software and applications.

Postediting time. Sanchez-Torron and Koehn (2016) mention, "the MT system with the
lowest BLEU score produced the output that was post-edited to the lowest quality and
with the highest PE effort, measured both in HTER and actual PE operations," (p. 16).
Moreover, Tezcan et al. (2019) affirm "time can be estimated with high accuracy when all
the translation errors in the MT output are known" (p.120). To sum up, the "PE quality is
largely independent of the profile characteristics measured" (Mitchell, 2015, p. ‫)׀׀‬. These
findings showed that the time spent in postediting is an indicator for assessing the quality
of PEMT. The time requires to do the posteding is controlled by some factors for example
the familiarity of the posteditor with the type of error reduces the postediting time. These
results are in line with Zouhar et al. (2021) who indicated that a translation with less
errors is an indicator of good MT system. On the contrary, they found no relationship
between MT system quality and time spent in PE.

Postediting efforts. Daems et al. (2017) say, "we find that most post-editing effort
indicators are influenced by machine translation quality, but that different error types
affect different post-editing effort indicators, a more fine-grained MT quality analysis is
needed to correctly estimate actual post-editing effort. Coherence, meaning shifts, and
structural issues are shown to be good indicators of post-editing effort" (p.1). Macken et
al. (2020) state "typing effort is the most frequently mentioned reason why participants
preferred working with MT (p. 12). Furthermore, Zouhar et al. (2021) state, "better MT
systems lead to fewer changes in the sentences in this industry setting" (p.1). Likewise,
"good MTQE information can improve post-editing efficiency and decrease the cognitive
load on translators" (Béchara et al., 2021, p. 1). These findings are consistent with Toral
et al.'s (2018) findings. They reported that postediting of MT reducing the cognitive
efforts by 29% in phrase-based statistical MT (PBMT) and 42% in NMT. The affordance
required from the translator to do the correction on MT is an indicator for evaluation the
service quality. Number of errors and the coherence between sentences are the backbone
of the efforts required for freeing the MT outputs from errors.

To answer RQ2, the quality of MT is interwoven by three indicators. The acceptability of

MTPE relies on the quality of the machine translation outputs to the clients. This was
reported by two research studies (Azer & Aghayi, 2015; Castilho & O’Brien, 2017). This
finding agreed AbuSa’aleek (2016), who set accuracy, coherence, and appropriateness as
main criteria for accepting MT to Islamic texts. Another indicators for measuring the
quality of PEMT is the time spent in post-editing. This finding is inducted from the
studies of Koehn (2016), Tezcan et al. (2019) and (Mitchell, 2015). This finding is
consistent with Zouhar et al. (2021) who indicated relationship between MT system
quality and time spent in PE. Finally, the quality of MT is also determined by the effort
required for post-editing in addition to the two previous indicators. Postediting as effort
an indicator of MT quality was reported in four of the analyzed studied. They are Daems
et al. (2017), Macken et al. (2020), Zouhar et al. (2021) and (Béchara et al., 2021). This
findings is confirmed by Toral et al.'s (2018) findings. Coherence, meaning shifts, and
structural issues play a role in determining post-editing effort.

Characteristics of PEMT outputs

The majority of studies compared between neural machine translation (NMT) and
segment machine translation (SMS) in various aspects whether the translation quality
outputs or the time required for postediting. MT and human translations were also
compared.

NMT vs MTS

NMT post-editing produces better translation outputs with less errors than the the
translation produced by using TM segments. Sánchez-Gijón et al. (2019) state, "NMT
post-editing involves less editing than TM segments, but this editing appears to take more
time" (p. 31). Furthermore, this lead to a less cognitive efforts of posteditors. Jia et al.
(2019 b) affirm, "post-editing output from NMT reduces the technical and cognitive
effort" (p. 90). The finding of Sánchez-Gijón et al. (2019) is confirmed by Zouhar et al.
(2021) who mentioned that NMT has the privileged over SMT regarding translation
quality. Similarly, Yamada (2019) reported that the output of NMT followed with PE is
more efficient with SMT with PE in case of number of errors.

MT vs HT

Yang et al. (2021) state that MTPE "is more efficient than human translation" (p.1).
Furthermore, "post-editing produces fewer errors than human translation. While the types
of errors vary, errors in terms of accuracy outnumber those related to fluency" (Yang et
al., 2021, p. 1). Eszenyi and Dóczi (2020) say, "human translators are not superior in
skills to the machine in the particular text type" (p. 119). Moorkens et al. (2018) state "a
neural MT system trained on literary data does not currently have the necessary
capabilities for a creative translation (p.375). Likewise, MT is accompanied by some
defects. Yang and Mustafa (2022) identify them as "confusing postediting standard,
inconsistent quality, ways to choose machine translation providers, and technical issues"
(p.1). These findings can be s reported that MT may produce acceptable outputs than
human translation. The superiority was prevailed in the time required, productivity and
the like. These findings disagree with Eszenyi and Dóczi (2020) who believed that MT
outputs are still far away from HT in some language pairs and genres. One of the
disability of machine in translating literary texts. To sum up, the superiority of MT over
HT may be limited to a special text type.

Conclusion, implication and recommendations for future research

Postediting is one of the state-of-the-art strategies which accompanies MT. Postediting

has become a must that translators apply on MT outputs. Studies showed that MT shifts
the task of translator from doing the translation into revising it from mistranslation. This
great shift let some researchers think that linguistic experts may do the task better
(Ramos, 2020). These findings set the responsibility of translation instructors to train their
students with postediting approaches, skills and ways of applicability. This is confirmed
by many studies (Eszenyi1 & Dóczi, 2020; O’Brien, 2002). Postediting per se needs to be
honed practically and theoretically. Students of translation or even EFL graduates should
be trained extensively on postediting. We cannot prevent them from using MT in their
life, we should build their abilities by providing them with the best practices of
postediting approaches, skills and knowledge which enable them to make use of their
university degree in their practical lives and be good individuals who depend on
themselves and do good for their societies.

Studies reported that NMT provides better translation outputs than SMT, and MT may
outperform HT in some text types and genre types. This field is connected with the daily
development of NMT software and applications. More research studies are needed to
classify the text-types which may be translated by MT in a way to meet the translation
quality and acceptability. Studies on NMT were conducted on language pairs from
English to Spanish, Japanese, German. Very limited studies were conducted on Arabic.
Therefore, a call to measure the NMT outputs between Arabic and English on various text
types.

Developing our students' postediting skills is the responsibilities of several entities. The
university professors at English departments which are the corner stone. They should
held symptoms and conferences along with the stakeholders, say in this situation,
companies and translation offices about the requirement or the characteristics of the
graduates. Taking this in mind, the university councils and Ministry of higher education
should also add posteding courses in all departments. Postediting may be one of the
learning outcomes that the university graduate should acquire in the studying years.
Therefore, courses on postediting should be inserted to the English departments
syllabuses.
References

AbuSa’aleek, A. O. (2016). The adequacy and acceptability of machine translation in

translating the Islamic texts. International Journal of English Linguistics,
6(3),185-193.
Almaaytah, S. A. (2022). Post-editing in translation: experiences and
development. Journal of Positive School Psychology,6(4), 8794-8803.
Alshehri, B. A., & Alowedi, N. A. (2023). Saudi translation agencies and translation
centers: a study of post-editing practices. World Journal of English
Language, 13(3), 118-128. https://doi.org/10.5430/wjel.v13n3p118
Aranberri, N. (2017). What do professional translators do when post-editing for the first
time? First insight into the Spanish-Basque language pair. HERMES-Journal of
Language and Communication in Business, 56, 89-110
Azer, H. S., & Aghayi, M. B. (2015). An evaluation of output quality of machine
translation (Padideh Software vs. Google Translate). Advances in Language and
Literary Studies, 6(4), 226-237.
Béchara, H., Orăsan, C., Parra Escartín, C., Zampieri, M., & Lowe, W. (2021). The role
of machine translation quality estimation in the post-editing workflow.
Informatics, 8(3), 61, https://doi.org/10.3390/informatics8030061
Carl, M., Gutermuth, S., & Hansen-Schirra, S. (2015). Post-editing machine translation:
A usability test for professional translation settings. In A. Ferreira & John W.
Schwieter (Eds), Psycholinguistic and cognitive inquiries into translation and
interpreting (pp. 145-174). John Benjamins
Castilho, S. (2016). Measuring acceptability of machine translated enterprise
content (Doctoral dissertation). Dublin City University.
Castilho, S., & O'Brien, S. (2017). Acceptability of machine-translated content: A multi-
language evaluation by translators and end-users. Linguistica Antverpiensia, New
Series–Themes in Translation Studies, 16, 120–136.
Castilho, S., Doherty, S., Gaspari, F., Moorkens, J. (2018). Approaches to human and
machine translation quality assessment. In: Moorkens, J., Castilho, S., Gaspari, F.,
Doherty, S. (eds) Translation quality assessment machine translation:
technologies and applications, (Vol 1.pp. 9-38). Springer, Cham.
https://doi.org/10.1007/978-3-319-91241-7_2
Chong, S.W., & Reinders, H. (2020). Technology-mediated task-based language
teaching: A qualitative research synthesis. Language Learning & Technology,
24(3), 70–86. http://hdl.handle.net/10125/4473
Chu, Y., & Liu, Z. (2021). Deficiency analysis of machine translation and post-editing
based on computational linguistics: A case study of the translation of government
work report. Saudi Journal of Engineering and Technology, 6(11), 426-431.
Daems, J., Vandepitte, S., Hartsuiker, R. J., & Macken, L. (2017). Identifying the
machine translation error types with the greatest impact on post-editing effort.
Frontiers in psychology, 8, 1282. https://doi.org/10.3389/fpsyg.2017.01282

Dunne, D. (2022). A post-editing approach to machine translation evaluation at the

document-level, (Unpublished MA thesis). The Univeristy of Dublin.
Eszenyi, R. &. Dóczi, B. (2020). Rage against the machine–will post-editing assignments
outnumber translations in the future?. In M. Fischer, R. Besznyák, & C Szabó
(Eds), Fit-for-market translator and interpreter training in a digital age, (pp. 119-
133). Vernon Press.
 Hearne, M., & Way, A. (2011). Statistical machine translation: a guide for linguists and
translators. Language and Linguistics Compass, 5(5), 205-226.
Herbig, N., Pal, S., van Genabith, J., & Krüger, A. (2019, May). Multi-modal approaches
for post-editing machine translation. Proceedings of the 2019 CHI Conference on
Human Factors in Computing Systems (pp. 1-11).
Hu, K., O’Brien, S., and Kenny, D. (2020). A reception study of machine translated
subtitles for MOOCs. Perspectives 28, 521–538.
https://doi.org/10.1080/0907676X.2019.1595069
Jia, Y., Carl, M., & Wang, X. (2019 a). Post-editing neural machine translation versus
phrase-based machine translation for English–Chinese. Machine Translation,
33(1-2), 9-29. https://doi.org/10.1007/s10590-019-09229-6
Jia, Y., Carl, M., & Wang, X. (2019 b). How does the post-editing of neural machine
translation compare with from-scratch translation? A product and process
study. The Journal of Specialised Translation, 31(1), 60-86.
Kasperė, R., Horbačauskienė, J., Motiejūnienė, J., Liubinienė, V., Patašienė, I., &
Patašius, M. (2021). Towards sustainable use of machine translation: usability and
perceived quality from the end-user perspective. Sustainability, 13(23), 13430.
https://doi.org/10.3390/su132313430
Kasperė, R., Motiejūnienė, J., Patasienė, I., Patašius, M., & Horbačauskienė, J. (2023). Is
machine translation a dim technology for its users? An eye tracking
study. Frontiers in psychology, 14, 1-14.
https://doi.org/10.3389/fpsyg.2023.1076379
Koponen, M. (2010, January). Assessing machine translation quality with error analysis.
In Electronic proceeding of the KaTu symposium on translation and interpreting
studies.
Koponen, M. (2016). Is machine translation post-editing worth the effort? A survey of
research into post-editing and effort. The Journal of Specialised
Translation, 25(2), 131-148.
Lopez, A. (2008). Statistical machine translation. ACM Computing Surveys (CSUR),
40(3), 1-49.
Macken, L., Prou, D., & Tezcan, A. (2020, April). Quantifying the effect of machine
translation in a high-quality human translation production process. Informatics,
7(2),12 https://doi.org/10.3390/informatics7020012
Mitchell, L. (2015). Community post-editing of machine-translated user-generated
content, (Doctoral dissertation). Dublin City University.
Moorkens, J. (2018). What to expect from neural machine translation: a practical in-class
translation evaluation exercise. The Interpreter and Translator Trainer, 12(4),
375-387.
Mrinalini, K., Sangavi, G., & Vijayalakshmi, P. (2016, November). Performance
improvement of machine translation system using lid and post-editing. In 2016
IEEE Region 10 Conference (TENCON) (pp. 2134-2137). IEEE.
https://doi.org/10.1109/TENCON.2016.7848403
O’brien, S. (2002). Teaching post-editing: a proposal for course content. In Proceedings
of the 6th EAMT Workshop: Teaching machine translation.
Popović, M. (2018). Error classification and analysis for machine translation quality
assessment. In: Moorkens, J., Castilho, S., Gaspari, F., Doherty, S. (eds)
Translation quality assessment. Machine translation: technologies and
applications, (Vol 1, pp. 129–158). Springer, Cham. https://doi.org/10.1007/978-
3-319-91241-7_7
Povilaitienė, M., Kasperė, R. (2022). Machine Translation for Post-editing
practices. Current Trends in Language Development, 24, 47–62.
https://doi.org/10.31392/NPU-nc.series9.2022.24.04
Pym, A. (2011). What technology does to translating. Translation & Interpreting,
The, 3(1), 1-9.
Ramos, L. (2020, October). MT Post Editing challenges: Training happy and successful
posteditors. In Proceedings of 1st Workshop on Post-Editing in Modern-Day
Translation (pp. 56-60).
Rossi, C., & Carré, A. (2022). How to choose a suitable neural machine translation
solution: Evaluation of MT quality. In Dorothy Kenny (ed.), Machine translation
for everyone: Empowering users in the age of artificial intelligence, (pp.51–79).
Berlin: Language Science Press. https://doi.org/10.5281/zenodo.6759978
Sánchez-Gijón, P., Moorkens, J., & Way, A. (2019). Post-editing neural machine
translation versus translation memory segments. Machine Translation, 33(1-2),
31-59.
Sanchez-Torron, M., & Koehn, P. (2016). Machine translation quality and post-editor
productivity. In Conferences of the Association for Machine Translation in the
Americas: MT Researchers' Track (pp. 16-26).
Stoyanova-Georgieva, I. (2021). Machine translation post-editing–Current situation and
the future of translator training in Bulgaria. Studies in Linguistics, Culture, and
FLT, 9(3), 63-75.
Temizöz, Ö. (2016). Postediting machine translation output: Subject-matter experts
versus professional translators. Perspectives, 24(4), 646-665.
http://dx.doi.org/10.1080/0907676X.2015.1119862
Tezcan, A., Hoste, V., & Macken, L. (2019). Estimating post-editing time using a gold-
standard set of machine translation errors. Computer Speech & Language, 55,
120-144. https://doi.org/10.1016/j.csl.2018.10.005
Toral, A., Wieling, M., & Way, A. (2018). Post-editing effort of a novel with statistical
and neural machine translation. Frontiers in Digital Humanities, 9.
https://doi.org/10.3389/fdigh.2018.00009
Vardaro, J., Schaeffer, M., & Hansen-Schirra, S. (2019). Translation quality and error
recognition in professional neural machine translation post-editing. Informatics,
6(41), 1-29.
Yamada, M. (2019). The impact of Google neural machine translation on post-editing by
student translators. The Journal of Specialised Translation, 31(1), 87-106.
Yang, Y., & Wang, X. (2019). Modeling the intention to use machine translation for
student translators: An extension of technology acceptance model. Computers &
Education, 133, 116-126. https://doi.org/10.1016/j.compedu.2019.01.015
Yang, Y., Wang, X., & Yuan, Q. (2021). Measuring the usability of machine translation
in the classroom context. Translation and Interpreting Studies, 16(1), 101-123
Yang, Z., & Mustafa, H. R. (2022). On postediting of machine translation and workflow
for undergraduate translation program in China. Human Behavior and Emerging
Technologies,2022, 5793054, 1-11. https://doi.org/10.1155/2022/5793054
Zouhar, V., Tamchyna, A., Popel, M., & Bojar, O. (2021). Neural machine translation
quality and post-editing performance. Conference on Empirical Methods in
Natural Language Processing. https://doi.org/10.48550/arXiv.2109.05016