Translators and machines: Working together

Abstract

The increasing use of language technologies has prompted an interest in their impact on cognitive processes and translation products. Potential issues related to human-machine interactions include working conditions, time and resource management, and emotional factors. Drawing on a large corpus of translation processes collected from professionals and students, we discuss the nature of translation as a cognitive and organizational activity. We argue that professional translators need to take increased ownership of language technology tools at every stage: in their development, their application, and their integration into organizational processes. This has implications for industry standards, models of translation expertise, and translation didactics.

Full text

Translators and machines: working together
Maureen Ehrensberger-Dow & Gary Massey
Institute of Translation and Interpreting, Zurich University of Applied
Sciences (ZHAW), Switzerland
ehre@zhaw.ch mssy@zhaw.ch
Abstract
The increasing use of language technologies has prompted an interest in their
impact on cognitive processes and translation products. Potential issues related to
human-machine interactions include working conditions, time and resource
management, and emotional factors. Drawing on a large corpus of translation
processes collected from professionals and students, we discuss the nature of
translation as a cognitive and organizational activity. We argue that professional
translators need to take increased ownership of language technology tools at
every stage: in their development, their application, and their integration into
organizational processes. This has implications for industry standards, models of
translation expertise, and translation didactics.
1. Language technology in the translation process
Professional translation has become a multi-activity task within a complex
system of client expectations, technological aids, information sources, and
organizational constraints. Reading and researching in the source language,
evaluating retrieved information, and writing and revising in the target language
impose heavy demands on bilingual cognitive resources. Translators also have to
juggle the competing concerns of loyalty to the source text, a high-quality target
text, and audience needs. All of this contributes to mental load, a construct
proposed to explain how various factors, such as time pressure or text
complexity, can affect translation performance (Muñoz 2012). The increasing
technologization of the professional translation workplace, in the form of
computer-aided translation tools (CAT) and machine translation (MT), relieves
translators of some routine tasks and provides easily-retrievable solutions to
recurring problems. It has contributed to increasing speed, improving
consistency, and reducing costs. However, there has been little investigation of
whether or how much these tools actually reduce the mental load of their users
during the translation process.
Competence in the use of language technology tools has become an integral part
of the job description of professional translation in the industrialized world. This

2 Maureen Ehrensberger-Dow & Gary Massey
is reflected in standards for translation services (e.g. EN15038 2006) as well as in
European recommendations for translator training (e.g. EMT 2009). Recent
models of translation competence (e.g. Göpferich 2009) identify the importance
of technology skills as well as cognitive/behavioral components and
psychomotor mechanisms clearly related to human-computer interactions. This
seems to align with industry expectations: Gouadec (2007/2010: 156) reports that
all of the 650 job vacancy advertisements for professional translators he
examined require skills in using TM or CAT systems.
The increased emphasis on language technologies in professional translation has
changed the translation process in some dramatic ways. At a cognitive level,
language technology tools have effectively extended translators’ memory by
externalizing it, thus decreasing the load on working and long-term memory (cf.
Pym 2011). However, the complexity of many of the newer CAT interfaces may
increase the cognitive demands on their users (cf. Hansen-Schirra 2012).
Furthermore, the use of translation memory (TM) tools may change the nature of
the translation task, encouraging translators to focus on the level of segments and
sentences instead of the text as a whole (cf. Hansen-Schirra 2012; Jiménez-
Crespo 2009).
The potential impact of such tools on cognitive processes goes beyond the
linguistic characteristics of the translation unit. In an investigation of sources of
disturbances in translation processes, Hansen (2006) identifies working
conditions, time management, use of translation aids, and emotions as important
parameters in the translation process. Language technology tools can influence
emotional state (cf. Beale/Peter 2008): Szameitat et al. (2009) report that delays
in computer responsiveness can affect task performance and cause negative
emotions, which potentially contribute to stress. Muñoz (2009; 2012) suggests
that typing mistakes might be an indication of stress and cognitive effort rather
than simply an artifact of a physical activity. These, in turn, can affect
concentration and the flow of the translation process when the translator
backtracks to correct them.
2. Investigating workplace processes
The field of cognitive translation research uses a variety of methods to gain
information about the internal processes and decision-making involved in
translation work. Techniques such as monitoring actions that take place on
translators’ computer screens, reconstructing the translation process to
understand individual steps and decisions, and asking translators to reflect on
what they do and why have contributed to a greater appreciation of the
competences involved in transferring texts from one language to another.
However, translation performance is affected not only by what happens in the
translator’s mind or on the computer screen, but also by how translators interact
with their technological, physical, and organizational environment.

Maureen Ehrensberger-Dow & Gary Massey 3
Translation studies has shown a growing interest in translation as a system that
involves not only multiple agents but also human-computer interactions (e.g.,
Risku 2010; O’Brien 2012). Humans and machines can reasonably be considered
to impact on, and adjust to, each other in order to respond to disturbances and
meet new demands. Much modern technology is designed to optimize human
performance, but systems can sometimes react inappropriately and actually be an
impediment. Alternatively, what might be considered a disturbance in some
contexts may actually be conducive to good performance in others. For instance,
Cades et al. (2010) point out that interruptions caused by certain physical
conditions in the workplace may not necessarily be detrimental to task
performance because people can adapt to them, whereas others may require
adjustments to the situation or to the machines and tools being used.
An ergonomic perspective provides an appropriate framework to investigate the
impact of various factors on the situated activity of translation. According to the
International Ergonomics Association (IEA)
1
, ergonomics comprises three
domains: the cognitive, the physical, and the organizational. Cognitive
ergonomics focuses on “mental processes [...] as they affect interactions among
humans and other elements of a system”, physical ergonomics on “human
anatomical, anthropometric, physiological and biomechanical characteristics as
they relate to physical activity”, and organizational ergonomics on “the
optimization of sociotechnical systems, including their organizational structures,
policies, and processes”. These descriptions can easily be applied to professional
translation, as discussed in the next section.
3. The ergonomics of language technology
In a longitudinal study conducted by our research team
2
, translation processes
were collected from beginners, advanced students, and professionals either at
their workplaces or in the controlled setting of our institute’s usability lab. The
students’ workplace processes were course assignments that they recorded with
screen recording software and submitted to the research team. The professionals’
workplace processes were screen recordings of their normal translation tasks
completed at their desks. The participants were shown the screen recordings of
their processes after completing some of the translations and commented on what
they saw themselves doing. They then participated in a semi-structured interview
about various aspects of translation and the tools they normally use, as well as in
an anonymous online survey about their workplace and heath. By examining
different sources of data (e.g. screen recordings, retrospective commentaries,
interviews, target texts), we have identified several ergonomic aspects of tool use
1
http://www.iea.cc/whats/index.html.
2
Information on the Capturing Translation Processes project and related publications
is available at www.linguistik.zhaw.ch/ctp.

4 Maureen Ehrensberger-Dow & Gary Massey
which could have adverse effects on translators’ performance and well-being (see
Ehrensberger-Dow/Massey 2014).
In many processes in our corpus, we have seen features of the text editing
software needlessly interrupting the flow of translation. A recurrent example is
when the autocorrect feature incorrectly changes a word, forcing the translator to
backtrack and undo the change. This is especially relevant for technical
translation, which involves terminology and abbreviations that may not be stored
in a text editor’s dictionaries. Although many software settings can be
customized to a translator’s needs, language settings are often carried over with
the source text. If a translator overwrites it (a common practice observed among
our professionals working outside a TM environment), adjustments may be
needed that slow down the process. In some applications, the automatic
spellchecker does not seem to recognize the target language, requiring repeated
changes to the language settings to avoid the distraction of passages underlined in
red.
Translators also mentioned becoming annoyed if their tools do not react as
quickly as they are used to. In some cases, they also reported that the tools might
be influencing their choices and decisions when, for example, they accept the top
entry in a concordance list. An interesting case study of the constraining
influence of tools is provided by a simple comparison of target texts produced
with and without TM. In the lab, with no TM available, professionals translated
into German an English source text that had three long sentences. As can be seen
in Figure 1, the shortest sentence was always translated as a single sentence but
the longest one as two sentences. By contrast, when the same professionals
translated source-text sentences of comparable length using TM in the
workplace, these were rendered as single sentences 50% of the time. The
sentence segmentation typical of most TM system settings (cf. LeBlanc 2013)
may unintentionally constrain creativity and the freedom to move away from
source-text syntactic patterns.
Figure 1. Source text sentences rendered as 1 or 2 sentences in the target text
Indeed, many comments made by professional translators suggest that language
technology tools are unnecessarily constraining their creative autonomy. Even
low-level decisions concerning punctuation have to be checked against parallel
texts, concordances, and clients’ style guides. Some translation tools and aids

Maureen Ehrensberger-Dow & Gary Massey 5
might be pushing translation into the direction of a search and match or
patchwriting task and away from interlingual transfer of meaning within a
multilingual’s mind. In this case, retrieval of information and ability to perform
will be highly influenced by changes in technology.
There is considerable evidence in our corpus that translators are juggling with
many sources of information as they work. A typical professional’s screen during
a workplace process might display a TM program, a concordance window, a
parallel text, and numerous other open tabs. Since it is logistically impossible to
determine the number of windows and tabs open in each translation process in
our corpus, we compared snapshots of the beginners’ and professionals’
computer screens 5 minutes into their workplace processes. While the students
tended to only have one window open (usually the target text) and about 5 tabs
visible in the command bar at the bottom of the screen (see Table 1), the
professionals often had at least two or three windows open and far more tabs
visible than the students, irrespective of the language pair. During the translation
process, the translators repeatedly switched between the windows and tabs,
suggesting heavy loads on their working memory to retain information between
switches or to remember which tab to open, which can affect performance.
Table 1. Number of windows and tabs open 5 minutes into workplace processes
Many issues are related to displays. The settings for some language technology
tools require translators to work in one half of the computer screen (e.g. either at
the bottom or on the right) and constantly shift between the halves to check
information. In certain processes in our corpus, recordings indicate that the
translators had trouble finding where they had been working on their text after
breaking off to research information or revise previous parts of their work. In
addition, computer screens seemed too small to have the internet browser open
next to a TM program, which would explain why professional translators
constantly switch between windows. A simple expedient like a browser button on
a TM interface could ease translators’ cognitive load, while greater familiarity
with shortcut key combinations would help to automatize routine procedures and
release cognitive resources for more demanding work.
4. The technologized translation workplace
Translators spend most of their day working with language and information
technologies. Professionals seem to rely much less on external resources than
students do, and their pausing behavior suggests that they are spending more time
consulting their internal resources (i.e. thinking) than the students are. As a

6 Maureen Ehrensberger-Dow & Gary Massey
result, professionals spend significantly more of their time using their keyboard
than their peripheral input devices compared with beginners and advanced
students, and are correspondingly faster at producing target text (see Table 2 for
data from the German-English and English-German lab processes). Translation
with language technology tools requires intensive interaction with hardware: an
analysis of a 20-minute lab process revealed that, in addition to typing
approximately 700 characters and spaces, the translator used the mouse wheel 78
times and made 106 mouse clicks.
Table 2. Use of computer peripherals and TT words in first 15 minutes of lab processes
Using a keyboard and other input devices such as a mouse and touchpad involves
more than just the hands or lower arms; the constant repetition of movement can
cause an overload of muscles of the upper extremities and back. Pineau (2011)
points out that the arrangement of letters on keyboards force hand distortion,
overly frequent finger extension, and imbalances between right-hand and left-
hand use. Since both source and target texts are usually in electronic form,
eyestrain due to long hours peering at a computer monitor can also become an
issue. In the interviews after commenting on their translation processes, the
professionals were asked specifically about satisfaction with their computer
workstations and user interfaces. Despite being basically satisfied, all of them
mentioned issues related to ergonomics, such as having no possibility to work
standing and computer-screen size. Complaints about user interfaces generally
focused on the limited space available for inputting text because of all the menus
and options. Surprisingly for such a screen-intensive task as translation, none of
the professionals in our study used two monitors, which would have solved a
number of the problems noted.
In addition, many professional translators work in offices that were not designed
for intensive text work. Contextual factors, such as ambient noise, inadequate
lighting, lack of ventilation, and people moving within translators’ fields of
vision, can influence translation performance and contribute to health problems.
In the anonymous survey done after we recorded processes at the workplace, all
of the translators reported a perceived impact of their work on their health, such
as burning eyes, concentration problems, headaches, nervousness, general
weakness and fatigue, neck pain, and back pain. All of them also mentioned that
there were problems with office air quality and ventilation. The most frequently
identified disturbances were noise from outside and inside the office (78% and
85% respectively), people moving around the office (78%), and email, chats, or
phone calls (71%). Not all disturbances to the translation process were viewed
negatively, however. Many of the translators said that they found interruptions
such as phone calls, questions from colleagues, and coffee breaks helped sharpen

Maureen Ehrensberger-Dow & Gary Massey 7
their concentration afterwards. Prudent distribution of breaks would seem the
natural good practice suggested by such observations.
Infrastructural and physical factors may be compounded by issues of
organizational ergonomics. A number of translation scholars (e.g. Grass 2011;
Olahan 2011) argue that, by largely failing to address human and organizational
aspects in the design and workflow deployment of language technology tools,
software developers and corporate LSPs have been increasingly disempowering
and alienating translators. The segmentation of the translation process into
management, terminology, pre-translation, revision, and other tasks can be
perceived as positive when it frees up human resources for work that people can
do better than machines, but compartmentalization can impact negatively on
translators’ self-concept and professionalization if it prevents them from making
informed decisions and taking adequate responsibility for what they do (cf.
Ehrensberger-Dow/Massey 2013).
5. Implications for industry standards, expertise, and didactics
Many of the comments made by the translators in our study reflect the
apparent priorities of EN15038 (2006: 11), in which translation is delineated as
an activity concerned with terminology, grammar, lexis, proprietary or client
style, local conventions and regional standards, and formatting; interestingly, the
target group and the purpose of the translation figures last on this list. The
question arises as to whether such quality standards may be over-standardizing
the profession and placing excessive contraints on the translator. If translators are
being constrained by the tools they are using and the system that they are
working in, it might prove very difficult for them to gain expertise. Constrained
systems run the risk of producing professionals who are very good at routine
work when all of the tools are operational, but who may not develop and
maintain the expertise needed to handle novel problems.
Understanding how translators use language technology should contribute to
optimizing translation workplaces, tools, and decision-making procedures as well
as motivating the need for individual adaptations to be built into technologies and
systems when indicated. Heightened appreciation of the importance of ergonomic
resources, tools, settings, equipment, and organizational systems should also help
translators and companies design more efficient and user-oriented workplaces,
tools, and workflows. The importance of ergonomic factors extends beyond the
various agents in the situated activity of translation. Ergonomic issues are highly
relevant for members of any professional group that operates at the human-
computer interface. Since the time spent at computer workplaces is increasing, it
is clearly in most institutions’ and companies’ best interest to understand which
ergonomic factors represent disturbances that seriously affect the performance of
their staff.
Finally, findings from research into how translators work with machines can feed
into an empirical, evidence-based approach to language service consultancy and

8 Maureen Ehrensberger-Dow & Gary Massey
the professional development of translators, and be directly applied to courses in
undergraduate and graduate translation studies programs. We believe that our
continuing investigations will enable us to identify good, better, and best
practices to diminish the detrimental effects of workplace constraints. This, in
turn, will improve our ability to prepare students and professionals for the future
challenges posed by human-machine interaction.
References
Beale, R. / Peter, C. (2008). The role of affect and emotion in HCI, in: Lecture
Notes in Computer Science 4868, p. 1-11.
Cades, D. M. / Werner, N. E. / Boehm-Davis, D. A. / Arshad, Z. (2010). What
makes real-world interruptions disruptive? Evidence from an office setting, in:
Proceedings of the Human Factors and Ergonomics Society Annual Meeting
54.4, p. 448-452.
Ehrensberger-Dow, M. / Massey, G. (2014/forthcoming). Cognitive ergonomic
issues in professional translation, in: Schwieter, J. W. / Ferreira, A. (eds.). The
Development of Translation Competence: Theories and Methodologies from
Psycholinguistics and Cognitive Science. Newcastle: Cambridge Scholars
Publishing, p. 58-86.
Ehrensberger-Dow, M. / Massey, G. (2013). Indicators of translation
competence: Translators’ self-concepts and the translation of titles, in: Journal of
Writing Research 5.1, p. 103-131.
EMT expert group. (2009). Competences for Professional Translators, Experts in
Multilingual and Multimedia Communication. Brussels: European Commission.
EN 15038. (2006). Translation services – service requirements. Brussels:
European Committee for Standardization.
Göpferich, S. (2009). Towards a model of translation competence and its
acquisition: The longitudinal study TransComp, in: Göpferich, S. / Jakobsen, A.
L. / Mees, I. M. (eds.). Behind the Mind. Methods, Models and Results in
Translation Process Research. Copenhagen: Samfundslitteratur, p. 17-43.
Gouadec, D. (2007/2010). Translation as a Profession. Amsterdam: John
Benjamins.
Grass, T. (2011). “Plus” est-il synonyme de “mieux”? Logiciels commerciaux
contre logiciels libres du point de vue de l’ergonomie, in: ILCEA Traduction et
Ergonomie 14. http://licea.revues.org/index 1052.html.
Hansen, G. (2006). Erfolgreich übersetzen. Entdecken und Beheben von
Störquellen. Tübingen: Narr Francke Attempto.
Hansen-Schirra, S. (2012). Nutzbarkeit von Sprachtechnologien für die
Translation, in: trans-kom 5.2, p. 211-226.
Jiménez-Crespo, M. A. (2009). The effect of translation memory tools in
translated web texts: Evidence from a comparative product-based study, in:
Linguistica Antverpiensia 8, p. 213-232.

Maureen Ehrensberger-Dow & Gary Massey 9
LeBlanc, M. (2013). Translators on translation memory (TM). Results of an
ethnographic study in three translation services and agencies, in: The
International Journal for Translation & Interpreting Research 5.2, p. 1-13.
Muñoz Martín, R. (2009). Typos and co, in: Göpferich, S. / Jakobsen, A. L. /
Mees, I. M. (eds.). Behind the Mind. Methods, Models and Results in Translation
Process Research. Copenhagen: Samfundslitteratur Press, p. 167-189.
Muñoz Martín, R. (2012). Just a matter of scope. Mental load in translation
process research, in: Translation Spaces 1, p. 169-188.
O’Brien, S. (2012). Translation as human-computer interaction, in: Translation
Spaces 1, p. 101-122.
Olohan, M. (2011). Translators and translation technology: The dance of agency,
in: Translation Studies 4.3, p. 342–357.
Pineau, M. (2011). La main et le clavier: histoire d’un malentendu, in: ILCEA
Traduction et Ergonomie 14. http://ilcea.revues.org/index1067.html.
Pym, A. (2011). What technology does to translating, in: The International
Journal for Translation & Interpreting 3.1, p. 1-9.
Risku, H. (2010). A cognitive scientific view on technical communication and
translation. Do embodiment and situatedness really make a difference?, in:
Target 22.1, p. 94-111.
Szameitat, A. J. / Rummel, J. / Szameitat, D. P. / Sterr, A. (2009). Behavioral and
emotional consequences of brief delays in human-computer interaction, in:
International Journal of Human-Computer Studies 67, p. 561-570.