Translator As A Text Editor
Translator As A Text Editor
Faculty of Arts
Department of English
and American Studies
English-language Translation
2014
Acknowledgement
I would like to thank my supervisor, PhDr. Jarmila Fictumov, for her kind
support and valuable advice that she gave me when I was working on this thesis. My
special thanks are due to my family for their inexhaustible patience.
LIST OF TABLES
Table 1: Translator Qualifications..............................................................................................................17
TABLE 2: FACTORS GUIDING THE CHOICE OF THE LEVEL OF POST-EDITING ..............................................35
TABLE 3: HUMAN INTERVENTION IN PE TYPES...........................................................................................42
TABLE 4: THE FUTURE OF MACHINE TRANSLATION AS VIEWED BY TRANSLATORS...................................45
TABLE 5: POST-EDITING TRAINING..............................................................................................................65
LIST OF FIGURES
LIST OF ABBREVIATIONS
TABLE OF CONTENTS
LIST OF TABLES........................................................................................................................................4
LIST OF FIGURES.....................................................................................................................................5
LIST OF ABBREVIATIONS......................................................................................................................6
TABLE OF CONTENTS.............................................................................................................................7
1
INTRODUCTION............................................................................................................................10
METHODOLOGY...........................................................................................................................14
3.1
RESPONDENTS...........................................................................................................................15
4.2
THE SECOND AND THIRD ERAS TM, GLOSSARY TECHNOLOGY AND WORKFLOW SYSTEMS 20
4.3
POST-EDITING...............................................................................................................................29
6.1
WHAT IS POST-EDITING?..........................................................................................................29
6.2
6.3
6.4
TYPES OF POST-EDITING...........................................................................................................34
6.4.1
6.4.2
6.4.3
6.4.4
6.4.5
6.5
7.1.2
7.1.3
7.1.4
7.1.5
7.1.6
7.2
10
7.2.1
Linguistic Skills...................................................................................................................49
7.2.2
Core Competences...............................................................................................................51
7.2.3
Instrumental Competence....................................................................................................53
7.3
7.3.1
7.3.2
7.3.3
Collaborative Translation...................................................................................................57
TRAINING OF POST-EDITORS...................................................................................................62
8.1
PREREQUISITE KNOWLEDGE.....................................................................................................68
8.2
8.3
COURSE OUTLINE......................................................................................................................69
RECURRENT ISSUES....................................................................................................................71
9.1
9.2
QUALITY...................................................................................................................................74
9.3
QUALITY ASSESSMENT..............................................................................................................75
9.4
PRICING.....................................................................................................................................77
11
CONCLUSIONS..............................................................................................................................84
INTRODUCTION
10
machine-translated texts (Champollion 2001; Gouadec 2007; Pym 2012). This line of
thought is still viewed as a highly radical one in the Translation Studies context,
however, it cannot be disregarded, as its proponents are no lone voices. The author of
this thesis intends to join their ranks predicting a radical shift in the translation practice
with translators becoming mere post-editors of machine-translated texts, and to manifest
how the shift has already been well on the way.
The most profound changes in the practice of translation have been taking place in
the last thirty years or so, and what is most striking about them is the rapidness of the
progress. Times change and translators working with pen and paper, or even the
typewriter, belong to a long bygone era. The role of the PRAT (pencil and rubberassisted translator) has been, with a few exceptions, taken over by the CAT (computerassisted) translator (Gouadec 2007: 109). Yet, computer-assisted translation does not
seem to satisfy the flow of progress. As early as the 1940s, scientists were striving to
automate the process of translation and people had been fiddling with the idea much
earlier; one of the first scholars to look into the matter of machine translation was
Joachim Becher in the 17th century (Krl 2003: 2).
Any automation is at first intended as a way to ease human effort and make
work easier, but history shows that it is usually economic interests that come to the fore,
while the well-being of humans is downgraded. Let us assess, for instance, how
Keyness prediction (1930: 369) of a 15-hour work week matches the reality of the early
21st century. It is economic factors, not human welfare, that drive progress and
influence peoples jobs, translation not excepting. If we accept that vested interests of
major players in the translation industry influence the development of the translation
profession, and there is ample evidence to support this claim, we may as well agree with
Gouadecs (2007: 295) subversive belief that all efforts are undertaken to get rid of as
11
much of the human component as possible in the translation process. This thesis will
demonstrate how the development of the translation profession is fuelled by economic
interests where translation is seen as a product, and where most emphasis is put on
generating profit. The author believes that turning translators into post-editors is a byproduct of this process of translation industrialization.
In 2001, Yves Champollion (the father of one of the first CAT tools namely
Wordfast) wrote that even though machine translation software was still in its infancy,
vast numbers of IT engineers were working on turning the translator into a proofreader
of computer output. His prediction appears to be provoking in itself, even without
mentioning its time specification Champollion claimed that this [turn] may take a
long time, but . . . a long time in the world of IT is 3 to 5 years (2001: 1). Meanwhile,
more and more researchers have been taking a similar position.
Even though Champollions prediction may not have been fulfilled as to its temporal
framing, I believe that the metamorphosis of the translator into a post-editor has been
gradually taking place and that it is worth looking into the reasons that may have led
Champollion, Gouadec and others to believe that translation as we know it will soon be
remembered only as a memento of the past. This thesis should serve as a way of
surveying this issue. The present thesis aims to achieve several main objectives. First of
all, it aims to explore the development of the translation profession from early history
until present times. Questions of who was the PRAT translator, what were his typical
work practices and how all of this changed over time are addressed in an effort to map
the development of the profession in order to support the main claim of this thesis, that
is to demonstrate that the translator is well on the way toward becoming a post-editor of
12
13
translation profession. The thesis surveys the advantages and disadvantages of machinetranslation post-editing and what impacts they have on the translation job. A brief
prediction of what may lay in store for the poor translator is made as well.
METHODOLOGY
The nature of the present thesis is largely theoretical and it draws mainly on theoretical
research in the areas of machine translation and post-editing as well as on reports
informing about practical implementation thereof. The issues discussed in the thesis are
accompanied by a small-scale quantitative survey in the form of a questionnaire
addressing language professionals who deal with translation and/or post-editing. The
questions were designed to get a picture of what translators based in the Czech Republic
make of machine translation and post-editing in order to illustrate some of the points
made in this thesis, and to predict the future development of the translation profession
based on the opinion of the main players in the translation process. The questionnaire
was distributed through two major Language Service Providers (LSPs) operating on the
Czech translation market, namely Moravia and esk peklady, which are both
renowned for using machine translation. Moreover, several acquaintances of the author
of this thesis were asked to take part in the research, however, the sample of participants
is random, the only two criteria for addressing the prospective participants were that
they needed to be active in the field of professional translation and that they could speak
English, since English was selected as the language of the questionnaire. The
questionnaire was anonymous. The questionnaire (see Appendix 1) was created using
Google Forms so that it could be distributed online and as such be easily accessible for
prospective respondents. Most of the questions are of the multiple-choice type, and
there are several please specify fields for the respondents to provide more complex
14
answers. There are eighteen questions altogether, divided into three parts. The first part
aims at gathering general information about the participants. The second part focuses on
post-editing experience and pricing, while the last section deals directly with questions
related to post-editing usage and participant-specific information. Most of the questions
are very simple in nature and were designed in order not to be excessively timeconsuming for the respondents. Since most of the questions are closely connected to the
participants experience with post-editing, those with no experience (those answering no
to question number 6 which asks whether the respondent has ever been offered a postediting job) are redirected to the end of the questionnaire after question number 6. This
approach enables to include the answers of people untouched by post-editing and
separately assess their fear (or a lack of it) from being replaced by machine translation.
The complete dataset collected in the survey is in the attachment section at the end of
the thesis (see Appendix 2).
3.1
Respondents
The total number of respondents is twenty-two translators. This number is not overly
high and as such represents only a small fracture of the Czech translation market. Any
conclusions made based on the data gathered from the questionnaire can therefore be
only tentative; a larger-scale research would be needed to replicate the results. However,
the data can be seen as representing a group of translators working in the Czech
Republic, and as will be discussed, it does not significantly differ from research
conducted in other countries.
The paragraphs below summarise the general information provided by the
participants. The answers to other questions than those mentioned in them are discussed
in relevant sections dealing with the issues they relate to.
15
Fourteen of the participants work as full-time translators, while seven translate parttime. One of the participants is neither full- nor part-time translator, he/she commented
that they occasionally take translation commissions. What is interesting is that only one
of the part-time translators has been offered a post-editing job, while seven of the fulltime translators (50%) have got an offer to post-edit. This might deserve further
research, as the numbers suggest a tendency among full-time translators to be selected
for post-editing jobs. This may be related to their experience with translation and
indicate a reluctance of language service providers to assign post-editing jobs to less
experienced translators. It is probably no coincidence that six out of seven part-time
translators mentioned having translation experience in the lowest range of 1-5 years.
Overall numbers for the participants experience can be seen in the picture below (see
figure 1).
The experience of the respondents with post-editing is not overly high. Five of the
translators admitted that they had post-edited only once, three translators sometimes
post-edit, while only one of the respondents claims to take post-editing jobs on a regular
basis. None of the translators taking parting in this survey has a post-editing experience
exceeding three years. Post-editing, therefore, does not constitute a large part of the
16
translators overall workload. The picture below shows that post-editing constitutes only
0-10% of translation jobs for most of the translators questioned (see figure 2).
3
6
10
1
5
14%
27%
45%
5%
23%
Table 1 presents an overview of the qualifications of the participants of the survey (see
table 1). The graph shows that most of the translators have a university degree either in
languages or directly in translation, very often a combination of both. Only one of the
participants has a degree in a non-linguistic field. This suggests that most of the
translators who took part in this research are recruited from among linguistic university
graduates of some sort. As will be discussed in the chapter on translators training (see
17
Chapter 8), it is these translators who should be the target of machine translation and
post-editing training.
This section deals with the historical development of translation from the view point of
technology used by translators. The evolution of the translation industry can be mapped
using a chart drafted in the TAUS (Translation Automation User Society) report
(2013b). Five subsequent and slightly overlapping eras can be distinguished (see figure
3). The first four eras are briefly discussed in this chapter, while the last era is returned
to in the chapter dealing with the possible future development of the translation
profession (see Chapter 10). Special attention is paid to the third era the era of
integration as it focuses on machine translation and its history, and as such is
intertwined with post-editing and current happenings in the translation industry.
4.1
As the origin of translation is often regarded the Biblical story of the tower of Babel
from the Old Testament which describes a fall into linguistic diversity (Robinson 2001:
22). There was no need of translation prior to this mythical disaster since all people
supposedly spoke a single language. When exactly men started to translate is unclear,
however, it is safe to assume that there must have been some kind of translation practice
which would enable communication between people of different linguistic backgrounds.
As soon as people started to migrate such a meeting and thus a need for a
communication intermediary was inevitable. There are no records of these ancient
encounters, but the first written mentions of interpreting date back to 3000 BC when the
Ancient Egyptians had a hieroglyphic signifying interpreter. The first known written
18
translations are those of the Sumerian epic Gilgamesh into Asian languages in the
second millennium BC. Other well-known translations from later periods include the
translations of Indian sutras into Chinese and the adaptations of Greek texts by Roman
poets (TranslationRealm 2014).
Throughout history, the practice of translation from the technological point of view has
not significantly changed. The ancient translators, as well as their Arabic, Renaissance
and early-modern followers, did not use any technology supposed we do not take ink
and paper and later pen and paper as great technological support. Therefore, the period
from the very first uses of translation up until about 1980 can be regarded as the first era
of translation the era of no translation technology (see figure 3.). It is hard to imagine
19
pen and paper translators working effectively in the globalised and profit-oriented world
of today, with some honourable exceptions, of course. However, it ought to be at least
borne in mind that these obsolete practices constitute the norm in what is by far the
largest part of translation history.
4.2
The Second and Third Eras TM, Glossary Technology and Workflow
Systems
Around 1990 the process of translation started to move from pen and paper towards
digital media. The amount of content to be translated grew, particularly as a result of the
growing use of personal computers. Translation demand was moving from documents to
the translation of software and then websites. According to Boucau, it is the 1990s that
introduced translation as a commercial activity on a truly global scale and the growth of
the business outstripped that of world trade as a whole (2006, cited in Hartley 2009:
106).
The transition from paper to digital content is technologically intertwined with
the use of CAT (Computer Aided Translation). This technology can be firstly understood
generally as MAHT (Machine Assisted Human Translation). Its aim is to boost the
performance of the human translator who is supported by digital tools such as
electronic dictionaries, various specialised terminology databases etc. The translators of
this era cease to make use of typewriters and printed dictionaries and move on to use
word processors and full-text search enabled by the electronic tools mentioned above.
Secondly, the term CAT can be viewed more specifically as specialised translators
workbenches. These are programs based on TMs (translation memories) which are
basically parallel corpora of previously translated sentences and are the key feature of
CAT. Therefore, computer aided tools are sometimes referred to as TMs.
20
21
part of the European Union. The translator could no longer rely on their linguistic
knowledge; market demands commenced their transformation into technical experts.
4.3
The idea of a mechanical tool which would enable automated translation has been the
dream for a long time. It is often found in modern science fiction the universal
decoder in Star Trek or the Babel fish from the Hitchhikers Guide to the Galaxy and
as mentioned by Somers, the idea itself antedates the existence of computers by several
centuries (2003a: 4). One of the first attempts to survey the thesis of mechanical
translation dates back to the 17th century when the German scholar Joachim Becher
created a ten-thousand-word dictionary with special codes for each entry (Krl 2003: 2).
When we move forward to actual experiments, the first two patents were issued in 1933
in France and in the Soviet Union, the former to Georges Artsrouni and the latter to Petr
Petrovich Smirnov-Troyanskii (Somers 2003a: 4). Even though their inventions proved
to be ahead of their time, they did not bring about any revolutions and were appreciated
only many decades later.
With the commencement of the Cold War came large volumes of investments in
machine translation research, and it is this period that is regarded as the starting point in
the history of MT (Somers 2003a: 4). Both the United States and Russia used the new
invention of the computer for code-breaking and Warren Weaver of the Rockefeller
Foundation and a British cryptographer Andrew D. Booth between 1946-47 developed
the idea of using computers for translation (Krl 2003: 2). The following years saw
further waves of governmental subsidies and funding worldwide, however, this was to
be changed after the publishing of a report by the Automated Processing Advisory
Committee (ALPAC) in 1966. The early optimistic expectations which predicted a swift
22
23
the marketing message forget about MT, it does not work, buy our TM products instead
(Meer 2005).
As mentioned above, CAT became the norm in the translation industry. They
brought considerable productivity and quality improvements and sped up the process of
translation. The companies which started using CAT as a competitive advantage to
reduce costs and offer better prices were gradually losing the cutting edge. According to
Zetzsche, the translation buyers started to take control of the translation process and
pricing as a result of using TMs and workflow automation (2012: 50). The inexorable
demands of the market can be seen as one of the factors which brought back the idea of
using MT as a productivity tool. Economics is always the same, and as Nogueira and
Semolini write [a]ny product or service, including translations, can be defined in terms
of three parameters: cost, delivery time, and quality (2010). Language Service
Providers (LSPs) were forced to reduce prices, the demand for translation was growing
rapidly, and MT logically started to be reconsidered as a possible solution. All machine
translation systems offer unbeatable speed, technological advances have significantly
reduced the costs, and quality has always been an impalpable notion in Translation
Studies.
The growing volume of content to be translated is another factor which has
favourably influenced the return of machine translation to the spotlight. Data collected
in the TAUS report shows that demand for applications and solutions based on machine
translation systems rose sharply between 2000 and 2010 (2013b: 36) and this growth is
unlikely to slow down. The same trend was acknowledged in the European Parliament.
Poulis and Kolovratnik document that the overall number of source pages translated by
the Directorate General for Translation increased from 43,963 in the first quarter of
2010 to 60,275 sources in the relevant period of 2012 (2012). It is clear that the EP will
24
have to adopt an advanced language technology if it wants to maintain its current level
of linguistic services and continue to tackle the challenges of multilingualism. Machine
translation is an ideal candidate. Moreover, it has been successfully implemented by
another body of the EU, the European Commission, since as early as 1976 (Garcia
2012: 297).
Both the growing volume of translation requests and the changing market
demands in terms of costs can be subsumed under the heading economic reasons.
Zetzsche identifies three other factors behind the rehabilitation of MT as a productivity
tool. The first is the dire need for automated translation revealed in the aftermath of the
infamous 9/11 events, secondly the technological advance in the form of statistical
machine translation which enables a relatively swift creation of MT engines for a
myriad of languages, and lastly the change of perception in relation to translation
quality (2012: 51). Even though Zetzsches first factor may be rather debatable, it is true
that it opened further government funding of MT in the USA, since it was recognised
that globalisation and previously unseen communication trends in cultural exchange
needed to be followed in real time and for a wide variety of languages. Any such
tracking is beyond the means of human translators. A progress in the development of
statistical and data-driven machine translation is a natural consequence of scientific
research in general, and of the globally scaled up funding. The third factor, that of
changing quality perceptions, can be completely agreed with. It was brought about
partly by the unparallel use of free online machine translation systems for gisting
purposes by which the users expectations of quality differ from the traditional perfect
translation, and partly by the spread of gisting practices in the corporate sector and
among LSPs.
25
The paragraphs above describe how MT came to be regarded as a viable tool for
providing translation services in the last two decades or so, and how it was adopted as a
reaction to the inability of CAT to cope with new technological, economic and social
trends. It should be noted here that this does not mean that there has been any
abandonment of CAT among translation professionals. Au contraire the rather stale
translation features which have been left untouched for almost a decade, except for a
few minor improvements, such as quality assurance of various kinds, are now enriched
by the integration of free and customer built MT engines. At first, CAT producers
started to add internal connectors to online translation engines like Google Translate and
Microsoft Bing (Zetzsche 2012: 51); nowadays this integration includes customer-built
MT engines which are often an integral part of the translation process. Almost any
translation environment used today first offers a fuzzy match from a TM, and if no
results are found, an MT output is suggested for revision by the human translator. This
choice of using the CAT environment instead of creating a new one appears to be a
clever move from MT developers. Translators are likely to embrace the idea of using
MT when it is a part of the workflow and environment they are already using and are
comfortable with. This fact has been also confirmed by the results of the quantitative
research in this thesis, where most translators do not feel a need for a special postediting tool to be developed (Question 17 Do you feel a need for a special post-editing
tool as opposed to the integration of machine translation into traditional CAT tools?).
Only one of the respondents would appreciate having a special tool for post-editing at
their disposal. Interestingly enough, this respondent has the most extensive experience
with post-editing, has a favourable attitude towards it, deals with it regularly, and PE
constitutes a fairly substantial part of his/her work (25-50%). Even though it is only one
person to suggest a special tool for PE should be developed, their experience with the
26
task may call for further research among professionals who deal with post-editing on a
more regular basis than most of the participants of the present survey. Similar research
could bring potentially valuable information to MT developers.
What does this mean for the job of the translator? Statistical machine translation
is expected to improve in the near future, and whether it will be used just as one of the
translators translation environment tools as suggested by Zetzsche (2012: 51), or as the
primary setup for translators to post-edit as suggested by this thesis, or even as the new
lingua franca as prophesied by Ostler (2011), there is not much we can do about it.
Machine translation is here to stay and the decision as to whether translators will be
forced out of the market by machines lies with the clients, not the translators (van der
Meer 2013; Nogueira and Semolini 2010). Translation buyers will take charge just like
they did in the case of CAT. It is clear that a human-driven translation process without
the use of MT technology will not be sustainable in the years to come.
The previous sections have documented the evolution of translation technology from the
pen and paper beginnings of the profession, through the streamlined use of personal
computers and CAT, to the return of machine translation and its integration into the
translational workflow. It is beginning to be clear that MT is going to play a crucial role
in the near future, as it is a hot topic in several fields and for a number of reasons. The
previous chapter identifies these reasons particularly as technological, economic and
social and is thus in accord with many other researchers and studies looking into this
topic. For example, Arnold et al. claim that MT is an important issue for social,
political, scientific and philosophical reasons (1994). This all is understandable, but how
is MT related to the idea of post-editing, and the discussion of MT to the central
27
argument of this thesis? Even though the idea of MT has always been predominantly
focused on fully-automated machine translation (FAMT), post-editing has been there
from early on. Garcia reports that the main traits of post-editing were mapped before it
could be even used in practice, that is by the mid1950s (2012: 300). The first research
into post-editing came as a by-product of MT evaluation, and the auxiliary nature of
dealing with post-editing can be traced in much of post-editing research history.
Research into post-editing (PE) can be tracked in the same waves as research into MT,
with a significant slowdown after the publication of the ALPAC report, and with a
renewed interest in the 1980s in the European Commission and in the 1990s globally.
Thus it can be stated that the concepts of MT and post-editing are closely related or
even intertwined. According to Garcia, it was also the wide availability of free online
MT output for average web users that brought much attention to post-editing (2012:
299). This can be understood in the light of the changing quality perception as described
above raw MT output (in large volumes) was ready to be either consumed
immediately, or it pointed at the need for post-editing. Nowadays, post-editing is
practised widely in the localisation industry and also among volunteer translators. It
attracts a lot of attention in research at universities around the world, and it is gaining a
multidiscipline overlap. State-of-the-art technologies are used in empirical research, for
instance eye-tracking or magnetic resonance imaging. Given the unavailability of
FAHQT (for more details, see Chapter 4.3), which is unlikely to change in the near
future, it is probable that post-editing will move to the fore of commercial and academic
interest for the reasons described above. Let us now have a look at what it actually is,
and how it can be put to use.
28
POST-EDITING
6.1
What is Post-Editing?
The term post-editing (sometimes also used in its non-hyphenated form postediting)
refers most commonly to a task related to machine translation, even though the term has
been used in various subfields of natural language processing (Allen 2003: 313).
According to Wagner, post-editing is different from traditional translation as it entails
correction of a pre-translated text rather than translation from scratch (Wagner 1985:
1). Looking at other definitions, however, this does not always have to be the case, since
the post-editor edits, modifies and/or corrects pre-translated text that has been processed
by an MT system from a source language into one or more target languages (Allen
2001: 1), which can mean tidying up the raw output, correcting mistakes, revising
entire, or, in the worst case, retranslating entire sections (Somers 2001: 138).
The task of post-editing should not be understood as a separate process of
editing, revising or quality assurance, but rather as an integral part of the automated
translation or localisation process (TAUS Report 2006: 2). The integration of postediting into the translation workflow has led to the creation of a new role in the
translation industry the role of a post-editor.
6.2
29
use and acceptance of half-finished texts. Allen (2003: 297-298) claims that in HT, the
question of half-finished texts has no relevance, since professional translators produce
finished, high-quality texts of publishable quality (with the exception of the few cases
where inexperienced or slapdash translators are involved), whereas raw MT output
constitutes a text which is inherently finished only partly or incompletely and may also
be called a quasi-text. When dealing with such texts, post-editors need to overcome
their sense for perfectionism resulting from their experience with professional human
translations. The question of acceptance of the fragmented or incomplete texts produced
by MT and the effort needed to post-edit them into a fit-for-purpose final product is
therefore one of the prime considerations in post-editing, but a non-issue in human
translation.
There are another two important differences between MT PE and HT mentioned
by Allen (2005: 3). The first one is that human translation is not perceived just as a
mechanical product, but rather as a piece of art crafted and fine-tuned by the translator
to emanate the best quality achievable given the relevant temporal requirements. The
post-editor, on the other hand, more often than not has to disregard any stylistic nuances
and provide a final text which is by no means a work of art, but more of a functional
product to fit the required purpose. The other difference is fairly similar and reflects the
more artisanal nature of HT as opposed to the more mechanical procedure of PE. The
translator in human translation does not simply follow a list of rules in a linear manner,
but has to grasp the meaning and all subtleties of the source text perhaps even more than
its author, whereas in post-editing the process is rather more mechanical and assemblylike. I suppose the computer-generated output is not a result of understanding of the
source text, it is just a computed translation. A translation for which the post-editor must
make certain that it reflects the meaning of the source text.
30
Finally, the difference between human translation and post-editing can be looked
at in terms of which human steps are involved in each of the tasks. A summary of the
steps is offered in a TAUS report on post-editing in practice (TAUS 2010). The
translator in HT first reads the source text, then he/she prepares an internal translation
(in their brain), and finally he/she types the target version using their working interface.
The post-editor, on the other hand, starts by reading the machine-translated raw output
which he/she then compares with the source text. The third step is to produce an internal
judgement on the quality of the offered solution based on a set of instructions from the
client or on some explicit guidelines followed for the specific post-editing job. After
that the post-editor has to swiftly decide whether the translation is appropriate having
the instructions in mind. Lastly, the post-editor either edits the text to improve it to
match the desired quality, or he/she needs to translate it from the scratch. The two
processes are displayed in the figure below (see figure 4).
31
Figure 4 clearly shows the main difference between HT and post-editing. In PE, the
post-editor starts with reading the target text, while the translator in HT starts by reading
the source text. This holds true even for the revision of human-translated texts
Guerberof Arenass research among professional translators suggests that the Source-toTarget direction is the most common practice when revising fuzzy matches from
translation memories (2013: 81). It follows that also revisers of HT (or post-translators,
see Chapter 6.3) work in a different sequential manner than post-editors. It should be
argued that translators and post-translators starting to work as post-editors are not used
to this working sequence and that may lead to some initial difficulties in post-editing
which can only be remedied by practice and training.
6.3
When speaking about post-editing and revising human translation, it should be at first
made clear that there is no consensus in the English-speaking word as to the
terminology. Mossop (2011: 135), for example, uses the term revision to cover a
range of activities in the process of looking over a translation to decide whether it is of
satisfactory quality, and making any needed changes and considers these activities on
the basis of whether they are performed by the translator, a second translator or a nontranslator, where the third category also encompasses the job of post-editing machinetranslated output. Gouadec distinguishes between a post-translator and a post-editor
where the former checks, proof-reads and revises work performed by a human
translator, whereas the latter deals with adjusting texts produced by machine translation
engines (2007: 114). This is a satisfactory working definition, and since the use of
terminology is not unified, let us adopt this distinction for the purpose of this thesis.
32
33
to the document being reviewed, the more favourably is their work viewed by
themselves and other members of the translation workflow. Should such a posttranslator transfer his skills and working habits into post-editing, this might, according
to Allen, lead to two possible results. Either the final product will be over-corrected,
which will result in a loss of the productivity gains expected of post-editing, or it will be
under-corrected with the effect of having an insufficiently reviewed final product. It is
clear that neither of these scenarios is desirable and that post-editing is distinctly
different from post-translation. Since post-editors and post-translators are expected to
carry out differing tasks with differing results, it is only logical for post-editors to
possess distinct skills from post-translators. The question of post-editing skills will be
dealt with in detail in a separate section.
6.4
Types of Post-Editing
Traditional human translation does not usually distinguish between different types or
modes of the activity. It could be safely said that a substantial part of HT is aiming at
high-quality translations. Other modes play only a minor role, for example summarising
the content of documents for the customer to see if they are worth translating in full.
However, one of the most notable differences between PE and HT is that post-editing
has several types which serve various purposes and which will be dealt with in this
section.
Loffler-Laurian makes a very basic distinction between fast post-editing and
conventional post-editing (1996, cited in Doherty and Gaspari 2013). The first type
requires only essential corrections and produces quick turnaround; conventional postediting, on the other hand, offers slower turnaround and requires more corrections
leading to a higher quality. This distinction is based on two factors only namely the
34
final quality and time/productivity and as such is insufficient to describe all the
possible applications in which PE may be used. Allen (2003: 301) provides a list of
factors that guide the type and degree of post-editing to be used. A summary of the list
is to be found below (see table 2).
As we can see, translation quality and turn-around time are only two of the several
factors on which the level of PE to be used is dependent.
The purpose or function of the text, that is its future use, is the most important
criterion according to which a more detailed division of PE is drawn. The two main
approaches to using MT texts are for inbound or outbound translation (Allen 2003:
301). Inbound translation, also referred to as MT for acquisition, indicative translation
or MT for assimilation, is simply translation to understand. The main purpose of such
translations is to get the idea across, to provide an understandable target text. Style
inconsistencies and grammatical errors are acceptable as long as the message of the text
is comprehensible. Outbound translation, also known as translation for dissemination, or
translation to communicate, is used for the translation of texts to be published. The raw
MT output has to be manipulated or corrected so as to ensure a publishable quality.
Both approaches to the use of MT texts can be further subcategorised according
to different levels of post-editing that needs to be applied as seen in the following figure
(see figure 5).
35
Figure 5: Machine Translation for Different Purposes (adapted from Guerra Martnez
2003: 19)
36
with HT. It is usually full of errors, but its main aim is to provide some basic
comprehension to its users.
Free Internet translation portals are not the only application of this approach to
MT. The other use of content gisting comes with companies offering customised
machine-translation solutions for corporate clients. These customised MT engines are
trained via user dictionaries, translation memories and alignment of older translated
documents and thus they offer superior quality compared to free on-line systems.
Another important advantage of similar systems is that they are usually integrated with
clients intranets, portals and self-service support centres, which means that they enable
a real time communication in cases where it did not use to be possible in the past. Some
of the resulting benefits of customised systems are mentioned in the TAUS report and
include reduced customer support costs for fewer contact centre calls, increased
efficiency of corporate staff, and increased security as confidential information is not
sent to free translation systems on the Internet (2006: 4). Customised machine
translation engines are by all means superior to their free online alternatives, however,
their use is still rather limited. For example, only two of the respondents of the
questionnaire accompanying this thesis admitted they post-edited output from customerspecific engines (Question 16 Where does the machine translation output you postedit come from?). Given the still rather negative attitude of some translators towards
MT PE, one of the biggest challenges of the machine translation industry is to sell the
idea of building and using customer-specific engines, as opposed to using free online
systems. With the growing use of customised engines perhaps even the attitude of
translators who do not see MT in a very favourable light may start to change.
37
Rapid post-editing provides a strict minimal editing of texts in order to remove only
the most blatant and significant errors (Allen 2001: 41). In this approach to MT output,
a minimal number of corrections is applied so as to provide a comprehensible text
removed from the most serious mistakes left by MT. Stylistic issues are not considered
whatsoever. It is used mainly for perishable documents with a short life span and
it came into existence to provide translations for urgent texts that are intended
merely for information purposes or for restricted circulation, such as working
papers for internal meetings, minutes of meetings, technical reports or annexes.
(Allen 2003: 302)
38
must be post-edited, because the information is expected for dissemination among third
party users. MPE is typical for the industrial and corporate sector, particularly for the
automotive and machinery industries. Guerra Martnez asserts that these are usually
technical texts or manuals describing the operation and servicing of machinery, and that
it is in their nature to be read in a non-linear fashion (2003: 20). The post-editor needs to
apply only a minimal number of changes to make the text comprehensible as required
by the text purpose and the client. For example, in the case of the technical texts
mentioned above where their readers often have to follow instructions in a step-by-step
sequence, any PE changes of cohesion should be made only to those parts containing
instructions to make the final text a comprehensible manual with a minimal post-editing
effort. Problems with cohesion in other parts of these texts would be left unattended.
Similar PE changes may differ across industries or textual types, and the
minimal amount of changes is hard to define. Allen (2003: 304) claims that minimal
post-editing is a fuzzy, wide-range category, since the decision as to the degree of
changes depends on the individual post-editor.
Minimal post-editing is a highly functioning approach, however, there are not
enough publicly available materials specifying the criteria to fully embrace it. Materials
of this kind do exist, however, they are unfortunately usually intended for internal use in
specific corporations only. This lack of clear guidelines constitutes one of the key issues
of post-editing in general and there are efforts to create and advertise PE guidelines (see
Chapter 9.2) so that they can be globally adopted, which would in turn lead to a better
understanding and use of PE both on post-editors and customers part.
39
40
6.5
The previous sections have outlined the different types of post-editing and how they
differ from each other. It is now time to discuss what implications the types of PE used
have for the translation profession and for human labour in general. The following table
41
(see table 3) summarises the types of PE and whether human intervention in the postediting process is required.
HUMAN
INTERVENTION
Content Gisting
Rapid post-editing
Minimal post-editing
100% MT
Full post-editing
NO
YES
YES
NO
YES
The table shows that there is no human intervention in content gisting and in 100% MT,
which means that there is no work for the post-editor in this type of MT PE scenarios.
This does not clearly have to be regarded as a threat to post-editing and translation jobs
in the latter case, since 100% MT is only applicable for a minority of translation
purposes as discussed above, and the near future is unlikely to bring a functioning
FAMT for dissemination purposes. The former case is slightly more complicated as
there are two distinct uses of content gisting. The first one concerns the use of MT raw
output from free on-line systems and it is the most frequent implementation of machine
translation. In 2012, for example, the Google Translate system had two hundred million
translation requests (Schulz 2013), and at the time of writing of this thesis the system
operates in 80 languages. Looking at the numbers, one would be inclined to think that
this is definitely an area in which machines are replacing humans, however, it must be
taken in consideration that a large proportion of these translation requests are for texts
that would not be otherwise translated, and MT use is here an alternative to no
translation at all. This is a generally accepted opinion and MT for content gisting
provided by free Internet systems should not be viewed as a threat to professional
human translation and post-editing either. Customised machine-translation solutions are,
42
on the other hand, a different matter. Here some of the work done by the MT system
was formerly done by human translators, post-editors, helpdesk specialists or other
professionals, and similar MT solutions can, in consequence, jeopardise prospective
human jobs. The translator is no longer needed, and [the] number of words is no longer
a cost driver (Moravia).
The three remaining post-editing types all require some kind of human
intervention and are, therefore, likely going to lie at the core of post-editors work.
These three types differ in terms of the degree of PE applied, with the least post-editing
effort invested during RPE, and the highest number of corrections in FPE. It should be
noted that the boundaries between RPE and MPE are blurred with certain overlaps.
There is no particular set of differences in the steps taken in RPE and MPE. The main
distinction between these two types is that of the intended use or purpose of the
translated documents. MT RPE documents are intended for internal purposes, while MT
MPE ones are to be published, just like FPE documents. In terms of human intervention,
it could be argued that translators/post-editors of varied statuses (see Chapter 7.3.2)
would be involved in the different degrees of post-editing regardless of the
publication/non-publication distinction. The partiality or fullness of the post-editing
process will be the deciding factor in job allocation. Since RPE and MPE require at
least some guidelines as to what should be post-edited and what is acceptable, I dare
suggest that they will be post-edited predominantly by in-house translators working for
corporations developing their own guidelines. Moreover, the high productivity of RPE
and MPE will not be hindered by a process of allocating and outsourcing jobs to thirdparty contractors. This does not mean that all of the RPE and MPE documents will be
post-edited by in-house translators, however, in the majority of cases freelance or
translation company translators are expected to carry out FPE jobs, because the jobs
43
44
Question 5: Do you think that machine translation will replace the human translator for
non-literary translations in the future?
Table 4: The Future of Machine Translation as Viewed by Translators
Strongly disagree
Disagree
Neither disagree nor agree
Agree
Strongly agree
7
3
11
6
2
0
14%
50%
27%
9%
0%
The previous sections have defined post-editing and its various implementation
scenarios. It has also been established that post-editing is different both from the
revision of human-translated texts and human translation. This chapter seeks to answer
the questions of who are the translators of yesterday and the post-editors of tomorrow,
and what skills are associated with professional post-editing as opposed to human
translation of texts. It is further established that it is the professional translator that is the
ideal candidate to become a successful post-editor.
7.1
45
46
the client. Terminological consistency, spelling, morphology etc. need to be taken care
of during this phase.
47
7.2
To arrive at a more specific categorisation of skills, a set of skills can be drawn from the
processes involved in PE. Rico and Torrejn group the competences and skills based on
Krings and Kobys model into core competences, linguistic skills and instrumental
competences (2012: 169). Their categorisation is a very useful one, as it encompasses
most of the skills and competences suggested by other researchers under only three
main headings (see figure 6). The individual categories of skills and competences are
discussed in the subsections below.
Figure 6: PE Skills and Competences (adapted from Rico and Torrejn 2012: 169)
48
49
post-editors. For instance, Vasconcellos and Lon in their research at the Pan American
Health Organization concluded that post-editing requires a trained professional translator
(1985: 125), who is by definition bi- or multi-lingual. They preferred trained professional
translators over subject-matter experts for post-editing jobs, however, they acknowledged
that such a translator should possess good knowledge of the subject matter vocabulary in
both languages, and a technical understanding of what is meant by the text (1985: 125).
Current research and practice, with just a few exceptions, show that scholars and
practitioners are in accord with the preference for the bilingual post-editor. It is still true that
post-editing may be carried out by translators, editors, bilinguals, and even monolinguals
(Doherty and Gaspari 2013), but the majority of authors seem to be in agreement that the
ideal post-editor is a professional translator trained for the task. For example, Joscelyne
explains that post-editing involves linguistic more than area specific skills and is
performed best by alert translators, familiar with machine output, working in a standard
translation environment (TAUS 2006: 2). Krings and Koby assert that
only a translator can judge the accuracy of a translation, [he is] the one best able to
pick up errors in the machine translation, he has a fund of knowledge about the
cross-language transfer of concepts, and he has technical resources at his disposal
which he knows how to use in the event of doubts. (Krings and Koby 2001: 12)
Similar statements must be agreed with, as the translator has excellent knowledge of the
source and target languages, and they also possess good revision skills which they use
to perfect their translations. Professional translators also meet other requirements to be
successful post-editors, as seen in the following sections.
50
51
in recent research. This trend is supported by the data compiled in the questionnaire
accompanying the present thesis. Nineteen out of twenty-two translators in total did not
feel threatened by machine translation in their profession (see figure 7). The three
translators who regarded MT as a threat to their profession have all worked in the
translation industry for ten years or longer, which matches the expected feelings of more
experienced translators discussed above. These experienced translators quite often see
the practice of MT PE as a degradation of the profession, as expressed by one of the
three translators. Nevertheless, a more positive attitude towards MT does not
necessarily mean that translators like it or prefer it over translation from scratch. When
answering Question 12 (Do you like machine translation post-editing compared to
translation?), six of the respondents expressed a clear preference for traditional
translation, while only two translators answered that they favoured MT PE. This might
suggest that translators view PE as a necessary evil and still prefer translating from
scratch.
52
7.3
The preceding subsections have demonstrated and suggested the skills a post-editor
ought to have in order to successfully fulfil the tasks related to the profession. A few
53
comments are needed to discuss the profile of current translators and future post-editors
from a slightly different perspective, looking briefly at who the translators are from a
demographic perspective and discussing the different job statuses of both translators and
post-editors.
54
budget. However, this trend could be reversed and many more men have been attracted
to the translation profession in recent years. According to Gouadec, this is a result of the
increasingly technical nature of ST materials, the widespread development of IT
translation tools, the diversity of the translation jobs and skills required to do them, and
the turn of many professionals away form technical fields into translation (2007: 88). It
would be wild guessing to claim that either men or women will tend to choose the postediting profession, however, the author of this thesis believes that a balance between
men and women may be maintained as a result of two contradicting forces. Firstly, postediting rates are generally lower (and sometimes may not be compensated by higher
productivity) than rates for HT, and as such may be again seen as a source of second,
supporting income, thus attracting more women. Secondly, however, given the more
technical nature of the skills needed in the translation profession, which is also valid for
post-editing, the number of women may be compensated by more technologically
minded men.
55
salaries may be considerably lower, they cannot choose translation jobs, as they have to
translate everything that arrives on their desks, and most importantly, they do not have
the level of freedom and independence as freelancers do. Freelance translators are selfemployed and work either directly for the client or as subcontractors for translation
agencies. Among the advantages of this status can be counted the independence and
possibility to accept and reject jobs and thus influencing ones own time and areas of
interest. A major disadvantage is connected to the long working hours which freelancers
usually have to work, since their income is dependent on how much they translate. They
also have to invest in special training and all the technological equipment necessary for
their job. The distribution of translators into these two categories is difficult to asses, but
the numbers are growing for both groups (Gouadec 2008: 102). Pym et al. in their study
for the DGT of the European Commission in 2012 summarise data from several surveys
and document that 50 to 89% of translators are freelancers (2012: 89). Usually, the
people working in salaried positions have to be more experienced and be able to offer
skills that freelancers do not possess, as the translation market is crowded with
newcomers, graduates and would-be translators, and the number of salaried positions is
limited. Gouadec claims that to get a salaried job, a translator needs expertise and
skills way beyond those required for just translating such as special subtitling or
dubbing, special IT skills, special terminology or language engineering skills, [and]
special post-editing skills (2008: 103, emphasis added).
This is where we return to post-editing per se, as this situation with translators
reflects the situation with post-editing. Allen maintains that the majority of experienced
post-editors work as in-house personnel, while a growing number of human translation
specialists are being recruited as external, freelance post-editors (2003: 299). This
means that most post-editing jobs are done by in-house translators who have the
56
57
58
59
60
61
phenomena clearly document the collaborative nature of the translation profession in the
21st century and should be considered in translator training. Collaborative mode of
translation is back and, to borrow OBriens words, the ability to translate in a
collaborative way is a skill that professional translators will need in the future (2011a:
20). Such collaboration will often include post-editing.
TRAINING OF POST-EDITORS
Looking into history, there have been plenty of models of translational competence and
requirements for translators. These were then usually taken as guidelines for translators
or as indicators of good practice, and the training of translators (once translators were
formally trained) often derived from these models. Virtually all of them carry the mark
of their era and the technology available at the time of their creation. For example, a
good translator, according to Dolet who wrote this in 1547, needs extensive knowledge
of both languages involved in translation, which was something new, since previously
translations were carried out in teams of usually monolingual experts (cited in Pym
2012). Nowadays, most of the dominant models of translation competence are multicomponential. The model developed for the European Masters in Translation can be
taken as an example of this (see figure 8). Based on this model a translation service
provider has to possess language, intercultural, thematic, technological and information
mining competences. Even though this is one of the most dominant models and
university education of many professional translators is based on it, it is clearly out of
date now. Admittedly, technology is taken into account in this model, however, skills
related to machine translation, and by extension to post-editing, are only represented in
the form of knowing the possibilities and limits of MT, which means, as Pym
mentions, that a translator should know about the existence of such technology, but in
62
reality does not need to have any skills related to it (2012: 7). Skills related to MT and
PE have been identified as belonging to the group of instrumental competences (see
chapter Post-Editors Profile and Skill-Sets above). This is clearly insufficient in the
light of the information presented by this thesis. According to Gouadec, a good
translator must have an absolute knowledge of what translation means, what it requires
and what it implies (2007: 150, emphasis added). The growing use of MT and postediting resulting from technological advance and (not only) economic globalisation are
becoming an integral part of translation and will make it impossible for translators/posteditors to work effectively without knowing about and having the skills to use machine
translation and related technologies. However, the increasingly dominant role of postediting and MT in the translation workflow is relatively new to the translation industry,
which means that there is no accord in how to train future translators/post-editors. The
present section endeavours to contribute to the debate on translator training.
Figure 8: The EMT Model of Translation Competence (EMT Expert Group 2009: 4)
63
Post-editing has been identified as a separate skill which needs to be acquired separately
(Belam 2003: 1). It has also been shown that PE skills are developed gradually
(Vasconcellos 1986: 145) and these skills need to be honed (Somers 1997: 201).
Moreover, TM and MT skills are reported to have a steep learning curve (Garca 2006:
98). The author of the present thesis believes that future post-editors and companies
working in MT and PE scenarios would benefit from translation graduates who would
have certain grounding in MT and PE, as their knowledge and skills related to these
technologies would already be developed and these people would have a basic
competence which could be further built upon. However, the current situation is such
that the majority of post-editors receive in-house or on-the-job training, and there are
not many full courses on MT and PE in either translator training programmes or
elsewhere which would provide a building of instrumental competence. This practice is
confirmed by the data gathered in the quantitative research conducted as a part of this
thesis (see table 5). None of the respondents learnt how to post-edit at university or in a
specialised course. Even the small number of respondents clearly indicates that postediting skills are not something that could be currently learnt in any other way than on
the job (at least in the Czech Republic). On the other hand, the respondents who have
post-edited at least once tend to be confident in assessing their knowledge of the
working of MT. Their answers to Question 15 (Do you think you know how machine
translation works?) revealed that eight translators believe they know how MT works,
while only one does not. It must be noted, however, that the concept of knowing is very
broad and may mean different levels of understanding to different people. More over, as
discussed above, mere knowing does not suffice. The data at least tells us that the
64
translators who took part in the research are confident about their knowledge of how
machine translation operates.
0
0
8
1
0%
0%
36%
5%
Even though the situation is improving, most universities with translator training
programmes still concentrate on the development of the traditional translation-aswriting capabilities (Hennessy 2008) when, however, the role of translation as writing
is diminishing. There are several reasons behind this. Firstly, constraints posed by
technical, economic or legislative situation at educational institutions may hinder the
wider inclusion of MT and PE in the curricula. Doherty and Moorkens, for example,
report on the difficulties encountered in running their lab sessions which were a part of
a module on translation technology (2013: 127-128). Any such course, they argue,
ought to provide students with experience of a variety of tools, however, limitations can
be encountered resulting from either institutional flexibility (instructors not having the
rights to install/update software) and technological restrictions (limited hard drive
space, for example). Admittedly, similar problems may hinder the teaching of MT and
PE slightly, but they are not insurmountable. Secondly, the reluctance of the academia to
reflect on the development in the business sector can be attributed to the well-know
clash between academic and vocational education, a clash that the author of this thesis
has personally witnessed. The commonplace arguments include
65
[on] the one hand the proponents of a more theoretical approach to syllabus
design [who] minimise the importance of developing practical marketable skills
[on] the other hand are the teachers who point out with equal justice that a
course of study which has divorced itself from the practice of the subject in the
real world is worse than useless to its students. (Belam 2003)
The former argue that they, after all, represent the traditional academia whose aim it is
to nurture critical thinking, not to provide mere, simple-minded practical training
similar to dealings at vocational schools. The latter, on the other hand, more often than
not have to first tackle the barrier built up from traditions and biases of the first group in
order to point at the changing needs of the world of today. It cannot be said that one or
the other approach is the best and that there is a fit-for-all solution, however, in the case
of machine translation and post-editing it should be possible to find an equilibrium
between these two approaches. Even teaching about MT and PE can offer room for a
discussion of serious, theoretical issues, such as questions of authorship, translator
visibility or ethics.
Fortunately, there have been efforts to change the status quo both in the business
sector and in academia. One of the biggest challenges in the training of post-editors is a
lack of clear guidelines, of a concise syllabus and methodologies which would inform
interested parties about what any such training should include. Research and discussions
have been in progress in independent, commercial and educational organisations. For
example, the Post-Editing Special Interest Group was established by members of the
Association for MT in the Americas (AMTA) and the European Association for MT
(EAMT) in 1998 (Allen 2001; 2003: 299). This special interest group set as its main
objective to create post-editing guidelines and develop a post-editing qualification
programme for trainee translators. Another huge contributor to the discussion on post66
editors work, training and MT in general is TAUS which was established in 2004 in San
Francisco. TAUS started as a forum for sharing ideas and experience with MT
implementation, and later, among other activities, commenced producing various
guidelines related to PE. At the time of writing this thesis, it is possible to enrol in PE
courses provided by TAUS in the form of e-learning. A scheme of the development of
TAUS activities is provided in the picture below (see figure 9).
Universities, as discussed above, were slower to embrace and include the teaching of
MT and PE, however, especially departments involved in MT and PE research took an
active role in the discussion of translators/post-editors training and skill-sets. These
include, to name just a few, Dublin City University, Universidad Europea, University
College London or the University of Exeter. The endeavours of these institutions are
usually well-documented, published, and presented in journals and at conferences. A
67
reasonable approach would be to use these as examples of good practice, and base
syllabus-design at other universities on their findings and experience.
In order to propose a list of areas that translator training courses should focus on,
let us now examine the content of some of the courses at the abovementioned
universities.
8.1
Prerequisite Knowledge
Most syllabus designers believe that the translator/post-editing student should acquire
certain special translation skills before the introduction into MT and PE. Students taking
MT and PE modules and courses are usually already familiar with basic translation tools
and terminology management, alternatively these areas ought to be covered prior to or
at the beginning of the MT and PE course. It should not be difficult to meet this
condition, since nowadays courses on CAT and related technologies are well integrated
into the syllabus of translation training programmes. This could be a problem if the MT
and PE course is taught too early, or as a part of general (i.e. non-translation oriented)
programmes, similarly to the course run at the University of Exeter reported by Belam,
which is taught for final-year undergraduates studying a degree in modern languages
(2003). General linguistic programmes are not, however, the focus of this thesis; for a
discussion of how MT courses can be beneficial to foreign language learners see Lewis
(1997).
8.2
68
students to take a more active role in their own learning, any MT and PE course should
contain a practical, hands-on part in which the trainees could put to use the knowledge
acquired in theoretical sessions. A report by TAUS acknowledges this, mentioning that
practical training is required to spot and correct typical machine mistakes as quickly as
possible (2006). Providing this hands-on experience, according to Somers, used to be
rather problematic for translator training institutions, since the prices of translation
software were oriented towards professional users, not to educational establishments
(2003b: 320). This is, nonetheless, no longer a problem; the prices of commercial
software have fallen down, there are usually demos, trial or free versions which can be
used for training purposes, and it is frequently possible to arrange special deals for
educational institutions. All the three universities mentioned above include practical
training into their MT and PE courses, and report promising results in the development
of IT skills, specific translation and post-editing skills, attitude towards MT and PE, and
the self-efficacy in relation to the use and knowledge of machine-translation
technologies (Belam 2003; Dohery and Moorkens 2013; Rico and Torrejn 2004).
There is little reason not to include practical sessions into an MT and PE course in the
training of translators.
8.3
Course Outline
69
70
RECURRENT ISSUES
Not everything is plain sailing for machine translation. Some of the problems have been
resolved over time, others have been revealed only recently after MT PE practice grew
in popularity, and some have been around since the beginnings of MT research. This
section aims to briefly comment on some of the frequently discussed issues in
connection to machine translation post-editing.
9.1
71
The available figures on productivity are still rather scarce, but growing research has
supplied more and more data which informs about the productivity of MT PE. However,
the data often presents very diverse results which can be interpreted as being affected by
several variables such as the language pairs, quality of source text, availability and
quality of legacy content for the training of the MT engine, or availability of customerspecific dictionaries. The general presupposition when speaking about productivity is
that it improves over time, once the post-editor gets more experience with the intricacies
of the task. This is not, however, validated by the data collected from the questionnaires.
Six translators claim that their productivity did not improve over time, as they answered
Question 14 (Has your productivity when post-editing increased over time?).
Unfortunately, the data cannot be considered relevant in this case. Not only because of
the low number of participants, but also because of the haphazard nature of their
answers to this question (see Appendix 2). Some of the translators who post-edited just
once reported an increased productivity, which means that the data has no meaningful
value.
The variables and final results are many, however, one very logical
presupposition must be met in the first place. For any post-editing to take place, it must
be more productive than traditional human translation. Unfortunately, there is no
consensus among interested parties as to what productivity actually is. OBrien assumes
that research carried out by academics and by businesses focuses on productivity from
different perspectives (2011b: 9). She believes that the former is only interested in the
ratio of quantity and quality to time (2011b: 2), while the latter is also questioning
how is productivity affected by the cognitive effort extended on the post-editors part.
The viewpoint of the first group is logical, it has been discussed that one of the reasons
why MT PE is on the rise lies in economics of incorporating it in the translation
72
9.2
Quality
73
Quality in relation to machine translation post-editing may mean two things. Firstly, it is
the quality of the raw MT output, and secondly, the quality of the final post-edited text
as expected by the client.
The quality of the final post-edited text should derive from clients expectations
and the chosen level/type of post-editing. A lot has been written about this in the chapter
Types of Post-Editing. What remains to be commented on is one of the most prevailing
issues in post-editing research and practice, i.e. the lack of guidelines constituting an
industry-wide standard for post-editors and customers to refer to. A great deal of work
in this respect is being done by TAUS which regularly publishes guidelines on postediting practice, including guidelines regarding PE quality. These guidelines aspire to
become the worldwide standard, unfortunately, there is still an information vacuum
among many LSPs, some of whom offer their post-editing services on a haphazard basis
in relation to quality expectations. Moreover, quality evaluation sometimes suffers from
the stubbornness of universities and LSPs who refuse to cease comparing the final
translation product with a non-existent perfect translation. The expectations of
translation quality and purpose have changed, so this traditional belief in an ideal,
perfect translation ought to change as well. Different levels of quality should be
recognised and standardised in quality assessment models specially designed with this
in mind, such as the one proposed by Huang who believes that HT and MT are separate
identities with different standards (2011: 8).
The MT output quality is dependant on the same variables as were identified for
productivity in the previous section. The better the raw MT output, the higher the
productivity on the part of the post-editor, since not so many changes will be required to
achieve the desired final quality. For this reason, a lot of effort is currently invested in
improving the raw MT output quality. One way of doing so is including Statistical
74
Postediting (SPE) in the translation workflow (TAUS 2010: 22). SPE is an automated
process through which the machine translation engine learns from the changes made by
a human post-editor to the raw MT output. The problem with the raw MT output quality
is that it differs from engine to engine, from domain to domain, from language to
language etc. It will take a long time for machine translation to be able to provide
comparable raw output quality for not all, but many varieties of content.
9.3
Quality Assessment
Another issue related to quality is quality assessment. The most frequently discussed
issue in this area is determining the MT raw output quality. However, assessing the
quality of the final translation is also problematic. There are different approaches to
measuring the quality of MT output. Some of them are based on a typology of errors,
others track the number and type of changes made during the PE process and others
predict the effort which will be needed for post-editing. Currently the trend has moved
from human evaluation of raw MT output towards automatic methods of evaluation.
This is represented by so-called automatic evaluation metrics like BLEU, TER or
METEOR. These metrics provide automatable measurements that deliver a ranking
(TAUS 2010: 32) usually expressed by a percentage or another numerical
representation. Unfortunately, there are several problems with such automatic metrics.
The numerical representations produced by them may easily be misunderstood, and the
metrics require a certain level of technical expertise, which is sometimes lacking in the
field of professional translation where a large part of the human force is recruited from
among graduates with a linguistic background. Even though some of the limitations of
these metrics have been addressed and are constantly being reflected in current research,
their biggest disadvantage lies in their need of a human translation against which the
75
raw MT output could be compared. As such, these metrics are not suitable for practical
use in predicting productivity and quality of raw MT output, but rather for tracking the
progress in the development of different MT engines and in their further tuning. This
also means that automated metrics are of no relevance to the individual translator/posteditor, but, as of now, are tools used by MT developers and major LSPs with sufficient
human and technical resources and expertise. These tools are used prior to or after the
post-editing phase itself, not during it, and thus are not helping the post-editors in their
work.
A promising solution can be detected in up-to-date development of MT systems.
Specia et al. inform about a new function of MT called Confidence Estimation (2009)
whereby MT engines produce so-called confidence scores. These scores are generated
by the MT system without any reference sentences (which are required in the case of
automated metrics). The confidence scores are created based on the information
collected by the MT engine during the translation cycle. The confidence scores then
reflect how the MT engine is confident that the MT output is of good or bad quality.
Similar confidence scores, provided they prove to work well and are correlated with
actual post-editing productivity, may provide valuable information for post-editors
working on a given task. The majority of translators (six out of nine) taking part in the
quantitative research in this thesis agreed that including confidence scores with each
segment prior to translating would be valuable for their work (Question 18 Would you
appreciate access to confidence scores for every segment to be post-edited?). Further
research of confidence scores technologies is therefore advisable. Translator community
would definitely benefit from a wider inclusion of confidence scores in the translation
process.
76
The process of reading the source text and gauging whether it is suitable for
post-editing, or if it needs to be translated from scratch, could be significantly
streamlined. This could in turn lead to further productivity gains and perhaps even to
translators improved opinions of machine translation in general. What is important is
that this will have to be done in a way that will be useful and natural for
translators/post-editors. For instance, integrating confidence scores of machinetranslated output into CAT environments which the translators are familiar with and
where the confidence scores could be suggested in a similar way to TM fuzzy matches.
Unfortunately, very little research has been done in this area so far.
As mentioned at the beginning of this section, main focus lies on the evaluation
of raw MT output, however, assessing the quality of the final product is not without
problems. It is not easy to gauge the final quality of post-edited translations consistently
and at a reasonable price. Automated metrics are inappropriate for this, since they are
limited by their need for reference translation. Yet, assessing the final quality of MT
translations is important to provide customers with sufficient evidence that they are
getting the quality that they paid for. Attila Grg of TAUS has proposed a new model
for evaluating quality having this in mind (2014: 22-26). He suggests that quality should
be measured by crowdsourcing the text to be assessed by a group of volunteer
evaluators. This approach has been shown to correlate well with professional human
evaluators and has the advantage of greater speed and lower price. This cannot work,
however, until an industry-wide standard in relation to translation quality is accepted.
9.4
Pricing
Speaking about pricing of machine translation post-editing is like walking on thin ice.
You never know if your next step will lead to trouble and there is hardly any going
77
back, since there are opposing opinions on MT PE pricing, and establishing a method
favouring one or the other side may lead to catastrophic consequences for future
practice. According to TAUS, the two most popular approaches to PE pricing are paying
for post-editing as fuzzy segment matches, or paying a fee based on the time spent
(2010: 32). Translators tend to favour the former option, as they can increase their
earnings by working faster. On the other hand, it is difficult to agree on a reasonable,
fair per word rate for many LSPs and translators/post-editors, so some still prefer to
choose the latter method. The translators taking part in the quantitative research of this
thesis confirmed the popularity of MT PE pricing based on an agreed per word rate (see
figure 10).
A further analysis of the data revealed that only one of the respondents sometimes
charges an hourly based fee for MT PE. The other translator in the other category
mentioned that he/she post-edited only once and that they were paid a per word rate.
Charging an hourly rate might sometimes be very reasonable, especially in cases where
the raw MT output is of uncertain or dubious quality. The preference for per word rates
of the translators in this research is therefore slightly surprising. Alternatively, this
78
might suggest that the reason why some of the translators took a PE job only once was
that they were not satisfied with the money they received for it. Unfortunately, it is
impossible to clarify this hypothesis from the quantitative data available.
Even though charging a per word rate seems to be typical of MT PE, it is rather
problematic. Four of the nine translators expressed their dissatisfaction with the pricing
when answering Question 11 (Are you satisfied with the price and the pricing method
for post-editing?). Translator number 12, who sometimes post-edits, even claimed that
PE is extremely underpaid. The greatest challenge in establishing a fair rate for postediting is that there are no industry-wide standards and recommended rates. An article
by Asia Online (a major MT developer) documents the early practices among LSPs,
who arbitrarily set a lower per word rate without any attention to the quality of the MT
output (2012). These arbitrary, low levels led to protests from post-editors and LSPs
which found themselves losing money. The questions of what level of fuzzy match
should the rates for post-editing be based on, and whether the level of effort required for
PE is similar to the effort when editing TM matches are therefore hot topics in current
research. OBrien, for instance, has demonstrated that the cognitive load for machine
translation matches is similar to fuzzy matches in the range of 80-90% (2007: 185).
Similar findings, of course, should be reflected in pricing schemes if they are replicated
in larger scale research and for more language pairs. In the meantime, other pricing
methods are being developed. For example, Asia Online bases their metrics on
productivity. They count how long it takes a human post-editor to edit an MT output
when compared to human translation only, and discount the productivity gain, adding an
incentive for post-editing. The translators are thus being paid less per word, but more
per hour in total (2012). Similar approaches may lead to a wider acceptance of PE
practices among professional translators. TAUS in its pricing guidelines suggests that a
79
combination of approaches should be used when setting rates for post-editing (2013a).
This should involve a three-fold approach using automated quality scores, human
quality review and productivity assessment (similar to the model used by Asia Online).
However, following these guidelines may be rather difficult to implement and it may
take a long time for the researchers to produce schemes that will be universal and quick
to use. Until then, it is best for any translator/post-editor to try and choose whichever
approach suits them best and cooperate with LSPs which will offer the desired pricing
scheme.
80
81
owned a smartphone by 2016, and that the students of today are the translators of
tomorrow (2014: 18). Applications designed specially for post-editing on mobile phones
already exist, and so do translation/post-editing scenarios in which they are used. For
instance, a new application called Kanjingo is being tested at Dublin City University in
cooperation with the organisation Translators without Borders (OBrien et al. 2014).
The volunteering translators contributing to a good cause are making use of the
converging technologies of machine translation, post-editing applications for
smartphones and collaborative framework provided by data sharing and cloud
technologies. The translation future is set to witness a lot more collaborative,
crowdsourcing non-professional translation.
These are some of the possible directions of future trends with machine
translation technologies which are certain to evolve, but what about the translation
profession and the world at large? The growing use of machine translation post-editing
will create an environment where the individual freelance translator will stand no
chance against the economic and technological resources of giant language service
providers, with a notable exception of few translators concentrating on niche markets
consisting of what the machine translation engine will not be able to translate
effectively. The freelance translator will have neither the economic means nor the
temporal capacity to invest into buying new technology, software and training. The
LSPs may be forced to push the prices down to the lowest possible level, which would
not be profitable enough for freelancers to make a living. Reverse auctions where the
lowest price wins may become the norm. The freelance translators and post-editors will
be forced to work as subcontractors of these huge companies. The best translators of
today will probably move to the role of post-editors working as in-house staff of the
LSPs, since they will have most of the required skills and will make a decent living by
82
83
up for them, and those for whom the practice of translation will not be creative enough
will move to different professions.
11 CONCLUSIONS
The present thesis strove to answer several questions related to the development of the
translation profession. This work provides a theoretical discussion of the translation
profession, and machine translation post-editing in particular, based on a wide selection
of both academic and industry-specific literature dealing with the topic. The thesis is
accompanied by quantitative research in the form of a questionnaire distributed among
language professionals working in the Czech Republic.
The theoretical discussion starts by mapping the practice of translation from a
technological perspective, where four distinct eras based on the technology used by
translators are identified. The first era goes back to the early history of translation
practice where the use of technology was limited to a few rudimentary tools. This era
spreads over most of the time during which translation was put to use, and the translator
of this time could be described as the PRAT translator (pencil and rubber). The second
and third eras lead the discussion through major technological advances toward the
fourth era which sees the comeback and gradual spread of machine translation
technologies. The last era is discussed in the form of a future outlook which predicts a
further implementation of machine translation technologies and post-editing practice.
Since it is assumed in this thesis that the future of the translation profession lies
in machine translation post-editing, a separate section is devoted to its various types and
implementation scenarios. The conclusion of this section tentatively suggests that posteditors will be engaged particularly in RPE, MPE and FPE, with in-house post-editors
dealing more with the first two types, while freelance agency workers will focus more
84
on FPE. The thesis further concentrated on drawing a profile of translators and their
skill-sets required of them to become future post-editors. Translators were seen as the
ideal candidates for taking up the post-editing profession, however, given the fact that
machine translation post-editing was identified as an activity that is distinctly different
from human translation and its revision, despite all the overlapping skills possessed by
translators and post-editors, this thesis argues that more efforts should be invested in
syllabus design and in including the teaching of and about post-editing into the curricula
of translator training programmes at institutions of higher learning. The present thesis
examines some of the issues encountered at universities running post-editing courses,
and it is pointed out that the experience gathered at such universities ought not to be
wasted and these courses should be used as examples of good practice by all course
designers. A number of future translators will be forced to deal with post-editing if they
want to stay on the translation market. Attaining basic knowledge of machine translation
and post-editing and the skills related to them may be profitable both to translation
graduates and their prospective employers/clients. The issue of collaborative translation
has also been discussed in a great detail. This thesis provided arguments which show
that the future of the translation profession may witness a return of more collaborative
approaches to translation as opposed to the individualistic approach that has dominated
the field from the Renaissance. Skills related to collaboration and teamwork have been
identified as being particularly relevant for future graduates of translation programmes.
Even though the quantitative survey included in this thesis implies a confidence
of translators into their own future, the main conclusion of this work is that the
translators of today will turn into post-editors of machine-translated texts. Machine
translation does not constitute a threat leading to a loss of jobs, however, the role of
translators is expected to significantly change in favour of post-editing. Translators
85
86
APPENDIX
1:
TRANSLATORS'
QUESTIONNAIRE
POST-EDITING
b) part-time
c) other
No
5. Do you think that machine translation will replace the human translator for nonliterary translation in the future?
-
Strongly disagree
Disagree
Agree
Strongly agree
87
6. Have you ever been offered a post-editing job? (by post-editing we mean
revising a machine-translated text)
-
Yes
No
I sometimes post-edit
I regularly post-edit
Please specify
0-10%
10-25%
25-50%
50-75%
11. Are you satisfied with the price and pricing method for post-editing? If not,
please specify why you are not satisfied and/or describe a better pricing method.
-
Yes
No (please specify)
88
In a specialised course
On the job
Yes
No
Yes
No
16. Where does the machine translation output you post-edit come from?
-
I do not know
17. Do you feel a need for a special post-editing tool (as opposed to the integration
of machine translation into traditional computer assisted translation (CAT) tools?
If so, please specify why you would appreciate a special tool for post-editing.
89
No
18. Would you appreciate access to confidence scores for every segment to be postedited? Confidence scores tell you how likely the machine translated output is a
precise translation. It is similar to the information about fuzzy matches in CAT.
-
Yes
No
90
91
Question
Q1
Q2
Q3
Q4
T1
Full-time
1-5 years
No
T2
Full-time
5-10 years
No
T3
Full-time
1-5 years
No
T4
Full-time
T5
Part-time
5-10 years
No
T6
Part-time
1-5 years
No
T7
Part-time
1-5 years
No
T8
Full-time
5-10 years
T9
Part-time
1-5 years
T10
Full-time
1-5 years
T11
Part-time
1-5 years
T12
Full-time
1-5 years
T13
T14
Full-time
Language certificates (FCE, TEFL etc.), University degree in Yes, definitely for some language pairs the shift is clearly from translation to post10 years or longer languages
editing of machine translation, at least in some industries such as IT.
T15
Full-time
No
T16
Full-time
Sad
T17
Full-time
No
T18
Full-time
No
T19
Part-time
1-5 years
T20
Full-time
5-10 years
T21
Part-time
1-5 years
T22
Full-time
5-10 years
Translator
No
No
92
Question
Q5
Q6
Q7
Q8
Q9
Q10
T1
No
T2
Disagree
No
T3
Strongly disagree
Yes
I regularly post-edit
25-50%
T4
Disagree
Yes
I sometimes post-edit
0-10%
T5
Disagree
No
T6
No
T7
Strongly disagree
No
T8
Yes
0-10%
T9
Disagree
Yes
0-10%
per word
T10
Disagree
Yes
I sometimes post-edit
0-10%
It depends, both
T11
No
T12
No
T13
No
T14
Disagree
Yes
0-10%
T15
Agree
No
T16
Agree
No
T17
Disagree
No
T18
Disagree
Yes
0-10%
T19
Disagree
No
T20
Disagree
Yes
0-10%
T21
Disagree
Yes
I sometimes post-edit
0-10%
T22
Strongly disagree
No
Translator
93
Question
Q11
Q12
Q13
Translator
T1
T2
T3
Yes
Yes, I like MT post-editing
post-editing requires much more attention, re-translation is necessary so
On the job
T4
you are paid significantly less for harder and usually longer work
On the job
T8
Yes
No, I do not like MT post-editing
For me, post-editing takes more time and energy that translation from
On the job
T9
On the job
T10
extremely underpaid
On the job
Yes
On the job
T20
Yes
On the job
T21
T22
Yes
On the job
T5
T6
T7
T11
T12
T13
Actually, none of the potential clients has yet accepted my rate which is
T14
T15
T16
T17
T18
T19
94
Question
Q14
Q15
Q16
Q17
Q18
Translator
T1
T2
From on-line machine translation engines (Google
T3
Yes
Yes
Translate, Bing...)
No
Yes
T4
No
No
No
No
T8
Yes
Yes
Yes
T9
No
Yes
Translate, Bing...)
No
From on-line machine translation engines (Google
No
T10
No
Yes
Translate, Bing...)
Yes
Yes
No
Yes
I do not know
No
Yes
No
No
T5
T6
T7
T11
T12
T13
T14
T15
T16
T17
From on-line machine translation engines (Google
T18
Yes
Yes
Translate, Bing...)
T19
From on-line machine translation engines (Google
T20
No
Yes
Translate, Bing...)
No
From on-line machine translation engines (Google
Yes
T21
No
Yes
Translate, Bing...)
Yes
T22
95
No
References
Allen, Jeffrey (2001) Post-Editing or no Post-Editing? International Journal for
Language and Documentation (8), 41-42.
Allen, Jeffrey (2003) Post-editing. In: Somers, Harold (ed.) Computers and
Translation: A Translators Guide. London: Routledge, 297318.
Allen, Jeffrey (2004) Case Study: Implementing MT for the Translation of Pre-Sales
Marketing and Post-sales Software Deployment Documentation. In Robert E.
Frederking, Kathryn Taylor (Eds.): Machine Translation: From Real Users to
Research, 6th Conference of the Association for Machine Translation in the
Americas, AMTA 2004, Washington, DC, USA, September 28-October 2, 2004,
Proceedings. Berlin Heidelberg: Springer, 1-6.
Allen, Jeffrey (2005) What is Post-editing? Translation Automation Newsletter (5), 13.
ALPAC (1966) Languages and Machines: Computers in Translation and Linguistics. A
Report by the Automatic Language Processing Advisory Committee. Division of
Behavioral Sciences, National Academy of Sciences, National Research Council.
Washington, D.C.: National Academy of Sciences, National Research Council,
1966.
Arnold, Douglas, Lorna Balkan, Siety Meijer, R. Lee Humphreys and Louisa Sadler
(1994) Machine Translation: An Introductory Guide. London, NCC Blackwells.
Asia Online (2014) Fair Compensation for Post-Editing Machine Translation.
http://www.asiaonline.net/EN/Resources/Articles/FairCompensationForMTPostEditi
ng.aspx (consulted 11.11.2014)
Belam, Judith (2003) Buying up to Falling down: A Deductive Approach to Teaching
Post-Editing. MT Summit IX Workshop on Teaching Translation Technologies and
Tools (T4) (Third Workshop on Teaching Machine Translation), New Orleans. 1-10.
Bell, Terena (2014) The how of Mobile Localization. MultiLingual 25(8), December
2014.
http://dig.multilingual.com/201412/?
emhash=c8398fd21c96d6b3249268a024c5de68 (consulted 07.11.2014)
Beninatto, Renato S. and Donald A. DePalma (2008) The Top 25 Translation
Companies
and
Some
Really
Big
Revenue.
http://www.commonsenseadvisory.com/Default.aspx?
Contenttype=ArticleDetAD&tabID=63&Aid=498&moduleId=391
(consulted
28.10.2014)
Bistu, Beln (2013) Collaborative Translation and Multi-Version Texts in Early
Modern Europe: Transculturalism, 1400-1700. Burlington, Vermont: Ashgate
Publishing Company.
96
Boucau, Ferdinand (2005) The European Translation Industry Facing the Future.
Brussels, EUATC. http://www.guilde.net/fr/news/euatc.pps (consulted 28.10.2014)
Boucau, Fernand (2006) The European Translation Markets. Updated Facts and
Figures 2006 2010. Paper Presented at the 2nd International Conference of the
European Union of Associations of Translation Companies (EUATC, 9 November
2006).
Carson-Berndsen, Julie, Harold Somers, Carl Vogel and Andy Way (2009) Integrated
Language Technology as Part of Next Generation Localisation. Localisation
Focus: The International Journal of Localisation 8(1), 53-66.
Champollion, Yves (2001) Machine Translation (MT), and the Future of the Translation
Industry. Translation Journal 5(1).
Cronin, Michael (2010) Globalization and Translation. In: Gambier, Yves and van
Doorslaer, Luc (eds.) Handbook of Translation Studies Volume 1. Amsterdam and
Philadelphia: John Benjamins Publishing Company, 134-140.
De Palma, Donald and Nataly Kelly (2008) Translation of, for, and by the People: How
User-Translated Content Projects Work in Real Life. Common Sense Advisory
Report.
Dsilets, Alain (2011) Wanted: Best Practices for Collaborative Translation. A Report
Published by TAUS.
DGT (2011)The Status of the Translation Profession in the European Union. Project
Plan,
July
2011.
http://isg.urv.es/publicity/isg/projects/2011_DGT/reports/TST_technical_offer.pdf
(consulted 28.10.2014)
Doherty, Stephen, Dorothy Kenny and Andy Way (2012) Taking Statistical Machine
Translation to the Student Translator. AMTA-2012: The Tenth Biennial Conference
of the Association for Machine Translation in the Americas, 28 Oct 1 Nov 2012,
San Diego, USA.
Doherty, Stephen and Federico Gaspari (2013) Effective Post-Editing in Human &
Machine Translation Workflows: Critical Knowledge & Techniques. Centre for
Next Generation Localisation, Dublin City University 5th December 2013.
http://www.qt21.eu/launchpad/sites/default/files/QTLaunchPad%20Post-Editing
%20Webinar.pdf (consulted 05.07.2014)
Doherty, Stephen and Joss Moorkens (2013) Investigating the Experience of
Translation Technology Labs: Pedagogical Implications. The Journal of
Specialised
Translation
(19),
122-136.
http://www.jostrans.org/issue19/art_doherty.pdf (consulted 30.10.2014)
Edmundson, Harald Parkins and David Glen Hays (1958) Research Methodology for
Machine Translation. Mechanical Translation 5(1), 8-15.
97
98
Hennessy, Eileen B. (2008) Navigating in a New Era: What Kind of Education and
Training
for
Translators?
Translation
Journal
12(4).
http://translationjournal.net/journal/46image.htm (consulted 30.10.2014)
Huang, Harry J. (2011) Intermediality and Human vs. Machine Translation.
CLCWEB:
Comparative
Literature
and
Culture
13(3).
http://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=1796&context=clcweb
(consulted 10.11.2014)
Keynes, Maynard John (1931) Essays in Persuasion. London: Macmillan.
Krl, Pavel (2003) Vyuit potaovch technologi v pekladatelskm procesu. ToP
(tlumoen-peklad) 69.
Krings, Hans P. and Geoffrey S. Koby (eds.) (2001) Repairing Texts: Empirical
Investigations of Machine-translation Post-Editing Processes. Kent: Kent State
University Press.
LanguageRealm
(2014)
The
History
of
http://www.languagerealm.com/articles/history-of-translation.php
02.10.2014)
Translation.
(consulted
99
Mossop, Brian (2011) Revision. In: Gambier, Yves and van Doorslaer, Luc (eds.)
Handbook of Translation Studies Volume 2. Amsterdam and Philadelphia: John
Benjamins Publishing Company, 135-139.
Mossop, Brian (2014) Editing and Revising for Translators, 3rd edn. Abingdon and
New York: Routledge.
Nogueira, Danilo and Kelli Semolini (2010) Will we be here Tomorrow? Translation
Journal 14(3). http://www.translationjournal.net/journal/53mt.htm (consulted
11.10.2014)
OBrien, Sharon (2002) Teaching Post-Editing: A Proposal for Course Content. 6th
EAMT Workshop Teaching Machine Translation, 99-106.
OBrien, Sharon (2007) Eye-Tracking and
Perspectives: Studies in Translatology 14(3).
Translation
Memory
Matches.
OBrien, Sharon (2011a) Collaborative Translation. In: Gambier, Yves and van
Doorslaer, Luc (eds.) Handbook of Translation Studies Volume 2. Amsterdam and
Philadelphia: John Benjamins Publishing Company, 17-20.
OBrien, Sharon (2011b) Towards Predicting Post-Editing Productivity. Machine
Translation
25(3),
1-26.
http://doras.dcu.ie/17154/1/Towards_Predicting_Postediting_Productivity_Final_2.p
df (consulted 11.10.2014)
OBrien, Sharon, Joss Moorkens and Joris Vreeke (2014) Kanjingo - A Mobile App for
Post-Editing. The 17th Annual Conference of the European Association forMachine
Translation, 16-18 June 2014, Dubrovnik.
Ostler, Nicholas (2011) The Last Lingua Franca: The Rise and Fall of World
Languages. London: Penguin Books.
Poulis, Alexandros and David Kolovratnik (2012) To Post-edit or not to Post-edit?:
Estimating the Benefits of MT Post-Editing for a European Organization.
Proceedings of the AMTA 2012 Workshop on Post-editing Technology and Practice
(WPTP 2012). http://amta2012.amtaweb.org/AMTA2012Files/html/9/9_paper.pdf
(consulted 11.10.2014)
Pym, Anthony (2012) Translation Skill-Sets in a Machine-Translation Age. Paper
Presented at the 16th International Symposium on Interpretation and Translation
Teaching. Graduate Institute of Cross-Cultural Studies, Fu Jen Catholic University,
Taiwan.
Pym, Anthony, Franois Grin, Claudio Sfreddo and Andy L. J. Chan (2012) Studies on
Translation and Multilingualism: The Status of the Translation Profession in the
European Union. DGT of the European Commission, July 2012.
http://ec.europa.eu/dgs/translation/publications/studies/translation_profession_en.pd
f (consulted 25.10.2012)
100
Rico, Celia and Enriue Torrejn (2004) Controlled Translation as a New Translation
Scenario: Training the Future User. Translating and the Computer (26), 1-24.
http://mt-archive.info/Aslib-2004-Rico.pdf (consulted 31.10.2014)
Rico, Celia and Enrique Torrejn (2012) Skills and Profile of the New Role of the
Translator as MT Post-Editor. Revista.Tradumtica: Tecnologies de la traducci
(10), 166-178.
Robinson, Douglas (2001) Babel, Tower of. In: Baker, Mona and Malmkjaer, Kirsten
(eds.) Routledge Encyclopedia of Translation Studies. London and New York:
Routledge, 21-22.
Schulz, Thomas (2013) Translate this: Googles Quest to End the Language Barrier.
Spiegel
Online
International,
13
September
2013.
http://www.spiegel.de/international/europe/google-translate-has-ambitious-goalsfor-machine-translation-a-921646.html (consulted 08.08.2014)
Somers, Harold (1997) A Practical Approach to Using Machine Translation Software:
Post-Editing the Source Text. The Translator 3.
Somers, Harold (2001) Machine Translation. In: Baker, Mona and Malmkjaer, Kirsten
(eds.) Routledge Encyclopedia of Translation Studies. London and New York:
Routledge, 136-140.
Somers, Harold (2003a) Introduction. In: Somers, Harold (ed.) Computers and
Translation: A Translators Guide. London: Routledge, 112.
Somers, Harold (2003b) Machine Translation in the Classroom. In: Somers, Harold
(ed.) Computers and Translation: A Translators Guide. London: Routledge, 319
340.
Specia, Lucia, Craig Saunders, Marco Turchi, Zhuoran Wang and John Shawe-Taylor
(2009) Improving the Confidence of Machine Translation Quality Estimates.
Proceedings of the Twelfth Machine Translation Summit. August 26-30, Ottawa,
Ontario, Canada. http://www.mt-archive.info/MTS-2009-Specia.pdf (consulted
11.11.2014)
TAUS (2006) Putting Machine Translation to Work. Report on the TAUS Executive
Forum. Washington DC: 23-24 March 2006.
https://www.taus.net/downloads/finish/56-public-reports/16-taus-putting-machinetranslation-to-work-april-2006 (consulted 30.07.2014)
TAUS
(2010)
Postediting
in
Practice.
A
TAUS
http://www.translationautomation.com/articles/ (consulted 28.07.2014)
Report.
101
TAUS
(2013b)
Translation
Technology
Landscape
Report.
https://www.taus.net/reports/taus-translation-technology-landscape-report (consulted
01.10.2014)
TAUS (2014)
02.11.2014)
History.
https://www.taus.net/history/history#2006
(consulted
Torrejn, Enrique and Celia Rico (2002) Controlled Translation: A New Teaching
Scenario Tailor-Made for the Translation Industry. 6th EAMT Workshop, Teaching
Machine Translation. November 14-15, 2002. European Association for Machine
Translation, 107-116.
Vasconcellos, Muriel (1986) Post-Editing On-Screen: Machine Translation from
Spanish into English. Proceedings of Translating and the Computer 8. London.
Vasconcellos Muriel and Marjorie Lon (1985) SPANAM and ENGSPAN: Machine
Translation at the Pan American Health Organization. Computational Linguistics
11(2-3), 122-136.
Wagner, Emma (1985) Post-editing Systran A Challenge for Commission
Translators. Terminologie & Traduction 1985(3).
Yngve, Victor H. (1954) The Machine and the Man. Mechanical Translation 1(2), 2022.
Zetzsche, Jost (2012) Translation Technology Comes Full Circle. MultiLingual
April/May 2012, 50-51.
102
RESUM
Tato diplomov prce se zapojuje do pevn teoretickho diskurzu za elem popisu
vvoje pekladatelsk profese z hlediska technologi, kter pekladatel pouvali
v rznch historickch epochch, piem se zamuje zejmna na souasnou
problematiku strojovho pekladu a editace text. Prce erp z poznatk teoretickho i
empirickho vzkumu v oblastech teorie pekladu, aplikovan lingvistiky a
matematick lingvistiky. Soust tto diplomov prce je kvantitativn vzkum
v podob dotaznku, jen byl en mezi pekladateli psobcmi na eskm trhu. Takto
zskan data jsou pouita jako podprn materil k teoretickm vahm, kter tato
prce pedkld. Prce si klade nkolik hlavnch cl. V prv ad se pokou zmapovat
vvoj pekladatelsk praxe od samch potk pekladatelsk profese a po souasnost
a tento vvoj pout jako argument pro hlavn hypotzu prce, a sice skutenost e se
pekladatel stanou editory text peloench technologi strojovho pekladu. Prce
dle namt, e se pekladatel tto zmn nemohou dnm zpsobem brnit a e jejich
jedinou monost jak zstat na trhu s peklady je pizpsobit se tomuto vvoji a osvojit
si dovednosti spojen s editac text. Dle se prce pokou o sestaven zevrubnho
profilu budoucho pekladatele a identifikuje dovednosti, je budou k vkonu tto
profese vyadovny. Prce tak apeluje na to, aby byla problematika strojovho
pekladu a editace zahrnuta do studijnch program pipravujcch profesionln
pekladatele, jeliko dovednosti spjat s tmito technologiemi se dle autora prce stanou
nedlnou soust vbavy kadho pekladatele, a je tedy teba vnovat se jejich rozvoji
u bhem studia. Prce se dle vnuje kolaborativnmu pstupu k pekladu, kter byl
v dob renesance nahrazen pstupem individualistickm. Autor pedpovd nvrat
ke kolaborativnmu pekldn. V neposledn ad jsou pedstaveny nejpalivj
103
104
SUMMARY
The present thesis describes the development of the translation profession in a largely
theoretical discourse focusing on the technology used by translators throughout the
history of the profession until now. The thesis deals particularly with the issues of
machine translation and post-editing. The discussion draws on theoretical and empirical
research in the fields of Translation Theory, Applied and Computational Linguistics.
The thesis is accompanied by quantitative research in the form of a questionnaire which
was administered to several Czech translators. The data gathered from the quantitative
research supports the theoretical arguments of the thesis. The thesis has several main
objectives. Firstly, it aspires to map the development of the translation practice from its
early history until the present time with the aim of predicting that the translator will
inevitably turn into a post-editor of machine-translated texts. It is argued that translators
cannot defend themselves against this turn, and that their sole fortification is to embrace
the practice of post-editing in order to stay on the market. Secondly, the present thesis
endeavours to draw a comprehensive profile of the future translator/post-editor and
identify the skills required for the profession. A tentative call is made to include the
teaching of and about machine translation post-editing into the syllabi of translator
training programmes, as it is argued that the skills related to these areas are best learnt
during training. Furthermore, the present thesis describes how the collaborative
approach to translation prevalent in its early days was succeeded by an individualistic
approach which was characteristic of the Renaissance. It is estimated that the
collaborative setup will be returned to in the near future. Finally, attention is also paid to
recurring issues with machine translation post-editing, and the thesis identifies areas for
further research.
105