A PROJECT-BASED APPROACH TO

TRANSLATION TECHNOLOGY

A Project-Based Approach to Translation Technology provides students of translation

and trainee translators with a real-time translation experience, with its translation
platforms, management systems, and teamwork.
This book is divided into seven chapters reflecting the building blocks of a
project-based approach to translation technology. The first chapter identifies the
core elements of translation environment tools and collaborative work methods,
while Chapters 2 and 4 review the concept of translation memory and terminology
databases and their purposes. Chapter 3 covers machine translation embedded
in the technology, and the other chapters discuss human and technological
quality assurance, digital ethics and risk management, and web-based translation
management systems. Each chapter follows a common format and ends with
project-based assignments.These assignments draw and build on real-time contexts,
covering the consecutive steps in the workflow of large and multilingual translation
projects.
Reviewing the many translation technology tools available to assist the
translator and other language service providers, this is an indispensable book for
advanced students and instructors of translation studies, professional translators,
and technology tool providers.

Rosemary Mitchell-Schuitevoerder holds a PhD in Translation Studies from

Durham University, UK, and has taught Translation Technology on the MA
programmes at Durham University, UK, and Newcastle University, UK. She is a
freelance translator of Dutch, Swedish, Norwegian, and Danish into English, and
English into Dutch.
TRANSLATION PRACTICES EXPLAINED
Series Editor: Kelly Washbourne

Translation Practices Explained is a series of coursebooks designed to help self-

learners and students on translation and interpreting courses. Each volume
focuses on a specific aspect of professional translation and interpreting practice,
usually corresponding to courses available in translator- and interpreter-training
institutions. The authors are practising translators, interpreters, and/or translator or
interpreter trainers. Although specialists, they explain their professional insights in a
manner accessible to the wider learning public.
Each volume includes activities and exercises designed to help learners consoli-
date their knowledge, while updated reading lists and website addresses will also
help individual learners gain further insight into the realities of professional practice.

Most recent titles in the series:

Note-taking for Consecutive Interpreting 2e

A Short Course
Andrew Gillies

Consecutive Interpreting
A Short Course
Andrew Gillies

Healthcare Interpreting Explained

Claudia V. Angelelli

Revising and Editing for Translators 4e

Brian Mossop

A Project-Based Approach to Translation Technology

Rosemary Mitchell-Schuitevoerder

For more information on any of these and other titles, or to order, please go to
www.routledge.com/Translation-Practices-Explained/book-series/TPE

Additional resources for Translation and Interpreting Studies are available on the
Routledge Translation Studies Portal: http://routledgetranslationstudiesportal.com/
A PROJECT-BASED
APPROACH TO
TRANSLATION
TECHNOLOGY

Rosemary Mitchell-Schuitevoerder
First published 2020
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
and by Routledge
52 Vanderbilt Avenue, New York, NY 10017
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2020 Rosemary Mitchell-Schuitevoerder
The right of Rosemary Mitchell-Schuitevoerder to be identified as author of this work
has been asserted by her in accordance with sections 77 and 78 of the Copyright,
Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised
in any form or by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying and recording, or in any information
storage or retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks,
and are used only for identification and explanation without intent to infringe.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data
Names: Mitchell-Schuitevoerder, Rosemary, author.
Title: A project-based approach to translation technology /
Rosemary Mitchell-Schuitevoerder.
Description: London ; New York : Routledge, 2020. |
Series: Translation practices explained |
Includes bibliographical references and index.
Identifiers: LCCN 2020003466 | ISBN 9780367138820 (hardback) |
ISBN 9780367138844 (paperback) | ISBN 9780367138851 (ebook)
Subjects: LCSH: Translating and interpreting–Technological innovations. |
Translating and interpreting–Data processing.
Classification: LCC P306.97.T73 M58 2020 | DDC 418/.020285–dc23
LC record available at https://lccn.loc.gov/2020003466
ISBN: 978-0-367-13882-0 (hbk)
ISBN: 978-0-367-13884-4 (pbk)
ISBN: 978-0-367-13885-1 (ebk)
Typeset in Bembo
by Newgen Publishing UK
For John, Thomas, Benjamin, Simeon
CONTENTS

List of figures xii

List of tables xiv
Preface xv
Acknowledgements xvi
List of abbreviations xvii
Glossary xix
Introduction for instructors xxiv
Introduction for students and translators xxxi

1 Computer-aided translation tools and translation project

management 1
Key concepts 1
Introduction 1
1.1 Translation environment tools (TEnT) 2
1.2 Compatibility 2
1.3 Translation memory 4
Food for thought… 6
1.4 File management 6
1.4.1 Filenames 7
1.5 From Microsoft to CAT tools 10
1.5.1 The ribbon 10
1.5.2 Keyboard shortcuts 11
1.6 The CAT tool 12
1.6.1 CAT tool features 12
1.6.2 Standalone and web-based CAT tools 13
1.6.3 CAT tools and cognitive ergonomics 14
viii Contents

Food for thought… 15

1.7 Translation project management 15
1.7.1 Volume and demands 15
1.7.2 Translation workflow 16
1.7.3 The translator in the workflow 17
1.7.4 Project management by the language service provider 18
1.7.5 Project-based assignments and project management 19
1.7.6 Human resources database 20
1.8 Cognitive friction 21
Project-based assignment 22
Concluding remarks 23
Further reading 24

2 The translation memory database 25

Key concepts 25
Introduction 25
2.1 Creating a translation memory database 26
Food for thought… 28
2.1.1 Segmentation 28
2.1.2 The concordance and consistency 30
2.1.3 The analysis feature and fees 33
Food for thought… 36
2.2 Metadata and subsegment matching 36
2.3 Boosting the translation memory 38
2.3.1 Alignment 39
2.3.2 Translation memory and reference files 39
2.3.3 TMX files 40
2.4 Formats 41
2.5 Other functions and features on the ribbon 44
2.5.1 Filters 45
2.5.2 Regex 46
Project-based assignment 48
Concluding remarks 49
Further reading 50

3 Integration of machine translation in translation

memory systems 51
Key concepts 51
Introduction 51
3.1 Artificial intelligence and machine translation 52
3.2 From statistical to neural machine translation in the CAT tool 53
 Contents ix

3.3 Matches from the translation memory and machine translation

compared 54
3.4 Machine translation – access and integration 56
Food for thought… 57
3.5 Translation memories train machine translation
engines 57
3.6 Adaptive MT engines 58
Food for thought… 59
3.7 MT quality in the CAT tool 59
3.7.1 Quality management 60
Food for thought… 62
3.7.2 How to evaluate machine translation 62
3.7.3 Evaluation models and metrics 63
Project-based assignment 66
Concluding remarks 67
Further reading 68

4 The terminology database 69

Key concepts 69
Introduction 69
4.1 Terms and terminology 71
4.1.1 Glossary 71
Food for thought… 72
4.1.2 Term extraction 72
4.1.3 Metadata 74
Food for thought… 76
4.1.4 Standardisation and quality assurance 77
4.2 Web searches 77
4.2.1 Search techniques 78
4.3 Corpora 79
4.3.1 Digital dictionaries 79
4.3.2 Multilingual data corpora 80
4.3.3 The creation of corpora 82
Food for thought… 83
4.4 CAT concordance 84
4.4.1 External concordances 85
4.5 Morphology and the terminology database 86
4.6 Termbanks 86
Project-based assignment 88
Concluding remarks 90
Further reading 90
x Contents

5 Human and technological quality assurance 92

Key concepts 92
Introduction 92
5.1 Quality and standards 93
5.1.1 Quality assurance and assessment 93
5.2 Quality in different TEnTs 95
5.2.1 Quality assurance in CAT tools 95
5.2.2 Quality assessment of MT technology 96
Food for thought… 97
5.2.3 Web localisation and quality 97
5.3 Revision in CAT tools 99
5.3.1 CAT revision of bilingual and monolingual files 100
5.4 ISO standards 101
Food for thought… 103
5.5 Revision and evaluation 104
5.5.1 Human evaluation 105
5.6 The post-edit of wholly or partly machine translated texts 107
Food for thought… 108
5.6.1 PEMT quality 108
Project-based assignment 109
Concluding remarks 111
Further reading 112

6 Digital ethics and risk management 113

Key concepts 113
Introduction 113
6.1 Intellectual property rights (IP) and ownership 114
6.2 Confidentiality and non-disclosure agreements 115
6.3 Business terms and conditions in relation to confidentiality and
digital security 116
Food for thought… 119
6.4 Data sharing and data protection 119
6.4.1 Google and GDPR 120
6.5 Digital ethics of shared TM and MT databases 121
6.6 Risk management 122
6.6.1 Our security and intellectual property rights in
translation projects 123
Food for thought… 124
6.6.2 Risk factors 124
6.6.3 Risk response 127
Project-based assignment 128
Concluding remarks 130
Further reading 130
 Contents xi

7 Web-based translation environment tools 132

Key concepts 132
Introduction 132
7.1 Integrated translation environment 133
7.1.1 Speech recognition 134
7.2 Servers and web-based CAT tools 136
7.3 Digital platforms 137
7.3.1 Translator platform 138
7.3.2 Marketplace platform 139
7.3.3 Localisation platform 139
7.3.4 Translation platform 140
7.3.5 Knowledge platform 141
7.3.6 Crowdsourcing platform 141
7.3.7 The portal 142
Food for thought… 142
7.4 From content to translation management systems 143
7.5 Translation management systems 143
Food for thought… 145
7.6 Localisation of webpages and search engine optimisation 146
7.7 Translation technology and the users 147
Project-based assignment 149
Concluding remarks 151
Further reading 152

Bibliography 153
Index 159
FIGURES

1.1 Translator’s parent folder 8

1.2 Subfolders generated by CAT tool 9
1.3 Ribbon in MS Word 10
1.4 A basic translation workflow chart 16
1.5 Translator’s workflow 17
1.6 LSP’s workflow in the cloud 18
2.1 File preparation dialog in SDL Trados Studio 29
2.2 Dialog box opened in concordance search in SDL Trados
Studio 2019 31
2.3 Concordance target results for the term ‘functional’ 31
2.4 Search and find dialog in SDL Studio 2019 32
2.5 Search and Find function Ctrl+F 32
2.6 Concordance dialog in memoQ 9.1 33
2.7 Analysis report (statistics tab in memoQ 9.1) 35
2.8 Alignment with join-up lines between source and target segments 39
2.9 Pseudo translated PDF file 43
2.10 Ribbon in MS Word 45
2.11 Ribbon in CAT tool 45
3.1 CAT tool and adaptive MT 54
4.1 Translation Results window in memoQ 9.1 70
4.2 Term entry in memoQ 9.1 75
4.3 Bilingual data corpora in Linguee stating that ‘external sources
are not yet reviewed’ 81
4.4 TAUS data cloud 82
4.5 The CAT concordance in operation 84
4.6 Termbank and lookup term in CAT tool (memoQ 9.1) 87
5.1 From i18n (internationalisation) to l10n (localisation) 98
 Figures xiii

6.1 Classification of technological and human risks in a

translation project 125
6.2 PDF of three sides of a carton with text sideways and
upside down 126
6.3 Risk response 127
7.1 Speech-to-text and text-to-speech software 136
7.2 File-based or cloud-based localisation 140
TABLES

2.1 MemoQ (M) and TRADOS 2007-like (T) counts compared in

MemoQ 9.2 Statistics 36
3.1 Edit distance 55
3.2 LISA model for quality assurance 64
3.3 TAUS (2019) Dynamic Quality Framework (DQF) uses
Multidimensional Quality Metrics 65
4.1 Example of term definition in SL 74
5.1 Positive and negative sides to revision in XLIFF or bilingual files 101
5.2 Random selection of criteria found in LSP evaluation
statements (2019) 104
5.3 Error typology grid applied to a CAT tool with integrated MT 106
6.1 Selected terms and conditions for translation services 117
PREFACE

A Project-Based Approach to Translation Technology is about technological translation

tools and how they learn through artificial intelligence. But the book is also about
users and how they can learn to manage these tools. The pedagogy, the process
of teaching and learning, is important in this book: it builds on the social con-
structivist theory in which knowledge is constructed collaboratively through social
interaction in class (Kiraly 1995, 2000, 2005, 2012, 2014). Kiraly argues that collab-
orative learning teaches us more about our learning process as well as the translation
process than, for example, a competence-based approach.This book moves on from
this premise and includes the management of workflow through technological
tools. Constructivist principles encourage interaction between professional activity
and academic reflection, a thread that runs through all the chapters (Mitchell-
Schuitevoerder 2013). Kiraly’s constructivism requires learner-centred activities in
which students are given a framework and then set their own objectives. Such
activities constitute the project-based end-of-chapter assignments in this book.
‘Learner empowerment’ is a constructivist notion suggesting that if learner inde-
pendence is initiated during the training, it will continue afterwards. This approach
is essential: incessant developments in translation technology and the expanding
range of tools requires the translator’s continuous adaptation to new models
and meanings. Therefore, it is important that ideas coming from the student are
‘scaffolded’, that they are given a real-time context, and are linked to supportive
materials by the instructor (Kiraly 2000). If students set their own objectives, they
must also know which tools and skills are required (Kiraly 1995).
A final comment about the importance of revision and evaluation in the
book: because students actively construct their real-time technological transla-
tion skill in the assignments, they need to be involved in the evaluation of their
own work to check whether they have met their objectives (Kiraly 2000). Suitable
assessment methods are required to find out whether the translation project and
process, including the use of technological tools, have been successful.
ACKNOWLEDGEMENTS

My gratitude goes first to hundreds of MA students at the universities of Durham

and Newcastle who attended my translation and technology classes. I observed how
they responded to the topic and my project-based style of teaching (at Durham
University). I would like to thank Federico Federici and Ya Yun Yalta Chen for
inviting me to teach. The project-based approach was Federico Federici’s brain-
child and my ensuing research path took me to conferences across the globe where
I heard and exchanged ideas with many of the authors quoted in this book.
My thanks go to the Institute of Translation and Interpreting (UK) for the wide
range of speakers they invited to their conferences and for their back-office support
to me as a translator and author.
I thank my colleague translators in the ITI North East Regional Group in the
UK for answering my questions about working practice.
A special word of thanks goes to Lucas Vieira Nunes who gave me the initial
wink and useful answers to my questions about technological tools used in training
institutions.
I would like to thank SDL Trados for giving me a personal licence to their
CAT tool.
I am very grateful to the librarian of Bill Bryson Library University of Durham
for giving me general access to their repository of (online) books.
I thank my friends: authors Jenny McKay and Simon Firth, editor Lynn Curtis,
terminologist Sue Muhr-Walker, and URL adviser Xinyi WANG, for their advice
and support.
And Lotte de Waal, thank you for your illustrations with a touch of humour.
Finally, my thanks go to the Routledge and series editors Kelly Washbourne,
Louisa Semlyen, and Eleni Steck for making the preparation of this manuscript a
smooth run. I received the support and answers I needed and I could work at my
own pace without the translation deadlines I am used to.
ABBREVIATIONS

AEM automatic evaluation metrics

AI artificial intelligence
API application programming interface
ASR automatic speech recognition
BLEU bilingual evaluation understudy
CAT computer assisted/aided translation
CDA confidential disclosure agreement
CMS content management systems
CPD continued professional development
DQF data quality framework
DQF-MQM data quality framework-multidimensional quality metrics
EAGLES expert advisory group on language engineering standards
EEA European economic area
EU/DGT European Union directorate-general for translation
GALA globalization and localization association
HR human resources
HT human translation
i18n internationalisation
INDD inDesign files
iOS iPhone operating system (now including all devices)
IP intellectual property
ISO international organisation for standardisation
KWIC key word in context
l10n localisation
LISA Localization Industry Standard Association
LQA language quality assurance
LSP language service provider
xviii Abbreviations

MS Microsoft
MT machine translation
NDA non-disclosure agreement
NGO non-government organisation
NLP natural language processing
NMT neural machine translation
PDF portable document format
PE post-edit
PEMT post edit machine translation
PM project manager
PO purchase order
POS parts of speech
Q&A question and answer
QA quality assurance
RBS risk breakdown structure
RSI repetitive strain injury
SaaS software as a service
SEO search engine optimisation
SL source language
ST source text
SR speech recognition
STT speech-to-text
T&C terms and conditions
TAUS translation automation user society
TBX TermBase eXchange
TEnT translation environment tool
TL target language
TM translation memory
Tmdb terminology database
TMS translation management system
TMX translation memory eXchange
ToB terms of business
TQA translation quality insurance
TTS text-to-speech
TT target text
TU translation unit
WCMS web content management system
WWW World Wide Web
XLIFF xml localization interchange file format
XML Extensible Markup Language
GLOSSARY

acceptability target text matches TL standards and end-user expectations

adequacy relates and qualifies the target text according to the source text
alphabetic language language based on roman letters
application programming interface a program that allows other programs
to use its functions
authoring rights claim rights to the writing of an electronic document
auto-assembled string words in the target segment put together by the TM
automated speech recognition cloud-based AI technology for speech to text
or text to speech
automatic evaluation metrics software programs that use a set of linguistic
criteria for testing against a human reference translation
automatic speech recognition cloud-based AI speech technology
autosuggest prescriptive writing in the CAT tool
bilingual evaluation understudy an algorithm to evaluate the quality of an
MT text
bootstrapped a glossary based on and derived from another glossary
code see language code/tag
cognitive ergonomics mental processes, such as motor response, which can
affect interactions among humans and other elements of a system
cognitive friction resistance encountered by human intellect when engaging
with a complex system of rules
collocates words that go together
Confidential Disclosure Agreement see non-disclosure agreement
content management system software application that allows users to collab-
orate in the creation of texts
context match a 100% TM match of a string of words
xx Glossary

Continued Professional Development professional training in a variety of

formats
copywriting writing original texts
corpus-based MT new data measured against data in the database
customisation Tmdb phrases or terms in client glossaries
dialog secondary window that allows users to perform a command
digital platform facilitates exchanges between multiple parties in the cloud
domain knowledge fields of specialism
domain-suffix last part of a web address to define the type of website
drop-down menu offers a list of options
Data Quality Framework multidimensional metrics for quality assessment
Data Quality Framework-Multidimensional Quality Metrics error
typology
editor see translation editor
encryption converting data into a code for protection
error typology measuring correct words against a benchmark translation
Extensible Markup Language documents encoded in a format that is
readable
false positives potential but unsuitable matches
filter searches for a specific item for a set purpose
fluency target text matches TL standards and end-user expectations
fuzzy match 70%+ correct
fuzzy match threshold proposal of matches only over 70%
gisting a quality that gives a rough understanding
granular tiny component
human resources personnel
human translation translated by a translator and not a machine
image-based image files, not editable
indemnity insurance an agreement to compensate for damage (through
incorrect translation)
InDesign Files image files, not processable in all CAT tools
intellectual property ownership of ideas, intellectual creation and translation
intellectual property rights legal right to claim ownership of ideas
interface where two electronic components exchange data
International Organisation for Standardisation group of experts to
develop international standards
internationalisation intermediary stage to remove culturally specific content
Key Word in Context how the keyword is used with other words
language code see language tag
language quality assurance see QA
language service provider organisation offering language services
language tag/code two letter endings to denote a language
leverage (of subsegment) maximise the value
linear approach read from beginning to end
 Glossary xxi

local search search in database

localisation adapt terminology to culture and language of target country
Localization Industry Standard Association produces a model for quality
assurance
logographic languages languages based on characters or symbols
lookup quick term (phrase) search in databases
machine translation text translated by machine translation engine
markup tag with information
match CAT tool term for TM proposal
natural language processing a computer language to understand human
language
neural machine translation recent development of statistical machine
translation
no match no equivalence found in the TM
Non-Disclosure Agreement signed document that promises not to divulge
content
Non-Governmental Organisation non-profit, voluntary group of people on
international level
non-translatables content that does not need to be translated
non-whitespace languages character-based languages
offline editor translation memory editor that works offline
on-demand software companies pay a flat monthly fee for use of software
open-source software that can be inspected, modified and enhanced by anyone
package ZIP archive with XLIFF, TMX, TBX files for translation
paradigmatic approach focus on the segment and its alternatives
parent folder structured folder in Windows File Explorer
part of speech category with similar grammatical properties
perfect match equivalence keyword
plugin small component in a larger piece of software and performs a
specific task
Portable Document Format uneditable image file
post-edit MT edit
post-edited machine translation machine translated text corrected
by humans
post-editor the linguist who post-edits MT
precision the fraction of words in the MT output that is correct compared
with the HT benchmark
pre-translate CAT function to search TM/Tmdb for suitable matches
preview separate window shows the target segments in context
project manager member of project management team with specific role
propagate CAT tool populates other occurrences in document
pseudo-translate tests suitability of file for translation
purchase order official document sent to translator with details about type,
word count, deadline, and agreed price for the translation
xxii Glossary

push and pull approach proposals and requests

quality assessment quality measurement
quality assurance CAT tool function checks for errors
recall the total number of correct words (i.e. precision instances) produced
by the MT
reference translation originally a human translation used to test the quality of
speech recognition
regex mathematical theory to match recurrent patterns
repetitive strain injury pain felt in muscles due to overuse
ribbon taskbar in MS Word and CAT tools
Risk Breakdown Structure categories of risk
Search Engine Optimisation a technique used to move websites to the top
of search engine rankings by means of search words or keywords
seed terms/list keywords on webpages
software as a service to access software in the cloud
speech recognition software to convert speech to text
speech-to-text speech converted to text on a digital device
standalone not web-based
standardisation technical specifications and criteria to ensure that materials
and products are interconnected and interoperable
statistical MT MT based on data from bilingual text corpora
status bar horizontal bar at bottom of screen
string group of words that operate together
syllabic language see alphabetic language
tag symbol that contains formatting information
term extract extracts keywords and expressions in a new source text or a
corpus of existing translations
terminology database database in CAT tool for manual entry of term pairs
terms and conditions general agreement between vendor and buyer of
service
terms of business see terms and conditions
text-to-speech artificial voice reads text from PC
transcreation translation+creation: a marketing term that refers to the
adaptation of ST to localised TT
Translation Automation User Society commercial organisation that
develops MT testing metrics
translation editor/grid the interface with grids with ST and TT segments
translation environment tool translation technological tools
translation management system supports complex translation projects with
integrated MT
translation memory key memory database in CAT tool
translation quality assurance see quality assurance
Translation Results window window that shows proposed matches
translation unit confirmed translation pair in the translation editor
 Glossary xxiii

usability MT quality criterion to determine level of usefulness

vendor linguist who sells their services to the LSP
web content management system web-based content management system
(see CMS)
web localisation making the text linguistically and culturally appropriate to
the target locale (country/region and language)
whitespace languages word-based languages
World Wide Web internet
INTRODUCTION FOR INSTRUCTORS

This book is designed for instructors whose students are new to translation tech-
nology. It can also be used for students and professionals who want to learn more
about translation technology and project management. This book is not a manual of
technological translation tools, nor is it a guide to translation project management.
If students have not yet been introduced to CAT tools, the suggestion is to throw
them in at the deep end. The manufacturers of CAT tools and other translation soft-
ware products provide detailed manuals, videos, Q&A pages, and knowledge base
portals. It is important to point students to the Help sections in the CAT tool. Don
Kiraly (APTIS Newcastle upon Tyne 2019) suggested that a 30-minute introduction
to the CAT tool by the instructor should be enough to get students started, on con-
dition that they use the CAT tool daily. The instructor’s colleagues should allow and
encourage students to deliver all their translation work in CAT tools. An introduction
to CAT tools for teachers may be helpful.The key to CAT tool competence is regular
use. A weekly instruction class is relatively unproductive, because it is not enough to
consolidate a large amount of newly learnt skills, however small they may be.
The project-based approach aims to give the students an all-round perspective
of translation projects, workflow, and of the teamwork required to complete large
multilanguage translation projects using translation environment tools (TEnTs).
The book does not aim to be prescriptive: suggested TEnTs may not be access-
ible and they could become obsolete over time. One can only use what is available
in the educational institution. There is the option to download demo versions on
personal devices. The URL address www.routledgetranslationstudiesportal.com/
in the textbook points to the (editable) Routledge Translation Studies Portal. The
double asterisks ** in the book are important markers that signify useful URL
addresses in the Portal.The asterisks are followed by the title of the relevant section
on the web page. If URLs are not accessible, you may be able to find alternative
sites from your location.
 Introduction for instructors xxv

The role of the instructor

The instructor is a guide and a supervisor. The project-based approach is student-
centred. The instructor decides with the students on a framework that is suitable
and feasible, makes suggestions and sets objectives, but does not determine the
outcome. Meanwhile, the amount of new experiences for students may be such
that clear guidance and a tight control of the steps in the process, especially in the
assignments, is necessary. The instructor is free to decide on the running order in
which chapters and assignments are presented to students.The instructor is the best
person to gauge the students’ competencies and skill levels. Equally important are
the availability of hardware, software, and access to the web that will determine
when or whether to discuss a chapter. A mix-and-match approach to the learning
material in the chapter is possible.

The learning process

The project-based approach draws heavily on the ideas of Kiraly (2000, 2012) and
Pym (1993). It is a constructivist way of learning in the classroom (Kiraly 2012) that
requires long-term, interdisciplinary, and student-centred activities. It assumes that
skills and competencies will be enhanced through social interaction in class. It
encourages the analysis of problems and troublesome areas, which should ideally
be managed collaboratively (Land and Meyer 2006). Pym (1993) underlines these
notions: the starting process should not be the translation process, but an analysis
of the problems ahead. If we start from a theory, we will be searching for matching
examples – an approach that does not encourage the application of critical thinking.
In this book the project-based assignments constitute the ‘problem’. Students and
instructors should start out by analysing them collaboratively. After the framework
and clear objectives have been set, the instructor becomes the guide and supervisor,
or an overall coordinator (Mitchell-Schuitevoerder 2014).
A project-based approach suits translation technology. Students need greater and
more in-depth skills, knowledge and hands-on experience before they enter the
profession. Humanities students are perceived to be more reluctant to engage with
technology and if there is an element of fear this could limit the student’s poten-
tial during and after studies (Doherty and Moorkens 2013). A translation project is
complex with many stages from translation to revision and completion, and many
more in between. Workflow [is important and] should ideally be discussed and
designed first in each project-based assignment.

Student-centred project management

We know that the CAT learning curve is steep and that much effort goes into gen-
erating the first CAT target translation.Why then should we add the layer of project
management? The answer is that the interplay between translation technology, situ-
ation, and context is best practised in project management. Situatedness brings the
xxvi Introduction for instructors

real world into the learning process (Rohlfing et al. 2003). A student who struggles
with the CAT tool and does not experience the interplay, who cannot understand
the context, and cannot visualise the situation, is likely to give up on translation
technology.They must learn to see the whole picture: the situation is created by the
source text with its characteristics and constraints (e.g. PDF formats), the context as
a construct that depends on factors, such as language pair and direction (are there
enough students in class with the same language pairs?), access to the internet, etc.
The context controls and challenges the user. A CAT tool presents its own bound-
aries and limitations and they too are part of the context.The student is confronted
with a situation and is challenged to act appropriately within it. They need to learn
to shape translation technology instead of being shaped by it; in the book they learn
about issues surrounding machine translation, translation quality in TEnTs, digital
ethics, translating in the cloud and this knowledge should help them make choices
and set parameters.
‘Situated learning’ gives the student a chance to reflect on the role and impact
of translation technologies and to see the bigger picture (Bowker 2015). We need
to create an authentic workplace in the classroom where students work and learn
collaboratively in preparation for a career in the translation industry. The ‘Food for
thought’ sections in this book invite discussion or reflection on the theory. The
project-based assignments combine understanding, reflection, learning, and team-
work in practice.

Project-based assignments
The project-based assignments in the chapters offer frameworks which can and
should be adapted by the instructor, in line with situation and context, and com-
petencies. The assignments challenge the students to remember, understand, apply,
analyse, evaluate, and create (Bloom’s revised taxonomy 2001). The instructor leads
discussions and offers support during planning and execution stages. After the
team’s self-analysis and evaluation, the instructor sits down with the students for a
final assessment of their collaborative translation project.

Cross-curricular or modular approach

If the degree course includes a terminology module, a cross-curricular/modular
execution of project assignments works well. Newly created glossaries – bilingual
or monolingual – and corpora can be uploaded in a CAT terminology database.
Cross-curricular/modular revision practices would support/could be supported
by CAT tool quality assurance. If practice translations (homework) are completed
in CAT tools and exchanged in bilingual doc. format, revised files with comments
can be re-imported in the CAT tool and resubmitted as a clean target file. This
working practice can be started early if the instructor’s colleagues are willing
to support a cross-modular/curricular approach to terminology and revision
with translation technology tools. By the end of the book, students should have
 Introduction for instructors xxvii

experienced most management and linguist roles pertaining to the creation and
completion of a translation project. They will also have tried and managed a var-
iety of technological features. They should have taken turns in managing multiple
translation projects. Project management and other business-focussed or career-
oriented skills must be consolidated in the final chapters. The instructor could
arrange a presentation on project management by a colleague or student from the
business school at any time.

Assessment
A project-based approach to translation technology is built on the premise that
what happens in the industry should be trialled in class. Errors are made, outcomes
may vary, and evaluation is essential to achieving improvement. Three types of
assessment in collaboration between instructor and students can be used as a starting
point (Robinson et al 2008):

• translation quality assessment

Students must not forget the ultimate objective of learning about translation tech-
nology, which is to improve translation quality. The submission of a final translation
by a team can be graded by the instructor.

• peer/self-assessment of translation quality and efficiency through project

management

The collaboratively agreed final team translation is the benchmark [translation]

by which students assess and rate their individual translations. Although the focus
in each project-based assignment is different, the students must recognise that the
goal of a quality TM, good reference materials, and terminology databases or revi-
sion practice achieved with TEnTs, is an error-free translation. It is easy to lose sight
of this goal and to concentrate on the success of the TEnT(s). Metrics and quality
standards discussed in the book make it easy to switch to rating the TEnT rather
than the translation!

• the individual student should assess their own translation quality and efficiency
achieved through collaboration and project management

Individual assessment should be of the student’s own translation against the bench-
mark translation.

• Project management assessment – both peer and individual – can be arranged

in collaboration with the instructor and should include the assessment of work
by the team and its individual members. It can be helpful to use a Likert scale
for rating activities such as:
xxviii Introduction for instructors

• learning from each other

• dividing/sharing work
• clarity of task briefs
• time management
• collective decision-making
• communication
• motivation
• responsibility of self/peers
• tool management
• personal learning

A heuristic approach to learning or problem-solving allows the student to dis-

cover and learn from their own experiences through analysis and synthesis.
Ultimately, students should be able to ask themselves: what should be my role
/ how can I solve the problem / should I act or refrain from comment / what
can I do in this situation and what is the most practical solution? Regardless
of the problem or the relevant technology, they should always adopt a can-do
approach.

Team setup
Students who have not yet worked in the translation industry may not neces-
sarily be familiar with concepts such as project management or project-based
translation. Both concepts are explained in Chapter 1 but students will need
guidance and support when setting up their teams to carry out a project-based
assignment. The assignments vary between two kinds of team setup: the pro-
ject team in which a team of students work together and a project manage-
ment team that manages and contracts students to do the work. The boundary
between the two types of team is not strict, and there may be situations where
there are not enough students to form teams and assignments will have to be
carried out collaboratively. Setting up a team is challenging and needs to be
well organised:

• Project management teams need enough members to manage the different

components of the translation project. If a project team is small, the members
may have to take on multiple roles. Students who are members of a team
may have double roles if they are contracted by other teams to translate.
Students should have a chance to try different roles. There are arguments
for and against continuing with the same teams. Variety is beneficial but
continuity can give better results. The decision may depend on whether the
translation project is short term or long term (see Shuttleworth 2017).
• Teamwork is vital. The establishment of teams may be dependent on the avail-
ability of students with similar language pairs. Diversity of target languages is
 Introduction for instructors xxix

common in real-time project management teams. Linguists are asked to check

for accuracy in languages they do not know.

Completion dates and source texts

There is a close link between completion dates and translation projects. The time
between request and delivery is often a matter of days. Ideally, students should
experience working under time pressure and it is wise to press for completion
within a week. If the assignment drags on for much longer, students lose interest
and become demotivated.
Trainee translator students do not generally have the knowledge and expertise to
choose suitable source texts for TEnTs.They will need guidance from the instructor.
There are several options:

1. the instructor prepares a range of URLs with suitable source texts containing
short sentences and repetition. The digital format of the text matters, too.
A text without layout requirements or HTML markup codes would be prefer-
able in the initial assignment;
2. the instructor invites an external ‘client’ to supply their source material for
translation. A good example is University College London (Shuttleworth
2017) where the translation trainee students collaborated with the Museum of
Zoology with a repository of 68,000 specimens. The museum was delighted
to have students translating for free and the students were highly motivated to
produce translated material that would be published.

Finding organisations around the campus or your locality willing to supply

e-materials in return for free translations by the students would undoubtedly be
appreciated by all parties. They could be museums, transport companies, commer-
cial companies, educational organisations, but also and particularly charities or not-
for-profit organisations … the list is endless. If they have a website or materials that
have been part translated, previously translated material would be ideal for text
alignment and consistency.

Hardware and software

Suitable hardware and software are essential, such as enough desktop PCs with MS
Windows, but when it comes to software, the instructor must decide what is feas-
ible. It is not necessary to introduce students to multiple CAT tools. If they under-
stand the basics of one, they may be able to adapt their skills to others. Software
manufacturers are known to offer CAT tool licences with a server version to edu-
cational institutions. There is a huge range of freeware and shareware tools. Web-
based TEnTs in Chapter 7 have the advantage of being comprehensive, including
TM and MT features.
xxx Introduction for instructors

Technology is challenging, particularly when it comes to compatibility, the

instructor needs ingenuity and lateral thinking.

Checklist for the instructor:

• source texts
• MS Word skills
• instruction to the instructor’s colleagues
• instruction to students on self- and peer-assessment (it includes revision, but
also personal and team skills)
• invitation to a business school teacher to give instruction on business and
workflow models relating to project management
INTRODUCTION FOR STUDENTS
AND TRANSLATORS

What the book plans to teach you

After thirty years as a freelance translator, I can honestly say it is a great profession.
Technology has only made it better for me, because I can now work anywhere
I choose. I open the computer-assisted translation (CAT) tool on my smartphone
and speak the translation into the mic of my phone or else tap on a collapsible
keyboard if I am in a public place. Amendments or additional paragraphs can be
translated and added immediately. After six years as a teacher to post-graduate
students of translation technology at two different universities in the UK, I have
seen how former students have moved on and progressed in the translation industry.
Recently, one student who is now a freelancer said that starting her career as a
project manager of a language service provider (LSP) was the best thing she could
have done. Another who worked with an LSP in the UK for 5 years, returned to
Shanghai to be a project manager using a translation management system (TMS)
either at home, at a hot-desk venue in the city, or at the LSP’s headquarters in
Dubai. Some have gone to work for LSPs abroad and returned as freelancers; others
have used their acquired business-management skills and secured jobs with tech
companies as software developers.
Fascinating as it has been to write a book on translation technology, it has also
sometimes been a frustrating experience because of the ever-changing dynamics
within the business. In the space of a year, long-established protocols and software
can be completely overtaken by new developments. The most recent case in point
was an observation I had made in one of the final chapters of this book that CAT
tools do not have grammar checkers. This function has duly been introduced by
one of the major tools, and undoubtedly other programs will follow suit.Translation
technology is all about change and how to manage it. The Routledge Translation
xxxii Introduction for students and translators

Studies Portal is therefore the site to consult regularly for updates and additions and
the names of products and tools.The URL address is www.routledgetranslationstud
iesportal.com/. In the book URL links on the Portal are marked like this: **. The
information in parentheses gives the title of the relevant section in the book on the
Portal’s web page.
Another concern has been deciding on the title of this book: outsiders inter-
pret ‘project-based approach’ as meaning project management and ‘translation tech-
nology’ as machine translation. The project-based approach refers to the style of
learning advocated and to the translation as a complex, frequently collaborative
undertaking: volume and time constraints on source documents often mean that
they must be split and shared among several translators. The translation technology
component is best explained as the software that is used by translators and LSPs in
large translation projects. The range of tools is infinite. The CAT (computer-aided
translation) tool is the main player, because it is the tool into which so many other
technological aids can be integrated. The machine translation engine is not the
main player, even though it likes to compete with and within the CAT tool. It is,
however, a very serious competitor both in terms of efficiency and improved trans-
lation quality. Watch this space!
The project-based assignments at the end of each chapter provide oppor-
tunities to practise translation technology skills, preferably within a framework
of collaboration, through the use of appropriate tools. If circumstances do not
permit organised collaboration, you will have to find other ways to collab-
orate: you can ask revisers anywhere in the world to revise target your files;
you can ask teachers to comment on your work in bilingual CAT files; you can
crowdsource; you can use digital platforms.
There may be several setbacks ahead of you while working through this book:
the tool may not be available or work well, there may not be enough peers with
the same language pair to facilitate collaboration or enough translators/revisers to
tackle the job; or missed deadlines. These problems crop up in the working lives of
professional translators too. This book aims to give you a better understanding, and
hands-on experience within the assignments, of how translation technology is
managed in the workplace. It also aims to help you use translation technology com-
fortably in your personal work environment.Translation technology is promoted as
being a translator’s best friend: manufacturers claim that it radically improves effi-
ciency. Their often-used mantra is less time, more profit, better quality. We will put this
to the test by using their translation technology tools and reviewing them through
the eyes of the different parties involved.
This book is not only written for trainee translators and aspiring transla-
tion project managers, but also for practising translators, project managers,
and instructors. Hopefully, practitioners within each category will appreciate
being given an overview of translation technology tools and their impact on trans-
lation work. The project-based assignments at the end of the chapters are set up as
frameworks and give users and readers of the book an opportunity to experience
 Introduction for students and translators xxxiii

different roles, those of freelance or in-house translator, as well as reviser or pro-

ject management team member ranging from project manager to human resources
manager, client manager, terminologist, technology adviser. The translation project
is viewed from the perspective of the software developers and the commissioning
client who buys the translation service. Conflicts of interest are inevitable and it
would be beneficial for all concerned if these could be aired and discussed during
or after the assignment.
The assignments are an important component of the book. If they are read
before an in-depth study of the chapter, they will give a preliminary indication of
the skills and competences needed. They may also need modification in line with
location, availability of tools, and language pairs. They should be planned together
with the instructor, who can advise on suitable documents and files for transla-
tion and who will advise on time allowance for completion. Time for reflection
must also be allowed: trainee and professional translators and managers should be
encouraged to think about the tools they use and the quality of the product they
hope to achieve. The ‘Food for thought’ sections are intended to promote discus-
sion of potentially controversial issues. Time, for example, matters to all of us but
in different ways. Clear deadlines must be set and met, and these should be tight in
line with expectations and practice in the translation industry. A short deadline is
important to the client who commissions a translation; having enough time to do
the work properly matters to the translator who must deliver. The book highlights
the paradox that although technology is supposed to help us reduce the time it takes
to complete a job, sometimes it forces us to progress faster than we feel entirely
comfortable with.
Other frustrations are that websites, webpages, and software programs are
updated regularly, and web links are modified. Some cannot be accessed all over
the world. For this reason, URL-links are not printed in the book but listed on the
Routledge Translation Studies Portal where updates can be made as neces-
sary. Technology companies and manufacturers come and go, the popularity of
their products grows and wanes, and therefore the book makes limited reference to
programs (see Portal).
Illustrations and data in the book are taken from the following CAT tool
programs: DVX3, memoQ9, and SDL Trados Studio 2019. I started with Trados
Workbench before the Millennium. I changed to DVX3, which was stable and
straightforward but it did not keep up with SDL Trados Studio and memoQ.
The competition between the latter two is fascinating to watch and experience.
Other tools such as OmegaT, Lilt, and Wordfast decided not to compete but to
move into the cloud where they are either open source or focus on machine
translation.
This book is not a manual on how to use technological tools, nor is it a
workbook. It is a practical textbook that aims to explain the ideas behind the
technological tools available, their features and functions. The application in the
project-based assignments depends on what is available where you study or work.
xxxiv Introduction for students and translators

Although there is a gradual build-up from chapter to chapter in introducing new

tools, it is possible to skip chapters or change their running order.
The book presumes a basic knowledge of CAT tools. Formal instruction
may not be needed because the programs have made so much instructive material
available, both within and outside the tool, that learning is best approached through
doing. One of the foundations of CAT tool competence is a sound knowledge of
MS Word and its functions.
The book touches on localisation, transcreation, and the technologies
surrounding the translation of webpages. However, it does not cover these areas in
detail. They are independent subjects, too extensive for a book that aspires to give
an overview of generic technological tools.The book does not mention audiovisual
tools for the same reason – they are specialist too. Nor does it explain machine
translation (MT) as a product. MT is discussed purely as a supportive technological
tool in relation to the CAT tool

Teamwork and code of conduct

The success of the project-based approach relies on good teamwork. The con-
struct chosen is that of a language service provider. An LSP may consist of an
in-house project management team with the following managers: (senior) pro-
ject manager(s), relation managers, marketing and desktop publishing managers,
recruitment managers, (human) resource managers, terminologists, technicians.
The LSP will have an extensive database of linguists, including freelance translators
and revisers.The project-based assignments require the formation of teams and the
creation of a database of translators with different language pairs. Each team needs
to set up a code of conduct to enable teamwork to progress smoothly. A code of
conduct consists of a set of rules that will be respected by all team members: a
commitment to meet agreed times, complete allotted tasks, maintain confidenti-
ality, and show respect to others (this does not mean automatic agreement with
other opinions, only that these must be respected). The rules should also include
listening to each other, and a commitment to resolve conflict and tackle problems.
There also need to be agreed measures to be taken should team members fail
to meet set standards – intentionally or unintentionally – required to fulfil the
project-based assignment.
There is no maximum or minimum number of team members. Under good
management a larger team will achieve more than a small team. Good management
means that senior managers will respect and listen to other team members and that
decision-making is based on mutual consultation.

Pitfalls
Rather than my listing a whole string of recommendations here for avoiding poten-
tial problems with hardware and software, translation texts and resources, language
newgenprepdf

Introduction for students and translators xxxv

pairs and direction, I suggest you discuss any specific concerns with your instructor
in advance of beginning assignments, but here are a few general tips and warnings:

• internet

If there is no access to the internet, the project-based assignments cannot be carried

out fully, nor can a translator or LSP do their work consistently. If the failure is
temporary, it may be possible to use smartphones as hotspots for personal laptops. It
is important that all files are kept backed up and available to meet this eventuality.

• texts for CAT tools

CAT tools work well with certain formats and love the repetition of terms and
terminological phrases.

• hardware

If educational organisations cannot purchase software or licences, there are

manufacturers that will offer licences for educational purposes. It is possible to
download demos on personal laptops for limited periods of time.

• if there is a shortage of linguists in a specific language pair or direction, it is

acceptable to source linguists outside your cohort.This is frequently a problem
for LSPs and must be managed.

Remember, problem-solving is a prime requirement for users of translation

technological tools!
1
COMPUTER-AIDED TRANSLATION
TOOLS AND TRANSLATION PROJECT
MANAGEMENT

Key concepts
• The translator, the language service provider (LSP), and computer-aided trans-
lation (CAT) tools have become partners in the translation process
• The computer-aided translation tool comes first among translation environ-
ment tools (TEnTs)
• The CAT tool is one of many TEnTs that can be linked or integrated
• The management of translation projects requires good teamwork, administra-
tive and project-solving abilities

Introduction
CAT tools are invariably used in managed translation projects, but they can also
be helpful in translation projects where the translator is working directly for the
client. The CAT tool was built to assist the translator, it is shaped by the trans-
lator, it can efficiently process the work of collaborating translators and revisers,
and it has become the tool used by project managers to help a team of linguists,
in-house or freelance, to deliver a high volume of good quality translations, in mul-
tiple languages, shared among many contractees, and in the shortest period of time.
Any difficulties associated with the tool must be solved by the users. Most likely,
the problems were created by humans in the first place and not by the technology
(Kenny 2016). This chapter builds on a basic level of CAT tool skills, good MS
Word competence, the availability of Microsoft (MS) Windows operating systems
and an appropriate CAT tool. In addition to MS Word and MS Windows expertise,
your administrative skills will be challenged: effective personal file management is
the precursor to successful CAT tool operations.
2 Computer-aided translation tools

1.1 Translation environment tools (TEnT)

The denotation ‘translation environment tool’ (TEnT) was coined in 2006 by Zetzsche,
a translating professional, who realized that when translators refer to ‘CAT tool’, they
are referring to the TM feature, which is only one component of a CAT tool. The
acronym TEnT is meant to cover a broader spectrum. Currently, there does not seem
to be much consistency in the usage of CAT or TEnT, so in this book CAT refers to
the translation memory program, TM stands for the translation memory database, and
TEnT covers a whole range of translation technology tools**(go to www.routledgetr
anslationstudiesportal.com/ – A Project-Based Approach to Translation Technology –
links: TheTool Box; GALA’s directory of language technologies). GALA (Globalization
& Localization Association) lists language technologies ad infinitum. Some examples
of lesser-known technologies are word-count software for all common file formats,
terminology management software for multiple dictionaries or glossaries, and software
that gives the user a clear picture of all their tasks and time expenses.The list continues
to grow because the industry will provide new tools to help us manage our day-to-day
operations and tools if they perceive a gap in the market. For example, translation
management systems (7.5) allow us to manage not only CAT tools but administra-
tion, accounting, human resources, project management, etc.
Zetzsche’s list (2003–17) of required technology tools for the translator heads
off with Microsoft Windows (MS) and Office. MS software is the first essential. In
a standalone and not web-based CAT tool, docx, Excel and PowerPoint files are
uploaded and saved in MS Office, then imported in the CAT tool and exported to
MS Office after translation. The translation, or target file, is then called a ‘clean’ file.
Sometimes the client or LSP also need the bilingual CAT work file, the so-called
‘dirty’ file or XLIFF file (2.4).
Other examples of TEnTs are dictionaries, bookkeeping/financial tools, conver-
sion tools (although CAT programs can convert Adobe PDF files (2.4), they do not
always retain the original layout), web browsers (Firefox, Chrome, Opera, etc.) and
programs for graphics and desktop publishing that enable the delivery of a ready-to-
use product. Beyond the basics, such as indispensable security software, there are PDF
fixing tools for recovery or repair, and screenshot readers which convert screen con-
tent to editable format. And then there are APIs, plugins (7.1), speech recognition
software (7.1.1) to name but a few, which can either be linked to the CAT tool or
integrated. Different CAT programs have different strengths and weaknesses, which
is why additional TEnTs can be helpful. Apparently, translators’ most popular TEnTs
lie in the field of terminology and will be discussed in detail in Chapter 4. And then
there are the comprehensive translation management systems, which automate many
parts of the translation process and include business management tools (Chapter 7),
popular among LSPs. The trend to integrate many TEnTs is gathering momentum.

1.2 Compatibility
Most CAT tools operate on MS Windows systems. Users of non-MS Windows
systems can either download virtual MS software or use CAT tools specifically
 Computer-aided translation tools 3

designed for their systems. Currently the leading programs are only Windows-friendly.
LSPs often prefer to contract translators who use MS operating systems for ease and
compatibility. Many MS-based TEnTs are a challenge for non-Windows users.
OmegaT was launched in 2000 to work on multiple operating systems and
others have followed. Whereas the big players in the MS Windows market are
expensive to purchase and to maintain (annual service agreements, upgrades),
OmegaT is an example of free open-source software for non-MS users. Open
means that it can be inspected, modified, and enhanced by anyone, either producer
or user. OmegaT’s compatibility with other CAT programs is supported through
filter plugins, which allow you to process XLIFF files, the standard workfiles in
CAT (2.4). XLIFF is an XML-based format, created to facilitate the exchange of
files between programs and users. Extensible Markup Language (XML) encodes
documents in a format that is both human-readable and machine-readable. The
demand to use iOS (Apple’s operating system) software and not only Microsoft
on desktops and laptops has increased so much that many suitable CAT programs
have been designed since OmegaT entered the market**(link: non-MS compatible
CAT tools). Leading CAT tool competitors are Microsoft only programs.
Not only hardware-cum-software, but also the format of source files is an
area where compatibility plays a role. CAT tools can handle major file types like
Microsoft Office files and webpages, but more complicated file types such as image
files are not always supported. For example, InDesign design files (INDD) are cur-
rently only supported by one CAT tool**(link: INDD compatible), and even then,
only after conversion in the cloud. XML files, in which text is created and stored
in databases rather than in document files, can be converted in CAT tools, but not
without limitations or applied filters. Once they are set up in the CAT tool the pro-
cess is straightforward and they will be delivered as a perfect image product, ready
for publication, if tags and codes have been applied correctly (2.4). PDF files can
be problematic in relation to formatting. Hence, there are conversion TEnTs to
make the import in a CAT tool straightforward (2.4).
A different kind of compatibility, or conflict, arises between web-based and
standalone CAT tools (1.6.2). When the translator is required to work in a web or
cloud-based CAT tool, such as a server managed by an LSP, the translator cannot
take ownership of the data they have entered in the TM database. If the trans-
lator cannot download the XLIFF file or work in, for example, the program’s off-
line editor**(link: offline mode), the translation units entered in the TM are no
longer the translator’s intellectual property. This loss of control often means that
translators cannot build their own TMs. The technical and ethical aspects of TM
sharing and online translation/collaboration will be discussed in more detail in
Chapters 2 and 6.
Lack of compatibility between CAT tools has brought much frustration to
working translators. The manufacturers are fully aware, and they improve com-
patibility to remain ahead of their competitors. Technology is often driven by
manufacturers, who do listen if users make their voices heard. The user-friendliness
of CAT tools today compared to the early programs in the 1990s has made the
current tools intuitive and much more accessible.
4 Computer-aided translation tools

1.3 Translation memory

In this section we will scroll through reasons for using a CAT tool. The technology
in the TM database will be studied in Chapter 2. But first, we should consider
the learning curve that appears to deter some practitioners from using CAT tools.
Over the years, the curve has become less steep for two reasons: the manufacturers
compete to make their software user-friendly; they have included clearly written
Help features and videos that can be accessed from within the CAT tool and the
internet has opened many opportunities for users and learners to find instruc-
tion and answers to questions. If you consider yourself a newbie and feel unsure
about how to use a CAT tool, demo versions can be downloaded for a month
and the manufacturers’ training videos will get you started. If you need more
help, there are translator platforms which offer advice through Q&A; translator
support is offered on mailing lists and in online support groups, and professional
organisations**(link: professional organisations) offer continued professional
development (CPD) webinars to members.
We owe the following definition for the TM to EAGLES (Expert Advisory
Group on Language Engineering Standards). It describes a TM as a text archive of
matching source and target units with appropriate metadata to support the ranking.
The definition also points to variations between programs:

A translation memory is a collection of multilingual correspondences with

optional control information stored with each correspondence. […] The
control information can include information about the source text […],
its date, author, company, subject domain. This information may be used
in ranking matches. […] A given query with a current source segment
may return a number of correspondences with matching stored source
segments.
CAT programs differ as to the information stored along with the raw texts
and the retrieval methods. […]
EAGLES 1996

In 2010, two decades after the software had entered the market and competition
had flourished, TAUS, the Translation Automation User Society, predicted a more
advanced future for translation memory**(link: TAUS).The EAGLES definition of
TM as a text archive from which translator select, has changed to programs which
are highly compatible.They use a ranking system to suggest and propose and it is up
to the user to reject or accept.TAUS wants to prove that algorithms, metrics, ranked
matches in TM, and machine translation (MT) are on a par. They believe that TM
will move into the cloud, like its MT sibling:

The opportunities for advanced leveraging using sub-segment matches

and a combination of statistical algorithms and linguistic intelligence
have unfortunately been ignored for too long by the mainstream industry.
 Computer-aided translation tools 5

TM will now finally become a smart tool, bridging the gap with its more
intelligent MT sister and significantly increasing the recycling of previous
translations. At the same time, TM will move into the cloud. Leveraging of
translations will be done in the cloud through web services links in desk-top
and enterprise translation tools. The combination of advanced leveraging
and the sharing of TMs in the cloud will boost translation productivity by
30% to 50%. And just as important, we will see greater consistency and
accuracy in translations.
TAUS Predictions Webinar, 21 January 2010

Why should we use TM? According to the industry’s mantra, we need TM

to reduce our time investment, to improve efficiency/productivity and profit.
They claim that the translator will spend less time recalling terminology they
translated previously, because matches are proposed each time the TM iden-
tifies repetition. Not only does the program increase our efficiency, but it will
also boost profit: less time means more money. Repetition has gradually become
a keyword with the negative connotation that creativity is irrelevant because
we need consistency. Indeed, the TM works extremely efficiently in texts with
high repetition rates, but repetition should not be the only criterion for using
TM. The CAT tool has many advantages over and above translations performed
outside the tool:

• clear side-by-side presentation of ST and TT

• segmentation of text helps prevent omission
• the search function facilitates searches in large texts and multiple texts
• power loss does not affect the TM, hence no loss of translated text
• autocorrection and accuracy checks support ‘tired eyes’
• well-structured revision features
• text sharing and third-party revision or collaboration across different computers
• the integration of many tools (dictionaries, glossaries, API machine translation)
in one program
• the history of previous matches

The listed features show that the CAT tool can be supportive in different genres
or styles of text and in a variety of formats. TM supports our personal working
memory; it remembers what might have slipped our mind.
The formatting function, a helpful characteristic of the CAT tool (1.6.1) is
a great time saver. It replicates the layout of the ST in the TT and it facilitates
the translation of file formats which cannot be processed in MS Word. The
CAT tool automatically converts the different formats to its standard bilingual
interface (2.4). Web formats are one example of the many new technological
features that have leveraged translation above the pure act of translating. They
have placed more demands on the translator. CAT tools can help (Melby and
Wright 2015: 665).
6 Computer-aided translation tools

Food for thought…

What value does the CAT tool add to the translation process?

1.4 File management

A file is a container in a computer system for storing information. Good manage-
ment of electronic files as soon as we receive them is crucial to learning how to
manage CAT tools. It prevents ‘loss’ of files in the many new folders set up by the
CAT tool. CAT programs have different systems of storing files, but they share the
creation of many new files and their way of naming them. For example, one source
file will generate the following files:

• source file
• bilingual doc file (for review)
• XLIFF file (the work file that can be shared with others)
• TM database
• terminology database
• target file

The total files and folders vary between programs, as well as the storage system.
The translator cannot control the generation of files but can control where they are
stored. Before importing the source file in the CAT tool, it is important to

• name the source file clearly, and

• to store it in a clearly marked folder
 Computer-aided translation tools 7

1.4.1 Filenames
Do not change filenames
The naming of a computer file should be done with care, clarity, and precision.
Once in the CAT tool the name cannot be changed. LSPs request a filename not
to be changed because a different name would prevent it from being imported in
their system.

Label files with short names or numbers and language tags

The filename should be clear and short and not exceed the width of the column
in File Explorer, where it is stored. Do not add spaces but underscore: AB_
2020_IT.
Numbers provide a better chronology than alphabetic ordering: 01_ASD
precedes ASD_01

01_ASD
ASD_01
BSD_04
BSD_15
BSD_7

Non-alphabetic languages
If the language is non-alphabetic and contains characters the computer will have
its own method to order filenames, by strokes or sounds, for example in pinyin. To
achieve a degree of conformity and homogeneity between languages, filenames
generally begin with numbers.

Language codes or tags

The ending of source and target files should always be with the relevant lan-
guage codes, also called language tags. Some languages have different
two-letter codes due to variations within a language. Language codes (ISO 639-
1)**(link: ISO standards) and country codes (ISO 3166)**(link: ISO standards)
are not the same. The filename 0123_en_GB indicates that the source text is in
British English.

Target language tags/codes

The CAT tool adds TL tags to filenames, or it creates folders with the relevant lan-
guage/country tags where it stores exported files. We can make the CAT tool send
our files to the folders where we want them. We need to be systematic and create
a good parent folder with subfolders (Figures 1.1 and 1.2) before we open the
CAT tool.
8 Computer-aided translation tools

Parent
folder

Subfolder
EN -FR

Subfolder
ST

Subfolder
TT

Subfolder
EN -IT

Subfolder
ST

Subfolder
TT

TM

FIGURE 1.1 Translator’s parent folder

Parent folder
A parent folder is the higher directory and can be filled with subdirectories or
subfolders. Figure 1.1 is an example of a parent folder with subfolders set up by the
translator who set up a parent folder with three subfolders:

TMs EN-FR and EN-IT

The translator’s TM folder in Figure 1.1 may have two TM databases for EN<>FR
and EN<>IT respectively. The language subfolders contain subfolders for ST and
TT. If CAT tools generate their own directories, it is best to keep your directory
simple and just create a folder as in Figure 1.2 where the translator has one folder
for CAT projects and one client folder (Agency F).

CAT tool folder

In Figure 1.2 the CAT tool generated seven subfolders for source text,
target text, review files, TM, reports, and retrofit (re-import of final target text).
If we set up a ‘project’ in the CAT tool, rather than a ‘single document’, we can
then add files to the project later.
 Computer-aided translation tools 9

CAT

Agency F

BackupBy
Retrofit

en_GB

External
Review

nl-NL

Reports

Retrofit

TM_eco_en-
nl.sdlproj

FIGURE 1.2 Subfolders generated by CAT tool

Tips
[…] symbol
If you are not familiar with operations, and you want to export or generate a target
file, do so slowly and click on the […] symbol next to the filename. It opens your
browser and allows you to select a folder where you want to store the target files.
By clicking on ‘next’ or ‘finish’ without opening your browser, the CAT tool takes
control and place files in its program folders or generates new folders. The general
rule of thumb is that if we click on Next, Next, Next, without much consideration
as to what is happening, we allow the CAT tool to make choices that tend to ignore
our directories.

Do not move files

Moved files can no longer be opened in the CAT tool because the CAT tool
classifies them as ‘lost’. The error message ‘file cannot be found’ is final and future
updates will not be possible.
10 Computer-aided translation tools

1.5 From Microsoft to CAT tools

A more than basic familiarity with Microsoft Windows and Microsoft Word is
extremely helpful. CAT tools replicate MS software with windows, dialog boxes,
ribbons, menus, toolbars, and functions. Dialog boxes are the small windows that
can be moved by clicking on the ribbon and dragging them to a position where you
want them. A good level of MS Word skills and knowledge of keyboard shortcuts
is useful too. There are many free MS Word skill tests on the internet to test or
improve your skill levels**(link: MS Word test).

1.5.1 The ribbon

CAT software and MS Word**(link: Microsoft glossary) have similar ribbons,
status bars at the bottom, windows, interfaces and dialog boxes. We can access
the ribbon by using the mouse (mousepad) or by typing shortcut codes.The mouse
has two buttons, the left click and right click, which operate different functions.
The left click is the main button by default (it can be reversed for left-handed
users) and we use it to select objects and double click to open them. The right
mouse button can open the contextual tabs on the ribbon and a drop-down menu
in a CAT tool segment. Generally, a right click will open a menu**(link: mouse
functions).

Right click

• opens drop-down menus (also accessible through tabs on the ribbon)

• opens a function icon on the ribbon which gives access to a toolbar with new
functions

Left click

• on an icon or the little triangle underneath opens a drop-down menu.**

(link: MS Word).

FIGURE 1.3 Ribbon in MS Word

 Computer-aided translation tools 11

The ribbon, in MS Word and the CAT tool, is a powerful command bar that has
organised the program’s features into a series of tabs at the top of the window.
Several MS Word core functions are quite similar to CAT tool functions:

• The Application button, top left, which presents a menu of file-related

commands, such as Save, Open, Close. It is advisable to make good use of this
button, particularly when closing a file (avoid the temptation to click the cross
in the right top corner – it is not a saving function).
• The quick Access Toolbar is the smallest toolbar (top left): it displays most fre-
quently used commands, such as Save.
• Core tabs are the tabs that are always displayed, such as Home and Review.
• Contextual tabs are displayed within certain types of files and remain grey if
not applicable.
• Galleries are lists of commands or options presented graphically as icons, for
example, on the toolbar with descriptors and icons, such as Thesaurus and the
icon of a dictionary.
• Dialog box launchers are buttons (or tiny arrows) at the bottom of some groups
of tabs that open dialog boxes with features relating to the group. ‘Font’ opens a
broad range of options, such as Strikethrough or Uppercase.

The functions on the ribbon operate the program. We tend to use the functions
we think we need, and ignore others. The CAT tool ribbon contains functions that
override the default status and make it more adaptable to our needs, for example,
our language pairs. Our productivity, quality, and efficiency will benefit from taking
time to become familiar with the ribbon.

1.5.2 Keyboard shortcuts

Many keyboard shortcuts**(link: keyboard shortcuts) are identical in MS and CAT
tools. Keyboard shortcuts open functions just like the mouse. Shortcuts are accurate,
immediate and appropriate, and they may even help us prevent repetitive strain
injury (RSI), associated with the mouse. An example of a keyboard shortcut in Word
and CAT:

Alt+A highlights the entire file or a smaller section

Ctrl+C copies the selection
Ctrl+X cuts, and
Ctrl+V pastes

Shortcuts help us work faster and are more precise than moving the mouse to
commands on the ribbon. If there is shortcut, it can be found by hovering over an
icon on the ribbon and the code will appear. Apparently, we save time by keeping
our fingers on the keyboard and it prevents clicking on the wrong icon or command.
12 Computer-aided translation tools

1.6 The CAT tool

LSPs generally expect their contracted freelancers to own a CAT tool and to
be competent users. In the following sections we will discuss what the CAT
tool does, the different types and the relationship between the translator and the
CAT tool.

1.6.1 CAT tool features

A CAT tool has four main features: firstly, it segments the text into units and
presents the source and target segments next to or below each other (2.1.1) in
a translation grid (also called translation editor); secondly, a source segment
and target segment form a pair, called a translation unit (TU); thirdly, the
translation memory stores TUs which are recalled as perfect or fuzzy matches
when the TM recognises a source segment (2.1); finally it employs an auto-
matic lookup in its TM and terminology databases to display search results for
insertion (2.1).
The following list of additional features was partly compiled by GALA
(Globalization and Localization Association), an international non-profit organ-
isation whose members specialise in language services, translation, and techno-
logical applications in the translation industry. All the functions and features are
incorporated in CAT tools to support the translation memory:

• Spellcheckers autocorrect the spelling of alphabetic languages, but most

do not support logographic languages based on characters or symbols. The
spellcheckers rely on delimiters such as spaces and punctuation to identify
words. In alphabetic or syllabic languages, they automatically highlight
errors and fix spelling and grammar mistakes. CAT programs rely on the MS
Word spellchecker and Hunspell, an open-source program which is used in
Google Chrome. The default spellchecker varies between CAT tools, but
either program can be used.
• The in-context preview allows translators to preview their translation in real
time in a separate window on their screen, which gives a better overview of ST
and TT than the editor’s segmentation window.
• Integrated machine translation is possible in CAT tools but must be enabled
in the CAT program. An API key (7.1) is usually required, so that you can
pay the provider of the MT engine for the data you send to the MT engine.
• Adaptive machine translation learns from user input in the CAT tool**(link:
examples of adaptive MT programs).
• A concordance is a text search tool that looks for examples of a word or an
expression in a string, i.e. within context, in the translation memory database,
when requested. It checks usage, not meaning. Results can be inserted in the
editor.
 Computer-aided translation tools 13

• Electronic dictionaries can be linked for additional term searches inside

the tool.
• The alignment function helps you build translation memories by aligning
source and target texts that were translated previously outside the tool.
• Terminology databases and term extraction tools.
• Word counts and match statistics, the basis of quotes for translation services.
• File conversion processes many different formats.
• Predictive typing as in smartphones.
• Quality assurance, which is an error-spotting function during the entire trans-
lation process (Chapter 5).

Most CAT tools support MS Office formats, such as .docx, .odt, .csv, .xlsx, plus
.html files and .xml. The translation industry has introduced the XLIFF file
(interchangeable CAT tool format) and TMX file (interchangeable format for
databases) which improve compatibility and homogeneity. Advanced CAT tools
also support software formats, such as .json, which is a language independent
format,Visual Studio and software for layout and building plans, such as InDesign,
Corel Draw and sometimes AutoCAD. The usefulness of CAT tools cannot be
denied when translating multimedia formats. They extract text for translation,
rebuild and reformat the file in the target language after the translation is finished.
The translator works in the standard bilingual CAT tool text display and is not
distracted by metadata or, for example, the autocompletion function in Excel (2.4
Formats). The spellchecker is automatically operable in the CAT tool, contrary to
Excel and other formats.

1.6.2 Standalone and web-based CAT tools

CAT tools are available for online or offline usage. Web-based (cloud-based)
CAT tools are accessed online and require a licence, which is often supplied
by the commissioning LSP for the duration of the project; standalone desktop
CAT tools must be purchased and the programs can be downloaded. Compared
with web-based CAT tools, they have the advantage that they do not depend
on internet connection stability. Although CAT tools were originally designed
to work in standalone mode, a server version has cloud-computing features. The
server is used in environments where collaboration is required, for example in
training centres or project management teams. LSPs may switch from servers
to comprehensive translation management systems (TMS). They automate the
translation process and give the project manager an overview of operations
ranging from translation projects to accounting, marketing, buying and selling
translation services (7.5)**(link: web-based and TMS systems). The advantages
and disadvantages of working with web-based tools are discussed in more
detail in 7.2.
14 Computer-aided translation tools

A new type of web-based CAT tool entered was launched in 2019, the smart-
phone or tablet version**(link: web-based and standalone parity). Obviously, the
phone version cannot show all the functions, but it works well as a backup for the
translator on the road who is asked to make quick changes or add a paragraph.
The developers of CAT tools are constantly bringing changes with modified or
new features. The similarity between the main tools has increased greatly, and so
has their compatibility. Many LSPs prefer translators to work on their servers and
if this trend continues it will be interesting to see if it will at all be necessary to
invest in standalone tools.

1.6.3 CAT tools and cognitive ergonomics

Despite all the improvements in TM software programs, there appears to be little
consideration for cognitive ergonomics, in other words, the well-being of the
user of TEnTs. The International Ergonomics Association (IEA) defines cogni-
tive ergonomics as ‘mental processes, such as perception, memory, reasoning and
motor response [i.e. physical movements], which can affect interactions among
humans and other elements of a system’ [i.e. human-computer interaction]. Surveys
carried out by researchers show that translators are concerned about the constraints
imposed on them by translation technologies (Ehrensberger-Dow and Massey
2014). These constraints have arisen due to technological developments and range
from lower rates paid for fuzzy and exact matches, added time pressure because
CAT tools surely support faster translations, and the translator’s irritation with tag
handling in highly formatted files.
In practice the tools we use should lighten our load, cognitively and phys-
ically. The benefits of computer-aided translation mean that we do not have to
search files for terminology, that windows and dialog boxes can be opened sim-
ultaneously on our screen, that HTML files and Excel sheets are converted in the
CAT tool and presented as an editable bilingual segmented page, and that files
for editing, generated by other translators, have a standardised format in the CAT
tool. However, the regular upgrades, new designs, features, and functions are often
found to be counter-intuitive by the user who must process and remember them
(Ehrensberger-Dow and Massey 2014).
There are reasons for some translators not to adopt CAT tools. They find that
multiple screens and functions, the requirement to manage external terminology
tools, dictionaries, and to integrate machine translation in the tool, make them
feel overloaded. The industry’s mantra, ‘speed, quality, and efficiency’, can be felt
as a curse, because if the use of a CAT tool means tighter deadlines, the tools are
seen to become counter-productive. It cannot be denied that the advancement of
translation technology is a controversial topic if we consider how TEnTs affect the
translator. Some tools may suit the LSP but will the ensuing work practice please
the translator?
 Computer-aided translation tools 15

Food for thought…

What advantages and disadvantages does the CAT tool bring to the translator?

1.7 Translation project management

LSPs used to be called ‘translation agencies’.Today they go under different titles; we
will use the LSP acronym. CAT tools play a central role in translation project man-
agement: they control the workflow, they help coping with volume, they support
the mantra of quality, efficiency, and profit, be it sometimes at the expense of the
translator’s sense of well-being.

1.7.1 Volume and demands

Since the Millennium, the amount of required translation has increased enor-
mously. Undoubtedly, the internet is the driver. Global communications gen-
erate masses of text for different purposes in an unprecedented range of formats.
Translation practices have changed accordingly. By 2011, translation buyers already
outsourced 90% of their translation work (Dunne 2011). Even organisations such
as the European Union cannot manage the vast quantities in-house and will out-
source to freelancers or LSPs. New formats have emerged, many new TEnTs have
been developed, expectations (speed and pricing) have changed and the translator
must manage the changes, or at least cope. Dunne (2011) makes the following
observations about change:
16 Computer-aided translation tools

• clients outsource because translation is not one of their competencies.

• large translation projects need more revision, terminology management,
localisation.
• book translations and manuals with tens of thousands of words are often shared
among several freelance translators.
• collaboration can be arranged through CAT tools which enables translator to
share TMs and terminology databases.
• real-time collaboration in the cloud is possible on servers or in translation
management systems, often managed by LSPs (7.2; 7.5).

Translations have become projects and translation is only one component in the
workflow. The translation project requires management of communications and
accounts in addition to the actual translation process. Sheer volume demands a
different approach to how we translate and deal with demands.

1.7.2 Translation workflow

The translation project consists of many stages, ranging from job offer, translation
to revision, and delivery. A simple workflow chart (Figure 1.4) shows the rela-
tionship between client, translator, and LSP. The chart can be extended: the trans-
lator may involve additional parties by sending queries to colleagues or posting
questions on mailing lists or forums. Even if an LSP is involved, a translator may
need to consult third parties if the dedicated project manager does not know
the source or target language. The client can often only explain, which leaves
the translator dependent on other sources to find equivalent terms in the target
language.
An LSP may have a team of managers with different roles under a senior pro-
ject manager. The number of stages and substages are infinite: the first stage will
most likely require briefing, and negotiating about time and remuneration between

Translator

Client LSP
Project
Managers

Proofreader

FIGURE 1.4 A basic translation workflow chart**(link: workflow diagrams)

 Computer-aided translation tools 17

source
text

target
translator
text
project
manager

revised
reviser
text

FIGURE 1.5 Translator’s workflow

client and translator. The LSP must source translators (and other linguists, such as
revisers). A purchase order (PO), which is the first official document with details
about type, word count, deadline, and agreed price for the translation service, needs
to be sent to the translator to complete the first stage.Translation is the second stage
followed by the third stage of revision; the final stage in the workflow includes
delivery and invoices from translators to the LSP, and from the LSP to the client. If
the contracted translator uses a CAT tool, stage 2 consists of:

receipt of file > preparation > ST import > translation > self-revision > return
to LSP > third-party revision > return to LSP > return to translator > export
of XLIFF file or generation of target text > return to LSP.

During the process additional nodes of interaction are generated through contact
with the project manager at different points such as delivery, return of revised trans-
lation, review by the translator and return to the LSP project manager (Figure 1.5).
If the translator translates in a web-based CAT tool, controlled by LSP, it will
produce a smoother cycle because the entire workflow is in the cloud (Figure 1.6).

1.7.3 The translator in the workflow

Complex translation processes require good project management. Translation
requests, additional paragraphs and modifications to documents need a prompt
response. Freelancers are their own project managers. Their challenge is to pre-
vent unreasonable time reduction: their productivity can be driven to unreasonable
volumes due to inflated expectations linked to technological advancement. As a result
of TEnTs, translators may be given ‘less time’ instead of the promised ‘more time’.
18 Computer-aided translation tools

client

LSP LSP

reviser translator

LSP

FIGURE 1.6 LSP’s workflow in the cloud

The time allowance for the following translation request is three hours from receipt
of email to delivery of the translation.The subject is technical (cosmetics) and requires
research. Note that self-revision is more effective the following day than immediately
after completion:

Good morning,

I hope you are well today. We are currently looking for a translator to help with
a translation for our Cosmetics client.
The text is about a new range of summer products for 2020; there are 414
words. Please take note of the source file in which you can also see the product
names for reference.
Would you have capacity to translate this by 2pm UK time today?
Email from LSP (2019)

Translation work has generally become more complex with less remuner-
ation (see 2.1.3) Good translation management is crucial for well-being and
satisfaction. It needs to be more comprehensive than the delivery of quality
translations.

1.7.4 Project management by the language service provider

Language service providers (LSPs) are well equipped to manage multilingual
translation projects. LSP project managers do not generally translate, they manage
projects and contract external translators, although they may also have in-house
translators. They manage sales, human resources such as vendors (linguists who
sell their services to the LSP), and marketing. They recruit vendors and will
ask for evidence of language pair competence, domain knowledge (fields of
 Computer-aided translation tools 19

specialism), tool use, and the types of service(s) they offer such as translation,
proofreading, post-edit of machine translation, desktop publishing, transcription.
Project management can be mapped into five processes (Dunne and
Dunne 2011):

• initiation
• planning
• execution
• monitoring
• signing-off

The processes contain nine knowledge areas:

• management of project integration

• scope
• time
• cost and pricing structure (2.1.3)**(link: word rates)
• quality
• human resources
• communications
• risk (6.6)
• procurement

A shorter formulation of the project manager’s responsibilities is as follows (Chan

2015:63):

• plan, execute, and close projects

• set achievable objectives
• build the project requirements
• manage cost, time, and scope of projects

Increasingly, LSPs automate the process and use either CAT tool servers or web-
based translation management systems.

1.7.5 Project-based assignments and project management

Each chapter ends with a project-based assignment for which you will need to
set up a project management team. We will now explain the construct. An LSP
operates through a project management team. Its members manage others: it
recruits, requests, monitors, and contracts linguists. In this book, you will be asked
to build your own project management teams to fully appreciate the benefits of
technology and experience the challenges of collaboration and teamwork. If in
class your circumstance do not allow you to recruit, the ‘project management
20 Computer-aided translation tools

team’ can be modified (reduced) to a ‘project team’ in which the members per-
form tasks themselves.Very often project team members then become project man-
agers because they need third parties to assist them in their roles and functions.
Thus, the project team may evolve into a project management team. There may
be circumstances when teamwork is not at all possible in class: the components in
the translation project can then be performed from the perspective of a contracted
freelance translator who collaborates with others.
In the real world, a project management team will ideally consist of linguists
with different native languages and language pairs. Team members have different
functions and can therefore be assigned different tasks. The teams usually have
one senior project manager, responsible for leading the team, ensuring that
the members have everything they need to complete their tasks to achieve
goals and objectives. The project management team will have any number of
project managers to manage the tasks through contracted external linguists
(freelancers).
The project-based assignment will always have a set time frame in which
members fulfil their roles. After completion the team is dissolved. Teams can recon-
vene, be modified, or reconfigure for new project-based assignments. If the team
works well, it can stay together. For the duration of the project the team members
work towards a set common goal of delivering the project within time and budget
constraints and according to agreed standards. Third parties are the stakeholders,
such as clients and service providers. If the service providers are translators, they
are contractees. The project managers may well spend much time overseeing and
managing their team of service providers. Completion and quality checks of the
translation should be performed by team members, such as revisers, quality assurers.
Project managers within a team are accountable to the senior project manager
and need to communicate effectively with their colleague project managers and
stakeholders to deliver a successful project.
Although there is a distinct difference in the distribution of tasks performed by
project teams or project management teams, hybrid project (management) teams
are quite acceptable.

1.7.6 Human resources database

A database is any collection of data, or information that is specially organised for
searches and retrieval by a computer. Human resources (HR) is generally under-
stood to be personnel. An LSP will undoubtedly have an invaluable HR database
consisting of linguists, the contractees, who carry out or revise translations. There
may even be a team member whose job it is to recruit linguists and manage the
database. The process of recruitment and human quality control needs a set of cri-
teria agreed by the team, based on qualifications and records of experience, lan-
guage expertise, and specialist fields so that freelancers can be vetted before they
are contracted.
 Computer-aided translation tools 21

1.8 Cognitive friction

Before we move to the project-based assignment where we will experience man-
aging a translation project, we should return to the translator and their rela-
tionship with technological translation tools. We have seen how the well-being
of the user can be affected by TEnTs. In addition to cognitive ergonomics
(1.6.3), the user may also experience ‘cognitive friction’. This is the ‘resistance
encountered by a human intellect when it engages with a complex system of rules
that change as the problem changes’ (Cooper 2004). This kind of conflict occurs
not only between the user and their tools, but also on another level, for example,
when the freelancer becomes part of a translation project and collaborates with
other linguists. Translators often disagree with, for example, the revision of their
translations, translators often argue about appropriate target terms, translators
protest against reduced rates of pay. The list can be continued. The translation is
a multilevel system (Ehrenberger-Dow and Massey 2014) with rules that play at
different levels:

• the rules that matter to the client are standard, style, price, and speed;
• the project management team is faced with hierarchical rules of seniority in
the team, which also affect external translators and revisers;
• the translator requires fair pay, sufficient time, helpful resources, and support.

Speed and rates are common causes of friction because the different parties do
not share the same objectives: the translator generally asks for more time than the
client/LSP is willing to give, and the same applies to rates.The contracted translator
may experience stress that they will lose their contracts with the LSP if they do not
abide by the rules.
Cognitive friction also exists on a technological level. What is the impact of
TEnTs on the translator?

• upgrades and new versions of their own CAT tools have a negative impact
on time
• translating in the LSPs web-based tools challenges routine and habit
• financial depreciation in relation to the pricing of TM matches
• reduced word fees for post-edited machine translation (PEMT)
• prohibited use of integrated MT in a CAT tool in relation to potential breaches
of confidentiality (6.2)

There has always been cognitive friction, but its nature has changed since the arrival
of TEnTs. LSPs find themselves in a difficult position between the translation buyer
and the translation vendor where the parties have a different understanding of time,
efficiency and quality. Good LSPs will try to satisfy the needs and wishes of either
party. Project managers must be skilled and diplomatic negotiators.
22 Computer-aided translation tools

Project-based assignment

Objective:
A hands-on experience of managing translation technology and human
resources in a translation project.

Assessment and outcome:

Peer/self-assessment of product (translation), process and performance. The out-
come should be the ability to manage the workflow in further assignments
with expertise and understanding, using different or additional tools.

Method:
The assignment is designed for teamwork, but collaboration between individ-
uals is possible.

Tools:
CAT-tool (a server version is not necessary at this stage)

Suggested resources:
Digital source text (for example a brochure for museums, cars, domestic
products, services, hotels and catering, promotional university websites,
health and safety in organisations, non-government organisations).
Consult your instructor to check if the text is suitable for a collaborative
translation in a CAT tool.

Language pairs and direction:

Multiple language pairs; both directions, if TM is reversable

Assignment brief
In this project-based assignment we will try our hands at managing a translation
project. The LSP team consists of different roles and the project offers different
 Computer-aided translation tools 23

perspectives: the project requires project managers (PM) and translators. The LSP
manages a translation project and contractees, but does not translate. The translator
managed by the LSP will be asked to translate or to revise.TEnTs play an important
part. The brief is as follows:
The client requests the translation of a brochure into multiple languages. The
LSP is asked for a quote and best turnaround time.The brochure has an approximate
word count of 3000+ words. Consider this request as an LSP and work through the
following stages as a team:

• Set up a project management team (or a project team if human resources

(contractees) are limited (see 1.7.4.1)), find a suitable source text and check
the availability of your contractees.The timescale needs to be decided with the
instructor.
• Draw up a quote for the client (samples of quotes, purchase orders, and invoices
can be found on the internet).
• In the team, decide on the steps from quote to delivery and design a workflow.
• Compose an email with a job offer and send to freelance translators in your
database (recruitment 1.7.4.2). Send the email stating the word count and the
deadline, with a sample of the source text in an attachment.
• Draft a purchase order (PO) and email it to the translators that have accepted
the offer. This email is a formal notification to the translator to proceed and
start the translation.
• Request contractees to return self-revised translations in two formats: clean
target file and an XLIFF work file (more details in 2.4).
• Self-assess your task as a project manager by scoring according to agreed cri-
teria, or in line with component(s) in the workflow diagram, or in relation to
the responsibilities suggested in 1.7.4. Discuss areas for improvement.

Possible pitfalls
• Unsuitable material for translation. The TM in CAT tools is effective when
there is repetitive terminology. Consult your instructor.
• Shortage of translators
• Role overlap
• Formats: it is advisable to use uncomplicated formats in this assignment, such as
docx. Other formats often require tags, which are discussed in the next chapter
(2.1.1 and 2.4).

Concluding remarks
This chapter has introduced you to a range of translation environment tools with
an overview of the the CAT tool, the TM, its features and functions. We have
considered the benefits and challenges of the CAT tool from the translator’s and
24 Computer-aided translation tools

the client’s or LSP’s perspectives.We have considered the similarities of ribbons and
commands in CAT tools and MS Office, and their usefulness. We have seen the
importance of good file management in respect to the CAT tool. We have learnt
how to manage translation projects that involve multiple target languages, managed
by LSPs who collaborate with the client, contracted freelance translators, take care
of accounts and quality standards in translations. We have looked at the different
roles in a translation project management team with project managers and in-house
or contracted translators and other linguists. You have been given an opportunity
to organise the setup of a project management team and to determine the roles
of its team members, create workflow charts and operate features and functions in
the CAT tool. All these activities help us reflect on not only our different roles and
contribution to the outcome and quality of translations, but also on the use of CAT
tools in the translation process.

Further reading
Dunne, Keiran J. and Elena S. Dunne (eds) (2011). Translation and Localization Project
Management. Amsterdam and Philadelphia: John Benjamins.
Ehrensberger-Dow, Maureen and Sharon O’Brien (2014). ‘Ergonomics of the translation
workplace: Potential for cognitive friction’. In: Deborah A. Folaron, Gregory M. Shreve,
and Ricardo Muñoz Martín (eds), Translation Spaces, 4(1): 98–118. Amsterdam and
Philadelphia: John Benjamins.
Melby, Alan K. and Sue Ellen Wright (2015). ‘Translation memory’. In: Chan Sin-Wai
(ed.), The Routledge Encyclopedia of Translation Technology, pp. 662–77. London and
New York: Routledge.
Walker, Andy (2014). SDL Trados Studio – A Practical Guide. PACKT Publishing.
Wright, Sue Ellen (2015). ‘Language codes and language tags’. In: Chan Sin-Wai
(ed.), The Routledge Encyclopedia of Translation Technology, pp. 536–49. London and
New York: Routledge.
2
THE TRANSLATION MEMORY
DATABASE

Key concepts
• The translation memory database can be filled to suit our individual needs
• The translation memory database accepts several methods to boost its content
• The translation memory database can be customised to suit our language
pairs
• The translation memory database includes features to improve translation
quality and productivity
• The translation memory database can affect profit margins

Introduction
‘Translation memory’ means TM database. In the translation industry the terms
‘TM’ and ‘CAT tool’ are often interchangeable: TM meaning CAT tool or vice
versa. We will try to avoid this confusion. The TM is a database consisting of trans-
lation units (TUs), which we have entered over time.We can increase the size of our
TM by importing TMX files (interchangeable TMs) from colleagues or LSPs for
our use or reference purposes (2.3.3). A TM needs to be managed intelligently to
be effective. The TM is more than a useful storage system. Regular edits and good
maintenance make the TM a resource that is customised to our own needs and keep
it up to date. The TM is the primary database in the CAT tool.
The TM’s special feature is that it automatically stores TUs when we confirm
a translated segment, whereas the terminology database (Tmdb) must be built
manually: we must highlight and add the terminology pairs we want to store.
The Tmdb is integrated in CAT tools. One CAT program prefers an external
Tmdb, because it can then also be used outside the CAT tool. In the tool it can
26 The translation memory database

be linked to the translation project and integrated like any internal Tmdb**(link:
External Tmdb).
The TM in a new CAT program arrives empty and it can take quite a while to
fill, depending on the amount of translation we do. If you are a student who uses
a public PC, the TM will be cleared each time the PC is shut down. Therefore, we
must be careful to export our TMs as interchangeable TMX files and save them in
our personal accounts for importation when we open a public CAT tool (2.3.3). If
we want to fill a TM before we start a new project, we can import resources (2.3.1)
or TMX files (external TMs).
In this chapter we will explore how we can best use and boost the TM to
suit our needs, even in the short term, and make it a superb tool to increase and
enhance quality, efficiency, and potential profit. Although translators and LSPs share
the objective of better quality and more productivity through the TM, translators
often cross their swords with LSPs about word rates when costing a project (2.1.3).
Manufacturers constantly update their programs, adding new features. The cloud
plays a crucial part in CAT tool development; it gives access to infinite new oppor-
tunities. In the project management assignment, you can experience TM features
and view them from different perspectives, through the eyes of the LSP and the
translator.

2.1 Creating a translation memory database

When the TM recognises an identical or almost identical source segment, the
target results or matches, appear in the Translation Results window. When the
TM performs a lookup, it compares the content in the source segment you are
about to translate against segments in the TM. It looks for source segments that
have enough content in common. It then presents the translation units (TU)
that it has found. The identical degree of the match between the source segment
and a TM segment is expressed in percentages. For example, if the content of a
translation memory segment matches the document segment exactly, it is a 100%
or perfect match. If a new segment and a translation memory segment match
precisely, including tags, numbers, punctuation, etc., it is a context match. Some
CAT tools identify a context match as a 101% match. A less than 100% match is
a fuzzy match. The default fuzzy match threshold generally is 70%. This cut-off
point is known as the ‘fuzzy match threshold’: there is insufficient usable con-
tent below this level.
You can accept or reject proposed matches in the Translation Results window,
edit fuzzy matches in the target segment after you have accepted them, or translate
from scratch in the target segment because no match was found.Your confirmed
translated segments are sent to the TM, which is updated when you confirm revised
TUs.Your entire TM can be edited at any time.
In the interchangeable TMX format, a TM can be shared with other translators.
It is a useful procedure to maintain consistency if several translators are working on
the same translation project. It flags up matches (in the Translation Results window)
 The translation memory database 27

each time it identifies a previously translated unit. A shared TM becomes more

effective when all the project translators are linked to a central server, because the
TUs of all translators are then confirmed in real time. A shared TMX offline can
only serve as an up-to-date reference at the time it is received and imported. LSPs
therefore use CAT systems with servers or translation management systems (TMS)
to maintain real-time consistency in translation projects with multiple translators
(7.2, 7.5).
We are not limited to the creation of only one TM in our CAT tool. In fact, it
is advisable to create several TMs, customised according to domain, subject field, or
client. A large TM may present too many matches, and it will need more mainten-
ance and editing. It is therefore advisable to use several smaller and customised TMs.
If we have one large comprehensive TM in addition to smaller TMs, it can be set to
read-only mode in the Translation Results window. The TM is complemented by
the terminology database (Tmdb). The TM stores whole segments and the longer
the string, the less likely the TM will find a perfect match, whereas term pairs are
concise and are more likely to generate matches.
The CAT tool has a pre-translation feature, which relies entirely on the
TM. We can ask the CAT tool to pre-translate a document before we start
translating. It looks up every segment in the TM and inserts the best match. If
there is no match, the program can use a fragment-assembled match or machine
translation (depending on our setting). It can also try to join or split segments
automatically if it results in better matches. Pre-translate is a powerful feature,
but it generally requires revision of the proposed TUs. Unless set otherwise,
the TM will propose matches in the Translation Results window. If we enable
autosuggest, the TM will guess while we are typing and we can accept its
suggestions on the fly.
Regardless of the quantities we store in our TMs, it is quality that matters.
Good work practice means regular maintenance. We can do so in the TM editor
where we use filters to remove old entries, duplicates, ‘blacklisted’ terms. In the
TM editor our eyes spot-check errors. False positives, for example, are TM
matches that are not suitable in context. Other important features are the repair
resource which re-indexes entries and the reverse lookup that reverses the lan-
guage direction in the TM for use in both directions. Reversed matches should
not be used indiscriminately: source segments in reverse generate a different set
of synonyms in the target language. For example, ‘express’ has different meanings
as a noun (fast train) and as a verb (to squeeze). If you translate the respective
target terms back into English, depending on the context, they may not bring
you back to ‘express’.
The TM can be a time-saver and ensure consistency in large files.The conditions
are that the source text is consistent and error-free, external TMs are maintained
and supplied glossaries are reliable (4.1.1, 4.1.4).We must be prepared to spend time
editing proposed matches to guarantee the quality of our own TMs. If an unsuitable
match is accepted, it will be propagated until we reject it.The TM is as good as we
make it, control it and maintain it.
28 The translation memory database

Food for thought…

What might be the advantages and disadvantages of using the Pre-translate

feature?

2.1.1 Segmentation
The CAT tool splits a source document into segments so that the TM can store
the TUs systematically for recall. File preparation, after importing a document
in the CAT tool, is an automated procedure that happens almost unnoticeably
in the CAT tool. ‘Project preparation’ is the process that is clearly displayed in
one of the CAT tools (Figure 2.1). The first task, ‘Convert to Translatable Format’
(Figure 2.1), is the segmentation process and converts the source text to a
segmented bilingual CAT tool format. By default, the segment boundaries are
spaces or punctuation marks, which works well in word-based languages such
as English where a full stop signifies the end of a sentence. The segmentation is
defined by rules specific to each source language and may vary between different
CAT tool programs. We can to some extent edit the rules to suit our needs, but
they are designed to support match searches and our intervention could prevent
recall. The inclusion or exclusion of one word in a new string will not generate
a perfect match. In non-whitespace languages (character-based languages in East
and South East Asia) segmentation has been traditionally defined by characters.
CAT tools are now enabling a choice so that words instead of characters can
be ticked as the basic unit for segmentation in Chinese and Japanese source
languages.
 The translation memory database 29

FIGURE 2.1 File preparation dialog in SDL Trados Studio

A smooth operation of the TM relies on segmentation. Researchers claim that

our eye movements are affected and constrained by segmentation in the source
language boxes. Pym (2008) and Dragsted (2005) argue that the translator only
sees what is in the box without realising syntactic or semantic relationships with
the preceding or following units or sentence(s) which leads to non-contextual
translation.
Manufacturers have done their best to address the perceived limitations of seg-
mentation. Firstly, the translator can join and split segments (depending on the
file type and the segmentation rules in the CAT program). Joining segments can
be helpful if cohesion is important within and between sentences, or if segmen-
tation prevents the insertion of a suitable term in the target segment. However,
when we join and thus extend SL segments, the TM may have more difficulty
finding matches because of longer strings. Another drawback is that splits or joins
insert (inconvenient) tags to maintain the ST layout (2.4). Secondly, the preview
feature has been introduced in many CAT tools to compensate for segmentation
and the suggested lack of context. The preview window allows the translator to
see their translations on the fly in monolingual format instead of having to wait
for the final export. Thirdly, it is possible for the translator to review and revise
the exported target text outside the CAT tool and then re-import the revised
monolingual target text to update the TM. Revision of a smooth unsegmented
monolingual text, which is kinder to the eye and the brain, can give better revi-
sion results. Errors are sometimes picked up in the generated translation that
30 The translation memory database

have been missed during a bilingual review in the CAT tool editor, the interface
with grids containing the segmented ST and TL units.
Despite improvements and new developments in the tools to help us deal with
segmentation, opinions on the ultimate translation quality are divided. Pym (2008)
claims that TM software may help maintain terminological consistency, but it requires
too much management to bring about great productivity gains. Meanwhile, LSPs
are great advocates of the TM. Their priority is to meet the client’s requirements
which is made possible through a shared TM in large projects. Furthermore, seg-
mentation structures source texts, improves repetition rates, guarantees consistency,
and leads to cost reduction (2.1.3).
Ultimately the translator is the decision-maker regarding acceptance or modi-
fication of segmentation rules. We need to be aware of syntactic differences in
relation to our language pairs so that we can change segmentation rules to suit
our needs (2.5.2). We can also control segmentation rules by being critical of the
way in which source texts are formatted before we import them. Editable source
texts can be formatted or reformatted by the translator in MS Word. For example,
we can use the formatting features in MS Word, rather than spaces, tabs, and hard
returns or line breaks. A reduction of these special characters prior to importing
the file will greatly reduce segmentation and reduce the number of format tags
(2.4). The translator should prioritise precision over recall, favour quality over
productivity (Bowker 2005). Precision refers not only to content, but also to
source formatting.

2.1.2 The concordance and consistency

The concordance feature makes it possible to look up specific words or word
sequences in our TMs. The dialog box, or concordance search window, presents all
the entries, including the surrounding words, in our TM(s). The purpose of a con-
cordance is to show how a term or phrase is used in context (see 4.4). It also helps
the translator retrieve a TU that is not shown in the Translation Results box, a ‘no
match’. This is helpful when the term or phrase was previously entered within a
string that is not identical and therefore will not even present a fuzzy match. Other
explanations for ‘no matches’ could be another layout and therefore a different set
of tags, an extra space, missing punctuation or other numbers. Seemingly minor
differences prevent matches.
In Figure 2.2 we see the concordance search feature on the CAT ribbon. One
click opens the dialog box with a choice between source and target searches
in the concordance. Note that there are also keyboard shortcut codes to the
concordance. The CAT concordance is bilingual and shows source and target
segments.
The highlighted results of a target concordance search of the term ‘functional’
are displayed in Figure 2.3. The translator can compare the five results and match
them with the source segments to determine why the second entry is ‘functional
 The translation memory database 31

FIGURE 2.2 Dialog box opened in concordance search in SDL Trados Studio 2019

FIGURE 2.3 Concordance target results for the term ‘functional’

dairy’ without ‘products’. The concordance contributes to consistency. As the SL

segments are identical in all matches, consistency would require the translator to
select ‘functional dairy products’ in the TL.
The CAT concordance is a popular feature. Translators are known to rely on
the concordance to find term translations in the TM instead of building a Tmdb.
The additional effort to add term pairs to the Tmdb, may be a deterrent, but the
reward is that Tmdb matches are presented together with the TM matches in the
Translation Results window.
A quick way to check for consistency is to highlight a term in either source or
target segments and press Ctrl+F (Figure 2.4).The dialog box then pops up in View
on the ribbon. Figure 2.5 shows that Search and Find is performed on the term
‘limited’. By pressing the enter key or clicking ‘Find Next’ the translator can search
for all the occurrences of, for example, the term ‘limited’ in the target segments of
the active file or project files.
The search function and use of the concordance are popular shortcuts to
check consistency. Good work practice requires consistency checks in the quality
assurance (QA) feature in the Review tab on the CAT ribbon. The QA feature
flags up inconsistencies and we will discuss this in more detail in 5.2.1. The bilin-
gual concordance is not designed to achieve consistency. It presents matches in
context to focus on terminology usage. It can however support consistency and
32 The translation memory database

FIGURE 2.4 Search and find dialog in SDL Studio 2019

FIGURE 2.5 Search and Find function Ctrl+F

is often used for that purpose. In Figure 2.6, the concordance dialog highlights
four search results for ‘supplied’ in the source language but there are only two
identical matches in the target language. Adjectival endings in the target language
prevent matches. However, if you highlight the stem of the word in the concord-
ance search, it will show results with and without the adjectival ending. A more
 The translation memory database 33

FIGURE 2.6 Concordance dialog in memoQ 9.1

professional search method is described in 4.5. Morphological changes are not

understood by the TM. This problem can be addressed in the Tmdb by adding
metadata (4.1.3).
The consistency of terms and phrases is important in technical documents.
It is important when the source document is split and shared between several
translators, each with their individual translation style. The TM can ensure that the
client will receive a document with consistent terminology, and translators benefit
greatly from consistency alerts or match proposals in large or updated documents.
In a shared project alerts and consistent match proposals are only possible if all
translators are working on the project in the cloud. Even then the project man-
ager must do a final consistency check and smooth over potential disagreements
between translators.

2.1.3 The analysis feature and fees

Translators run the analysis/statistics feature before they start to check the word
count. They will see the amount of repetition which should speed up the job,
because the TM will propagate perfect matches. LSPs use the feature to cost the
34 The translation memory database

translation project (Figure 2.7). They expect a translator to offer a discount for
repetitions, 75% discount for 90–100% matches, 50% discount for 70–90% fuzzy
matches and no discount for (fuzzy) matches below 70%. Alternatively, the LSP sets
up a Purchase Order (PO) and chooses a different breakdown against their TM and
it is up to the translator to accept or reject the translation job. Some translators will
reject discounts altogether. Their arguments may be that they purchased an expen-
sive CAT tool for their benefit and that even 100% matches need to be checked in
context. They risk not being offered the job.
The analysis gives word and character counts and other statistical data, such
as the number of new words and repetitions in a file, before or after translation.
In Figure 2.7 the ribbon at the top gives the name of the project, file(s), active
document, or part of the document from the cursor, and counts. It gives a status
report of what has been translated and the amount of repetition in the ST. It
also gives the option to calculate the word count in another program, as CAT
tools vary. The variance is caused by the method chosen to identify and count
numbers or tags, and different segmentation rules. The analysis in the lower half
of Figure 2.7 shows that there is repetition in 15 segments and in 37 source
words. A calculation in the other tool showed the same number of segments
but less repetition, a lower source word count and fewer words with no match
(Table 2.1).
There are also language-specific variances in pricing. For example, 1000 Chinese
characters will usually be translated into about 600–700 English words; 1000 English
words will be translated into about 1500–1700 Chinese characters (depending on
the nature of the text). A Dutch target text generally scores a 15% higher word
count than the English source text, and this variance within language pairs is not
uncommon. It is due to a different syntax. In other languages a letter or character
count per line may offer a more reliable fee basis. Fees are generally based on the
source text because these counts are available at the initial negotiation stage. Ideally
fees should be set per language direction.
In German, a language that joins many noun compounds into single long words,
fees tend to be based on standard lines consisting of 55 characters including spaces
between words. Fees can also be based on hourly rates, page rates, and flat fees
(minimum fees).
The analysis feature in the CAT tool is convenient for costing, both for translators
and LSPs.When CAT tools first entered the market and introduced match matrixes
in the analysis feature, the agencies began to ask contracted translators for reduced
rates for repetitions. Many translators felt let down because they saw their increased
profit margins fall below the level prior to CAT tools. Translators argued that fuzzy
matches needed editing and that revision included a critical review of context
matches and therefore there should not be any discounts. One of the manufacturers’
selling points was that CAT tools would increase productivity and increase profit.
The question many translators ask is ‘whose profit’.
 newgenrtpdf
The translation memory database 35
FIGURE 2.7 Analysis report (statistics tab in memoQ 9.1)
36 The translation memory database

TABLE 2.1 MemoQ (M) and TRADOS 2007-like (T) counts compared in MemoQ 9.2
Statistics

Segments Source words

M T M T

All 281 281 3553 3532

Repetition 15 15 37 35
Translator confirmed 281 281 3553 3532
No match 266 266 3516 3497

Food for thought…

What are arguments to defend a 100% payment of fees for perfect and fuzzy
matches?

2.2 Metadata and subsegment matching

It is important to add metadata, in other words data about data, which can be
done in the Tmdb but not in the TM that only accepts metadata during the
project setup. When we create a new project and add or create the TM, we can
add information such as date and name, domain, client, and deadline. These data
are visible in dialog boxes while we translate in the translation editor. They are
important if several TMs are open. Several and different matches may appear
 The translation memory database 37

in translation results. We can see in which TM, associated with which client
or domain, they occur. Displayed dates are significant too and may remind us
that it is time to clean a TM if obsolete terms present themselves. Although
we cannot add metadata to the TM database on the fly, we can use the Tmdb
for this purpose. It may be useful to know the part of speech (POS), gender
or number, so that we can account for declensions and inflections. The TM
searches in authentic strings, while the Tmdb searches term pairs or phrases,
which we can modify, extend, make plural, etc. Fuzzy matches are only generated
if there is a high level of similarity. Although it is possible to lower the fuzzy
match percentage in the settings, it is not recommended. The TM will propose
many unhelpful suggestions. If translators find the TM’s recall disappointing, it is
because matches are part of longer strings in segments. In this respect, the recall
by Tmdb is superior (Flanagan 2014).
Software developers are working hard to develop subsegment matching in the
TM to improve match results.TAUS, the Translation Automation User Society**(go
to www.routledgetranslationstudiesportal.com/ – A Project-Based Approach to
Translation Technology – link: TAUS), was set up in the Netherlands in 2009.
The association consists of major translation buyers and other organisations and
companies interested in the advance of translation technology, especially machine
translation. It offers webinars to educate its members and other stakeholders, and
conferences to generate networking possibilities. TAUS has developed a frame-
work to test (machine) translated data and to collect and disseminate data among
members (Zetzsche 2019 (297th Journal)).
TAUS has launched a new product called Matching Data. Its developers recog-
nise that adding metadata to the TM could in fact be counterproductive if metadata
cause TUs to be locked into one domain. Their solution to inadequate matching
lies in receiving huge quantities of parallel language data from different sources and
owners, which are then transformed into unique corpora, i.e. bodies of text, such as
glossaries.They are not only domain-specific (e.g. associated with a client’s product,
field or science) but are also customised to individual search requirements. In
Matching Data, the matches are based on granular subsegments, which means they
focus on much smaller units, with more lexical or morphemic detail and descrip-
tion. TM manufacturers have adopted this approach. Many TMs can now search at
subsegment level in the TU to give a partial perfect match, rather than a fuzzy match
which needs post-editing. This kind of subsegment leverage (Flanagan 2014) has
been given different names by different CAT tool manufacturers: ‘Deepminer’,
‘Uplift’, or ‘Longest Substring Concordance’. The TM looks for a matching string
of words within another TU and that substring is not affected by any surrounding
words which would previously have prevented a match.
The identification process is straightforward, but CAT programs operate dif-
ferently when their respective TMs reassemble or auto-assemble new and longer
strings, and results vary. ‘Uplift’ which matches subsegments in the TM works better
if the TM is smaller and more specific.The program claims that Uplift saves us using
their external terminology database. This might be true if the searched terms are
38 The translation memory database

stored in the TM, but if they are new or stored in external resources the suggested
match may be less perfect without access to terminology and reference features.
CAT tool developers are trying to remedy this short coming, and some CAT tools
give the translator more control over resources that can be used in addition to
the TM. ‘Deepminer’ was the first form of subsegmentation. The TM ‘mines’ for
subsegments and uses statistical data to analyse TM content (DTA=Dynamic TM
Analysis). DTA was called ‘Guess Translation’ in another CAT tool. What all CAT
tools do is perform a concordance search, i.e. find subsegment pairs in source and
target strings and present the subsegments combined with the non-matching word
strings in the Translation Results window. This procedure allows the translator to
select the target subsegment, insert it, and translate the remainder of the string. The
feature Auto-assemble was also a product of data analysis. It was a laudable attempt
to create good matches, but the edit distance was high. Autosuggest is not unlike
predictive writing in smartphones: it recalls potential matches in the TM, triggered
by the translator’s keystrokes. Subsegmentation in CAT tools is work in progress.
CAT tool upgrades bring new features.
Obviously, the quality of subsegment matching will depend on the quality
and specificity of the TM. A random TM will give random matches, whereas a
customised, domain-focused, well-maintained TM will deliver higher quality
matches and a lower edit distance. Again, it is the translator who holds the controls.
Sophisticated features need data, and they are provided by the translator.

2.3 Boosting the translation memory

Any translator using a CAT tool and expecting that it will immediately speed up
their translation, may be disappointed. A new CAT tool arrives with an empty
TM. The building process can be slow if, for example, your translation work is
non-specialist, you have many different clients in different domains, or if your
texts have little in common. Students, who use their institution’s CAT tools sev-
eral hours a week and are given texts from different domains to provide trans-
lation challenges, may find themselves in a similar situation: the TM will not
provide many matches.
A repeated warning for users of public PCs in training centres regarding TM
creation: public PCs shut down without storing data. They do not allow you to
build a TM.You must remember to export TMs in the transferable and exchange-
able TMX format at the end of a session and store the files in personal folders.
Despite these precursors of data loss, CAT tools have good features to boost the TM.
They enable us to import monolingual or bilingual reference files which will give
us bilingual or monolingual ST/TT matches in the Translation Results window.
We can import external interchangeable TMX files to boost our TM; alignment
is the pairing of source and target texts that have not been translated in the CAT
tool; and importing reference materials within the project is another helpful feature.
TM edits and maintenance are important, but so is the preparation of translation
projects, which includes the need to boost the TM prior to translating.
 The translation memory database 39

2.3.1 Alignment
The alignment feature in the CAT tool turns previously translated documents and
their source texts into translation units (TUs) so that we can add them to a TM.
Performing alignment to boost the TM can be useful but it is time-consuming.
CAT tool developers have tried to facilitate and automate the alignment process.
It has become more intuitive and one CAT tool claims it is now integrated and
automated. Different programs have different names for the alignment feature. In
one CAT program (memoQ 9.1), we add reference files to a LiveDocs corpus and
the program will then align them automatically. It claims that the aligned results are
correct in most cases, but that human revision is needed each time a match presents
itself to ensure good results. It is advisable to make yourself familiar with the
alignment feature before the need arises. Automated alignment is a great improve-
ment compared to the manual adjustment of join-up lines between segments which
used to be the standard method in CAT tools (Figure 2.8).
CAT tools offer five types of alignment: alignment with review, which means
checking the join-up lines between segments or review in the translation editor on
the fly, alignment of single files and multiple files. The fifth type is the monolin-
gual review (5.3.1): the translator can revise an exported clean target file.When the
reviewed target file is re-imported, the TM performs an automatic alignment and
updates changes. The alignment of source and target texts, and external reference
material is not the only way to boost a TM, we can create TM content with existing
public translations or public, bilingual corpora (4.3).

2.3.2 Translation memory and reference files

The phrase ‘reference material’ is commonly used by LSPs. The translator will
receive additional files with information or terminology. LSPs tend to send
packages, a CAT tool term for ZIP archives that are created in the CAT tool
containing: project files for translation, TM, Tmdb, analysis report, and reference
files. The package is imported and opened in the translator’s CAT tool and after
completion it is emailed or uploaded as a return package. Competition and the
need for compatibility has made packages interchangeable between programs.

FIGURE 2.8 Alignment with join-up lines between source and target segments
40 The translation memory database

Packages can also be prepared on a server. The translator is given a temporary

licence for the relevant CAT tool. This is a trend which might outstrip the use of
personal CAT tools (7.2).
Reference material can be contextual, containing extra-linguistic information
to explain the topic, images of products, applications and packaging or illustrations
to help the translator understand the subject of the source text. It can also be a lin-
guistic resource with relevant terminology, glossaries (mono or bilingual), or previ-
ously translated TTs. CAT tools can now import and process reference material in
suitable formats such as MS Word (PDF files are image files and would first require
a conversion to editable text) without performing an alignment. Keywords in con-
text are then presented as reference material in the Translation Results window.
Colours and icons are used to differentiate between references and translation
matches.
Reference files can be included in a package, or sent by email, or made avail-
able online. Programs deal with the import of reference material in different ways,
either as a separate feature after or during import when you tick ‘reference’ in a
drop-down box to identify them as non-translatable files. The benefit of importing
reference files in your CAT tool is that the TM does the searching and presents
relevant matches, explanations, and definitions on the fly, which saves the translator
having to scan or highlight relevant phrases in reference files before or during trans-
lation. The benefit of reference material supplied to the translator by client or LSP
depends on its relevance and quality.

2.3.3 TMX files

A TMX file is a TM that is exported in an interchangeable format which can
then be imported in another CAT tool. TMX files can also be exchanged between
colleagues or team members working on the same job if they are not sharing a TM
on a server. It should be noted that shared TMX files are best used in Read Only
format. If a proposed match is accepted by the user, it will then be entered in their
own TM, contrary to importing an entire TMX without checking the content.
Many open-source TMX files are available on the web (some can be downloaded,
but with the same warning: to be used as Read Only files):

• MyMemory is a TMX file**(link: external TMX) with a collection of transla-

tion memories from the European Union and United Nations, produced by a
company called Translated.net. The TMs can be downloaded in TMX format,
and they are also available as plugins in CAT tools.Translators have reported on
websites that MyMemory TMX does not come without errors and could be
compared with MT regarding quality. If it is used in the CAT tool as an online
plugin, it is in effect an MT tool.
• The Directorate General for Translation has compiled TM databases in EU
languages **(link: external TMX)
• For other multilingual corpora see Corpora (4.3)
 The translation memory database 41

Boosting a TM is tempting when your TM is empty. Alignment, importing ref-

erence files, external TMX files are methods worth exploring but expectations
should not be high and time investment is best limited to prevent disappointment.
Small, customised TMs work best, but we must create, edit, and manage these
ourselves.

2.4 Formats
Source text files for translation come in many different formats (up to 50) and
CAT tool manufacturers compete to keep up with changes and the interchange-
ability of new formats. CAT tools have made it possible to translate files that were
previously inaccessible or not readable on translators’ desktops. There are image
files or files with such complex layouts that reformatting the TT would put the
translation job beyond the scope of the translator and the budget of the client.
Before accepting an unusual file format, it is advisable to check if your CAT
tool can process it. You may receive files in formats with formatting instructions
hidden in tags, which must be observed in the target segments. Tags are entered
in the TM as part of the string. Tags affect not only the word count but also the
identification of strings. The TM may not identify a string as a match due to
an added or missing tag. Tags are considered inconvenient, their insertion takes
extra time, but they are the best way to retain difficult formats. Some formats are
easier to manage within the CAT tool because of its standardisation of all formats
in the translation editor. Formats such as Excel spreadsheets, web-based HTML
files, PDF files, and PowerPoint files all conform to the segmented grid in the
editor. The only difference is that their formatting must be kept by means of tags.
There are format tags in the SL segments and they must be inserted in the target
segments by the translator.

MS Excel
The translation process of an MS Excel file format is more convenient in a CAT
tool than in Excel: the text does not hide behind other cells, furthermore, it can
be spellchecked, the word count feature operates and the predictive aspect is absent
in the CAT tool. Although Excel spreadsheets were not designed for translations,
they have become popular with LSPs to manage short or fragmented source texts
such as instructions or specifications that need translating in multiple languages.
All the target languages can be assembled in one spreadsheet. When working in
Excel, the translators must insert translations in the respective language columns,
and if their language is in the P column, it is difficult to see the source text in the
A column without moving the scroll bar. The CAT tool imports and converts the
Excel format to its clear bilingual editor interface and exports the translation in its
original Excel format. One CAT tool (DVX3) has a function that excludes red text
(fonts) in spreadsheets. If the translator copies the source text to their column and
changes the font colour to red in the A column, the source text will remain in the
42 The translation memory database

A column and not be overwritten in the CAT tool and the target text is exported
in the required column.

HTML
HTML (HyperText Markup Language) files have a computer ‘markup’ language
which although readable, has tags with all the information needed to define struc-
ture and layout for the text on the World Wide Web. The tags are placed within
these symbols <>, which is what the eye can see. The CAT tool reads all (hidden)
instructions and presents them with its own set of tags (also called codes) in the
source segments. To replicate the original layout, the CAT tags must be included
in the right places in the target texts. Translators complain about tags, because they
require extreme accuracy and cost extra time. If the syntax of SL and TL is very
different, the insertion of tags in the right place can be demanding. The CAT tool
flags up error warnings if tags are forgotten or misplaced and the file cannot be
exported until the omission is corrected. CAT tool manufacturers are working
hard to simplify and automate the insertion of tags. It would be impossible to post
HTML target files on the web without the markup and we must therefore respect
the CAT tool tags.

XLIFF and TMX

XLIFF (text files) and TMX (TM databases) are exchange formats, which means
they are interchangeable between desktops/laptops and compatible with a variety
of CAT programs. The terminology database is interchangeable in TBX format.
Interchangeable files are helpful when collaboration is required between translators
who are not working on a server or in the cloud. Despite the interoperability of
TMX, we need to be aware that CAT programs have different ways of doing things
such as segmenting the ST and storing the information in tags. Consequently, what
is considered a segment in one CAT tool might not be in another and the imported
TMX file may not present anticipated matches due to a different segmentation pro-
cess or tag coding systems, or simply because the translator joined or split sentences
prior to conversion.
Target files are exchanged in XLIFF format and have a similar problem: there
are so many different file formats and so many different programs that although
most tools can process XLIFF files, some loss of data or features during the
transfer process is possible. If we want to be sure that our CAT tool can handle
the imported XLIFF file, we can apply the ‘pseudo-translation’ feature in the
CAT tool. It translates a source text in a pseudo target language to test if the
XLIFF file can be handled and processed (Zetzsche 2003–2017). The bilingual
XLIFF in the CAT editor in Figure 2.9 shows a pseudo translated PDF file in
dollar symbols. It looks fine and has imported well. However, the export gives the
following message:
newgenrtpdf
The translation memory database 43
Pseudo translated PDF file
FIGURE 2.9
44 The translation memory database

Adobe Acrobat Reader could not open ‘xxx.pdf ’ because it is either not a supported
file type or because the file has been damaged. For example, it was sent as an email
attachment and was not decoded correctly

The pseudo-translation feature is useful because it generates a timely warning

that the CAT tool cannot process the format.

PDF
The Portable Document Format (PDF) is one of the most widely used file
formats. It is an easy method to secure a file and change it from an editable format
in MS Word to an uneditable image file. It is a problematic format for the trans-
lator. Translation in a CAT tool is not always possible and if it is, sentence sequen-
cing and format may change. If the PDF file is text-based you will find that
you can search and copy text, but if it is image-based this is not possible and an
Optical Character Recognition (OCR) program is needed to convert the file and
make it editable.
It is important to understand why CAT tools have more difficulty converting
some files, and why the success of the TT layout is not guaranteed. The clearer
the text, the better the conversion result, but a scanned text file saved in PDF
format is enough to make a translator and their CAT tool despair. The pseudo-
translation feature is helpful to see if the file can be exported and what the target
file will look like prior to attempting a translation. If a translator receives many
PDF files, it is worth considering the use of a high-quality OCR program to
convert PDF files prior to importing them in the CAT tool. It guarantees the
retention of format and sentence structures. In comparison, CAT tools can pro-
cess other common formats extremely well, such as PowerPoints, and formatting
results are good.

2.5 Other functions and features on the ribbon

The ribbons in MS Word and CAT tools operate in similar ways (Figures 2.10 and
2.11). They have tabs, icons, and drop-down boxes which vary slightly between
CAT tools. A thorough investigation of the ribbon will bear fruit in the long
term. A minimalist approach to using a TM, i.e. how to make it work with the
least effort, may be efficient in the short term, but will not give much satisfaction.
The program arrives in a default mode and needs customising and personalising.
Each translator’s requirements are different, depending on text genres, domains,
and language pairs. We will discuss filters and regex, which allow us to personalise
the TM. And if this is not possible because we are not the administrators but users
of the CAT tool, it is good to study the menu, so that we know which features
and functions are available. We could then ask the administrator to change a few
settings to suit our needs.
 The translation memory database 45

FIGURE 2.10 Ribbon in MS Word

FIGURE 2.11 Ribbon in CAT tool

2.5.1 Filters
We have stressed the importance of maintaining and editing the TM. CAT
programs have their own servicing mechanism and we should apply the TM
repair tool regularly. It does most of the ‘maintenance work’ for us by re-indexing
the database and we can set the edit tab in the TM to remove duplicates. This
process is called setting up filters. The TM editor has many filters and deserves
our attention:

• Edit filters can be used to search for a specific source/target term or phrase in
the TM.
• We can check modification dates in ‘history’ to recall previous versions for
comparison.
• We can set or add filters. For example,TM maintenance is made easier if we set
the filter to show duplicates, because they can then be deleted on the fly each
time they appear in the Translation Results window.
• We can set a filter to capitalise certain words in our languages or we can tell it
not to capitalise.

All these functions and more are available in the TM editor. When a TM slows
down because of its size, or if we use several TMs simultaneously in a project to
recall as many matches as possible, filters help improve matches and prevent false
positives.
46 The translation memory database

Some programs have a filter that sets penalties. They produce percentages in
the margin and alert us to matches which may be 100% identical but include a
tag or a space. If we ignore this penalty it will affect the format. A percentage of,
for example, 90% will alert us to differences and allow us to accept or edit the
proposed match. We can add or remove penalties depending on our requirements.
Furthermore, if one of our clients requests the term ‘item’ (ST) to be translated
as ‘product’ (TT) and another client prefers ‘article’, we can set penalties to alert
us that our clients use different terms. Penalties are also useful when we import a
TM from a colleague. A penalty set for the entire external TM will tell us that its
search results are not our matches. Penalties are not ‘punishments’ but alerts that
matches need closer scrutiny. Filters give us a means to control the TM regarding
the matches it recalls.

2.5.2 Regex
Our texts are full of patterns, such as dates, which are ordered differently per country
and language. Regex in the CAT tool is set up to apply transfer rules for different
formats of date and time, currencies, metrics, numbers, and email addresses. The
transfer rules will adapt dates from, for example, US 03.13.22 ST to UK 13.03.22
TT. Regex helps the TM recognise the string and it will match the digits in the
source and target segments on the fly.
Regex, which stands for ‘regular expressions’, is a mathematical theory on
which pattern matching is based. MT and CAT could not operate without
regex. It is not about regular language expressions but about matching recurrent
patterns. Why is regex important to us? If we recognise and understand pattern
matching, we can change segmentation rules and improve our error checking in
the CAT tool’s QA (quality assurance) functions. Here follows an example where
the TM will not recognise the sentences as identical because they lack automatic
numbering:

This is a new model.

(a) This is a new model.
1) This is a new model.
B) This is a new model.
1.1. This is a new model.

Matching is improved if we change the segmentation as follows:

This is a new model.

(a)
This is a new model.
1)
This is a new model.
 The translation memory database 47

B)
This is a new model.
1.1.
This is a new model.

If you go to segmentation rules in the CAT tool you can add the following regex:

^\(?[a-zA-Z0-9]+\)[\s\t]*

It means: look for all segments that start with any lowercase letter or
uppercase letter or any number between 0 and 9, which repeats itself one or
more times, is preceded or not by a left parenthesis, is followed by a right paren-
thesis, then by a space character or a tab character, which repeats zero or more
times. To change this rule we then add #!#, which means apply a segment break
here. Regex may not immediately appeal to linguists, but translators who have
learnt a few codes find it extremely helpful and an excellent way to customise
their TMs.
Regex codes consist of metacharacters, which are standard. Here are some
examples:

? makes the preceding character optional

\d a number character (digit)
\w a word character (letter)
\s a space character
\b a word boundary
[4,] four or more of the preceding character
| ‘or’
* zero or more of the preceding characters
+ one or more of the preceding characters

If you want to check for inconsistencies, and you realize/realise that you may have
confused GB and US spelling, you can click on Find and Search for one spelling
and then again for the other spelling. The regex metacharacter | enables one
search: realiz|se.
Many regexes are pre-set: if boxes in the CAT tool settings are ticked, cur-
rency in the English group should appear in the target format as 1,234.56 and
in the German target format as 1.234,56. This feature is automatically enabled
when we select source and target languages. However, we may have a client who
requires a space instead of a dot as the decimal separator, in for example, 2 000,
and it is helpful to set the regex accordingly in the custom list in auto-translation
rules**(link: regex).
CAT tools explain how to create a regex list in their Help sections and although
it requires some code writing, it is a skill that can dispel irritation: it prevents
48 The translation memory database

the TM from producing repeated errors or fuzzy matches or what we call false
positives, or unwanted matches in Search and Find. MT and TM systems both rely
on regexes to match TUs; MT needs regexes as a fully automated system. CAT tools
and MT systems are gradually merging, based on the same modus operandi. The
next chapter discusses integrated and adaptive MT in CAT tools, which would not
be possible without regexes.

Project-based assignment
Objective:
Collaboration in a translation project and the management of one or more
shared TMs either online or offline

Assessment and outcome:

Self/peer assessment of the collaboration with special focus on TM/TMX
management, the exchange of XLIFF files and their impact on consistency
in the translation

Method:
The assignment is designed for a project management team but collaboration
between individuals is possible

Tools:
A desktop CAT-tool to test the exchange of TMX and XLIFF files. A server
version is not necessary but would facilitate real-time collaboration between
collaborating translators.

Suggested resources:
Digital source text and digital reference materials (monolingual for reference,
bilingual for alignment)

Language pairs and direction:

Multiple language pairs; both directions if TM is reversable
 The translation memory database 49

Assignment brief
Set up a project management team (1.7.4.1) to include a senior project manager, a
relations manager, who is responsible for negotiations with the client and translators,
a resource/reference manager, and other managers as appropriate.
Your translation project involves a client, contracted translators, and one trans-
lation with a high word count plus adequate reference materials. The source text
for translation into one or more languages is split and shared between two or more
translators per language pair/direction. The main objective is to build a useful TM
(to be shared as a TMX) and supply reference materials to maintain consistency in
the translation. Consistency can be measured according to key terms in the TMX
and reference materials.
In this assignment most time should be spent on project and TMX preparation
within the team.
The following actions are suggested, the order is not set:

• distribution of team management roles

• recruitment of translators
• selecting the source text (PDF files are not suitable for this assignment) in con-
sultation with your instructor
• performing initial alignment(s) within the team to build the TM/TMX before
sharing
• sourcing suitable digital reference files
• if using a server, creating packages with source, reference files and TMX/
TBX files
• creating POs and briefs for translators, including the analysis/statistics report
• offering a price scale for repetition according to analysis/statistics report
• checking consistency in returned translation
• TMX files from translators should be imported in the master TM
• the master TM should be edited, including removal of duplicates.

Possible pitfalls
• duplicates in the TMX
• poor quality of the source text. Manipulate the ST if needed. Consistency in
the ST can be helpful.
• import and export of TMX files impact (sub)segmentation
• loss of data in TMs and TMXes during migration – keep backups
• forgetting to attach a TM(X) to the project

Concluding remarks
This chapter has illustrated how the TM as the main component in the CAT
tool can improve translation quality and increase your productivity. The industry’s
50 The translation memory database

mantra has been reviewed in the light of quality and consistency, productivity and
quantity, and the impact of the TM on the profit margins of manufacturer, LSP, and
translator.The CAT tool is an extremely sophisticated and comprehensive program.
The TM is however relatively inflexible. We cannot make alterations or additions
in the TM on the fly, as it can only accept our TUs, store them and recall matches
with high similarities. Improved translation quality depends on the user’s mainten-
ance of TM quality. The TM can only increase productivity if it is well filled and
customised, possibly accompanied by a range of smaller customised TMs, rather
than one big master TM. Furthermore, if we do not use filters and regex, and we
must change smart apostrophes to curly apostrophes manually after completion, we
are not using the tool efficiently. If we are asked by the client to use terminology
consistently and there are multiple translators involved, we must check matches
from an external TMX file before we insert them. We need to know their origin,
date, and version history. If our TM is boosted with TMX data, it might be prefer-
able to view it in Read Only mode. Alternatively, we can create a new TM for the
client and import the TMX data into the customised TM. Productivity and quality
are ultimately controlled by the translator and not by the CAT tool.

Further reading
Dragsted, Barbara (2005). ‘Segmentation in translation. Differences across levels of expertise
and difficulty.’ In: Target, 17(1): 49–70. John Benjamins. https://doi.org/10.1075/
target.17.1.04dra.
Flanagan, Kevin (2014).‘Subsegment recall in Translation Memory – perceptions, expectations
and reality’, JoSTrans, 23.
Garcia, Ignacio (2015). ‘Computer-Aided Translation’. In: Chan, Sin-Wai, The Routledge
Encyclopedia of Translation Technology, pp. 68–87. London, UK: Routledge.
Gintrowicz, Jacek and Jassem Krzysztof Jassem (2007). ‘Using regular expressions in trans-
lation memories’, Proceedings of the International Multiconference on Computer Science and
Information Technology, pp. 87–92.
Zetzsche, Jost (2003–17). A Translator’s Tool Box for the 21st Century: A Computer Primer for
Translators. Winchester Bay: International Writers’ Group.
Zetzsche, Jost (2019). ‘The 297th Tool box Journal’. The International Writers’ Group.
3
INTEGRATION OF MACHINE
TRANSLATION IN TRANSLATION
MEMORY SYSTEMS

Key concepts
• The translation industry has embraced technology in response to high volume
translation requiring a fast turnaround
• Artificial intelligence (AI) and the further development of machine translation
(MT) meet the high demands
• Translation technology developers have invested significantly in research to
improve MT
• We can combine MT search engines with the TM in the CAT tool to improve
and speed up our translation work

Introduction
Machine translation (MT) is an important TEnT. In this chapter we will evaluate
MT’s added value to translation when it is integrated in the CAT tool. The MT
function requires little preparation or action. When it is enabled in the CAT tool,
its proposals pop up in the Translation Results window for the translator to accept
or reject. TM matches take priority and MT matches may appear at the bottom of
the list, depending on the CAT tool.
The MT database cannot be edited: it is part of the external search engine.
We can only accept, reject, or post-edit proposed matches. Our discussion of MT
is specific: MT as a TEnT, integrated in the CAT tool and potentially adaptive.
For example, if we accept and edit an MT match, the TU enters the TM when
it is confirmed. Next time, the previously edited MT match will be proposed as
a TM match. Developers have caught on to this and designed closed circuits in
which the MT learns from the TM and becomes adaptive.This is not possible in an
52 Integration of machine translation

independent MT engine such as Google, which learns from huge corpora and not
from the individual translator’s CAT tool entries.
It is important to understand the operation of TM and MT in the CAT tool.We
must also know how to estimate and evaluate MT quality in the CAT tool. If we
have a better understanding of MT features as a TEnT and use the tool appropri-
ately, we can reduce the need to post-edit (PE) after the translation is completed.
We can do the editing on the fly. We will therefore discuss MT as a TEnT in a sup-
portive capacity, its integration in CAT tools, the quality it delivers, and the evalu-
ation of its quality through models and metrics. All this can be tried and tested in
the project assignment.

3.1 Artificial intelligence and machine translation

Artificial intelligence (AI) is a machine’s ability to perform the intelligent tasks
we do without thinking. Machine translation (MT) is an AI system. Through AI,
machines can solve problems, interact with our immediate surroundings and think
creatively. The following four AI categories are relevant to MT: machine learning,
deep learning, neural networks, and natural language processing. Machine learning
is the science of programming computers to act in an automated way. It uses
systems that can learn from data, identify patterns, and sets of rules, thus creating
algorithms to make independent decisions with limited human invention. Deep
learning is a subcategory of machine learning. It analyses huge amounts of data to
complete tasks like speech recognition (7.1.1), translation, TM and terminology
databases, and more. Neural networks drive the analysis and consist of a set of
algorithms modelled after the human brain, designed to recognise patterns in data.
Algorithms cluster and classify data.They can be trained. And finally, there is natural
language processing (NLP) which enables computers to understand and process
human languages. NLP teaches computers to use speech and understand language.
Computers do not yet understand the nuances of language, such as body language,
facial expressions, tone of voice, and intonation, but considering that humanoid
robots can give witty, sarcastic replies and understand random, ambiguous human
utterances, we anticipate improved performance in MT.
More and more CAT programs are paired with machine translation engines and
offer adaptive MT based on algorithms within the tool. They learn from the data
provided by the translator in the cloud, in standalone or web-based CAT tools.The
learning process is called deep learning, which digs deeper into TU segments. It
can offer improved suggestions for fuzzy matches by piecing together subsegments
(2.2). Up to now, MT suggestions in CAT tools have often led to much time spent
on error repair. The roll-out of neural machine translation (NMT) in CAT tools
has lifted MT output to such a level that post-editors cannot always identify errors
in a monolingual review because of smooth and fluent NMT output. The diffe-
rence between NMT and statistical MT (SMT), its predecessor, is discussed in the
next section.
 Integration of machine translation 53

3.2 From statistical to neural machine translation in the

CAT tool
Corpus-based statistical machine translation uses a database with infinite pieces
of text, monolingual, bilingual, and multilingual, and it applies statistics to present
matches from its huge database at random.The SMT translation decoder is an algo-
rithm that searches for the best equivalent in its database. Zetzsche (2019 (298th
Journal)) compares the search technique to looking for suitable ‘LEGO bricks’: you
know what you want to achieve and pick the pieces that might produce the best fit.
You slot them into place in a top-down movement. SMT works like that: its chunks
give us the correct meaning, but the sentences are awkward with incorrect word
order or syntactic relationships. NMT works differently: it is a deep learner and uses
neural networks to understand and identify the best sequences for the translation of
full sentences instead of phrases. It has an attention-based encoder/decoder model
in which the encoder takes the source sentence and all available knowledge (meta-
data) and generates/decodes a target sentence. The ‘attention’ mechanism looks for
suitable alternatives. It is not always helpful because it can pay too much attention
to a term or phrase. Furthermore, the system tends to struggle with proximity, the
closeness of pairs. Nevertheless, it supersedes NMT when it comes to sequences,
because the preceding and following words and phrases help determine the out-
come. The sequential relationship is quite different to dropping and forcing indi-
vidual LEGO bricks into place.
SMT generates a higher level of word accuracy but NMT scores higher in
word order and fluency and adequacy, two MT descriptors. Toury (1995) used
the term ‘adequacy’ and ‘acceptability’ (called ‘fluency’ in MT) when the focus
moved from text-based to context-based or functionalist approaches in the 1980s.
‘Adequacy’ relates and qualifies the target text according to the source text, and the
‘acceptability’ of a translation follows TL standards and end-user expectations. The
descriptor for SMT matches is more likely to be ‘adequacy’, whereas NMT is best
assessed in relation to ‘acceptability’ or ‘fluency’.
Slight mistranslations in NMT are difficult to recognise and require more editing
time. NMT is less efficient in specialised domains. It uses general data up to a cer-
tain level and then switches to domain-specific data. NMT has a fragmented view
of a text, which explains inconsistencies in word matches. Its strings operate well
at sentence level but do not respond well to the predictive features (2.2) in TMs
or translation management systems. The NMT wants to predict, based on what is
available in the database.
Although NMT gives superior results, its non-interactive nature does not
facilitate integration in CAT. The translator must review all the ready-made MT
matches and accept/reject or edit. In contrast, the TM offers matches which have
been entered by the translator or are stored in an external or imported TM.
TMs can be maintained, edited and improved at any time, whereas the data-
base of the MT engine cannot be edited by the translator. In the CAT tool, TM
matches supersede MT matches. The TM will indicate the quality of the match in
54 Integration of machine translation

CAT tool

custom
MT

terminology translation
database memory

FIGURE 3.1 CAT tool and adaptive MT

percentages (2.1), whereas MT matches are not rated or scored. This is an imbal-
ance that is detrimental to MT matches. If algorithms were given more domin-
ance, the TM might be able to select from the MT database, but NMT requires
such a large amount of training data that it would be too large for the TM to
hold and process.
Most CAT tools enable the integration of NMT and various programs list the
NMT engines that can be used. Although the integrated MT comes second in the
CAT tool, TM matches and MT matches are recycled in each other’s databases. If
the TM does not leverage a match, the MT will generate one, which will then be
entered in the TM after it has been edited or confirmed by the translator. It will
populate the segment next time as a TM match. In an adaptive MT system (3.6),
TM matches enter the MT through deep learning (Figure 3.1). The searches of
TMs and MT engines in their respective databases follow a similar pattern, but the
qualitative outcome is quite different.

3.3 Matches from the translation memory and machine

translation compared
When we enable the MT engine in the CAT tool, the search function scans the TM
and MT databases. If we have enabled matches to populate the target box, then by
default, the TM is given priority over an MT match. If there is no context or per-
fect match, the TM may present a fuzzy match. We know that fixing fuzzy matches
below a certain threshold of around 70% is not viable, and it is not recommended
to lower the threshold level. MT matches do not have a scoring system: they may
need serious post-editing. MT would benefit from a similar scoring system. For
example, if MT matches were rated gisting quality, which gives us only a rough
understanding of the meaning, we could decide to reject them like fuzzy matches.
The TM does the rating by matching new TUs with previous TUs in the database.
If MT could be scored, it would preclude much unnecessary post-editing (Garcia
in Chan 2015: 81).
 Integration of machine translation 55

TABLE 3.1 Edit distance

The president confirmed that employees should be informed

The CEO said the employees should be told
S S D I S

CAT systems have shown improvement in MT match accuracy. They can now
apply another layer of metadata to segments to indicate whether an MT translation
will reach an acceptable level of quality (TAUS). Web-based CAT tools can predict
which MT engines are likely to give the best integration into CAT workflow. This
development is called machine translation quality estimation and its development
is still in its early stages.
TAUS (2019) compares the different levels of productivity between TM and
MT. The study shows that segments translated with TM have a shorter edit dis-
tance on average and are translated faster.The minimum edit distance between two
strings of words is the minimum number of editing operations that are needed
to transform one to match the other. Editing operations are generally insertion
(I), deletion (D), and substitution (S). The example shows how the MT transla-
tion (line 2 in Table 3.1) compares with the ‘ideal’ reference translation (line 1 in
Table 3.1):
The edit distance is determined by the weighting given to each edit distance. If
S (substitution) scores 1, D (deletion) scores 2, and I (insertion) scores 3, the edit
distance in this sentence is 8.
DQF, the TAUS Data Quality Framework, looks at the distribution of
MT+PE+TM+HT in hybrid TM/MT processes. Any process will require some
post-editing (PE): MT matches, TM fuzzy matches, and human translation (HT)
without matches. What TAUS discovered was that 100% matches in the TM were
more reliable than any MT match, and that fuzzy matches which by default have a
correctness threshold level of around 85% are less reliable than MT: the lower the
fuzzy match percentage, the more reliable the MT match. Their study of efficiency in
the different processes shows that a combination of all processes (MT+PE+TM+HT)
is given an efficiency percentage over 45%, TM achieves a percentage of nearly 45%
in suitable word matches and HT is given the highest percentage of time spent on
each segment. TAUS’ findings indicate that the edit distance in TM is longer than
in MT and yet TM productivity is higher. A suggested explanation is the translator’s
familiarity with their TM and the awareness of the level of editing that is required in
fuzzy matches according to their percentages. TM fuzzy matches are, however, a grey
area because of the different match rates. TAUS concludes that there is no clear point
where MT efficiency proves to be higher than TM efficiency.They also found that the
target language plays a role in quality and output. If English is the source language, MT
productivity in Western European languages can be up to three times higher than in
Asian languages. In terms of productivity, MT comes close to the lower fuzzy matches.
56 Integration of machine translation

We can draw several conclusions: the TM match supersedes the MT match in

time and quality, and while MT matches and TM fuzzy matches compare well, a
combination of all processes in which HT has the final word gives the best results.
Language pairs and text genre affect MT quality: TM efficiency and time may
vary between different language pairs, depending on syntax or script. Current
developments in TM data training and adaptive MT (3.6) aim to leverage the inte-
gration of MT in the CAT tool and may resolve the imbalance between TM and
MT matches.

3.4 Machine translation – access and integration

MT is free/open-source software, which means that it may be used freely for
any purpose (Forcada in Chan 2015: 152). MT needs to be free/open-source,
because engines used in CAT tools are statistical corpus based and depend on
the availability of data, ideally in sentence-aligned parallel texts (2.3.1). However,
open source does not mean that anybody can interfere with the software, because
it is legally copyright protected and relies on experts and skilled computer
programmers to make changes. If software is free of charge, it is called freeware.
CAT tools list appropriate MT engines for their programs and ask you to enter
the API key code to link up with the cloud-based MT database of your choice.
Some engines are free of charge, others charge per character and require API
subscriptions.
The challenge for free/open source MT technology is its integration in CAT
tools. As mentioned in the previous section 3.3, TAUS made two pertinent
observations: TM matches override MT matches and MT suggestions are not
rated. MT would also benefit from being more user-friendly and unified across the
different search engines (Forcada in Chan 2015: 164). In sum, MT in the CAT tool
would benefit if:

• TM and MT segment matches were more aligned (similar rating scores). MT

matches could have a benchmark like fuzzy matches in the TM. Fuzzy matches
are compared with previous TUs and the differences are highlighted and rated
in percentages.
• MT matches could be compared with a potential and good reference transla-
tion and the difference (edit distance) could be shown in percentages.
• Differences between MT search engines could be overcome by an integration
of MT systems on a single platform.

Advances seem to come from the CAT program manufacturers who have
developed adaptive MT in CAT tools to leverage MT as a TEnT that learns from
the TM. But before we discuss adaptive MT in CAT tools and translation man-
agement systems, we will discuss how MT search engines learn from TMs in the
CAT tool.
 Integration of machine translation 57

Food for thought…

Why are many translators less enthusiastic about MT than the companies who
make the engines?

3.5 Translation memories train machine translation engines

A custom or self-build MT engine learns from the data provided in the TM and
Tmdb. The three databases are combined in a package where they are interactive
and create a closed circuit. The MT engine observes the translator’s input which is
used to update MT suggestions. In a translation management system (TMS) (7.5),
the translator interacts with the MT by accepting, rejecting, or editing suggested
TM matches, thus refining MT while translating. The same happens in a CAT tool,
if an API or plugin is used to access any of the standard MT engines. When the
translator accepts or modifies an MT proposal, it is added to the TM and will pre-
sent a prioritised TM match next time. The difference between integrated MT in
a CAT tool and the closed-circuit TMS or self-build MT system is that an API or
plugin in a CAT tool does not enable the translator to build or edit the MT search
engine. In both systems, data are kept confidential, either in the closed circuit or
through APIs and plugins that do not allow data to be kept or stored (6.4, 6.5).
Obviously, TM inconsistencies are replicated in the MT. If we maintain our
TMs well, consistent TM data have a positive impact on the quality of MT output
(Moorkens et al. 2013). Consistency is not only determined by output but also by
input, by the source text. A study on ‘data laundering’ (Moorkens et al. 2013), the
recycling of data, investigated if modified MT input in the TM would enhance
58 Integration of machine translation

consistency and results. It appeared that modifications were possible in either dir-
ection, and could have an upgrading or downgrading effect, the latter caused by
repeated changed word order in the MT.
Quality and edits are the responsibilities of the user. Confirming unedited or
poorly edited matches will have a negative impact on the TM and consequently
on the MT, regardless of which database presented the match in the first place.
The conclusion in the study was that TM data performed better than the baseline
engine, which correlates with TAUS’ findings (3.3).
A research group at the University of Trento Italy (Farajian et al. 2017) focused on
MT input and discovered that the integration of NMT in a translator’s workflow seemed
to work reasonably well if source domains were controlled, but that quality deteriorated
if there were multiple domains, large domains, and hence greater data diversity. They
also found that NMT worked well when sentences were processed from domains close
to the training data. Hence the success rate is likely to be higher when translators work
in a limited number of domains, feed their own data into an integrated MT in their
CAT tool, or build their own MT engines in translation management systems (7.5).
Adaptive and self-build MT engines enable translators to influence, control, and edit
MT content and quality before matches are confirmed and propagated.

3.6 Adaptive MT engines

Adaptive MT is a new category of artificial intelligence (AI). It learns from the
translator’s entries in real-time and customizes the data it finds in the TM via inter-
changeable TMX files. It can also learn from monolingual files. In addition, it learns
from the translator’s entries in the Tmdb and generates improved matches. As we
discussed in 3.5, the TM stores the translator’s TUs and offers perfect or fuzzy matches
for reuse, based on similarity with previously translated content. The adaptive MT
engine is smart and learns from our edits, our style, our terminology and syntax.
Modern MT**(go to www.routledgetranslationstudiesportal.com/ – A Project-
Based Approach to Translation Technology – link: adaptive MT engines), a three-
year EU funded project (2015–17), was financially supported by the biggest EU
Research and Innovation program, Horizon 2020, with nearly €80 billion of funding
available over seven years (2014–20). ModernMT is a new open-source machine
translation platform. The system did not make any changes to the baseline-engine
but used a technology called ‘example-based adaptive NMT’, in which the adaptive
part of the technology was new. A TM layer (or a very small TM) is the first to
receive the translator’s request. The NMT uses similar segments to generate a more
suitable suggestion (Farajian et al 2017).The NMT adapts its engine on the fly.This
kind of system removes the need to train an MT engine. The MT can be a large
generic engine which then adapts the parameters of the translator’s request while
the translation is happening (Zetzsche 2019 (297th Journal)). It is a much more
advanced method than the statistical MT LEGO brick building approach (3.2).
Adaptive MT engines are cloud based. In a translation platform system (7.3.4)
the translator can access one controlled interface via a plugin or API and thus
 Integration of machine translation 59

collaborate with other translators. The interface integrates the components MT,
TM, and Tmdb and the engine provides the best suggestions based on the three
components. It states the source of the proposed matches.The topic of adaptive MT
engines is continued in translation management systems (7.5).

Food for thought…

How do you see the benefits of MT as a TEnT when it is integrated in a CAT tool?

3.7 MT quality in the CAT tool

How can we define the quality of machine translation integrated in the CAT tool,
where it is a shared quality with other databases? Quality itself is a relative concept, for
example, the quality of a translation can be defined as being ‘fit for purpose’ or it can
be measured against a benchmark, which presents an additional quality. A ‘standard’
with set boundaries is much more measurable. Hence, in the 1990s, the EAGLES
working group (Experts Advisory Group on Language Engineering Standards)
suggested the following purpose-based classification of translation qualities:

• Raw translation, which may contain minor grammatical or syntactic errors

without impeding comprehension; to be used for large amounts of scientific
material
• Regular quality translation, which transfers the information grammatically
correctly but may be lacking in style correctness; to be used for technical
manuals
60 Integration of machine translation

• Extra quality translation; both fluent and idiomatically and accurate, culturally
correct in the target language; to be used for advertisements and literature
• Adaptation, which is not a direct translation of an original text; to be used for
press releases and advertising
EAGLES 1996

Statistical MT cannot be measured against standards, because its translation model

relies on a fluid corpus and it is not rule-based like its predecessors. It lacks a frame-
work which could facilitate standardisation. Instead it takes its data from bilingual
texts in a phrase table.The target phrases can then be scored for the phrase with the
highest score to be picked and proposed in the CAT tool as the best match. SMT
uses a language model of monolingual target texts and compares the segment with
the highest score with a segment in the monolingual text. This method confirms
that the proposed segment is likely to be an error-free translation. Proposals are
based on algorithms and probability and not on standards. Neural MT also relies on
statistical data and does not offer scope for measured standards.
It is not easy to measure the quality of machine translation in the CAT
tool. MT is neither outcome nor product, but a process. We want to find out
how MT as an additional TEnT can help us improve quality. In the CAT tool,
integrated MT matches are post-edited. In adaptive MT the generated output
is corrected by the translator, and the MT system ‘learns’ the correction so that
errors will not recur. If we were to measure the quality of proposed matches
and the respective post-edit time of integrated MT versus adaptive MT, we
would expect to see a marked difference in favour of adaptive MT. At present
measuring the quality of MT in the CAT tool is still an area with work in pro-
gress, although we can adopt ISO standards (5.4), or standards set by the trans-
lation industry (TAUS, BLEU in 3.7.2). Only the latter standards are specifically
designed for integrated MT.

3.7.1 Quality management

Since 2015, many LSPs adhere to the ISO standard for translation (5.4), which
measures the quality of the translation process. The ISO standard does not apply to
machine translated texts. Good practice expects any commissioned translation to be
sent out with a translation brief, a job instruction to ensure that the translator has
understood the requirements and the purpose of the translation. The term ‘quality’
in translation briefs is generally not detailed, if at all mentioned.The most common
qualification is that the translation must be delivered ‘error-free’. In the following
example of a translation brief, the purpose of the translation is not mentioned, nor
its quality. A CAT tool will automatically perform the suggested quality checks.
Quality management is one of the most difficult aspects of a translation project,
partly because our individual concepts of quality differ, and a project with mul-
tiple languages is managed by project managers who may have different language
pairs and concepts. The following translation brief should not give rise to much
 Integration of machine translation 61

disagreement, but whether it will automatically guarantee good translation quality

is a different matter:

• The font used throughout the document is Helvetica Neue

• Please translate all text that appears in the main slides and the speaker
notes panel
• Please translate the ‘Section’ headings that appear in the slide thumbnails panel
• Please do not translate words that are brand names or product names
• Where possible, please try not to adjust the width, height or position of the text
boxes. If the translated text doesn’t fit then please reduce the size of the text
• If your language does not display properly in the Helvetica Neue font then
please substitute for Arial or a similar sans serif font
LSP translation brief 2015

Views on quality differ. Muzii (2012) draws an analogy between purchasing a

car and buying a translation. He suggests that when we buy a car, we want to
know what the vehicle can and cannot ‘do’. In a new car, we expect a different
quality to that of a second-hand car, yet both should be structurally sound and
do the job. External imperfections in a showroom car will not affect its per-
formance and can be addressed by the recipient. So why shouldn’t translations
be offered at different qualities and different prices? Educating the client about
the quality of translation is often suggested as a way forward. Muzii suggests
that translation providers should ‘profitably’ signal what they offer in such a
way that the buyer can understand and will accept the price tag. Translation
providers need to be clear about the translation’s quality, its purpose (who is
going to read the translation and for what purpose), and its price (if adapted to
its purpose – the price should be lower for gist, higher for publication). Offering
quality is fine, but not without a detailed understanding of the quality concept
by both parties, providers and recipients (Mitchell-Schuitevoerder 2015). How
high should we raise the bar for MT quality in the CAT tool? Instead of pro-
posing a de facto level of perfection, maybe we could determine adaptive MT
quality in accordance with its purpose (gist or quality), the level of difficulty in
the source text and per domain.
And finally, how does the client relate to quality? The client who has
commissioned the translation needs to stipulate to what level of quality a translation
is to be made. Negotiations between translator and client (often through an LSP)
are primarily about time and cost. But shouldn’t translators or LSPs, as intercultural
communication experts, advise the client which translation technology is available
and what is best suited to the source text or the purpose of the target text? And
shouldn’t we discuss the client’s preferred terminology, and our translation approach
(which combination of MT / TM / PEMT / HT)? If we decide that MT needs
to be enabled in the CAT tool (due to high volume and time constraints), collab-
oration between client and translator is needed. Shouldn’t the client provide the
sources and data which the MT engine needs to learn from? If translator and client
62 Integration of machine translation

work together and there is a better understanding of the applied technology, the
translator and their TEnTs are more likely to deliver a better-quality translation.

Food for thought…

How would gisting quality impact the translator’s work method?

3.7.2 How to evaluate machine translation

Despite continuous MT improvement, the developers, manufacturers, and
organisations, such as TAUS, seem to focus on machine-led evaluation systems
(Wilks 2008). Apparently, evaluation systems are currently perceived to be the key
to improving quality and to reducing time spent on post-editing (3.7.3). Before we
discuss quality estimation in detail, we will first determine if the translator can esti-
mate MT quality in a CAT tool.
The evaluation of MT output is not straightforward. There are no absolute
benchmarks, and it is multidimensional: MT is a TEnT designed to assist the translator
in the three areas of quality, time, and efficiency, but none of them have clear boundaries.
Moreover, evaluations and assessments are affected by the MT engine used, the evalu-
ation model and metrics employed, the translator’s proficiency, and PEMT, which is a
subjective human-led activity. Furthermore, evaluation criteria for human translation,
e.g. about style, are not transferable to the assessment of machine-translated text quality.
Hence there is a shift in the current trend of MT evaluation towards automatic methods
and metrics, which measure the quality of MT in relation to the quality of a similar
 Integration of machine translation 63

HT piece. Metrics are based on text similarities.The BiLingual Evaluation Understudy,

BLEU (Papineni et al. 2001), was one of the first automatic MT metrics. It was an IBM
research report that worked on the premise: the closer the MT output to HT output,
the better.The analysis is based on n-grams, which in the field of computational linguis-
tics and probability are a contiguous sequence of n-words in a piece of text: one word
is a 1-gram, two words is a 2-gram, etc. The system looks at frequency of occurrence
and assigns probability to an n-gram occurrence with other n-grams. Probability makes
it possible to predict the next words and correct errors.
The automated approach to MT evaluation probably brings as many challenges
as human evaluation. First, MT engines vary depending on the amount of data sets
that have been used for engine training. Too much terminology data can give com-
peting data which leads to output inconsistency. And data quality is also affected by
the source of the input: TMs or glossaries. Second, MT cannot yet master style, such
as a shift from formal to informal register within a sentence. Third, different search
engines give different results: a generic engine (Google) is suitable for generic texts,
whereas appropriate domain terminology is needed for specialised domains.
These challenges can to some extent be addressed by the user. The reliability
of evaluation models is increased if the MT user checks the domain specificity of
the MT engine. Domain-specific engines and open-source databases for self-build
engines are available (Forcada 2015). Glossaries from clients for MT data training
must also be checked: they require a high quality to give good quality results.
Furthermore, expectation management should be part of the equation: if the client
expects high results, they must provide domain-specific data. In return, the supplier
of the translation must be able to estimate the anticipated quality level of the trans-
lation based on the input quality of data and source text. Data and ST management
have become part of the translator’s job. The responsibility to deliver good quality
does, however, not only lie with client and translator, but also with the manufacturers
of MT programs. The industry should be willing to align APIs and not make users
pay for individual APIs. The type of error management set up by manufacturers and
applied by the engines would benefit from uniformity across different MT engines.
In conclusion, evaluation models must give the translator clear indications of
MT error levels in CAT tools, in line with TM match levels. An increased awareness
of domain-specific terminology and a realisation that one engine does not fit all
domains or texts would improve MT integration. Other points of importance for
translators to note are that MT engines give different results for different language
pairs. In fact, some language combinations use pivotal engines which means that
automated translations between two languages run via another engine, e.g. Italian
into Japanese may run via an English pivotal engine. A good understanding of MT
engine output (and source of input) can greatly improve the quality process and the
reliability of quality estimation and evaluation.

3.7.3 Evaluation models and metrics

The main goal of translation quality assurance (TQA) is to recognise a specific level
of quality. Automatic evaluation metrics (AEM) are software programs that use
64 Integration of machine translation

a set of linguistic criteria for testing against a human reference translation. AEMs can
make use of the information retrieval concepts precision and recall (Doherty 2017).
Precision is the fraction of words in the MT output that is correct compared
with the HT benchmark and recall is the total number of correct words (i.e. pre-
cision examples) produced by the MT. This is called error typology, whereas
other AEMs measure the edit distance according to the minimum number of edits
needed to match MT output to the reference HT. Of course, we could question the
objectivity and reliability of the ‘golden standard’ human translation, which is used
as the MT reference (Doherty 2017). Two translators will rarely generate identical
target texts, and two reviewers will seldom produce the same review.
The following models, LISA (Localization Industry Standards Association) and
TAUS, use metrics that are meaningful measurements and calculations to improve the
engine and estimate the quality of the output. In this book we are not measuring pure
MT but comparing MT and TM output in CAT tools in integrated and adaptive mode.
Nonetheless, we can still apply the following metrics after we have generated a ‘golden’
benchmark translation in HT. The important feature is that the metrics are weighted.
The LISA Quality Assurance (QA) model is used by companies or LSPs to structure
feedback from reviewers on translation quality in the localisation industry. Errors are
weighted with one point for a minor error, 5 for major and 10 for a critical error:
Weighting means that if we record two minor mistranslations, they will generate
a score of 2. But if we record two critical mistranslations, they will generate a score
of 20. Scores for all segments in the task are totalled to give a final score for the task.
If the overall quality of the task falls below a predefined threshold, the task will fail
the LISA check. It is important that scoring models are agreed to make sure that all
reviewers score to the same standard.
The error typology template is used by LSPs to manage their quality program.TAUS
has developed a Dynamic Quality Framework (DQF) which uses Multidimensional
Quality Metrics to standardise translation evaluation. Their objective is that buyers
and providers of translation can compare and benchmark MT productivity and
quality. For translators it would be useful if we could apply metrics to a sample trans-
lation to give a client an estimate of quality, time, and cost.With the DQF plugin we
can track our translation quality and productivity in a CAT tool.The plugin collects
data from our CAT tools and sends a report to visualise our data.

TABLE 3.2 LISA model for quality assurance

Minor Major Critical

Doc language
Mistranslation 1 5 10
Accuracy 1 5 10
Terminology 1 5 10
Language 1 5 10
Style 1 5 10
Consistency 1 5 10
 Integration of machine translation 65

The new metrics are analytic metrics that try to explain why translations are good
or bad, whereas holistic metrics merely state the quality level. In the DQF, TAUS sets
new standards to define errors. It creates a taxonomy, a more accurate classification
of translation errors, and a scoring method to produce numeric indications. A root
cause analysis also refers to the source text and its quality. Its metrics require human
intelligence, it is not a one-size-fits-all metric, it is not an automatic approach, nor a
BLEU reference-based score. It renders a TQ qualification for any quality manage-
ment system so that it is in accordance with ISO standards (5.4). It includes validity,
verification, and reliability as properties of a metric.
The DQF-MQM (Dynamic Quality Framework-Multidimensional Quality
Metrics) is an error typology used by TAUS like the LISA Quality Assessment. It
is extensive:
The category Verity (Table 3.3) means that translations should be rendered according
to locale conventions, for example, a copper wire in a product in one country might
not be copper in another. The category Style tends to be subjective and is an area of
disagreement among translators.The best rule is to aim for a minimal set of error types.
In fact, it is best to avoid checking style if not relevant (e.g. in manuals or specifications).
The emphasis of metrics tends to be on error management and error repair.
Good MT translator management should consist of domain-focused engine selec-
tion, TM data input and editing, and ST quality checks to enhance output quality.
At the beginning of the chapter, MT was presented as a rigid, inflexible TEnT but
we have discovered many ways of managing input, which should give us better
results in integrated (adaptive) MT and reduce PE time within the CAT tool. The
following project-based assignment gives us an opportunity to trial this aspect of
translation tool management.

TABLE 3.3 TAUS (2019) Dynamic Quality Framework (DQF) uses Multidimensional
Quality Metrics

Locale Accuracy Design Verity Terminology Style Other

convention
formats

Address Addition Length Culture- Inconsistent Awkward

specific with
references termbase
Date Overtranslation Local Inconsistent Company
formatting use of style
terminology
Measurement Undertranslation Markup Inconsistent
Currency Improper exact Missing text Third-party
TM match style
Shortcut key Mistranslation Truncation/ Unidiomatic
Text
expansion
Telephone Omission
Untranslated
66 Integration of machine translation

Project-based assignment

Objective:
Ability to estimate and assess output quality and efficiency of integrated TM
and MT in a CAT tool

Assessment and outcome:

Team-based assessment of data collection after trialling integrated TM and MT
procedures. You need to create a data presentation in a bar chart according
to discussed metrics models

Method:
The assignment is designed for project teams with one language pair. A clear
distribution of tasks is necessary to provide rich data for metrics; good
data records will increase the reliability of results and restrict the impact of
variable data

Tools:
CAT tool with integrated or adaptive MT; TMS system; APIs

Suggested resources:
Several source texts from different domains with high word counts to try
different MT search engines

Language pairs and direction:

One language pair; both directions if TM is reversable

Assignment brief
The project assignment is about testing and assessing integrated MT in the
CAT tool. You will need to set up a project team and appoint a project manager.
As a team you prepare a quote for a 6000-word translation (a deadline of four
workdays) which gives the client an estimate of the anticipated quality levels and
 Integration of machine translation 67

cost of integrated MT, depending on the level of PEMT required. It is an analytical

assignment, which can be performed by a project team or by an individual.You are
trying to assess translation quality. The focus is on MT data in the translation pro-
ject. There is less focus on team management, although task distribution within the
team is important.
This project-based assignment provides the framework to assess quality and effi-
ciency of integrated or adaptive MT and must be shaped by your team with a con-
sideration of the following:

• The tools, CAT and MT (adaptive or integrated) + APIs available to you

• The selection of source texts and distribution within the team
• The procedures or combinations of TM/MT/PE/HT employed within the
team and a special focus on PEMT
• The selected metrics model for evaluation
• Choose and develop an evaluation model based on error typology or edit
distance (3.7.3) (or a combination)
• Create a metrics model before you start: your own or based on BLEU, LISA
• Selection and preparation of texts (see pitfalls): some should be highly suited
to MT, and one should be less suitable. They should have similar word counts.
MT suitability can be tested by running samples through freely available MT
engines
• Allocation of integrated/adaptive TM/MT translation tasks with agreed
deadlines. STs can be shared or split among team members. TMX files must
be used
• Experience MT integration/adaptation in different domains
• Comparison of error typology and distance edit models
• Bar chart (s) (MS Excel) to illustrate MT/TM quality and/or efficiency
• A quote for the client based on edit distance and/or quality (Tables 3.1, 3.2,
or 3.3).

Possible pitfalls
• TM(s) with little data
• Poor MT data
• Text suitability for MT: the evaluation may be less dependent on data than the
suitability of the selected source texts. Test samples first

Concluding remarks
The objective of this chapter is to help you understand the processes used by MT
engines and the conditions that make them work well in a CAT tool. We have
explained how neural machine translation is a significant improvement after statis-
tical machine translation, particularly in fluency, syntax, and word order, although
not necessarily in terminology. You have had a chance to experience post-edits
68 Integration of machine translation

of MT matches in the assignment. PEMT will be discussed in more detail in

Chapter 5.You have learned how CAT tools can integrate MT and give you access
to the respective MT engines through APIs or plugins.We have examined how MT
operates and provides matches from its own database or by learning from your TM
databases while you are translating, which is called adaptive MT. We have reviewed
how you can improve MT quality through good TM maintenance, by checking for
high ST quality and matching generic and specialised MT search engines appro-
priately to ST domains. In addition, good management of integrated or adaptive
MT gives you a powerful TEnT when you import client glossaries or external
TMs, provided you apply regular TM edits. We have also seen how the industry
has adopted a new approach to quality enhancement and typical MT error repair
through models for error typology and edit distance. The models give an insight
into the conditions that will give an acceptable to fair level of MT quality in the
CAT tool and reduce our editing time for poor matches. Metrics help us produce
estimates of quality and time investment in a potential translation project. All
combined, you should now have an idea how MT, as a TEnT, can give translation
work added value. The project team in the project-based assignment is a vehicle to
experience how large translation projects incorporate MT. The assignment is not
so much about managing team members and their operations. It is more about
delivering the required level of quality through integrated or adaptive MT in a
CAT tool.

Further reading
Doherty, Stephen (2017). ‘Issues in human and automatic translation quality assessment’.
In: D. Kenny (ed.), Human Issues in Translation Technology, pp. 131–48. London,
UK: Routledge.
Forcada, Mikel (2015). ‘Open-source machine translation technology’. In: Chan Sin-Wai
(ed.), Encyclopedia of Translation Technology, pp. 152–66. London and New York:
Routledge.
Garcia, Ignacio (2015). ‘Computer-aided translation’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 68–87. London and New York: Routledge.
Hutchins, John (2015). ‘Machine translation. History of research and applications’. In: Chan
Sin-Wai (ed.), The Routledge Encyclopedia of Translation Technology, pp. 120–36. London and
New York: Routledge.
Kit, Chunyu and Billy Wong Tak-ming (2015). ‘Evaluation in machine translation and
computer-aided translation’. In: Chan Sin-Wai (ed.), The Routledge Encyclopedia of
Translation Technology, pp. 213–36. London and New York: Routledge.
Moorkens, Joss, Stephen Doherty, Dorothy Kenny, and Sharon O’Brien (2013). ‘A vir-
tuous circle: laundering translation memory data using statistical machine translation’.
In: Perspective Studies in Translatology. DOI: 10.1080/0907676X.2013.811275
TAUS (2019). DQF BI Bulletin – Q1
Wilks,Yorick (2008). Machine Translation. Its Scope and Limits. New York: Springer.
Zetzsche, Jost (2019). ‘The 298th Tool Box Journal’. The International Writers’ Group.
4
THE TERMINOLOGY DATABASE

Key concepts
• The terminology database is a comprehensive database in the CAT tool
• A terminology database is usually internal and integrated in the CAT tool (but
linked to the CAT tool if external)
• The terminology database complements the TM, especially when morpho-
logical changes affect matches
• The quality of online terminology resources is variable
• Corpora and concordances are indispensable terminological resources

Introduction
The terminology database (Tmdb), also called termbase in some CAT tools, is
one you build yourself within the CAT tool. A good Tmdb optimises, refines, and
customises TM recall. When the TM does not give anticipated matches, the Tmdb
can assist. Terminological pairs must be added manually. They are not stored auto-
matically like TUs in the TM. We create or open a Tmdb at the start of a project
and add new entries by highlighting the source and target term pairs or phrases.
We access the ‘add term’ function on the menu with a right click, a shortcut code,
or through an icon on the ribbon.To add metadata, such as parts of speech (POS),
synonyms or definitions, we use a different shortcut code or click on full entry
in the drop-down box. When the term recurs, matches appear in the Translation
Results window. They are marked with different icons or colours to show that
they are Tmdb results. The Translation Results window in Figure 4.1 (top left)
shows from top to bottom: two red TM matches with red ticks (1–2), three blue
Tmdb matches with green ticks (3–5), and four EuroTermBank results with their
typical icon (6–9) (4.6).
70 The terminology database

Translators give a higher reliability rating to matches proposed by the Tmdb

than TM or MT. Possible reasons could be the specificity and accuracy of Tmdb
matches, because they are not part of an auto-assembled string (see subsegment
matching in 2.2), or because they are part of added reference material supplied
by client or LSP and compensate for what is not yet stored in the TM. When
translators look for a suitable target term, the TM only recalls what has been entered
and the Tmdb presents what has been added. Therefore, if they do not know the
meaning or translation of a term, they may consult digital dictionaries, use search
engines on the internet, or post queries on language or subject-specific translator
network websites.The CAT Tmdb is often underused or not well filled. Many users
are content with the sophisticated search options within the TM. Undeniably, the
building and editing of a Tmdb requires considerable effort and time, but so do
online terminology searches, and web hits need checking.

FIGURE 4.1 Translation Results window in memoQ 9.1

 The terminology database 71

In this chapter we will touch on the vast quantity of available terminology

resources available to translators. We will explore good terminology work practice
and show how to integrate terminology in the CAT tool to give it the prominent
place it deserves beside the TM.

4.1 Terms and terminology

Translators are keen to receive reference materials from client or LSP; the latter
will try to obtain suitable material not only to support their translators, but also to
ensure terminological consistency. We will therefore discuss the resource qualities
of the Tmdb first and then explore how the Tmdb and the TM can be reconciled
to optimise the CAT tool. Whereas the TM is used to memorise content and assist
the translation process with the required level of consistency, the Tmdb is a more
specific guide to help translators deal with domain-related terminology in a trans-
lation project. If the translator has access to clearly defined translations of terms in
the CAT Tmdb, it will improve quality, maintain consistency, and increase the CAT
tool’s efficiency, partly through fine-tuning the TM.

4.1.1 Glossary
A glossary can be defined as a language dictionary in one or more languages
containing all the terminology of a domain usually preferred by an organisation.
A vocabulary is a list of terms with synonyms and definitions (or explanations) in
two or more languages, relating to a specific subject field, but not pertaining to the
activities of a company or organisation like the glossary. A good example of an online
glossary is ‘Electropedia’, which is produced for the International Electrotechnical
Commission by the International Electrotechnical Vocabulary, a leading global organ-
isation that prepares and publishes International Standards for Electrotechnology**(go
to www.routledgetranslationstudiesportal.com/ – A Project-Based Approach to
Translation Technology – link:glossary). Electropedia is multilingual, currently offering
terms in 14 languages, accessible to all. Another example of a glossary created by
an organisation is the Online Glossary on Governance and Public Administration
developed by the Division for Public Institutions and Digital Government (DPIDG),
of the UN Department of Economic and Social Affairs in collaboration with
CEPA**(link: glossary), Classification of Environmental Protection Activities:

The purpose of the Glossary is to provide United Nations Member States, and
all other interested parties, with a common definition of the basic terms and concepts
related to governance and public administration used in United Nations
documents, and DPIDG. In particular, the Glossary aims at improving the
clarity of the intergovernmental deliberations of the United Nations itself; and
at assisting Member States to better implement United Nations resolutions
by providing a more unified understanding of governance and public admin-
istration terminology. This Glossary contains non-legally binding definitions
72 The terminology database

of terms most frequently used in governance and public administration in

(a) United Nations documents and (b) in the work of DPI.
https://publicadministration.un.org/en/About-Us/UN-Glossary

The UN glossary is monolingual and contains definitions only, specifically designed

to set terminology standards which will be recognised by all member states. The
multilingual standardisation of terms is important to the translator to prevent the
random use of terminology (see ISO standards in Termbanks 4.6 and 5.4). Definitions
are needed to define a concept in clear words. For example, ‘domain’ is a technical
term (and not a personal field or area) with different meanings in the fields of
computing, physics or biochemistry. Definitions should be part of our personalised
CAT Tmdbs and customised per client. Manufacturers have preferences: what may
be called an ‘item’ by one, is called a ‘product’ by another. Such refinements can be
added to the Tmdb, but not to the TM.

Food for thought…

Why is a glossary, a potential source for a CAT Tmdb, more suitable than a
vocabulary?

4.1.2 Term extraction

Term extraction is the process of extracting domain-relevant terms from a corpus,
a body of texts or words. Imagine we are sent mono or bilingual reference material
for a translation project; then searching for relevant terms, adding definitions and
 The terminology database 73

target terms, or pairing terms in a bilingual text can be time-consuming. After

the search, we need to identify the terms and create a corpus, such as a bilingual
glossary, in which to store them. Creating the glossary in a suitable file format (txt,
csv) means that it can be imported in the Tmdb.
Before we extract terms, we should at least know how to identify and cat-
egorise them. A term is a word (simple term) or a multiword expression (complex
term) of a concept within a specialist subject (domain). A concept is an idea that
corresponds to its essential features. It will cover many varieties, such as pencil,
piece of chalk, pen, all products of the same concept, the writing tool.The termin-
ology of a subject is the set of special words and expressions used in connection
with it (Cobuild). LSPs generally ask a translator about their ‘specialisms’, meaning
terminological expertise, before they contract them. Terms behave in a distinct
manner and have relatively fixed contextual surroundings (4.4). In the English
language these are typically noun phrases, such as ‘adapter socket wrench manu-
facturer directory’, or verb phrases, such as ‘to be in/out’, which in cricket ter-
minology refers to whether the players are batting or not. They are difficult to
translate. Other features may be typographical, e.g. italics for scientific names, an
acronym, such as WYSIWYG (What You See Is What You Get) or initialisms, for
example, UN (United Nations, but known as UN), or the official title of a position
(CEO – chief executive officer), organisation (FC – football club), department
(DS – department of state) or unit (ARMAD – Armoured and Mechanised Unit
Air Defence).
In a CAT tool, we can extract term pairs by highlighting and adding them to
Tmdb on the fly. Several CAT tools**(link: Term Extraction) offer monolingual
term extraction as a regular part of the translation process or as a separate feature to
build up glossaries or Tmdbs (Zetzsche 2003–2017). Term extract is a very useful
but often underused feature. It extracts important words and expressions in a new
source text or a corpus of existing translations. The selection is data-driven and
uses algorithms. Consequently, a term list may contain a quite a few false positives,
which are irrelevant results due to multiple meanings. An extracted term list, how-
ever, enables a translator to research important terms before translating the text and
to include them in a Tmdb if they appear to be correct and consistent.
The CAT tool with an external Tmdb (SDL Trados Multiterm) uses existing
translated documents saved in the CAT tool and checks the frequency of terms
at subsegment level to create a term list. In CAT programs with an integrated
term extraction editor, we can run term extraction each time we add new files. It
will simultaneously search our Tmdbs. It is advisable to edit and maintain the lists
constructed by the CAT, because they are created as proposals and may need some
reconfiguration. Filtering (2.5.1) is also recommended to prevent the extraction
of syntactic words without terminological significance, such as articles (the, a/n).
Before we apply the useful function of term extraction, we need to understand
how to create a reliable and usable Tmdb. The CAT tool can do the job, but it
works better if we add metadata. They operate as a filter and reduce the number
of useless matches.
74 The terminology database

4.1.3 Metadata
If we compare the translation memory with the terminology database, there is a sig-
nificant difference in the amount and type of metadata either can store, and our control
over them. Metadata in the TM tell us when and by whom a segment was translated
or edited. This can be useful when comparing previously translated versions, which
are all stored in the TM.When we set up the project, we can state the domain, client,
and deadline. The Tmdb, in contrast, is filed manually and metadata are added by the
translator: SL/TL variant(s), fields of application, part of speech (POS), definitions,
sources, usage, and context (2.2, 4.4).When a Tmdb match appears in the Translation
Results window, we can open a dialog with a right click to view added metadata. If
the Tmdb is well filled and edited we call it a ‘high-end termbase’ (Melby 2012).This
means that the concept entries have been checked (not taken from the web without
quality control), and are accompanied by equally checked information, the metadata.
These data help the translator choose appropriately if there are several target options
for the source term. This work practice has a positive impact on the quality of the
TM and reduces TM edit time if appropriate terms are selected immediately and con-
sistently. Table 4.1 shows how a comprehensive terminological entry in the database
removes ambiguities and supports terminological consistency in a translation project.
It gives an example of a definition of a special breed of cat in the Tmdb of a 20,000-
word-encyclopaedia-on-cats translation project:
The window ‘Create term base entry’ (memoQ 9.2) in Figure 4.2 opens after
you have highlighted a source and target term pair in the translation editor and
clicked on ‘Add term’. It has entered the following metadata:

Termbase(s) en-nl-nl
Term entry supplied (duplicates 3)
Languages English Dutch

TABLE 4.1 Example of term definition in SL

SL term Calico cat

SL contextual definition Sunnyvale: Rare male calico cat turns up at

shelter. […]
Calicos are not a breed; their name comes from
the animal’s unique multi-coloured coat.
SL source www….
Concept Type of coat/ fur

TL term …….
TL contextual definition …….
TL source www....
Translation equivalent (full or partial) …….
 The terminology database 75

FIGURE 4.2 Term entry in memoQ 9.1

The translator completes the following:

• entry – source term(s): [+] for additional terms, [|] for change, and [-] for
removal
• matching:
• matching ‘50% prefix’: morphological changes (4.5) such as verbal
prefixes or endings are not immediately identified as matches by the
TM. The ‘50% prefix’ is the default setting. In this case, the Tmdb will
look for ‘supplies’ and ‘supplied’ but not for ‘supplying’. The default can
be changed to fuzzy, exact, or custom
• case sensitivity: this can be set to ‘yes’ or ‘no’, if you would like a term to
contain small letters or caps or both.
76 The terminology database

• usage enables us to add forbidden terms, meaning how a term should not be
translated
• grammar enables us to add POS, gender, and number. For example, the term
‘research’ can be a verb or a noun
• definition (Table 4.1)

Definitions are sometimes needed to clarify the meaning and standardise the
term (4.1.4, 5.4), particularly in technical texts. A good and helpful definition will
meet the following criteria:

• Predictability – the definition inserts the concept into a concept system: a

carrot is a vegetable
• Simplicity – the definition is clear and concise and no longer than one sentence
• Affirmativeness – the definition states what the concept is
• Non-circularity – the definition does not begin with the term and does not
use words which are dependent on the concept
• Absence of tautology – the definition is not a paraphrase of the term, but gives
a description of the semantic features
• Part of speech – the definition begins with the same part of speech as the term
you are describing
Pavel Terminology Tutorial**(link: term definitions)

Food for thought…

Which metadata would the translator find helpful in a TBX file provided by the
LSP to assist a translation on fuel converters? Rank them in order of importance.
 The terminology database 77

4.1.4 Standardisation and quality assurance

Consistency of term translation is enhanced if the Tmdb is managed well. It is
best to follow agreed terminology standards and to perform regular terminology
maintenance. We have discussed term definition criteria (4.1.3) which need to be
met, although they do not guarantee agreement among the users. The industry
relies on technical specifications and criteria, rules and guidelines to ensure
that materials and products are interconnected and interoperable. This is called
standardisation. We need to apply similar standards to terminology so that we
agree which technical terms will be accepted as standard. Terminology standards
for preferred terms and definitions ensure a common understanding of the key
concepts. In other words, the standard definition of a ‘tree’ should be such that it
is understood to be a tree anywhere in the world regardless of size, appearance
or characteristics.
ISO standards**(link: ISO terminology) can be applied to multilingual or bilin-
gual terminology databases (5.4) and termbanks (4.6). But how are ISO standards
of terms determined? Terms need to pass a reliability assessment; they should be
able to cope with emerging knowledge and not be obsolete on the day they are
standardised. Acronyms and names of school examinations, for example, are often
reviewed, modified, or replaced. Our CAT terminology databases need to be
maintained and kept up to date accordingly.
There are terminology management tools available to help us check consistency
and relevance. They also provide quality assurance and a certain level of termin-
ology management **(link: terminology management). We can import bilingual
CAT formats, such as TMX, XLIFF in the terminology management tool that
will then run checks for completeness, consistency, number, and tags. Results are
presented in a common results window. The tool can also perform Quality (QA)
checks in the CAT tool. Chapter 5 gives more details on the quality assurance (QA)
function in CAT tools and its reliability.

4.2 Web searches

In their terminology searches, translators make prolific use of online mono/
bilingual dictionaries, specialised glossaries, comparable and parallel corpora
(4.3), MT, social networking and cloud technology (Wang and Lim 2017).
Junior translators are the most frequent users of web resources and their
search techniques are more sophisticated compared to senior translators when
researching the translation of difficult terms. In their study, Wang and Lim
(2017) found that senior translators rely more on experience and memory and
have a more linear and syntagmatic approach to reading the ST. The senior
translators read from beginning to end, compared to junior translators who are
more paradigmatic and focus on the segment and its alternatives. The para-
digmatic approach is more likely to lead to misinterpretations, which supports
Pym’s negative views (2008) on the impact of technology on human transla-
tion (segmentation 2.11). These findings also highlight the importance of good
78 The terminology database

search techniques and validity checks of our web hits. When we add a target
term retrieved from the internet to the Tmdb, we must be sure of its validity
and accuracy. Furthermore, the insertion of target terms in the CAT editor
needs to be based on our understanding of the entire source text, and our trans-
lation benefits from a linear, syntagmatic approach.

4.2.1 Search techniques

The World Wide Web (WWW) is open source, it can be accessed free of charge,
it can be modified, censored, and manipulated by users. A translator who needs to
know the term of a consumer product in the target language and cannot find it in
a bilingual dictionary will search the web. However, the internet was not designed
for terminology searches. Hence our online term searches must be sophisticated.
Web searches for terminology can be enhanced if we choose appropriate search
engines best suited for our term searches. There are many search engines, but the
main categories are threefold:

• (General-purpose) search engines are software applications that help search for
websites in WWW and then access and display them. Each search engine has
its own special features which also vary per country
• Specialised search engines show index pages for special topics only, which may not
be shown in general-purpose search engines
• Meta-search engines are search tools that operate as aggregators and use other
search engines’ data to produce their own results**(link: meta-SE)

Our search techniques can be made more term specific by using the operators
AND, ANDNOT in the search box of the search engine.These operators are called
Boolean logic. Its use has distinct advantages and immediate results:

• AND narrows the search by retrieving only documents that contain the
keywords we enter, e.g. ‘tea AND coffee’ excludes ‘tea and chocolate’,
• OR expands the search by returning findings with either or both keywords, for
example, ‘nursery OR day care centre’, and NOT limits the search to the first
keyword only, for example ‘coffee ANDNOT tea’.

The validity and accuracy of our term hits on the web are important. Boolean logic
is a good filter and makes hits more specific. Another filtering method is to check
the source and only accept term hits from reliable URL website addresses. The
URLs presented to us by the search engine tell us about the origin of the web-
site and help us determine a level of reliability. A domain suffix**(link: domain
suffix) is the last part of a domain and defines the type of website we may be
about to access. ‘Com’ domains are commercial websites, whereas ‘org’ domains
are used by organisations, and ‘co.uk’ means that the business is run in and from
 The terminology database 79

the UK. Domain suffixes are not entirely reliable and the actual website must
be checked to be sure that it was not hijacked by a less trustworthy body for
their own purposes. The quality of the website does not necessarily guarantee
the quality of its terminology and linguistic checks are necessary. We can check
the quality and usage of term hits by inserting them in linguistic corpora and
concordances.

4.3 Corpora
‘Corpora’ (plural form of ‘corpus’) are large structured sets of texts or words.
A terminology database in a CAT tool is a corpus of words, like a dictionary; the
TM is a corpus of phrases or sentences. Aligned texts and reference files consti-
tute corpora in a CAT tool (Alignment 2.3.1 and Reference files 2.3.2). Digital
corpora can be external or integrated in the CAT tool. In the following sections
we will examine some external corpora and their TEnT qualities. Glossaries and
external terminology databases (interchangeable TBX files) can be imported in
the Tmdb.

4.3.1 Digital dictionaries

Dictionaries are corpora. Printed dictionaries have always been associated with
standards. Standards, however, vary and are closely linked to name, tradition, repu-
tation, or price. A reputable dictionary will ‘sample’ a term before it is entered,
which means that the lexicographers will look at one or more screenfuls of rela-
tive citations to see how the term behaves grammatically and lexically before it is
entered or after an existing entry has been revised. With the vast rise of digital dic-
tionaries, unpublished and free of charge, we cannot be sure that similar sampling
procedures are applied, and consequently the quality of many digital dictionaries
must be determined by the user.
With never-ending and ever-changing data available on the web, digital dic-
tionaries offer new and excellent opportunities for lexicographers to compile
corpus-based dictionaries (Müller-Spitzer and Koplenig 2014). They can be
updated online without printing costs, the entries can be linked to collections
of texts (4.4) and multimedia, such as audio files, graphs, and images. In add-
ition, there is the new aspect of collaboration, the entire internet community
can contribute, with Wikipedia as an example. Wikipedia has become a popular
website, used as an immediate source of information. However, its open-source
character, meaning that anybody can add, remove, and modify content without
any validation or control, makes its content unstable, if not unreliable. The study
by Müller-Spitzer and Koplenig (2014) showed that the following criteria were
considered important in digital dictionaries by their respondents: reliability of
content, up to date, and to a lesser degree, accessibility and clarity. Reliability of
content means that all details represent actual language usage and that all details
80 The terminology database

are sampled and validated on a corpus. Maintenance by a well-known publisher

or well-known institution was only given an importance rating of 8% (Müller-
Spitzer and Koplenig 2014).
In conclusion, the quality of a digital dictionary is difficult to assess. Digital
dictionaries are like websites: we can determine their source and origin by
checking their URL, but without an ISO standard, they do not have a quality
label. If bilingual dictionaries are machine translated, they are even less likely to
have been validated. If we want to import their glossaries in a CAT Tmdb, we
will find that digital dictionaries are generally protected and it is not possible
to download and import them in the CAT tool. They must be treated as inde-
pendent TEnTs. Only glossaries downloadable as text (rtf) or XML files can be
imported as reference material in some CAT tools and will propose matches
during the translation process.

4.3.2 Multilingual data corpora

Many large organisations make their multilingual corpora available online for
general use and sharing, which means that users can upload and download ter-
minology to and from the websites. There are some very impressive data sources
available on the web, which we can use as TEnTs. The programs use web crawlers
to search for translated content online, including e.g. EU multilingual data and
then match it with web-based dictionaries**(link: multilingual corpora). In
Figure 4.3 we see the results if we enter a source term or an entire phrase in the
tool: it displays a vast amount of complete target segment matches. Some matches
have warning symbols, such as ‘this translation can be wrong’. The website of the
source is listed below the match. The tool also states if translation pairs have not
yet been reviewed.
Multilingual corpora with a more reliable quality are provided by TAUS. Its
Data cloud**(link: multilingual corpora) is an extensive collection of transla-
tion memory data from large translation buyers predominantly in software and
IT industries (Zetzsche 2003–17). It is a neutral and secure repository plat-
form and gives translators, language service providers, developers, researchers,
and stakeholders access to a number of services, such as Search, Upload (your
own data for sharing), Discover & Download (only if you have uploaded, and
acquired credits), Account History, My Data, and Feedback. Its collabora-
tive plan gives you credit points if you upload data and you gain credits in
return which enable you to download. The Data cloud is accessed through
the web or API, and accessibility depends on a membership plan. Once cor-
pora are created, they are offered as ‘libraries’ to other potential purchasers.
The TAUS Data search engine in Figure 4.4 has produced 100 entries plus
translations for the term ‘unilateral’. These occur in compounds and phrases,
ranging from ‘Unilateral effects’ to ‘Anencephaly and unilateral anophthalmia’
and ‘Unilateral leg pain, and/or swelling’. The i (information) icon after the
 The terminology database 81

FIGURE 4.3 Bilingual data corpora in Linguee stating that ‘external sources are not yet
reviewed’

first entry in Figure 4.4 gives information on industry (Legal Services), type
(Standards, Statutes, and Regulations), Data owner (European Parliament), Data
Provider (TAUS).
Google offers us multilingual data corpora by letting us search the web in the
requested language. It gives a random set of links to the search term in context.
Other kinds of open source multilingual corpora programs**(link: multilingual
corpora) contain words and phrases which are too new or too rare to appear
in a dictionary or standard corpus. Similar to TAUS Data cloud, they operate
as a concordance in Key Word In Context (KWIC) style (4.4). There are many
other free downloadable parallel corpora of texts (for alignment) and terms avail-
able to fill your tools, but this does not guarantee the quality of your translations
(Zetzsche 2003–2017).Their web mining for material is done automatically and
82 The terminology database

FIGURE 4.4 TAUS data cloud

manual corrections are not necessarily carried out.**(link: multilingual corpora)

Downloadable data deserve a thorough check before we use them: is the source
reputable or is it a DIY corpus; is it bootstrapped, i.e. a glossary based on and
derived from another glossary?
This brings us to the conclusion that the quality of online multilingual data does
not necessarily meet guaranteed standards or required criteria. It does not mean that
we cannot use them, but it implies that we must have a system in place to check
their usage and accuracy. In the following section we will have a brief look at how
corpora are built and how we can potentially build our own prior to importing it
in the CAT tool.

4.3.3 The creation of corpora

An established dictionary published online may have a history of
200 years.**(link: online dictionaries). It can then draw on a wealth of reliable,
tested, and informative material. Its lexicographers will use language databases,
i.e. corpora, in multiple languages to extract terms and try and test usage. The
corpora ensure that language in the dictionary is authentic and accurate. New
data is fed regularly from spoken or written material, from television, radio,
websites, newspapers, journals, and books. The Bank of English is part of a
larger corpus with approximately 650 million words. The full corpus**(online
corpora) contains 4.5 billion words from all English-speaking countries. If we
want to create our own corpora to build our own terminology databases, we
 The terminology database 83

can extract terms from reference materials by ‘bootstrapping’ our corpora onto
existing and available web corpora. For this there is the webBootCaT tech-
nology and there are BootCaT tools**(link: creating corpora for Tmdb). The
designers (Baroni and Bernardini 2004) of the webBootCaT built a tool that
does not need downloading but can be created by using the Google search
engine. The basis method is that you first select a few seed terms (keywords),
then send queries with the seed terms to Google and then collect the pages in
Google’s hits. The vocabulary in your created corpus of texts can be extracted
automatically in the tool. The webserver of the designers will hold the corpus
and load it into a corpus query tool, such as Sketch Engine** (link: corpus
building) where it is investigated and analysed. The terminology database you
have created must be converted to suitable formats (CSV) to import them in
CAT tools. The corpus query language in the term extracting tool is a special
language that looks for complex grammatical or lexical patterns to establish
suitable search criteria. We can see the outcome of its searches in the concord-
ance, where our queries are presented in context.
In the following section about the concordance feature we return to the
parameters of the CAT tool which can give us more control over quality in our
databases.

Food for thought…

How could standards and quality in multilingual data corpora be made more
transparent?
84 The terminology database

4.4 CAT concordance

The concordance feature in the CAT tool is very useful. It is intended for term
or phrase searches and is often seen as a substitute or an alternative for the Tmdb.
It allows you to access the TM when searching for individual terms or phrases
within strings. It may explain why some translators prefer its convenience and
the opportunity to search the TM rather than having to build a Tmdb. The CAT
concordance retrieves all the entries of a specific word or phrase from the TM
and displays them bilingually so that context-based information can be given of
source and target segments. In the CAT concordance we must take account of
subtle differences due to segmentation or morphological changes which affect
what is retrieved by the TM (4.5). The CAT concordance is not only bilingual
but also parallel as it gives multiple translations for one source term or phrase
(Zanettin in Chan 2015). A concordance does not give any definitions, but it
recalls nuances, and different shades of meaning in context. This is helpful to the
translator when having to decide on a good match. If the Translation Results
window does not propose any matches, the translator can retrieve term pairs
through the concordance function. If the term pair is embedded in a string that
is different to a previously confirmed string, the TM does not propose a match.
The concordance function, however, ignores the string and focuses on the term
pair (Figures 2.3 and 4.5).
The limitation of the CAT concordance is that it can only recall what has
been entered in the TM. If the Tmdb presents matches from imported reference
files, which have not yet been confirmed and entered in the TM, the concord-
ance will not offer any results. External concordances are useful TEnTs to help
the translator determine accuracy and usage of terms or phrases in context.

FIGURE 4.5 The CAT concordance in operation

 The terminology database 85

4.4.1 External concordances

There are two types of external concordance: those which enable us to create a
concordance through reference files, and comprehensive web concordances that we
may use to check our target terms.
The World Wide Web has become indispensable to the translator, but it has its
drawbacks. As we pointed out in our discussion of corpora in 4.3, mined termin-
ology on WWW is uncontrolled, it lacks background, testing, quality assurance,
and agreed standards. Nevertheless, it is an unrivalled source of corpora for spe-
cialist terminology and Tmdbs. Source texts with specialist terminology require
two different kinds of source material, i.e. linguistic reference material to under-
stand concepts and meanings, and a confirmation that we have selected appropriate
target terminology, if retrieved from e-dictionaries or the web. Concordances or
KWIC (Key Word in Context) tools provide solutions to both.
An external concordance**(link: external concordance) can automate the pro-
cess of finding specific reference texts on the web and collate them in one corpus
(Olohan 2004). The process is as follows: first the user provides a list of key terms
and phrases in the program, which then sends them as queries to a search engine
that returns a list of potential URLs. The user selects suitable URLs, and the
program retrieves the webpages and converts them to plain text or saves them in
‘txt’ format. Some CAT tools accept the text files as reference materials, and they
will then be incorporated in CAT concordance searches. The external concord-
ance can retrieve and process many texts in a short period of time which makes it a
very useful TEnT. Its external nature may have an additional advantage. Reference
materials may not need keeping. They are useful to understand the ST or find
translations for the TT as a one-off. TUs are then entered in the TM, which makes
new terminology entries retrievable in the CAT concordance without having to
store the reference material.The CAT term extraction editor (4.1.2) can add them
to the Tmdb.
Digital dictionaries**(link: e-dictionaries) contain billions of words. If we
search high-frequency words like ‘make’, the British National Corpus (BNC)
(latest edition 2007 and no longer used for new dictionary material) may give
hundreds of millions of results (known as ‘citations’). If translators want to know
more about actual usage, they can use online corpus search software which
displays the searched term or phrase in combination with other words, called
collocates, and the software gives the frequency of the collocations. A good
external concordance can assist a translator in the following ways: show the
most typical combinations of the word (or phrase), show synonyms (thesaurus),
compare with another word and find examples of usage. In a parallel concord-
ance we will find examples of alternative translations. We can apply filters to the
tool, so that it customises and collates our searches. Because sources are provided
for each citation, we can tell the program which sources are preferred. In sum, a
concordance TEnT helps us select and control our web hits through its examples
of usage.
86 The terminology database

4.5 Morphology and the terminology database

It can be frustrating when neither TM nor Tmdb recognises a term due to an
added prefix or suffix, while you are convinced you have previously translated and
confirmed it in one or both databases. This is a TM problem related to morph-
ology, i.e. the form, shape, or structure of the word. If a derivative is not recognised,
translators can try to resolve this by using filters (2.5.1) or highlighting the core
of the word without the prefix or suffix and search in the concordance feature.
Zetzsche (2019 (268th Journal)) made a study of TEnTs and their morphology
support, i.e. the ability of the underlying technology to recognise that different
forms of a word belong to one root version. TEnT manufacturers have created
different solutions to prevent TM non-recognition associated with morphological
change. A translation management system (7.5) uses AI to discern morphological
changes in words. Some CAT tools require the user’s application of an asterisk or
pipe character to tell the tool that following characters should not exclude pos-
sible matches, such as ‘support*/|’ must include ‘supported’, ‘supporter’. A CAT
program with an external terminology database has incorporated morphology
for a range of languages. The manufacturer decided not to use the external Tmdb
but moved the process into the cloud where it is combined with the program’s
MT service**(link: morphological change). The user can select an existing web-
based Tmdb and add new term pairs. Searches in the Tmdb differentiate between
linguistic and character-based fuzzy searches. The former is based on the mor-
phological core of the term and the latter is based on searches of three letters or
more, comparable with asterisk and pipe-based searches. Especially interesting is
the manufacturer’s use of third-party tools, i.e. an existing web-based ‘morphology
engine’, instead of developing their own engine. It should be noted that if the web-
based Tmdb is shared, it could impact confidentiality (6.5).
Zetzsche (2019 (268th Journal)) stresses the need for morphological support
in all CAT tools. Particularly languages such as German, with compound words
would benefit from improved recognition. It would also increase the accurate veri-
fication of terms in the quality assurance (QA) function (5.2.1). The CAT tool
may flag up false positives during a QA check and mark morphological changes as
supposed inconsistencies or false matches. Until all CAT programs support mor-
phological changes, we have the option to use the filters and change fuzziness
settings. Although it is not advisable to lower the percentage of fuzzy matches,
it could be useful in the read-only mode during a revision as it is more likely to
include some morphological changes. Obviously, this is a second-best approach
until the software developers find solutions in the CAT tool. It may well be an AI
solution, as part of translation management systems.

4.6 Termbanks
Large external terminology databases are often referred to as termbanks**(link:
termbanks). They have different modes of operation within or outside the CAT
 The terminology database 87

tool. If an integrated termbank in a CAT tool presents termbank information, or

matches, in a results window while we are translating, it is called a ‘push approach’,
whereas if we use the concordance or consult external terminological material,
it is called the ‘pull’ approach (Warburton in Chan 2015: 656). One CAT tool
performs push lookups through the EuroTermBank online terminology service,
which is a databank with more than 600,000 entries of more than 1,500,000
terms. Figure 4.6 shows how a click on the Quick Access tab displays the Term
Lookup icon in the CAT program. Because the Euro TermBank is integrated in
the tool, the user can toggle between ‘Search memoQ term bases’ and ‘Search Euro
TermBank’. One unsuitable result for ‘power’ is shown in the Lookup term dialog.
The Translation Results pane does not give a Tmdb result and the remaining option
is to click on the Online Lookup in the Lookup term dialog. The subject field was
not selected, which accounts for an inappropriate result found in the subject field
of ‘Communications and Education’. The TM, however, gives an acceptable 101%
context match. The different features complement each other well, depending on
the smart use of the translator.
External termbanks can be accumulated and aggregated on behalf of
governments, multinational organisations, or international umbrella organisations,
such as the UN, which rely on ISO standards. It is possible for registered users to
download, but not to upload or share, terms in contrast to many online corpora
run by web-based companies which make a profit through sharing and advertising.
TERMIUM Plus®, for example, is maintained by a translation bureau under the
auspices of the Canadian government. It is one of the largest termbanks in the
world and it gives access to millions of terms in English, French, Spanish, and
Portuguese**(link:Termium). It complies with ISO standards on Terminology
(5.4). The majority of the termbanks are independent standalone tools not
integrated in CAT tools.

FIGURE 4.6 Termbank and lookup term in CAT tool (memoQ 9.1)
88 The terminology database

Termbanks are valuable components for us to use as terminological resources.

Several CAT tools have the Term Extraction editor feature (4.1.2). Unfortunately,
many termbanks and other terminology resources lack integration in the
translator’s workflow and the translator will need more than one screen to open
several windows simultaneously. Standalone term extraction tools remove this
problem and may suit translators who are keen to tap into as many resources as
possible and build their own customised Tmdbs. The process of term extraction
tools is explained in 4.1.2. There are several on the market which differ in the
amount of metadata that can be added**(link: term extraction). Once termin-
ology is extracted from the files you have uploaded into the tool, the term list can
be imported in the CAT tool in an interchangeable format. One extraction tool,
set up as part of an EU-funded project, runs the extracted term list against any
of the following resources to query and check term pairs: your own CAT Tmdb,
EuroTermBank, IATE (EU terminology database), TAUS corpus, and a statistical
database of aligned web terminology (Zetzsche 2003–2017). Once the query pro-
cess is completed, matches are displayed in the Results dialog of the CAT tool.
Today’s translator cannot complain that they are short of terminology resources.
Our content and quality control over the TM are determined by quality and quan-
tity in our Tmdbs, more than we realize. But we must first create and build ter-
minology databases.

Project-based assignment

Objective:
A hands-on experience of creating, building and sharing a CAT terminology
database (TBX) in a translation project to ensure quality and consistency in
all translations across the project

Assessment and outcome:

Self- and peer-assessment of the impact of TBX on consistency in a shared
translation project:
 The terminology database 89

• Assessment of quality and consistency in the revised project trans-

lation in comparison to your collaboratively agreed benchmark
translation
• A file which lists all terminology resources used and consulted by
team members and contracted translators (the checklist needs to be
created for all, completed and returned)
• Checklists are shared and compared among the teams for assessment

Method:
The assignment is designed for:
• A project team with several terminologists, revisers and a project man-
ager who create a TBX for sharing. You translate the text in-house.
A revised translation on completion of the project will provide the
basis for your assessment
• A project management team with terminologists, revisers and project
managers who split the source text between multiple contracted
translators and provide them with a TBX and monolingual/bilingual
reference texts
• Collaboration between individuals to create a TBX prior to or during
the translation of a shared source text

Tools:
CAT-tool (a server version is not necessary), access to online concordances,
termbanks, WWW and other terminology supporting TEnTs discussed in
the chapter

Suggested resources:
One digital source text with domain-specific terminology which can be
split between multiple translators in a CAT tool. Consult your instructor.

Language pairs and direction:

One language pair; both directions if the TM is reversable

Assignment brief
A client requires the translation of a text in a highly specialised domain. They con-
tact a language service provider (LSP) and ask for a quote and best turnaround time.
The source text has an approximate word count of 500 words per available translator.
The client cannot produce a glossary. The source text is split between translators.
Teamwork is necessary to test the outcome of a shared translation supported by
good terminology resources. The following minimum requirements for resource/
reference materials and TBX apply:
90 The terminology database

• The TBX file for translators contains extracted terminology from mined
sources
• The TBX file includes metadata if you use a CAT tool with an integrated
Tmdb
• Reference files (mono/bilingual) are supplied to translators for alignment in
the CAT tool

Possible pitfalls
• Inferior quality of mined terminology. If ISO standards are not associated with
the resources, how do you check quality?
• If TEnTs are not available or require a subscription/licence, use available
resources, glossaries and dictionaries and create your own quality Tmdb/
TBX file
• Incompatibility of file formats of reference files/glossaries. Txt files or Excel
files (saved as CSV) can be imported in most CAT tools or Tmdbs

Concluding remarks
Translation without terminological support would at times be impossible. We have
seen that terminology occupies a database in the CAT tool in which term pairs must
be added manually, unlike the TM database which accepts all confirmed TUs auto-
matically. One CAT tool comes with an external Tmdb which gives the impression
that terminology is a separate entity outside the CAT tool. The objective of this
chapter is to demonstrate that the two databases are inseparable. The Tmdb is a
benign, clever, supportive, and indispensable friend of the TM. Where the TM fails
to present a match, despite subsegmentation, the Tmdb can do so, with data support
from many other TEnTs, either in the cloud or through digital terminological
resources. The volume of multilingual corpora we can draw from is infinite, ran-
ging from digital dictionaries and WWW term searches to databanks. Our TEnT
findings can be added to the CAT Tmdb. When they become confirmed TUs in
the TM, they can be checked for consistency and usage in a CAT concordance
or external concordance. Various CAT programs accept and integrate termino-
logical reference materials to varying degrees, but one aspect stands out: a high-end
terminological database greatly assists the operation of the TM. The translator is
accountable for the quality of the CAT Tmdb, which is determined by our working
practice and standards.

Further reading
Müller-Spitzer, Carolin and Alexander Koplenig (2014). ‘Online dictionaries: expectations
and demands’. In: Ebook Package Linguistics, pp. 144–88. Berlin: De Gruyter,
ZDB-23-DSP.
 The terminology database 91

Olohan, Maeve (2004). Introducing Corpora in Translation Studies. Oxon: Routledge.

Pavel Terminology Tutorial, https://termcoord.eu/wp-content/uploads/2018/04/Pavel_
Silvia_and_Nolet_Diane_Handbook_of_Terminology-002.pdf.
Warburton, Kara (2015). ‘Terminology management’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 644–61. London and New York: Routledge.
Zanettin, Federico (2015). ‘Concordancing’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 437–49. London and New York: Routledge.
5
HUMAN AND TECHNOLOGICAL
QUALITY ASSURANCE

Key concepts
• Translation technology tools offer quality assurance; translators offer revision,
evaluation, or assessment of translations
• Translation quality assurance in the CAT tool and revision are
complementary
• The final quality check is performed by the user of the CAT tool
• ISO standards apply to the translation process rather than the translation
• A post-editor’s task is complex and deserves more recognition

Introduction
In the previous chapters we presented translation environment tools (TEnT) as
invaluable to the translator for ‘better quality, more efficiency and productivity, and
higher profit’. Ultimately, improved quality, productivity, and profit is determined
by our efficient management of the tools. In this chapter we will try to discern a
balance between human assessment and its technological counterpart, which is
referred to as translation quality assurance (TQA), quality assurance (QA), or lan-
guage quality assurance (LQA) respectively, depending on the CAT tool program.
QA does not only mean quality assurance but can also mean quality assessment.
There is a difference:

• Quality assurance is the function we use when we enable the CAT tool to
resolve errors or flag up warnings to prevent errors. QA in the CAT refers to
the checking of terms, spelling, grammar, non-translatables, omissions, tags.
The process, also called ‘verification’, uses default specifications, such as the
 Human and technological quality assurance 93

spellchecker, terminology verifier based on our Tmdb, and tag verification to

check that tags have not been moved or omitted.Verification already happens
on the fly while we translate and can be requested again on completion. It will
then list all identified ‘errors’ in a separate window.
• Quality assessment (sometimes called ‘quality evaluation’) measures the
extent to which a product complies with quality specifications. Because of its
analytical nature, it is something a linguist may be asked to do; it is beyond the
scope of the CAT tool.

In our discussion of translation quality, we will include ISO quality standards,

which are external requirements. It is important that we examine how CAT tool
and translator work together, recognising that the final decision must always be
delivered by the latter.

5.1 Quality and standards

Quality in relation to human and automated translations was defined in
3.7.1 and standards were discussed in 4.1.4. In translation, neither quality nor
standards can be absolute, they are fluid: translation quality is related to its pur-
pose and rated by the user, and translation standards are benchmark translations
by which other translations are assessed or evaluated. But what makes a trans-
lation a benchmark translation, and who was/is the creator? We currently have
ISO standards for translation (5.4), agreements which bring translation close to
agreed standards. They do not lead to standardisation, in contrast to consumer
products that rely on technical specifications and criteria, rules and guidelines.
These will ensure that materials and products are consistent and exchangeable,
regardless of source or manufacturer. In Europe, where you can drive through
several countries in one day, traffic signs and notices are standardised: red and
round signs prohibit, triangular signs are warnings, and rectangular signs give
information. The designs may vary, but they are recognised by all road users.
TMX and XLIFF files are good examples of standardisation in CAT tools: they
make files exchangeable and operable in different CAT programs. The stand-
ardisation of translations may not be possible, or desirable, because it would
prevent the translator from dipping into the beauty, variety, and richness of
language. Standards, however, and quality can certainly be agreed and achieved.
In the following sections we will try to clarify the differences between quality
assurance and quality assessment in TEnTs and the role of the translator in
quality control.

5.1.1 Quality assurance and assessment

Quality assurance (QA) is a useful function in the CAT tool and many LSPs
make QA a mandatory requirement prior to the translator’s delivery of a
94 Human and technological quality assurance

translation. QA employs the TM database and checks, using statistical grounds

and algorithms (Munday 2012). A translator’s quality assessment is based on
evaluative criteria, but it also uses emotions and intuitions to make selective
choices as to what is right or wrong. Our emotions emerge from our response
to the ST or TT (Munday 2012), which affects the assessment in a subjective
way. None of this is to be found in the CAT QA, which systematically checks
the accuracy of the translation process. QA cannot check the product; it is a
technological assurance of quality of the translation process. We could say it
checks the quality of the TM and its operator (you) but leaves it to the trans-
lator or reviser to perform a quality assessment of the final product according
to agreed quality standards.
Organisations and companies work with goals, objectives, and values to assure
quality in procedures and assess quality in products. Their criteria are meas-
urable, they use KPIs (Key Performance Indicators) and their business goals are
SMART: Specific, Measurable, Achievable, Relevant and Time-bound. Consumers
can file claims if they do not receive the anticipated quality or standard in the
products. Employees attend annual appraisals, which are reviews of their perform-
ance against agreed objectives. In academic organisations students are accustomed
to scores, percentages, and grades, with set criteria and descriptors that express
grades in words.
Can we apply such forms of evaluation to translation work? And if so, do we?
All parties involved in a translation project recognise that translation errors are
made but there is not much agreement on the criteria used to categorise and rate
translation errors. Sometimes there is disagreement on whether an error is in fact
an error. Most LSPs in the UK will choose to meet ISO standards, which require
revision by a third person. Nonetheless, freelance translators are advised and some-
times required to take out an indemnity insurance, which protects them against
claims by the client for faulty or substandard translations.
Although assessments and error correction seem to be preferred methods
to deliver improved translation quality, it appears that more effort is made to
protect against defects than to set measurable translation standards. LSPs guar-
antee translation quality assurance to the client by confirming that they will
meet ISO standards. The processes of quality assessment vary between LSPs:
self-revision and assessment or third-party revision and assessment. If it is a
self-assessment, the translator will confirm that self-revision (often incorrectly
called proofreading) has been carried out, including a check of all pre-translated
text and CAT tool verification; that specifications have been followed; that
queries have been raised with a project manager and subsequently resolved. If
it is revision of a TT, a third-party linguist will return the revised TT with a
completed assessment form.
Self-assessment and revision statements are sent to translators for signing. They
do not guarantee that the translation is error-free. They only confirm that revision
and assessment have been carried out.
 Human and technological quality assurance 95

Ideally, we need a basic revision model like translation job model contracts,
which are made available to translators by professional organisations (7.3.1), with
reference to technological tools and methods that can assist good translation quality.
The revision model should include guidelines for self and third-party revision, but
also for full use of the QA function available in CAT tools. Quality assurance and
quality assessment are essential parts of the revision process. Revision is a critical
stage in the translation project.

5.2 Quality in different TEnTs

Quality assurance is easier to understand if we accept it as a CAT tool
function: default error checks against confirmed entries in the databases. The
following sections are about quality assurance (QA) of the bilingual document
in and by the CAT tool and our quality assessment of MT and localisation as
translation tools.

5.2.1 Quality assurance in CAT tools

The QA function in CAT tools is an active feature that checks the quality of
the translation in progress and at any time when it is enabled by the translator.
It cannot provide a human evaluation. It does not analyse target sentences or
consider linguistic context; it measures against data in the database. It is corpus-
based, phrase-based, and statistical: if the match is not 100% correct, some CAT
tools place error warnings, and spelling errors are marked with a red underscore.
Omissions are flagged up as inconsistencies by means of warning symbols. In a
final check the QA lists all ‘inconsistencies’ in a separate window for the translator
to edit or ignore.
QA was designed in the CAT tool to support the translator. Boring jobs,
repetitions, numbers are notorious for typos and errors. QA checks that your
numbers are correct, that you have not typed two spaces, that you have used your
client’s preferred terminology in a custom Tmdb, that you have been consistent,
that you did not miss or misplace tags. And if the warning error is inappropriate,
you can tick an ignore box. QA should be used for its good qualities. LSPs often
make it a requirement to perform a QA check before delivering the target file.
The use of regex and filter functions (2.5.1 and 2.5.2) can reduce error warnings
significantly.
QA can be carried out at any time and repeatedly during the translation process.
Human revision can unfortunately generate new typos or inconsistencies and it is
recommended to run a QA as a final check before exporting the file. The function
in the CAT tool is not a revision feature; it is an error spotter. As such, it is an indis-
pensable tool for a revising translator and fulfils revision parameters much better
than we perhaps realise. QA can have some pedantic and irritating habits (if repeti-
tive, use the filter!), but its value should not be underestimated.
96 Human and technological quality assurance

5.2.2 Quality assessment of MT technology

We recognise that MT is not 100% reliable; errors are obvious in the target lan-
guage. Neural machine translation (NMT), however, delivers correct syntax and
linguistically correct phrases in the TL, but this does not guarantee correct equiva-
lence between SL and TL. Its apparent linguistic correctness in the TL makes it
more difficult to spot errors.
Poor MT quality has been addressed by manufacturers, with varying degrees
of success. ISO standards exclude machine translated texts. Other standards are
needed. MT services have opted for a standard-based approach that sets a refer-
ence translation (human-produced) as a benchmark. Originally, the reference
translation was a human translation used to test the quality of speech recog-
nition. The human translation has been adopted in the machine translation
industry as a benchmark to assess machine translated output and is called the
reference translation. The industry has set an absolute quality standard for MT
based on TQA metrics (3.7.2, 3.7.3) with BLEU scores (Hernandez-Morin
et al. 2017), and we call this error typology. The question we may ask ourselves
is whether any reference translation is a quality benchmark or an evaluation
standard.
There are many ways of addressing the quality problem in translation. TAUS
defines two levels of quality: ‘good enough’ or ‘near human’. This is very much the
industry’s approach that compares MT with HT. The industry would like both to
have equal quality instead of HT, the reference translation, always being the accepted
benchmark. Academic researchers and manufacturers approach MT quality differ-
ently. Academic studies focus on user tolerance of MT output, in other words,
what the user finds acceptable determines the quality (Bowker 2007). Doherty and
O’Brien (2014), however, take a different view and assess MT as a tool. ‘Usability’
is their quality criterion, which means that the evaluation is based on how well
the MT translation can be used to achieve a goal. Apparently, manufacturers use
acceptability and usability tests to determine a standard (Hernandez-Morin et al.
2017). These tests consist of yes/no questions put to the user. Here we might wish
to question the level of objectivity.
A more subtle typology approach to the assessment process consists of two
steps: first we assess the ‘acceptability’ of the target text and then the ‘usability’ of
the TT in relation to the ST (Daems, Macken, and Vandepitte 2013). Gouadec’s
view (Daems et al. 2013) is that translation errors have different effects on
the user, and they also have different levels of criticality. Any form of human
assessment is bound to be subjective compared to automated assessments. We
must, however, bear in mind that automated assessments have been set up
by humans. Therefore, it is worth stepping back and assessing the automated
quality assessment in CAT tools not from a right or wrong perspective, but
rather in degrees of helpfulness to the translator who ultimately determines the
quality.
 Human and technological quality assurance 97

Food for thought…

Discuss the reliability of a reference translation as a quality benchmark.

5.2.3 Web localisation and quality

Web localisation constitutes such a substantial part of translation technology that
it must be mentioned. LISA (Localization Industry Standard Association) gave a
clear definition of localisation:

Localisation involves taking a product and making it linguistically and cultur-

ally appropriate to the target locale (country/region and language) where it
will be used and sold.
LISA 2013

Web localisation (often written as l10n = 10 letters including first and final letters)
requires a different take on translation quality. Hypertexts can be presented in the
tool in a nonlinear fashion, which means that what the translator sees is quite
different to how it will be read by users.The content is often split and shared among
translators and it is difficult to adopt a holistic approach to the text (Jiménez-
Crespo 2013: 30). It is even more important that the localising translator has a
greater awareness of digital genres, so that they know what kind of localisation is
required, particularly if the topic is not immediately clear from the segmented text.
98 Human and technological quality assurance

Compliance with local norms and conventions will make or break the quality of
the translation in the eyes of the reader.
We use special localisation tools to localise websites**(go to www.routledgetra
nslationstudiesportal.com/ – A Project-Based Approach to Translation Technology
– link: localisation tools). They are like CAT tools and have a TM. Their filters are
prepared for the most common types of files found in software and websites. They
separate text segments (called ‘strings’) from the source code so that the coding
is not visible to the translator. It makes translation more straightforward and the
codes are not exposed to unintended changes that could damage the product.
Another important feature of localisation is that files are ‘internationalised’ (inter-
nationalisation – i18n), an intermediary stage in which the source file has been
delocalised. The ST has had linguistic and culturally specific features removed so
that translators can start their localisation from a neutral file ( Jiménez-Crespo
2013). Figure 5.1 illustrates how a UK clock, which shows 10 o’clock at night-time,
is internationalised and localised to a 24-hour clock, used in many other countries.
Localisation is a growing industry which should not be ignored by any translator.
Dedicated localisation tools, some of which can be integrated in the CAT tool,
may be better suited to digital material. Web content is becoming so heavily coded
that even CAT tools struggle. Localisation requires the translator to have a good
understanding of metalanguage and adaptation.The client may internationalise web
texts before they are sent out for translation, simply to standardise the process and
standardise strings in future updates.
Quality management in localisation is comprehensive.The industry’s approach is
to use QA procedures to guarantee that quality requirements are met. It uses quality
control (QC) procedures from ST to TT delivery to check the quality of product
or service (Jiménez-Crespo 2013: 104). QA procedures include the localisation of
links and images hidden in the website structure.
Quality in web localisation is of interest because of its fluid nature and diversity.
One-size quality cannot fit all localised texts. In this respect the ‘corpus-assisted
approach to quality evaluation’ (Bowker 2001) deserves mentioning. We have seen
how many methods despite metrics are subjective such as the human reference

FIGURE 5.1 From i18n (internationalisation) to l10n (localisation)

 Human and technological quality assurance 99

translation. Localised text with much metalanguage cannot be assessed according

to a reference text. Bowker suggests that the corpus-based approach is useful at an
earlier stage in the localisation process because it can identify quantifiable data in
strings and recurrent patterns. The relationship between the linguist and localisa-
tion is an interesting one. Localisation tools can automate web texts to the extreme,
but there is always the point where quality comes into the equation and human
input is needed. We will return to localisation when we discuss digital platforms in
Chapter 7.

5.3 Revision in CAT tools

The best-known parameters of revision are accuracy and completeness. They
are checked specifically by the CAT QA: new matches are compared against
confirmed matches in the database. Spelling and formatting errors are identified
through the spellchecker and against matches found in the database. If the ST
contains ‘hidden text’ that is not visible, it appears in the source segments of the
translation editor. It cannot be overlooked by the translator. ‘Hidden text’ is a fea-
ture in MS Word, which includes not only footnotes, or tracked changes, but also
text from any other document inserted by the creator. The CAT tool eye does
not miss anything during the conversion phase when importing the document.
Other revision parameters are logic, facts, customisation, terminology, mechanics,
layout, and typography (Mossop 2014). The CAT QA can accommodate them in
different and unexpected ways:

• Logic – The translator is expected to deliver a coherent target text, in language

and content. This is the parameter of logic. The QA cannot check content
logic but it can highlight inconsistencies in the TT, and even in the ST. If
there are terminological inconsistencies in the ST, such as item, article, product,
and you decide to be consistent in the TT and use one term, QA will alert
you to this perceived inconsistency. Source texts often contain errors, which is
why some CAT programs allow edits in the source segments to prevent TM
corruption. Coherence is part of logic: sequences must be logical. Numbers
(plural/singular) and tenses must match. This kind of linguistic error is picked
up in a QA and classified as an inconsistency. It is matched against correct pairs
in the database.
• Fact – QA checks texts for factual errors: its observance of number matches is
important.
• Customisation and terminology – We have discussed the importance of
customised terminology when creating TMs (2.1): customisation of phrases
and terms in line with, for example, client glossaries is checked carefully against
the Tmdb in QA.
• Parameters of mechanics, layout and typography – They are catered for by
tags containing formatting instructions and make sure that the target text
100 Human and technological quality assurance

is delivered in an identical format. QA flags up error warnings when it

detects typographical errors, such as different font, missing punctuation or
capitals.

What the QA does not check is smoothness or cohesion which relate to style and a
smooth flow of words. We already explained (4.2) how segmentation can interrupt
the translation process if the translator does not take a linear approach to the whole
text. If segmentation affects the TT in spite of precautions, the final monolingual
revision of the TT must address style parameters.

5.3.1 CAT revision of bilingual and monolingual files

Self-revision can be performed in the editor of the CAT tool, in an exported
bilingual rtf file or in the final monolingual target file. Third-party revision can
be carried out in an exported XLIFF file, in an exported bilingual (review)
file, and of course in the final clean target file, using track changes. Third-
party revision is preferable in XLIFF or bilingual docx/rtf files because the
reviewed file can be re-imported in the translator’s or LSP’s CAT tool to update
the TM.
From a technological and a human perspective, there is a difference between
revising XLIFF and bilingual docx/rtf files. Revision in either type of file
enables updates of the databases. There are advantages and disadvantages to
revision in either file format. Layout and formating are best checked in XLIFF
files where you can insert and check format tags. In bilingual rtf/docx files, tags
might be inadvertently moved. Tags should always be checked in a final QA in
the CAT editor. Technological features, such as glossaries or term matches are
not accessible in bilingual files. However, from a human perspective, a bilin-
gual file offers more clarity with less information and fewer windows to dis-
tract the eye. Table 5.1 compares the positive or negative impact of several
features in XLIFF and bilingual files on self-revision. Changes made in bilin-
gual files must be updated in the TM and Tmdb after re-importation. It is
good practice if the translator receives a third-party revised file to accept and
reject changes in the editor of their CAT tool, so that databases are updated.
Time constraints sometimes prevent LSPs from returning revised files to the
translator.
What stands out in Table 5.1 is that ‘clarity’ in the bilingual file features posi-
tively. We all know how easy it is ‘to see yet not to see’. We can be selective in
what we see. Segments in both types of file seem to function as boundaries and
restrict what the eye sees and the mind takes in. Hence it is understandable that
translators like to do the final revision, a monolingual read, in a clean target text.
This mode is not suitable for revisers who need to see S and T segments side by
side. A revised clean target file can be re-imported in the CAT tool to update
the TM.
 Human and technological quality assurance 101

TABLE 5.1 Positive and negative sides to revision in XLIFF or bilingual files

XLIFF Bilingual Comments

file doc file

TM checks + - TM not accessible in bilingual files

Tmdb checks + - Tmdb not accessible in bilingual files
Clarity - + Only one view
Tags + +/- In bilingual files tags are reduced in
number and quality, which can
be visual gain but detrimental to
formatting
Additional CAT features, + - Not accessible in bilingual files
such as concordance
Clarity of revision comments - + Bubbles in XLIFF; text in bilingual

A final comment about monolingual revisions: after hours of revising your own
work (it is estimated we can revise approximately 1000 words per hour), we become
tired.We read but can no longer take in what we read. Other than leaving the desktop
and taking the dog for a walk or making a non-essential trip to the supermarket, there
are a couple of tricks that can wake us up and renew our critical faculties, for example,
changing the background colour of your file, or reading white letters on a dark grey
background. The other trick is to make yourself alert by moving to a different place,
such as taking the laptop to another room or even sitting on the stairs.The latter may
well be uncomfortable and will make you want to get the job done!

5.4 ISO standards

Third-party revision in addition to personal checks is required to meet ISO standards.
The CAT tool’s features for revision (Quality Assurance 5.1.1) are not sufficient. The
International Organisation for Standardisation, ISO (with over 160 member countries)
is an independent, non-governmental international organisation with a membership of
national standard bodies.Through its members it brings together experts to share know-
ledge and develop voluntary, consensus-based, market relevant international standards,
also called benchmarks**(link: ISO). ISO 17100:2015 is the internationally recognised
standard for translation services and was created in 2015 to respond to the market
need for a universally accepted benchmark. It is a process standard and does not define
the quality of completed work; it sets quality standards for the translation process. The
standard does not apply to interpreters, nor does it include machine-translated work.
ISO 17100:2015 is an ISO norm for translation services. It includes several
pertinent recommendations. These affect not only the translation service provider
(TSP) but also the client’s specifications (Mitchell-Schuitevoerder 2015):

ISO 17100:2015 provides requirements for the core processes, resources, and
other aspects necessary for the delivery of a quality translation service that
102 Human and technological quality assurance

meets applicable specifications. Application of ISO 17100:2015 also provides

the means by which a translation service provider (TSP) can demonstrate
conformity of specified translation services to ISO 17100:2015 and the cap-
ability of its processes and resources to deliver a translation service that will
meet the client’s and other applicable specifications. Applicable specifications
can include those of the client, of the TSP itself, and of any relevant industry
codes, best-practice guides, or legislation.The use of raw output from machine
translation plus post-editing is outside the scope of ISO 17100:2015. ISO
17100:2015 does not apply to interpreting service.
www.iso.org/iso/catalogue_detail.htm?csnumber=59149:2015

In other words, the ISO 17100:2015 standard specifies translation steps such as:

• Translation (including a translation check by the translator themselves)

• Revision by a third party
• Review (an optional step, designed in order to assess the suitability of the translation
against the agreed purpose, domain, and the recommended corrective measures)
• Proofreading (an optional pre-publication check) of the monolingual final
target text
• Final verification

The following definitions taken from the Oxford English Dictionary (2019) and
Merriam-Webster (2019) give us an accurate definition of the above terms used in
the revision process:

• evaluation the determination of the quality of something; assessment

• assessment the process or means of evaluating academic work; an examin-
ation or test
• proofread to read (text, esp. in proof) in order to find and mark errors for
correction
• spot check a check or inspection made on a random or representative sample
of a group, or one conducted without warning
• revision critical or careful examination with a view to making corrections,
amendments, or improvements
• review the action of looking (again) over something, as a book, text,
etc., with a view to its correction or improvement; revision; an
example of this
• edit to prepare (written material) for publication by correcting, con-
densing, or otherwise modifying it; change (text) on a computer

ISO 17100 Translation Services Management Standard has superseded the old
European quality standard BS EN 15038 for language service providers that was set up
by the European Committee for Standardisation in 2006.The former BS EN standard
continues to be embraced by many language service providers as a stamp of approval,
which indicates that their products are delivered with an acceptable quality. The
 Human and technological quality assurance 103

translation industry’s initial enthusiasm for the BS EN ‘fit for purpose’ standard may be
due to a lack of consensus within the industry (Gouadec 2007; Pym 2010a) as to what
a standard should be. One of the differences between ISO 17100:2015 and the earlier
BS EN 15038:2006 ‘fit for purpose’ standard is that the ISO now has an additional
competence requirement which means that the translator must be able to translate into
the target language using appropriate style and terminology (Mitchell-Schuitevoerder
2015). Another key difference is that there is a greater focus on the customer:

• identification of key customer requirements prior to production

• handling and processing of customer feedback and
• the delivery of the service as a whole

Furthermore, this International Standard applies to LSPs without internal translators

but with freelance contracted translators, LSPs with employed in-house translators,
but also to individual translators who collaborate with other translators.
The reliance on revisers by LSPs does not necessarily guarantee good prac-
tice. The translator needs to take full responsibility for the quality of their work
in accordance with reasonable and agreed standards. Furthermore, the translation
industry requires good, recognisable standards, which should be negotiated between
client and translator in a spirit of fair play. Too often the ball is kicked into the
translator’s court. And finally, as the use of raw output from machine translation plus
post-editing is outside the scope of ISO, the question arises who will set and agree
MT standards, especially when the target text emerging from a CAT tool is likely
to be a product of machine and human combined (Mitchell-Schuitevoerder 2015).

Food for thought…

Create three sets of revision standards: for human translation, for CAT translation,
and machine translation. Are there any major differences?
104 Human and technological quality assurance

5.5 Revision and evaluation

In practice, there does not appear to be much agreement on descriptors used for
third-party evaluation to meet ISO standards. ‘Proofread’ is a common term used in
the translation industry: ‘Please proofread the translation according to listed criteria.
Use a score from 1 (lowest) to 5 (highest) in the supplied evaluation statement’ (LSP
2019). The LSP expected the reviser to revise (not proofread) a translation and score
an evaluation of the translation based on error correction in the target text.Table 5.2
contains a random selection of third-party revision descriptors used by LSPs. The
first column shows the most frequently used descriptors and the third column lists
those that were only found once. The fourth column gives two scoring systems. It
was found that if a scoring system is used, it is generally numerical and not weighted.
The descriptors in Table 5.2 suggest that the reviser is expected to revise and
correct: ‘proofread’ is generally a misnomer for translation revision and is more
appropriate for desktop publishing services when checking the final proofs. The
term ‘proofread’, often used by LSPs, is the least applicable to translation evaluations
because it only requires the marking of errors for correction. If the revision pro-
cess is managed well, a translation is sent to a third-party reviser and returned by
the LSP to the translator who can accept or reject track changes in the file. EU/
DGT**(link: quality control) distinguishes two types of quality control: revision is
a bilingual examination of target language content against source language content
according to its suitability for the agreed purpose, and review means monolingual
examination of target language for its suitability for the agreed purpose.
In the translation industry, evaluation and revision are performed seemingly
without any clear or universal model.Translators may be asked to do tests before they
are entered in the LSP database.Yet they may be asked to revise peer translations or
evaluate peer tests without specific qualifications. They are unlikely to be tested on
revision or assessment competences and skills, contrary to translation skills, which
are frequently assessed by LSPs. Revision now is an ISO requirement, but there
needs to be an acceptance that a translator is not necessarily a competent third-
party reviser.

TABLE 5.2 Random selection of criteria found in LSP evaluation statements (2019)

Assessment criteria Assessment criteria Assessment criteria Scoring systems

(most frequently) (frequently) (infrequently)

accuracy lexis cohesion score 1–5 (1=low 5= high)

comprehension syntax coherence seriousness of mistake:
grammar register organisation of work 1. very serious
2. serious
punctuation orthography
3. not too serious
spelling accentuation
TT fluency
terminology
 Human and technological quality assurance 105

Agreed criteria for revision are crucial. Third-party revision and self-revision
require different mindsets: third-party revision is best performed with circumspec-
tion and respect for the translator who has undoubtedly translated to the best of
their ability. We do not know the circumstances under which the translation was
carried out, for example, was there time pressure? What was the quality of the ori-
ginal ST? Were reference files sent to the translator? Was it a shared translation?
A reviser should check a translation without comparing or modifying according
to personal preferences. It is better to query when there is doubt and leave it
to the translator to review their own choices and decisions, especially if they are
not evident errors (Mossop 2019). He suggests revisers should ask themselves the
following questions after completing their revisions:

• Was each change I made needed? If so, did it adequately correct the
problem?
• If the change was adequate, how would I justify it (does it fit the checklist
supplied by the LSP)?
• Have I missed any errors?
• Have I introduced new errors?
Mossop 2019: 205

In heavily revised translations it is not unusual to find that the reviser has uninten-
tionally introduced new errors. If Track Changes is used, it is best to hide the red
lines during the revision process to keep the text readable.
Revisions and evaluations require skill, competence and experience. They are
potentially subjective: they should be carried out according to appropriate criteria
and returned to the translator for acceptance or rejection, ideally in the CAT tool
to update the TM (and Tmdb).

5.5.1 Human evaluation

From a study it appeared that there was a significant divergence between different
reviewers (Guerberof 2017). They generally agreed on poor and good translations,
but not on the classification of errors. CAT tool data were taken into the equation
and there was more agreement on the number of no match errors than on Fuzzy
and MT matches. The grid in 5.3 is based on LISA QA (see 3.7.3) and the error
typology is applied to translations performed in CAT tools.
Overcorrection in the grid is an interesting but often ignored feature that consists
of revision or post-edits performed by the translator that were unnecessary and
may have led to incorrect renditions. The findings (Guerberof 2017) showed that
different linguists who reviewed the same translations confirmed a trend in which
the no match category showed the highest number of errors and that Language
and Style were the grey areas where nobody agreed. LSPs are keen to use the cat-
egory Language, but they do not make a clear distinction between Grammar and
Syntax. Style is often not included in an evaluation statement, because it can be too
106 Human and technological quality assurance

subjective. Furthermore, terminology errors were low in MT matches and style

errors were low in fuzzy matches. From other studies it appears that the initial MT
quality is important, and not the amount of post-editing. PEMT has a superior
quality compared to translation from scratch (Nunes Vieira 2017).
Revision and translation assessment, or evaluation, are closely related
to quality and standards. Dynamic quality evaluation models for translation
(Table 5.3) break with the standard error typology, in line with the fit-for-
purpose principle commonly used by LSPs. Error typology is used as a quality
safeguard (O’Brien 2012). Quality is closely linked with customer opinion
and yet its evaluation is based on static evaluation models (O’Brien 2012).
Existing models may be a form of security, because LSPs apply them to large
volumes of translation generated by large numbers of freelance translators in
their HR database, all with different levels of translation competence and skills.
The LSP must meet ISO requirements: each translation must be reviewed by a
linguist other than the translator. The main concern of the LSP is cost and time.
Quality is measured according to the number of errors, and action is taken if
the number of errors is considered too great. Action may consist of not giving
the translator further work (O’Brien 2012). In a worst-case scenario, the trans-
lator will not be paid.
What can a translator do to improve quality? How should a translator prac-
tise self-revision? A translation should not be delivered containing grammat-
ical errors, inaccuracies, term inconsistencies, or omissions (Lawrence 2018).
The translator needs to know what level of quality is required. Commercial
translations (advertising) have an optimisation problem: either you make the
translation as good as you can in a fixed amount of time, or you decide how
good it needs to be and get it to the required standard as quickly as possible
(Lawrence 2018). It does not mean that you should cut corners. Instead, think of
your work method. When do you do your research? Do you skim read the ST

TABLE 5.3 Error typology grid applied to a CAT tool with integrated MT

Type of error No match MT match Fuzzy match Totals

Mistranslation
Accuracy
Terminology
Language
Consistency
Country
Format
Style
Over-correction
Totals

Source: Guerberof 2017: 201

 Human and technological quality assurance 107

before you start and mark your problem areas, or do you begin without reading
the ST and stop and start each time you come across an unfamiliar term? Do you
confirm each segment, or do you send segments to the TM in draft mode? What
is your best method and what is your quickest way to revise? How is your CAT
tool set when you re-import revised bilingual documents? Does it import the
file with Track Changes enabled, which means the translator must review each
modification before confirming? If time is short, is it acceptable to skip some
steps by, for example, not checking the revisions in the editor but simply clicking
on accept and confirm all. After all, the CAT file history function records all
your changes: you can compare them after you have confirmed them, if there is
any doubt. Your work method has a significant impact on your time and prod-
uctivity, as well as its quality.

5.6 The post-edit of wholly or partly machine

translated texts
Post-editing refers to the correction of machine generated translations by humans.
The edit consists of the removal of major errors and the improvement of fluency
(linguistic quality) in machine translated texts. Our focus is on MT engines used in
CAT tools, preferably in adaptive and integrated mode (3.4, 3.5, 3.6). It automat-
ically lifts the quality and expectations and virtually removes a PEMT requirement
if the editing is performed on the fly, i.e. while the translator translates in the CAT
tool and modifies proposed MT matches. This type of PEMT on the fly is a fine
example of human and technological revision combined. The results are good.
We therefore need to define the difference between the act of translating and
post-editing and recognise where they overlap. First, the post-editor is not only
post-editing but also translating. It is not possible to modify a target sentence
without generating a translation in your mind. How can the post-editor make
choices and changes without knowing what they should be? Thus, the post-editor
is not only an editor but also a translator. The editing process is not simple. It can
be broken down into four actions: deleting, inserting, replacing, and moving words
(Zetzsche 2019 (298th Journal)). Replacing is paradigmatic, i.e. you take out the
wrong word and replace it. The other actions are syntagmatic, i.e. you upset the
sentence structure by deleting, inserting and moving words, and must then repair it
(Zetzsche 2019 (298th Journal)). And because the MT engine does not understand
metaphors or metalanguage, the resulting errors add to the post-editor’s cognitive
load. The post-editor’s cognitive effort is undervalued (also financially) if we con-
sider the many thought patterns needed while post-editing.
A typical PEMT difficulty created by neural machine translation is that NMT
tends to be fluent and it is difficult to spot errors in the machine translated text
without a thorough scrutiny and comparison of ST and TT sentences (Zetzsche
2019 (298th Journal)). This also adds to the post-editor’s cognitive burden, and the
overall tendency to reduce word fees by more than 50% for post-editing does not
recognise or reward the post-editor’s effort.
108 Human and technological quality assurance

Food for thought…

Can you agree on the editorial corrections needed in a machine translated para-
graph and a suitable financial reward on an hourly or a word basis? Discuss and
offer suggestions.

5.6.1 PEMT quality

The quality of a post-edited product is best understood if it is expressed in terms
of fluency (linguistic quality) and adequacy (translation accuracy) (Nunes Vieira
2017). Both are closely related to the quality of the MT product that is delivered
to the reviser. Fluency and linguistic quality especially, smoothness and readability,
suffer if the edit requires many corrections.There is a clear link between the human
editing process and the quality of the edited MT product. The quality depends on
the amount of temporal, technological and cognitive effort made by the reviser
(Nunes Vieira 2017). The three types of effort are closely related, for example, the
more thinking effort, the longer the editing time. The technological effort refers
to the technological operations that need to be performed in the CAT tool. These
efforts require time and consideration. They also apply to the revision of human
translation in a CAT tool where the final linguistic quality shows less improvement
when the number of errors is high.
Recent versions of major CAT tools measure the edit distance, either PEMT or
self-/peer-revision in the CAT tool. The so-called Levenshtein distance used in the
CAT tool is an algorithm and it measures the minimum number of single-character
 Human and technological quality assurance 109

edits (insertions, deletions, or substitutions) needed to change one word to another.

Here is an example in which the edit distance is 3:

bitten > bittin (1 change); bittin > bitin (2nd change); bitin > biting
(3rd change)

Three-digit changes have been made. In PEMT, the edit distance can be used to
measure the post-editing effort: fewer edits indicate less effort on the post-editor’s
part – and payment will be less as the post-editor is only paid for changes made. If
this approach is taken, many aspects of PEMT that are in fact similar to translating
from scratch in a CAT tool, such as job preparation, reading and comparing of
source and target segments, topic research and checking supplied glossaries or style
guides, final QA and delivery and administration, are overlooked.
The differences between MT post-editing and HT revision are largely
determined by the purpose of the translations. For example, if an MT text is not
for publication and only for gist, there is less emphasis on high quality and time
and cognitive effort will decrease. Another difference relates to the predictability
of errors in MT, such as terminology inconsistency, lack of cohesion in gram-
matical number, gender, or punctuation. These typical errors make post-edits
easier and quicker, whereas errors in HT are less predictable: they vary from
translator to translator. It is difficult to predict how remuneration and fees will
continue to be calculated. It is to be expected that time and effort in PEMT are
put under pressure depending on the purpose of the translation and that quality
may not be the main priority. On the positive side, adaptive MT in the CAT tool
allows us to make corrections to proposed matches before they are confirmed
in the TM. This should automatically reduce the need for PEMT and improve
quality levels.

Project-based assignment

Objective:
To measure efficiency (post-edit time) and quality during the revision process
and after revision or post-edit
110 Human and technological quality assurance

Assessment and outcome:

Self-/peer-assessment through revision and QA procedures with special focus
on the use of appropriate translation quality criteria

Method:
The assignment is designed for a project management team but collaboration
within a project team or between individuals is possible

Tools:
CAT tool (a server version is not necessary but would facilitate collaboration
between contracted linguists and team members)

Suggested resources:
One digital source text of approximately 1000 words

Language pairs and direction:

Multiple language pairs and/or directions but there must be multiple users of
each pair or direction

Assignment brief
The assignment incorporates steps discussed in previous chapters and requires
you to manage them through collaboration: ranging from the recruitment of
required linguists to the sending out of Purchase Orders (PO) (1.7.2) and trans-
lation briefs (3.7.1), the distribution of files for translation and revision, and the
receipt of invoices from contracted linguists. The new component is human revi-
sion combined with QA features in the CAT tool. In your team you must agree on
revision and assessment criteria. You arrange the translation of your source docu-
ment and third-party revision.
The project-based assignment provides the framework to manage and assess
translation quality and must be shaped by your team with due consideration of the
following:

• Establish separate sets of revision and evaluation (scoring) criteria for

your contracted revisers, which are appropriate for your chosen digital
source text
• Create and send a clear job description, requesting QA from translators before
they return target files
• Create and send personalised POs to translators and revisers with agreed fees
• Fees may vary per linguist (concerning language pairs), provided the total cost
does not exceed the budget agreed by the team
• Fees for post-editing may be calculated according to the edit-distance
function in the CAT tool
 Human and technological quality assurance 111

• Manipulate your source text and send out different versions to translators or
revisers, as appropriate, for example:
• prepare and send a clean ST to translators
• prepare and send a clean ST with a TBX or glossary to translators
• pre-translate the ST with MT in the CAT tool and send to revisers
• pre-translate the ST with MT but clear 50% of the target segments and
send to translators
• Prepare files for revisers in the following formats: XLIFF, bilingual doc or clean
target files
• Return revised files to the translators for rejection/acceptance
• An accounts PM manages incoming invoices and creates a spreadsheet
• Compare and discuss revisions and evaluation statements; compare the quality
of target files translated and revised through different methods; discuss and
compare the target quality of manipulated source files after revision. Show
your findings in tables or diagrams

Possible pitfalls
• Quality – too many criteria will make it difficult to measure your objectives
• Time – the time factor plays a significant role in revision and editing. If the
revision of a 1000-word file exceeds one hour, the reviser should be given the
option to return the file to the PM, who will then contact the translator and
request self-revision. In PEMT files with reduced fees (lower than translation),
post-editors should be advised about the time allowed and the quality required
• Profit and fees – in this assignment fees are important and must be negotiated
with the different parties including the client. Fees can be variable according to
the type of service required but the PM team must stay within a budget when
contracting their linguists. The budget is decided after the price of the transla-
tion service charged to the client is set.

Concluding remarks
This chapter discusses translation standards and quality, evaluation or assessment
and review, revision and edits without the requirement for a benchmark translation.
Standards are necessary, but they cannot be standardised, nor should we say that
we are fully in control of translation quality. The arrival of the internet and TEnTs
have accelerated the time to complete and deliver translations. Reduced time will
not necessarily improve translation quality, but if the end user is content, maybe we
should accept that the quality must suit the purpose. It is time for us to reconsider
our ideas of standards and quality benchmark translations. ISO 17100:2015 fails to
set standards for the translation product but sets them for the translation process.
We must accept that boundaries are shifting (Jiménez-Crespo 2013:112–13), for
example, between professional and non-professional translations (in crowdsourcing
112 Human and technological quality assurance

7.3.6). There is bound to be a noticeable difference between human and machine

translation. PEMT blurs the boundaries between MT and HT. Quality is partly
inconsistent due to the fragmentation of source texts: the translator no longer
receives complete and coherent texts, but updates; or the ST arrives partly pre-
translated; STs are often split and shared among multiple translators.
The best standards are those which are agreed by all parties in the translation
process: the client, the translator the reviser, the terminologist, and other linguists.
LSPs can play a major role in this process. Their best tool is transparency as to what
is required from all parties, and the level or integration of human and techno-
logical input in the translation. The interaction between TEnTs and the translator
is quite striking: in some cases, the final intervention comes from the translator
(in PEMT) and in reverse, technology provides the final touch through a QA. We
must recognise the level of interdependence between humans and technology in
the field of revision and accept that quality is often determined by the purpose of
the translation.

Further reading
Cronin, Michael (2003). Translation and Globalization. London and New York: Routledge.
European Commission. Directorate-General for Translation (2015). DGT Translation
Quality Guidelines. Brussels/Luxembourg. [Online] https://ec.europa.eu/translation/
maltese/guidelines/documents/dgt_translation_quality_guidelines_en.pdf [accessed
October 2019].
Jiménez-Crespo, Miguel A. (2013). Translation and Web Localization. London and New York:
Routledge.
Mossop, Brian (2014; 2019). Revising and Editing for Translators. Manchester: St Jerome
Publishing.
Nord, Christiane (1991). Text Analysis in Translation. Amsterdam-Atlanta: Rodopi.
Nunes Vieira, Lucas (2017). ‘From process to product: links between post-editing effort and
post-edited quality’. In: Arnt Lykke Jakobsen and Bartolomé Mesa-Lao (eds), Translation
in Transition. Between Cognition, Computing and Technology, pp. 161–86. Amsterdam/
Philadelphia: John Benjamins Publishing Company.
O’Brien, Sharon (2012). ‘Towards a dynamic quality evaluation model for translation’, The
Journal of Specialised Translation, 17. [Online] www.jostrans.org/issue17/art_obrien.pdf
[accessed May 2014].
6
DIGITAL ETHICS AND RISK
MANAGEMENT

Key concepts
• Intellectual property rights of translations need a clear definition about
ownership
• Confidentiality is at risk when web-based TMs and terminology databases
are shared
• Collaborative translation environments challenge digital ethics
• Non-disclosure agreements could potentially put the translator at risk
• Security and risk in the translation project are the responsibility of all parties

Introduction
This chapter is about our consideration of ethics in the digital translation industry,
where terms of business should govern the ways in which translators, LSPs, and
the client collaborate. The contractor trusts that good work will be delivered and
the contractee expects that it will be remunerated fully and in time. Purchase
orders (POs) can give both parties some legal standing. POs carry agreements
between LSP and contracted linguists. Invoices from linguists will include their
terms and conditions about payments.
Digital ethics relate to digital materials: how we store them, protect them, what
we do with them and what happens to them once they are in the cloud. Published
and printed materials are copyrighted. They are protected and cannot be sold for
profit, wholly or partly, without permission from the owner, the author, and pub-
lisher. We must now ask ourselves how we can protect our translations in digital
format against modification and other forms of violation, which could impact the
quality and consequently our reputation. And the second question is: how can we
114 Digital ethics and risk management

practise confidentiality of translated contents, either imported or exported? TMX

files are shared and our databases can be used to feed machine translation engines.
How do we keep sensitive materials out of the public domain?
As translators we may wish to check if LSPs have drawn up ethical pol-
icies for their collaboration with contractees, but we should first raise our own
awareness of privacy and intellectual property rights in relation to our trans-
lation work. The chapter also discusses questions about risk management and
our response to risk, whether we are project management teams or freelance
linguists.

6.1 Intellectual property rights (IP) and ownership

Intellectual property (IP) is a broad term used to protect your ideas (for example,
patents) and intellectual creation (such as design). A patent is a licence to protect
your idea or product for a set period. It gives you the sole right to make it, use it
or sell it. IP can be split into six main categories: patents, copyright, database rights,
know-how, for example, specialist technical knowledge, designs and trademarks of
established products or brands. Database rights are like copyrights: any electronic
database is protected for 15 years from the moment of its creation. It does not need
to be registered. Our TMs, TMX, and TBX files are arranged methodically and
systematically, they are accessible electronically and therefore meet the criteria to
be copyright-protected. When our digital files are shared with a client or LSP, it is
within a translator’s right either to restrict usage or to charge a fee. IP ownership
is not straightforward. If the work is collaborative, the LSP may have a valid claim
to your database. The LSP could argue that e-files are part of the translation job
and that they are included in the agreed fee. Copyrights are rarely discussed. LSPs
send agreements to be signed by the contracted linguist, and the translator must
sign in order to be given the go-ahead and start the translation. Many agreements
are biased and weighted in favour of the client and LSP. The copyrights of the
translator are barely mentioned if at all (Figure 6.1). Our signatures are not unlike
Faustus’ signature when he sells his body and soul without recall. A signature can
make the signee (the translator) accountable if confidentiality is breached or a poor
translation causes damage to the client. But should all the blame be ascribed to one
person only?
Collaborative translation environments and complex translation projects are
challenging because they involve shared resources and materials. Considerations of
confidentiality, ownership, copyright, authoring rights, associated legal matters,
costs, trust, quality, and reliability (Drugan and Babych 2010), all deserve more
attention. Unthinking signatures under terms and conditions without recognition
of IP rights, or realising that copyright should be attributed to multiple parties, or
that copyrights may have to be shared, could put all parties involved in a translation
project (client, linguist, and LSP) at risk.
 Digital ethics and risk management 115

6.2 Confidentiality and non-disclosure agreements

It is indisputable that confidentiality in digital files must be respected.The client and
LSP will require guarantees that confidential details in the source text are kept secure.
Any translation service involves the submission of potentially sensitive material to
other parties and transfers between computers. The client and the LSP will aim to
protect the content of ST/TT files from distribution and exposure to other parties.
They will therefore ask the linguists involved in the translation service to sign a con-
fidential disclosure agreement (CDA)/non-disclosure agreement (NDA). If
the translator breaches the agreement, the commissioning party could seek financial
compensation from the translator. Therefore, linguists are encouraged to take out an
indemnity insurance to protect themselves against potential claims.
There has been a shift in translation ethics, the focus seems to have moved from
the translation to translators and linguists (Chesterman 2001, 2018; Baker 2018). It
means that we view translation ethics from the translator’s perspective, the translator
and ethics, rather than the traditional view of the translation and ethics, which was
about the way in which we could best translate sensitive material. The questions
now are more likely to be: ‘Is it safe for me to translate this?’ and ‘What happens
to my work after I have delivered it?’ The range of sensitive material is extremely
broad and includes product promotion prior to the launch, minutes and agendas
of business meetings, surveys, police reports. There is very little material that is not
sensitive other than what has already been published in the public domain.
LSPs are most certainly aware of risks to confidentiality and take appropriate
measures. Confidentiality does not only apply to, for example, financial or legal
texts, but also to commercial products. Because the launch of a new product is
prepared well in advance, promotion material needs to be localised prior to the
launch and the client is keen to keep the information secret from their competitors.
It is the responsibility of the LSP to prevent a leak of data in the ST/TT. They
therefore ask their contracted linguists to sign a Non-Disclosure Agreement, which
may contain some of the following conditions:

1. Audio-visual material for translation may not be shown or used for any other
purpose than translation
2. The linguist promises to take measures to protect confidentiality by preventing
digital copies from being passed on
3. The linguist promises not to discuss the client or the product or mention that
they work for the client
4. The linguist must sign that all files and materials will be deleted, removed, or
destroyed after completion and delivery of the translation
5. Logos, ideas, designs, notes, etc. shall not be used by the linguist

The second condition is intended for the linguist, but is it also signed by the LSP
and any project manager dealing with the digital files? And how does this condition
116 Digital ethics and risk management

relate to TM databases, or TMX files, or TMs on servers, or TMs linked to adaptive

MT? The fourth condition does not specifically mention the translation memory
database(s) used by the linguist.
Terms and conditions in NDAs are legally binding once signed and offer
little protection to the translator. If there is an unintentional leak of data, for
example by posting a query on a platform or forum, it is important that the
translator can prove that all conditions have been met. In practice it may be a
difficult thing to do. It is important to be aware of the potential consequences
of NDAs before signing them, and to ask questions about the retention of your
TM databases.

6.3 Business terms and conditions in relation to

confidentiality and digital security
The NDA is a client’s/LSP’s safeguard against potential risk (6.6.2). LSPs increas-
ingly send their terms and conditions to contractees.Translators are advised by their
professional organisations to submit their own terms and conditions (T&C)**(go
to www.routledgetranslationstudiesportal.com/ – A Project-Based Approach to
Translation Technology – link: T&C of professional organisations). Table 6.1 gives
a selection of terms and conditions relating to digital security and confidentiality
only. What stands out in the T&Cs of either party (LSP and translators) is that the
conditions of the two parties do not add up. The two parties claim copyright but
do not include provisions for shared copyrights, which is likely to be the case in a
translation project with an author and a translator. Owner rights of digital files, such
as terminology and TM databases, are defined by the professional organisations.
The deletion of files ordered by LSPs does not allow the translator to retain proof
of their work if it were to be violated. There is little mention of digital distribu-
tion through file transfers between the different parties, or uploads and downloads
through web-based tools. Electronic transfers of materials between different parties
are numerous; the translator is only one party in the process and should not be held
solely accountable if there were to be breach of confidence.
T&Cs manifest in diversity. Some streamlining would be beneficial to all parties.
The starting point could be a basic model that concentrates on the ownership
and copyright of the source text. This might require or encourage the client to
take ownership and responsibility for the whole process from translation to final
version in print or on the web. The client could thus take control of deletions and
modifications to the TT for layout purposes, without causing any damage to the
translator. The translator can claim copyrights for TM and terminology databases,
but if the client wishes to have ownership after completions to protect confidenti-
ality, this can be agreed with the translator in exchange for a fee. And finally, there
is much existing legislation that gives mutual copyright protection when multiple
parties are involved. The wheel does not need to be reinvented for the translation
industry.
Table 6.1 does not display a harmonised approach to terms and conditions
There is hardly any mention of the translator’s intellectual property rights. There
 newgenrtpdf
TABLE 6.1 Selected terms and conditions for translation services

TRANSLATOR LSP TRANSLATOR LSP

ORGANISATION Confidentiality Confidentiality ORGANISATION Digital Digital security
security

The translation remains the The translator guarantees to The translator shall do their best
property of the translator give the contracting LSP all not to use or store confidential
unless agreed otherwise in intellectual property rights information in an externally
writing. associated with their services. accessible computer or electronic
information retrieval system
without security software and shall
prevent unlawful access by third
parties.
Copyright must be ensured by The translator gives the LSP all Confidential information requires
the client before the source copyrights and other intellectual the use of software to protect

Digital ethics and risk management 117

material is given to the property rights. confidential information and the
translator. translator must ensure it is updated
regularly.
The parties agree that a third In normal circumstances the If a translation memory is used, All data transmitted for the
party may be consulted translator shall not pass the translator licenses the use of translation and particularly
over specific translation confidential information to the translation for this purpose. translation memories remain the
terminology. others or use it for their own LSP’s property. They must not be
benefit, even after the contract used for any other job.
has ended.
Confidentiality should not apply
to the following:
- Information that is or becomes
available to the general public,
Continued
 newgenrtpdf
TABLE 6.1 Cont.

118 Digital ethics and risk management

TRANSLATOR LSP TRANSLATOR LSP
ORGANISATION Confidentiality Confidentiality ORGANISATION Digital Digital security
security
- and terminological glossary
entries compiled by the
translator for and during the
translation service(s). Unless
the client and translator
agree in writing that the
glossaries become the client’s
property after payment to the
translator. The conditions of
confidentiality will then apply.
The client’s documents are only All clients’ texts and translations The translator must ensure updated
considered confidential if this are confidential and should not antivirus software is used so that
was stated by the client. be transmitted to unauthorised files are returned virus-free.
third parties.

Source: ITI, ATA and global LSPs 2016–19

 Digital ethics and risk management 119

is an urgent need for guidelines as to who owns the TM before, during and after
the job is completed, the client or the translator (Berthaud 2019). In our digital
work environment, confidentiality is challenged in source texts, target texts, file
transfers, interchangeable file formats, CAT tools with or without MT, and many
other cloud-based tools. T&Cs are meant to protect the client or the translator,
but there needs to be more specific information as to who and what are being
protected and where. Confidentiality is a broad concept and a statement like ‘the
client’s documents are only considered confidential if this was stated by the client’
(Table 6.1) is an obvious starting point. The same applies to copyright and own-
ership which can and should be defined in an initial agreement. A good point of
departure would be for the translator to inform the client which digital tool is used
and the level of security practised.

Food for thought…

Discuss and create a workable set of terms of business for translators in relation to
their translation memory databases.

6.4 Data sharing and data protection

We continually share data, on our phones, laptops, and tablets. How much thought
do we give to confidentiality and the protection of information? General Data
Protection Regulation (2018) is an EU law on data protection and privacy for all
individual citizens of the European Union and the European Economic Area. It also
120 Digital ethics and risk management

addresses the transfer of personal data outside the EU and EEA areas. It not only
affects the way in which we communicate digitally, it regulates how businesses should
handle and process data. LSPs have become keen to show that they are GDPR com-
pliant and ask their contractees to sign that they agree. GDPR has given a significant
boost to uniformity in work practice within the EU/EEA area and beyond.
Large global organisations outside the EU/EEA area need to be GDPR compliant
because the legislation affects their business to and from the EU/EEA area. TAUS, the
independent data collection and sharing network, collects language data, metadata, and
personal data. Language data include source and target texts, metadata are associated
data, including language pairs, data about industry sector and content type; personal data
include information that will identify the person (name, email address, IP address). Since
GDPR,TAUS has introduced a different method for selecting data, it removes all meta-
data, including, for example, company names. It is important for companies to know
that their company name is no longer attached to any of their language data uploads.
GDPR has affected LSP work practice in several ways. LSPs store a large amount
of personal data, such as bank details of their contractees. They must assure that
such details are stored safely and that they have permission to store them. Group
emails must not show names other than that of the addressee. LSPs now ask their
contractees to take precautions regarding the materials they store on their devices,
and may ask for encryption of files and emails, or password protection. There is
a growing awareness on all sides that it is essential to treat all transferable data
associated with translation projects with great care.

6.4.1 Google and GDPR

Google, a large global company, well known for its MT engine, is developing
fast and its developers are aware of security and confidentiality issues. They claim
Google is GDPR compliant:

Google has been certified compliant with ISO 27018 for G Suite and Google
cloud Platform ISO 27018 (cloud Privacy). ISO 27018 is an international
standard of practice for protection of personally identifiable information (PII)
in Public cloud Services.
Google does not claim any ownership in any of the content (including text
and labels) that you transmit to the cloud Translation API.
Google 2019

Google Cloud Translation, contrary to Google Translate which stores data for
training, claims their API (7.1) is secure and that data uploaded to their MT engine
will not be stored and reused:

When you send text to Cloud Translation API, we must store that text for
a short period of time in order to perform the translation and return the
 Digital ethics and risk management 121

results to you. The stored text is typically deleted after a few days, although
occasionally we will retain it for longer while we perform debugging and
other testing. Google also temporarily logs some metadata about your Cloud
Translation API requests (such as the time the request was received and the
size of the request) to improve our service and combat abuse.
Google 2019

Clients and LSPs are concerned about the insertion of ‘we must store that text for
a short period of time in order to perform the translation and return the results to
you’, and in their terms and conditions LSPs may prohibit use of the MT engine.
The difference in GDPR compliance between open-source Google Translate and
API-controlled Google Cloud Translation is not always recognised by all parties in
the translation industry. In LSP terms of business (ToB) MT engines may be listed as
‘third parties’ that have unauthorised access to content. Furthermore, the difference
between CAT tools with secure integrated close-circuit MT engines, and CAT
tools with secure APIs, or APIs to open-source MT engines deserves more delinea-
tion in job descriptions. In her study, Berthaud (2019) discovered an unwillingness
to engage with digital ethics relating to MT. Some of her responding LSPs did not
know or want to know whether their translators use MT. This kind of ignorance
creates ethical issues. LSPs must know which tools their contractees use and know
the level of security and GDPR compliance necessary to deliver subcontracted
translations to clients with confidence. The impact of MT on translation quality is
recognised and a much more graduated view of quality in machine translated trans-
lation is growing. The digital ethics of Google Cloud Translation deserve a closer
examination rather than an unqualified decree by LSPs that MT should not be used
by their contracted translators.

6.5 Digital ethics of shared TM and MT databases

Statistical machine translation databases would not exist without our uploads.When
we download, we use shared data – source, authenticity and quality unknown.
Hence, the condition in a professional organisation’s ToB that ‘copyright must be
ensured by the client before the source material is given to the translator’ is per-
tinent. The translator has a right to know who owns the copyright of the source
text before starting the translation process and potentially uploading and sharing
(Drugan and Babych 2010).When you translate on a server with a shared TM, how
do you know if the proposed match is an original match? How many edits were
performed on the TM content before it reached your desktop? Should we query
the authenticity of an ST or a shared TM before we become party to the translation
project? A code of ethics to guide us from the moment we are commissioned to
the point of delivery and subsequent payment is essential to prevent the translation
project from becoming a thick pea soup in which no party can claim copyright or
ensure confidentiality.
122 Digital ethics and risk management

TMs and MT are merging in CAT tools and may soon be inseparable. Adaptive
machine translation (3.2) learns from our entries in the CAT tool. Programmers
and manufacturers insist that all data belong to us and are kept safe, in other words
not shared with other users. Neural Machine Translation uses encryption. Data
centres and all operations are ISO 27001:2013 certified, which means that they will
not use our translation for training purposes**(link: adaptive MT confidentiality).
Any content uploaded by users to the NMT engine, new or edited, is not shared.
Once or twice a year the NMT engine will, however, use our data together with
multi-billion-word corpora to improve the performance of their engine. Apparently,
it does not impact confidentiality or security because it is not possible to recon-
struct authentic sentences from NMT training data. Zetzsche (2019 (297th Journal))
queries whether this process does not raise a digital ethics issue considering that
other major MT systems decided not to use data for training purposes after the
introduction of GDPR legislation, if APIs**(link: adaptive MT confidentiality) are
used. If there is concern about confidentiality, an organisation can use a model based
on their server to give the utmost privacy. Translators do not have this option and
must trust that their entries in the MT corpus cannot be used by third parties.
Whatever our feelings about the digital ethics of integrated and adaptive MT
support in CAT tools, the NMT engine needs our TM entries before it can produce
and generate matches. If there might be a confidentiality requirement in the ST, we
can disable the MT function in the CAT tool.The programmers and manufacturers
claim that MT has been trained on legally acquired data, wherefore it is safe to use
without breaches of confidence. If we accept commissions with an awareness of eth-
ical risks in shared digital sources and resources, and we make our commissioning
LSPs and clients aware of our and their authoring rights, combined efforts will be
a step in the right direction. Confidentiality in digital materials is just as important
as our intellectual property rights.

6.6 Risk management

Risk is part of any translation project run by a team: perhaps the contractees do
not appear to be up to the job, or the responsible project manager has left the
team and the client is threatening action because the deadline has not been met.
Translators experience risk by not meeting the deadline due to hardware failure,
problems relating to ST, incompatible file issues … and if the source of the problems
does not even lie with the translator, they will feel aggrieved if their earnings
are disappointing in relation to their time and effort expended. We need to iden-
tify risk, learn from past experiences and find strategies to prevent harmful risk.
Gouadec (2007) suggests that detailed job descriptions supplied by the LSP could
reduce risk to both parties. Pym (2010a) builds on Gouadec’s proposed job descrip-
tion and suggests the following workable job description:

• list of materials (ST and reference files),

• function information (readership and quality profile),
 Digital ethics and risk management 123

• task information (deadlines, fees, format/CAT tool, terms of payment), and

• agreement on translator options (in/formal address, spelling, cultural adaptation).

Pym questions Gouadec’s opinion that job descriptions may lead to risk reduc-
tion, and suggests that high-risk jobs may benefit, but low-risk jobs could be left to
the discretion of the translator. Undoubtedly, translation projects requiring NDAs
need detailed job descriptions, but in real time, the focus tends to be on quick
recruitment and prompt agreement between LSP and contractees and on a smooth
transfer of digital files between users. Important details about readership and target
audience of the translation job in hand are often only supplied on request. The LSP
is not always informed by the client.
There seems to be an unspoken gentleman’s agreement that a good collaboration
between LSPs and linguists depends on trust, communication, and understanding.
Clients cannot be part of this relationship, if they do not know enough about trans-
lation to draft a translation job description. Our conclusion must be that in the
digital era a gentleman’s agreement can be a risky agreement. All parties need to
increase their awareness of security risks and breaches that could adversely affect
translation work, file storage and transfers.

6.6.1 Our security and intellectual property rights in translation

projects
Breach of security and risk factors often pass unnoticed until the damage is
done. Translations move around in electronic transfers, uploads, and downloads
24/7. Files may be sent to an incorrect email address. Confirmations of receipt
are not always sent. LSPs may request that delivered translation be deleted
from our devices to secure content entrusted to them by the client. Security
requires appropriate software against threats, malware, computer viruses,
hacking, and more, on all our devices. Lack of security may affect the content
of ST/TT.
We need to ask ourselves if intellectual property rights will protect our
translations against misuse? Are our rights in translated material regulated
by law? Does translation carry property rights (Wong in Chan 2015: 249)?
Copyright gives the creator the sole legal right to print and publish literary,
artistic, or musical material. But who is the creator: the author of the source
text or the translator of the target text? Are only literary translations protected
(Wong in Chan 2015: 249)? Translations are protected on paper and in elec-
tronic publications, but what happens to the translator’s property rights when
they are uploaded to the web? What are the legal regulations when we share
files and databases? Property rights can be shared but they must be discussed
and negotiated. Risks can have a negative impact on content. They should be
identified before we start work so we can take suitable measures. We can do
this as individuals, but ideally in collaboration with other parties involved in a
translation project.
124 Digital ethics and risk management

Food for thought…

We take risks even if we also see a positive outcome. What kind of digital risks
would an LSP be prepared to take which might affect a translation positively or
negatively?

6.6.2 Risk factors

Risk management is more than risk response, it includes risk avoidance. Risks are
not always predictable, but it is possible to recognise their potential and take steps
to reduce the possible impact. Common risks are the incompatibility of hardware
or software, loss of a file, the use of computers in public buildings with shared hard-
ware. In public buildings it is advisable to save files in a generic interchangeable
format, such as TMX and TBX, and store them in different locations, in the cloud,
on a USB device for immediate transfer, or temporarily in the Inbox of a personal
email account.
Risks need triggers to make them happen and they need circumstances in which
to occur (Lammers in Dunne 2011: 211–32). We may not be able to anticipate or
control circumstances, but we do recognise triggers. If a translation is shared among
several translators, and one or more translators miss the deadline, the final translation
cannot be delivered to the client on time. In response, the LSP could investigate
what caused a translator to miss the deadline, rather than penalise the translator,
financially or by refusing to offer future work. Other questions the LSP could ask
are: ‘Was the deadline reasonable?’ ‘Did the translators’ expertise match the type of
ST sent to them?’
Typical management risks (Figure 6.1) that may impact a translation project are
technical and project-management related (Lammers in Dunne 2011: 211–32). In
response, we can apply some of the categories in the Risk Breakdown Structure
(RBS) proposed by the Project Management Institute (US) to translation project
 Digital ethics and risk management 125

Project
Technical
management

Requirements Planning

Complexity Priorities

Technology Estimates

Resources Communication

Quality Control

FIGURE 6.1 Classification of technological and human risks in a translation project

Source: PMI Risk Breakdown Structure (PMI 2004)

management. Although we will consider it from a project management team per-

spective, the risk categories are just as pertinent to the individual translator as they
are to the organising team.
Technical challenges (RBS) for translation projects (PMI 2004) are:

• Requirements and complexity

Clients invest much time in their websites or design of products and packaging
and expect the translation to be delivered in an identical format ready for print.
Source files come with text wrap around the images. Cartons with instructions
and illustrations are provided as PDF files. A PDF of a carton contains much
small print, some of the text appearing upside down.The CAT tool conversion
could impact text chronology (Figure 6.2).
• Technology
Hardware and software (in)compatibility, inadequate IT skills, incomplete
or corrupted file transfers, power failure – the list is infinite. Can the trans-
lator or the LSP manage automated workflows on platforms, servers with
multiple agents with different equipment and in different time zones? The
following email contains a job offer to a translator which consists of a
126 Digital ethics and risk management

FIGURE 6.2 PDF of three sides of a carton with text sideways and upside down

graphics file that cannot be processed in a CAT tool and requires specialist
tools for conversion:

Dear translator,
Would you be available to translate a brochure? The source texts are EPS files and we
would require the translations to be supplied as EPS files so that we can place them
into the artwork for printing.
Email from language service provider, 2018

Should the job offer have been sent to the translator? The translator may take it
on without realising that the files are incompatible with their CAT tool…. It is
interesting to read threads on translator mailing lists and forums. Quite possibly the
technical Q&A threads outnumber terminology queries.

• Resources
Good resources include reference files, glossaries, and more, but also appro-
priate human resources (specialised translators) with appropriate tools. How
do LSPs prepare translators for collaboration in a shared translation project?
• Quality
The translation brief must state the purpose of machine and human translated texts,
target audience, locale (for example, GB or US English), etc.What does the transla-
tion brief state about quality? Is the application of QA in the CAT tool adequate?
 Digital ethics and risk management 127

6.6.3 Risk response

Challenges for translation project management (PMI 2004) from the LSP
perspective:

• Planning, prioritising, and estimates: Planning related risks are primarily

linked to the time schedule, which in turn depends on prioritising tasks. For
example, is it wise to book a reviser when a translator has not yet been found?
The other major risk is cost and time estimations: how do you cost a transla-
tion project when the client keeps returning with additions or modifications?
How do you cost a revision that takes much longer than expected? How do
you reward linguists when their cost is taking you over budget and you risk
losing the client?
• Communication: Emails are a common method to contact translators and to
respond to queries. Some are automated, others are not personalised (they start
with ‘Hi’ and are sent to many translators at the same time), and some direct the
translator to a portal where they can find information about a potential transla-
tion job. The risks are that contacted linguists do not respond to non-personal
emails or invitations to view jobs on (password-controlled) portals (7.3.7); the
risks are that project managers lose the plot due to the volume of incoming
emails from linguists with queries and attachments; confusion of job numbers
and contracted linguists.
• Control: The selection of risk factors suggested above relate to quality, scope,
efficiency, and profit. They affect agents and tools. The question always is how
can a project management team stay in control of large translation projects?
How well are they informed by their client and how well do they inform their
contractees?

Once risk and identified risk factors are understood, course of action can be as follows
(Figure 6.3): it can be decided to avoid certain situations at all costs and take preventative

avoidance

mi!ga!on
Risk response

transfer

acceptance

con!ngency

FIGURE 6.3 Risk response

Source: PMI 2004
128 Digital ethics and risk management

measures, or it is decided that risks are acceptable, either because their impact is low
and not worth the investment of time and effort, or because the risk is worth taking.
Alternatively, potential serious risks can be mitigated by introducing some measures that
are not too demanding on time and effort, or there is a transfer of risks by using, for
example, a different translator. If risks are taken seriously, it is wise to design contingency
plans so that if plan A does not work, a quick transfer to plan B can be made.
Risk management is indispensable and should be practised by anyone involved
in the translation business. Understanding risks is the first step, but the implemen-
tation of appropriate measures can also be a risk. Very often a balance is needed
between the investment of time and resources to mitigate or avoid risks and the
potential gain if those investments are made (Lammers in Dunne 2011: 211–32).
In other words, the translation provider, LSP, or translator, must prioritise risks
and make choices. Risks are not always negative; they can have a positive out-
come and be worth taking. The other much-needed balance is between the client’s
expectations and the actual role of the translator. Technology has made translators
and other linguists easily accessible, which is good news. The downside is that
working hours are not and cannot be respected if collaboration takes place across
time zones. The accessibility of MT engines and translators on the web carries the
risk that translators and MT are juxtaposed and perceived as equal but different
modes. Furthermore, translation technology is considered to lighten the translator’s
load, without any realisation that the opposite can be true: time pressure due to
increased volume, shared projects with fluctuations in consistency and quality, and
breaches in confidentiality arising from poorly regulated electronic transfer, storage
and sharing of databases.These are problems that need to be dealt with by freelance
linguists and LSP teams. Raised awareness and transparency as to what constitutes
breaches or risks would be a step in the right direction.

Project-based assignment
Objective:
Identifying digital risk and taking appropriate measures to prevent breach of
confidentiality in project-based digital translation
 Digital ethics and risk management 129

Assessment and outcome:

Peer and team assessment of risk management in a translation project

Method:
The assignment is designed for a project management team but collaboration
within a project team or between individuals is possible

Tools:
CAT-tool, and MT engine in the CAT tool (a server version would facili-
tate an examination of security breaches and/or risk in web-based shared
databases and files)

Suggested resources:
One manipulated digital source text (you may have to artificially create con-
fidential issues) of approximately 1000 words; a (manipulated) TMX and
TBX file.The ST must contain confidential issues to check the suitability of
your safeguarding measures

Language pairs and direction:

Multiple language pairs and/or directions

Assignment brief
Call your project managers and other team members together to discuss a previous
translation project: you have received a serious complaint from the client who has
discerned a breach of confidentiality. Alternatively, the press release you were asked to
translate has been leaked and consequently a competitor has announced their immi-
nent launch of a similar product. Some of the terminology that was provided to you
by the client was leaked – intentionally or unintentionally. The client has threatened
legislative action.

• You must now identify the risk factors that may have caused breach of
confidence, e.g.:
• Inadequate Terms of Business on all sides
• Your human resources database does not tell you what TEnTs your free-
lance translators use; whether they use MT in their CAT tools
• You did not ask your contractees to return XLIFF files, which contain
much metadata (e.g. about possible use of MT)
• Other risk factors
• Identify and discuss your risk responses to prevent recurrence
• Set up a new translation project, apply your preventative measures and check
to what level you have managed to avoid potential breaches or risks that were
experienced in the previous project
130 Digital ethics and risk management

• Conclude with a debrief for your team and discuss how well your contractees
responded to your new set of requirements and how well you managed risks

Possible pitfalls
• Too many potential risks – it is not possible to prevent and avoid risk entirely.
The identification of too many risks may cause you to lose focus, become
demotivated to keep agreements and cost too much time and effort
• Being too risk adverse or too risk tolerant. Try to find a workable balance
• GDPR begins in the way you communicate with your contracted linguists and
adherence must be stressed in the job description sent to your contractees

Concluding remarks
In this chapter we have discussed the digital ethics of intellectual property, copy-
right in digital materials, in TM databases and in translated digital files. We know
the risks surrounding digital ethics, we can identify the problems, but the solutions
are not always apparent. We see conflicting Terms of Business sent by LSPs and
translators. If there is an issue, the parties may prefer not to look for answers
because it requires too much effort, and blame goes to the party that signed. It is
not easy to define confidentiality, but we could accept that material is confidential
only if the client or LSP state its confidentiality. We could also check the security
of the digital sources and resources we have received. If we are likely to receive
confidential material, our TEnTs require security software and strong passwords to
prevent unauthorised access.
Our reputation could be damaged if the target text or TM is modified by a third
party and no longer matches the authentic source text which is under the client’s
copyright. Copyright also applies to attachments in emails. Copyright is difficult
to define in shared tools: the materials can be uploaded or downloaded innumer-
able times, modified, edited, revised. We must be aware of potential risks when
sharing files.
Digital ethics affect the working practice of all parties involved in a transla-
tion workflow: client, LSP, and contracted linguists. It may be appropriate to sign
a non-disclosure agreement. However, conditions that remove the rights of the
translator to their translations, should be questioned. All parties must know which
TEnTs have been used and there needs to be multilateral agreement on TEnT
usage. A global recognition of GDPR would be a solid step forward towards a
working code that is understood by all.

Further reading
Chesterman, Andrew (2018). ‘Translation ethics’. In: Lieven d’Hulst and Yves Gambier
(eds), A History of Modern Translation Knowledge. Sources, Concepts, Effects, pp. 443–8.
Amsterdam: Benjamins.
 Digital ethics and risk management 131

Drugan, Jo and Bogdan Babych (2010). ‘Shared resources, shared values? Ethical implications
of sharing translation resources.’ In: Ventsislav Zhechev (ed.), Proceedings of the Second
Joint EM+/CNGL Workshop. Bringing MT to the User: Research on Integrating MT in the
Translation Industry. Available from: https://pdfs.semanticscholar.org/4acd/2c229ef9dfa3f
a903911ed7447e62f726edc.pdf.
Kenny, Dorothy (2010). ‘The ethics of machine translation’. Proceedings XI NZSTI National
Conference.
Lammers, Mark (2011). ‘Risk management in localization’. In: Keiran J. Dunne and Elena S.
Dunne (eds), Translation and Localization Project Management, pp. 211–32. Amsterdam and
Philadelphia: John Benjamins.
Moorkens, Joss (2019). ‘Uses and limits of machine translation’. In: Ethics and Machines in an
Era of New Technologies. ITI research e-book.
Project Management Institute (2004). A Guide to the Project Management Body of Knowledge.
(PMBOK® Guide). 3rd ed. Newtown Square, PA: PMI Press.
7
WEB-BASED TRANSLATION
ENVIRONMENT TOOLS

Key concepts
• Translation environment tools move into the cloud
• Collaborative translation projects can be managed centrally when using servers
and translation management systems
• Digital platforms offer opportunities for translators and LSPs to collaborate, to
recruit linguists or source jobs and exchange knowledge
• Translation must adapt to different reading techniques of web-based material
• LSPs and translators have different views on the impact of web-based transla-
tion technology on quality, efficiency and profit

Introduction
This chapter is about the interaction between the translator and web-based tools. It
describes new trends in translation technology and examines the technologies that
may be helpful in our work. The TEnT market evolves, new tools come and go.
Translation management systems, for example, incorporate CAT features and are
gaining popularity among LSPs, because the comprehensive system enables them to
manage the workflow in the translation project from beginning to end. LSPs expect
contractees to work online in their translation management system. The translator’s
challenge is to be up to date with any web-based TEnTs we may be required to use.
We will also visit several digital platforms to find out what is on offer there. Digital
platforms are virtual meeting places where vendors and buyers of translation can
discuss matters, ask for support, look for answers, translate, localise, offer or bid for
translation jobs. They are gaining popularity among freelance translators as a virtual
meeting place where jobs are advertised, or as a resource centre and knowledge base.
The ownership of translation tools is changing too. New trends show that it is
unnecessary to invest in an expensive CAT tool with its annual maintenance fees
 Web-based translation environment tools 133

paid to the manufacturer. There is the option to subscribe to web-based CAT tools
or machine-assisted translation systems**(go to www.routledgetranslationstudies-
portal.com/ – A Project-Based Approach to Translation Technology – link:
machine-assisted translation systems) with integrated CAT functions. You are not
tied into a subscription but can opt out when required. If, however, you feel com-
fortable with your CAT tool and gain much benefit from your TM and termin-
ology databases, it is worth considering APIs and plugins which give access to many
web-based tools. In the project-based assignment you will be given an opportunity
to test translation project workflow in different web-based tools. The objective is
that your experiences with a variety of TEnTs give you the skills and competence
to try new ones and to be ready for changing trends.

7.1 Integrated translation environment

Translators have free online access to huge databases. In 2.3.3 we discussed the
downloadable TMX file with a collection of translation memories from the
European Union and United Nations.The database is managed, curated, picked, and
produced by a commercial company**(link: online TMX database). Database inte-
gration is possible in several CAT tools**(link: CAT tools with TMX integration).
It would also be useful to have integrated access to our preferred glossaries and e-
dictionaries. Switching from the CAT tool editor to open glossaries or dictionaries
in other windows on our screen(s) is inconvenient. The manufacturers, however,
like to keep control of what can be linked to their CAT programs. Dictionaries, for
example, are limited to the ones they make available on their open servers. If we
want to integrate our own glossaries, we must convert and save them in a suitable
format (CSV spreadsheet format in Excel) and import them in our Tmdb.
Meanwhile, the range of APIs and plugins in CAT tools, which gives us access to
other web-based TEnTs, is growing rapidly. The range varies greatly between
programs. API (Application Programming Interface) is a piece of software that
allows other programs to use its functions. To integrate an MT engine in a CAT
tool, you will need an API to access the MT engine you would like to use. The
API itself is free but the user is charged on a monthly basis for the number of digits
generated by the requests to the MT engine. In CAT tools we can choose from
a range of MT engines to suggest matches while we translate in the translation
editor. We pay the MT engine provider for usage, not the CAT tool manufac-
turer who offers the convenience to integrate the MT engine in their CAT tool,
unless they have developed their own baseline (generic) engine and adaptive MT
model**(link: integrated adaptive MT model).
APIs are independent connectors between different programs, whereas plugins
are offered by the CAT tool manufacturer for integration of other software in their
own programs**(link: plugin to external TMs). A plugin is a small component in a
larger piece of software and performs a specific task.The API is an external program
that can be used for access within different CAT tools or other programs, whereas a
plugin is meant for usage in a specific CAT tool and is usually made available by the
developer. MT plugins may come with an API. A wide range of different plugins
134 Web-based translation environment tools

can be downloaded free of charge. The following list of plugins shows a selection
which are provided for and by a variety of CAT tools:

A range of CAT tool plugins

File filter for complex file formats

Glossary converter to convert a glossary into a Tmdb or vice versa
Invoices and quotes for price calculation, quoting, invoicing, with a built-in time
tracker
Lookup performs web or local searches for terminology
Match bleeper gives audio hints for available matches for visually impaired
users
MT plugins for major CAT tools
Resource lookup to receive matches from the CAT tool’s server TM (API
needed)
Sketch Engine Concordance integration in the interface
Terminology to access external terminology portals, e.g. EuroTermBank
TM finder searches in the TMs of another CAT tool on your computer
or a range of public TMs**(link: plugin to external TMs)

In sum, plugins are developed to enhance the CAT tool by integrating external
programs. APIs, sometimes called connectors, are a convenient way of integrating
tools into your CAT tool but require a fee to be paid to the provider. Plugins
appear to be under-used by translators (Nunes Viera and Alonso 2018). Lack of
awareness may explain why translators tend to export their translations for a spell
and grammar check outside the CAT tool (most CAT QAs include spellchecks,
not all include grammar checks**(link: CAT grammar check). The plugin was
not initially designed for the translation market but has found a new niche in the
translation industry and we will undoubtedly see the range increase. It is good to
‘shop around’ because quality varies greatly, and so does cost. Some plugins are free,
others require a licence or API. Several CAT tools now have the microphone icon
for speech recognition (SR). It is not a new TEnT but it is to CAT tools. It may
require the use of a smartphone as a microphone**(link: CAT tool with ASR).

7.1.1 Speech recognition

Speech recognition started as a dictation tool and has become sophisticated
speech technology. Automatic speech recognition (ASR) now is a cloud-based AI
speech technology, which is available in speech-to-text (STT) and text-to-speech
(TTS) software. Both types are useful to the translator: STT generates a transla-
tion when you speak your translation in the microphone. There is no direct need
for a keyboard while you are speaking your translation if you make use of spoken
 Web-based translation environment tools 135

commands to make the STT do what you want it to do, such as ‘move down’ to
the next segment. A wireless microphone means that you do not have to sit down
and stare at your screen. Sight translation (also practised by interpreters) can be
done from your ST on a piece of paper while walking around, if you wish. TTS
allows you to listen to your translation read by an artificial voice. TTS presents an
alternative and very different way of revising your translation. It also invalidates
the criticism that segmentation in CAT tools reduces translation quality, because
TTS helps you ‘see and hear’ your translations in context. The TT is presented to
you as a coherent text. TTS helps you assess the level of fluency in your transla-
tion (Ciobanu 2019).
ASR, STT, and TTS are forms of weak AI, they do not operate like NMT, they
are not creative, they cannot produce novel utterances, but rely on historical data,
i.e. the exact data previously entered by the speaker. ASR can however be trained
by you, they can be corrected on the fly, and they can learn shortcut codes for
functions, which you would normally type on your keyboard. Of course, ASR
becomes strong AI when it is combined with NMT: a speaker at a conference
who uses ASR will see machine translated speech-to-text appear on the audience’s
tablets without time delay. NMT uses the ASR generated data to generate the
target text.
A good quality ASR program can achieve a very high level of perfection and
can be used in the CAT tool. It needs to be told and trained how to move from
one segment to the next and how to open your frequently used functions such as
concordance, lookups, find, etc. If you spot an error while you are translating, you
can correct it in real time or leave it and do an ASR or manual keyboard revision
later. Some translators claim that they can increase their translation speed by 500%
(Ciobanu 2019). This increase can only be achieved if the translator does termin-
ology research in advance, if speech recognition errors are few, if the text has a
smooth narrative with full sentences to facilitate recognition, and if the translator
can ‘sight translate’. Pauses and incomplete sentences or phrases hinder the tool
from recognising speech matches in the corpus. A paid SR program can recognise
over 86 languages, including character languages. Free online SR programs recog-
nise around 30 languages.
In the CAT tool it is possible to combine STT and the keyboard without con-
flict. If you want to increase your productivity, the ASR tool needs to be taught
commands. Your tone of voice is important for the ASR to recognise when it
is a command and not a phrase for translation. There are a few CAT tools with
integrated STT**(link: CAT tools with ASR). They call the function ‘dictation’,
and it can be enabled with or without an app.
There is still much scope for CAT tool developers to improve the audio feature,
but also for translators to understand the benefits of ASR. Adaptive technology
should help ASR work with other integrated tools. For example, MT and ASR do
not yet speak to each other, in other words, give each other priority. Consequently,
it causes conflicts when both are used simultaneously. Integrated audio data in our
tools would also offer opportunities for different kinds of immediate communication
136 Web-based translation environment tools

FIGURE 7.1 Speech-to-text and text-to-speech software

between collaborating linguists.The potential benefits of automatic speech recogni-

tion for translators are numerous.

7.2 Servers and web-based CAT tools

Major CAT tools now have server versions, which are designed to integrate all the
translation stages for a smooth workflow. LSPs manage complex translation projects
on servers or in translation management systems (TMS) (7.5). A CAT tool server
is often referred to as TMS. The difference between a server and TMS is that the
server remains a CAT tool without additional TMS features, such as administration
and accounting. TMS is part CAT tool but places more emphasis on MT and inte-
gration of the other tools.
When users connect to cloud-based apps, servers, and TMS, these are also
called software as a service (SaaS). It means that the apps or programs are
accessed and not owned by the user. LSPs use servers as a web-based CAT tool
to facilitate collaboration between project managers and contractees, the work-
flow in the translation process, apportioned files, pre-translation, the locking
or hiding of segments (to prevent revision or modification), and the sharing of
TMX or TBX files. The translator and reviser can download and upload their files
from and to the server or perform their tasks on the server, which is the usual
method, and they will not retain the translation after it is completed. Servers, run
by LSPs, are desktop-based CAT tools or cloud-based programs. Both types of
server, desktop (purchased) and cloud-based (leased), come with an agreed pool
of licences for LSPs to supply to contracted linguists for the duration of the trans-
lation project. Licences are not required for contractees who have their own CAT
tool. They are given password access to the server. Alternatively, LSPs sometimes
email project packages to the contractees: they contain XLIFF and TMX files
compatible with most major CAT programs. The translator or reviser returns the
files in a ‘return package’.
The server gives the LSP many advantages: it keeps the translation project
together, makes smooth collaboration possible, and prevents the misdirection or
loss of emails with file attachments. The LSP’s TM will absorb entries from many
translators. The downside is that a huge TM becomes unmanageable or slow and in
need of much editing. TM management is a significant task.
 Web-based translation environment tools 137

The advantages of the server for the contracted translator are significant too.
The translation process on the server is shorter and quicker because the document
files are ready for translation without the need to save, store, and import into the
CAT tool. They do not need exporting either. If collaboration between multiple
translators in a translation project is well organised, the TM and Tmdb will update
in real time and support consistency. The translator can see what colleagues have
entered in the database.Well-edited and customised TMs and Tmdbs, plus reference
material, are great resources for the translator. A temporary LSP licence supplied
to the translator means that there is no direct need for the translator to purchase a
CAT tool. If the translator would prefer to pay monthly for a licence rather than
buy a CAT tool, it is possible to lease a web-based CAT tool**(link: web-based
CAT tool).
There are, of course, disadvantages to translating on a server or web-based CAT
tool for the translator, such as potential unfamiliarity with the CAT program, an
inability to link or integrate your usual TEnTs, such as dictionaries, your TM and
Tmdb databases, or MT API. If the web-based CAT tool has an offline editor, the
file can be downloaded as an XLIFF file from the cloud and translated in the off-
line editor. This allows you to work beyond a wifi signal. It is possible to down-
load the file from the editor into your own CAT tool before returning it to the
web-based tool via the offline editor**(link: web-based CAT tool link to personal
CAT tool).This procedure does not allow you to benefit from matches in the web-
based TM: they are only accessible online. When translating in the offline editor,
matches can be checked once the translation is uploaded to the web-based CAT
tool. It is not a straightforward method for the translator, whereas the LSP enjoys
an automated workflow without having to manage attachments in emails. Working
on the server raises two questions: whose are the property rights associated with
the translation and TM entries, and how can the translator make convenient good
use of their own resources?
Web-based CAT tools are becoming more sophisticated: as translation man-
agement systems and in portals or on platforms. At this level of sophistication, the
main market consists of companies, organisations, and LSPs. The translator risks
becoming a confused user with the reins taken from them. These developments are
important and in the following sections we will discuss various aspects of working
in the cloud.

7.3 Digital platforms

A digital platform is a technology-enabled business model which facilitates
exchanges between multiple parties, such as clients, agents, and linguists in the
translation industry, who do not necessarily know each other. The parties either
offer or require services that need to be managed efficiently and offer quality. They
are typically easily accessible to all. In everyday life, we are familiar with platforms
for advertising, marketplace, media, social media, knowledge, and crowdsourcing.
Translation platforms are similar but have different faces depending on their
138 Web-based translation environment tools

purpose and user groups. We will concentrate on two types: the business platform
(and portal) for translation and crowdsourcing where translations are produced and
the platform where translators meet, which can be a mix of advertising, market-
place, social media or be knowledge based.

7.3.1 Translator platform

When paper or digital resources do not provide the required term translation, free-
lance translators frequently turn to human resources. Their requests need not only
be about terminology but also for information about LSPs, such as their payment
reliability, CAT tool problems, etc. Translator platforms have emerged on social
media, as web groups and through professional translator organisations**(link: pro-
fessional translator organisations). There are specialist web groups and mailing lists
where translators with the same language pairs, CAT tools or specialisms can post
their queries and receive or find answers.
Professional translator and interpreter organisations in all countries aim to
promote not only high standards in the translation industry, but also to support
their members in many different ways: with advice (practical, legal, etc.), training
and career development, professional examinations and qualifications, infor-
mation, regional, language, and specialist support groups. Their websites may
act as an interface between those who need and those who offer translations.
Translators must apply for membership, which will be granted if they meet the
required standards. Full membership may require the candidate to pass a trans-
lation examination set by the organisation and to demonstrate experience as
a practising translator. Members will then have their contact details, working
languages, and specialisms added to a website which can be accessed by any
individual or organisation that wishes to commission a translation. To safe-
guard standards and quality, the organisations may ask their members to record
their CPD (continuing professional development) activities, which consist
of training in any format, for example, by attending professional webinars or
conferences, by authoring features in journals, organising translator events, etc.
Increasingly, LSPs are requesting their contractees to send them a list of recent
CPD activities.
Translator platforms are primarily supportive and intuitive digital platforms
without financial support or top-down organisation. A different kind of translator
platform is Translators without Borders (TWB), which is a non-profit organisation
that offers language and translation support for humanitarian and development
agencies and non-profit organisations on a global scale. Their work is supported by
volunteers, grant funders, donors, and sponsors. They provide language services as
one of the components where an emergency has arisen. They work with volunteer
translators worldwide who may provide translations or train translators in the crisis
area to deal with the situation themselves. The organisation runs predominantly
on a platform**(link: non-profit translator platform) with only three offices across
the globe.
 Web-based translation environment tools 139

7.3.2 Marketplace platform

A very different kind of digital platform is the marketplace. It is a platform that
can be used by translation buyers to post translation jobs or browse a translator’s
directory and it is a platform where translators can apply for jobs or bid for them.
It is called a marketplace platform because its main purpose is offering and buying
translation services. Buyers may choose to use the platform to attract multiple
translators for one job. It enables them to select the most suitable translator in terms
of pricing, availability, and specialism. This kind of platform (generally password
protected) can ask competing translators registered with the agency to bid for the
job, and the buyer, an LSP, can select the translator with the lowest bid, which gives
the LSP more profit. It is distinctly different compared to the previous platforms
because of its financial and business/marketing aspects.
The first marketplace digital platform was set up in 1999. It has seen a succession
of other marketplace platforms**(link: marketplace platforms). They have gradually
become more than a commercial marketplace and are now virtual translator com-
munities. They offer a wide range of TEnTs that can be purchased (with or without
discounts): conversion tools, EU TMX files and TBX files with specialist glossaries,
software downloads, and more. Translators appreciate the discussion forums and the
online glossaries. Membership is graded, either free or paid. The platform operates
according to a reputation system: the more recommendations a party receives from
agents or linguists, the more stars they have, the better their perceived reputation.
Like any marketplace, vendors and buyers can be good, bad, and indifferent. On the
platforms it is recommended that you check the status of any organisation or indi-
vidual requesting a translation service and that you do not post personal details or CVs
without protection against identity theft. It is a problem when translators post their
CV on the platform to find that their data, including their professional qualifications,
are stolen and to see them offered in fake profiles by an unprofessional translator (a
scammer), usually to win bids. Fraud can also be committed by individuals who present
themselves to the translator as an LSP, for the translator to find that their translation is
accepted without any hope of payment or recall because the contact details are fake.
These situations can be prevented by checking ‘blacklists’ (often on the platform) and
testing all details and website data about the buyer, prior to accepting a commission.

7.3.3 Localisation platform

Web localisation and quality were discussed in 5.2.3. Tools for localisation can
be linked to the translator’s CAT tool (or integrated), but the platform takes the
automation process one step further. On a localisation platform source content
and translations are automatically synced to a global content database that can be
accessed by users at any time. Different teams can work and communicate dir-
ectly on the platform with translators and project managers. It is an open and
connected platform which integrates into a translation project managed by an LSP.
The platform is a business model that offers on the fly localisation so that translators
140 Web-based translation environment tools

FIGURE 7.2 File-based or cloud-based localisation

working on live content in the development cycle, can keep up with language
updates.The translator can either push content for translation, i.e. upload files to the
platform, where localisation will be performed, or use the pull approach (4.6) and
consult the platform database.
When localising website material, the choice is between a file-based solution
or a cloud-based solution. Figure 7.2 shows potential workflows in a simple local-
isation activity. Analogue twelve-hour clocks as used in the UK must be localised
to digital 24-hour clocks. The manufacturer may deliver the ST in an i18n version
and the translator can perform l10n in a CAT tool (file-based) by using an API.
Alternatively, the process can happen in the cloud (cloud-based) in TMS (7.5).
A cloud-based platform requires an API to obtain access, and payment is based
on usage**(link: localisation platform). In this case, the localisation function is
automatically integrated in translation management systems (7.5).

7.3.4 Translation platform

A translation platform is a fully automated platform without translators. Translation
platforms rely heavily on MT to be fully automated. Web-based CAT tools are
offered as a translation platform to companies and businesses whose websites
need localising. The web-based CAT tool on the platform can pre-translate
documents with NMT and the client can then turn to an LSP, or directly to a
 Web-based translation environment tools 141

translator, for PEMT and/or translation of any segments beyond the scope of the
tool**(link: online open source adaptive NMT platforms).

7.3.5 Knowledge platform

A knowledge platform is a platform that enables an organisation, such as CAT tool
developers to manage formal learning and to share knowledge in a single location.
The exchange of knowledge may also be between users of a CAT tool if the owner
of the platform has enabled the posting of messages in a Q&A section. Whereas
manuals and Help features were the way to help users understand and learn how to
use CAT tools in the early years, the Help function in the CAT tool is now more
likely to take you to the manufacturer’s ‘Knowledge base’, where you can find
answers, watch supporting videos, and where you are led to the appropriate Help
section that offers solutions in response to your query. If you have signed up to
the CAT tool, you may be sent invitations to attend webinars to learn more about
new features in the tool. Many digital platforms have a knowledge-sharing aspect
on their platform, which is a helpful development for CAT tool users.

7.3.6 Crowdsourcing platform

Cronin (2012) makes the following observation about crowdsourcing: it is a
consumer-oriented translation activity in which the consumer becomes the pro-
sumer (producer+consumer). This style of translation activity upsets the traditional
norms in which the LSP is productivity oriented. Crowdsourcing happens when
an individual or an organisation use a platform to appeal to crowds for free or low-
cost translation submissions. Instead of asking traditional service providers, they
encourage the online community to deliver translations.The platform makes it pos-
sible. Facebook was launched in 2008 and two years later their website was available
in 75 languages (Drugan 2013: 174). The best way to describe this phenomenon is
that an online community of friends decided to produce the Facebook translations
for the benefit of the community. Their act of translation was selfless and not for
financial gain. In 7.3.1 we saw how an NGO uses a platform for a good cause,
which would not be possible without the willingness of translators to translate free
of charge. It is a type of crowdsourcing that is called cause-driven (Jiménez-Crespo
2013: 193–7). Although the main difference between translator and crowdsourcing
platforms seems to be that crowdsourcing involves non-professionals, there are
now commercial companies that have embraced crowdsourcing. One company
with major LSPs introduced a paid crowdsourcing platform in 2013**(link: paid
crowdsourcing platforms), which is called ‘outsourced crowdsourcing’ (Jiménez-
Crespo 2013: 193–7) or ‘people-powered’. The company uses a crowdsourcing
platform which operates in a similar way to the marketplace platform, the main
difference being that responding translators need to be registered with the
crowdsourcing company. The company claims that it delivers ‘high quality results’.
It does not give details to confirm that it is ISO certified, including QA and third-
party revision. The emphasis is on convenience and speed.
142 Web-based translation environment tools

Crowdsourced material is predominantly web-based. Readers have a different

approach to reading webpages, it is nonlinear, superficial (skimming), and accelerated
(Pym 2010b). Sometimes the reader is looking for keywords and a ‘gist translation’
is all that is needed. Quality does not matter. Interestingly, the collaborative aspect
of crowdsourcing does in fact improve the quality of lay translation through col-
laborative correction (Pym 2010b). It would be wrong to say that a crowdsourcing
platform is inferior and for amateurs only. Although crowdsourcing on a commer-
cial basis is growing, it does not receive much attention from academics (O’Hagan
in Kenny 2017), whereas the collaborative workflow with translation and revision
is resourceful and supports quality in such a way that even the professional should
consider the values of voluntary collaboration, such as peer revision.

7.3.7 The portal

The operations on a translator portal are like those on a platform. The main diffe-
rence is that a portal is dedicated to a company or LSP. It is a website that provides
access to, for example, an LSP and its functions. First you log in to a portal, create a
password and complete a personal profile. The LSP will then warn you by email if
there is a job available which you can accept (or reject) on the portal and download.
When the translation is completed you will upload it to the portal. Invoicing has
become an automated feature on the portal, and it is not necessary to create your
own invoice. The portal is not a TEnT, it is an administrative tool. Large LSPs may
even use their portals as marketplace platforms and ask you to bid for the offered
job, as explained in 7.3.2.The styles of portal and their intent depend greatly on the
ethics of the companies or agencies that run them. They are not regulated.

Food for thought…

Discuss the different digital platforms and their benefits from the translator’s
perspective.
 Web-based translation environment tools 143

7.4 From content to translation management systems

Before we discuss the translation management system, which shares characteristics
with the translation platform, we must consider what gave rise to its devel-
opment. The volume of documents for translation continues to increase, and
so does competition. It means that LSPs and clients want to manage volume
efficiently and keep prices down. Digital content and volume begin with the
content management system (CMS), usually managed by the client, which
is a software application that allows users to collaborate in the creation, the
authoring and editing of texts. It enables the production of digital content,
such as webpages, websites, blog posts, etc. CMS is morphing into WCMS (web
content management system): content no longer consists of hard-copy files but
includes emails, mobile apps, social media, websites and more. Because of the
complexity of its content, CMS (WCMS) has become a management system,
which originally began as a business presentation system with PowerPoint files
for product presentations. Today, the entire content is digital, including multi-
media, and requires smart exposure on the web with SEO (Search Engine
Optimisation (7.6)).
Translation happens in the TMS, which can be either connected to the WCMS
or operate as an independent system.A marketing organisation will want to optimise
successive communications in different formats for laptops, tablets, and smartphones
and in many different languages, which can all be managed collectively in the TMS.
Optimisation increases the number of hits on the web and the SEO requirement is
an additional challenge for the translator, regardless of the system used.
It is understandable that clients prefer to outsource comprehensive content
management, including translation and localisation, instead of managing it them-
selves. They move their entire digital presence to global CMS platforms. LSPs have
jumped on the bandwagon and through API connectors they can align client con-
tent across multiple platforms globally in the relevant languages. Connector packs
are sold as ‘integration solutions’ and they enable LSPs to manage the clients’ CM
systems in their translation management systems. Here is an internet advertisement
for a TMS that can be connected to a CMS:

Request, manage, track, and view the progress of your translations

With the client portal you have on-demand access to SaaS tools and reports for
requesting, managing, tracking, and viewing the progress of your translations.
It’s easy to use and navigate and provides enterprises with a portal they can share
with internal teams for requesting translations and checking the status of translation
projects. Includes API for custom system integration.
WWW advertisement of a translation management system 2019

7.5 Translation management systems

A translation management system (TMS)**(link: TMS systems) supports complex
translation projects. Its main features are workflow management, translation memory
144 Web-based translation environment tools

and terminology database management, integrated machine translation (MT), and

administrative functions. A TMS is more than a CMS, because of its translation fea-
ture. An update of a company’s website content in the TMS can be assigned to LSPs
or teams where the members are not necessarily located in the same office or even
at the same location, but in different countries and time zones. Many TMS programs
include APIs for a range of MT providers and use adaptive MT. Businesses can use
TMS as a plugin in their CMS, which is useful if content needs to be localised.
Localisation is highly automated in the TMS and will be done without linguists.
There are several types of TMS, but the two main types are desktop-based or
software as a service (SaaS) cloud-based systems. SaaS is also known as on-demand
software which provides the same or enhanced capabilities compared to in-house
software. The main benefit is that companies pay a flat monthly fee for their use
of the software and can leave the service any time. The provider selling the service
is responsible for security, maintenance, and global accessibility. TMS functions are
increasingly included in major CAT tools on their servers. They are language cen-
tric, with the typical functions needed by project managers, translators, and revisers.
SaaS versions have the advantage of being totally cloud-based, available 24/7 with
browser-based access. They do not need installing but offer a licence model where
the user pays for usage (Shuttleworth in Chan 2015: 680–1).
The objective of a TMS is to automate the translation process. It makes a com-
plex translation project more controllable and avoids repetitive tasks. A project man-
ager can oversee multiple projects, multiple freelancers, check delivery times, check
progress, and manage terminology requirements. If the TMS has an on-demand
platform or a portal, the project manager can oversee progress because all tasks
are automated. The TMS has many attractions for the LSP for a variety of reasons.
Online systems do not require software installation, or licences for subcontracted
translators. They are said to be intuitive and easy to manage by the user. The TMS
is generally integrated with a global portal-like platform (7.3.7) which allows the
translator to receive, perform, and deliver translation jobs irrespective of their loca-
tion or software. Purchase orders and invoices are managed on the platform. TMs,
glossaries, source files, and other project materials are readily available to the trans-
lator. Downloads, uploads, and emails are no longer necessary when working in the
cloud. All materials are accessed on the platform. When you, as a contracted user,
complete your translation or revision job, you simply change the status button from
‘in progress’ to ‘completed’ to inform the project manager that the job is done.You
can then move on to the invoice tab and click on a ready-made invoice for you.
With a click, your invoice is sent, and your translation is in the cloud.
The TMS has the following business functions to suit the project manager:

• project planning
• workflow management
• integration of TEnTs for large-scale translation activities
• coordination of all contributors within, outside and across organisations
• automation of repetitive tasks
• reduction in file management and transfer
 Web-based translation environment tools 145

• lower risk of errors

• huge data access
• smooth workflow in integrated system
Sargent and DePalma 2007; Chamsi in Dunne 2011:
61–2; Shuttleworth in Chan 2015: 550–62

The TMS has the following translation features to suit the translator:

• segments and terms entered by one translator can be reused by all

• it facilitates collaboration with colleagues
• translation platform systems centralise databases, maintenance, simplify and
automate tasks
• it is an interface between the translator and adaptive MT
• it offers choice between CAT tool ownership and TMS subscription
Garcia in Chan 2015: 80

The main selling point of TMS is adaptive MT: whereas CAT tools offer MT
access as a plugin, the primary TMS focus is on its access to an MT engine.The TM
and Tmdb databases are integrated in the TMS. The TMS needs the two databases
to make the MT adaptive. The MT engine learns from your entered TUs in the
TM and adjusts as you translate. TMS developers believe that TMS tools will offer
such high-quality MT that the user will not even have to post-edit (Zetzsche 2017
(281st Journal)). The translator edits on the fly, confirms and adds TUs to the TM,
and the MT learns from the TM. The other major difference between CAT tools
and TMS is the latter’s inclusion of business management functions.

Food for thought…

Discuss potential implications of the translation management system regarding cog-

nitive ergonomics (1.6.3, 1.6.8).
146 Web-based translation environment tools

7.6 Localisation of webpages and search engine

optimisation
The localisation of commercial webpages includes SEO (Search Engine Optimisation)
in the target language (TL). SEO is a technique used to move websites to the top of
search engine rankings by means of search words or keywords. Keywords, also called
seeds, are the words the website visitor is likely to type in the search box. A seed list
consists of baseline keywords and phrases most relevant to a domain. In US women’s
fashion, for example, the search volume of ‘dresses’ is 1,830.000, cocktail dresses
rank 450.000 hits and plus size dresses rank 301.000 hits. In another country, the
ranking order will be entirely different and ‘cocktail dresses’ may not be a term used.
The localisation and translation of commercial websites do not only concentrate on
keywords but also on metadata, which are the data about the webpage under the
URL address. Metadata must also be optimised. Another point of consideration is
that countries use specific search engines that rely on different algorithms to rank
websites. For the translator, it is advisable to use the search engine in the target
country/language to discover the ranking order of terms.
LSPs tend to list SEO skills under translator competencies. There are however
many tools to assist the translator, ranging from translation management systems to
plugins, Google’s tools, and more. The objective of this book is not to cover the
field of localisation and SEO, it is rather to give an understanding of the nature of
webpage translation and an awareness of supportive tools.
When translating a webpage with SEO, the translator has two objectives: first,
the keywords must be such that the search takes the visitor to the website, and
second, the keywords must hold the visitor’s interest and keep them reading and
clicking. SEO translations require thorough keyword research, an understanding
of marketing, in-depth SEO knowledge, and good copywriting skills. There are
many keyword-research tools available on the web, some Google based. ‘Google
Analytics’ is a tool that gives a better understanding of the customer’s search
behaviour. For example, readers of webpages generally stay less than 10 seconds
on the page, their reading is nonlinear and has taken on a form of accelerated
power browsing, with a strong visual component (Cronin 2012). The way readers
approach web-based material is changing: skimming and scanning of web texts
is becoming the norm. It affects the way texts are composed and subsequently
translated. Reading for gist and less interest in text quality must be taken into
consideration by author and translator.
SEO optimisation in Google is performed as follows:

• Google uses crawlers, programs that systematically search the World Wide Web
to collect data. Based on those data, Google creates ranking algorithms
• Algorithms determine the ranking order of a website: at the top of the
page, at the bottom, or on pages 2, 3, etc. In order to push a website up the
ranking order, visits can be influenced through suitable website content,
appropriate length, relevance and keywords or keyword phrases
 Web-based translation environment tools 147

• The Google analytics tool checks the following:

• how long and how many customers stay on the site
• bounce rate (do customers look at one page or move on to other pages?)
• metadata: the information put in and under the URL address
• website security: the lock symbol confirms a secure website, [i] means
insecure
• backlinks to other websites and backlinks on those webpages to your
website.
ITI webinar UK by David García Ruiz 2019

Although SEO based on a seed list is not a creative form of translation, the client
who needs web translation may ask for transcreation rather than translation,
including SEO.Transcreation (translation + creation) is a marketing term that refers
to the adaptation of ST to localised TT. One LSP (2019) referred to transcreation
as a ‘creative and adaptive translation’. Another LSP (2019) described transcreation
on their website as a ‘hybrid of innovative, culturally adapted content and straight-
forward translation, in which the sense or feeling of the original must be retained in
the target language’. The question worth asking is whether these features have not
always been part of translation.

7.7 Translation technology and the users

The European Union of Association of Translation Companies (EUATEC) held
a survey in 2019 about the expectations and concerns of the European lan-
guage industry**(link: 2019 Language Industry Survey report). They received
1404 responses from 55 countries, many outside Europe, from companies, LSPs
and translators. The objective of the survey, which was also held in 2017 and
2018, is to gain a sense of mood among translators and LSPs and to iden-
tify trends and concerns. We will focus on findings that relate to the use of
web-based TEnTs:

• What clients want

• Technology improvement wish list
• Technology investment plans
• Tool familiarity and usage
• CAT user rights and ownership

What do clients want? The designers, developers, and manufacturers of CAT tools
believed their tools delivered the mantra of less time, more profit, better quality. LSPs
and translators agreed that quality of service came first, followed by responsive-
ness, meaning prompt delivery, the quality of the translation, and flexibility. It is
interesting that price, or rather low cost, was not considered a top priority for the
customer by translation providers. LSPs and translators disagreed about the lan-
guage focus which they thought was expected by the client: translators reported
148 Web-based translation environment tools

domain specialism, whereas the LSPs believed the client was looking for a broader
spectrum of language services.
When it comes to technology improvement wishes vary. Translators would like
more ease of use of their tools and a lower cost of ownership, whereas the LSPs
would like to see better integration, better MT quality. They agreed on the cost of
ownership of tools, which both parties considered too high.
The gap between LSPs and translators widened in terms of tool-investment
plans: LSPs scored 62% for machine translation (cf. 16% for translators) and 52%
for automated workflow (cf. 6% for translators). QA tools were not popular with
translators, contrary to LSPs, which suggests that translators are not (yet) convinced
that QA tools offer added value.
Tool familiarity was on a par across the board: MS Office, major CAT tool and
Google Translate were among the top five. The difference between familiarity and
usage, however, was significant. Google Translate dropped from high familiarity to
50% among LSP users and to 75% among translators as a frequently used tool.
LSPs do not appear to recognise the different levels of confidentiality risks between
open-source web-based MT engines (high risk) and API access with adaptive
use (low risk). In the survey, concerns for breaches of confidentiality and inferior
quality were high among LSPs. The following communication from the LSP to
their contractees confirms these concerns:

Instructions to the translator:

… It is not permitted to copy or partly copy the source texts we send you and
to enter any part into machine translation engines on the internet. Nor is it
permitted to upload the source texts, wholly or in part, via MT functions in
TM software to those machine translation engines. …
LSP communication to contracted translators 2019

Translators gave a higher rating to Linguee, a web concordance, compared to LSPs.

The major CAT tools came first among both parties, LSPs and translators, in terms
of usage. TMS tools were more frequently used by LSPs. They hardly featured as a
frequently used tool among translators.
The 2019 survey results for CAT user rights and ownership in respect to transfer
to the client had not changed compared to the two previous years. Translators
objected to transfer, whereas the larger LSPs did not hesitate to pass TM rights on
to their clients. The survey showed a sharp rise in the level of transfer of TM rights
by LSPs. Among translators there was a consistent decline in the transfer level of
TM user rights, followed by the transfer level of TM ownership. The transfer levels
of terminology user rights and ownership showed a very slight increase among
translators, whereas among LSPs, they showed a sharp decline. This finding is not
entirely surprising: terminology databases remain potentially underused. The reli-
ance on the TM and the concordance feature, and the effort required to build a
terminology database or glossary, may be a root cause.
 Web-based translation environment tools 149

Two main points stood out in the survey: translators are not habitual users of
TMS, and although translators covet their TMs, they are seemingly generous with
their terminology databases. LSPs show remarkably little interest in ownership and
transfer of terminology databases. Both points are closely related with ethics and
property rights. Contracted translators that work in web-based tools, such as the
TMS, do not have ownership of their work.The transfer of user rights is very much
beyond the translator’s control, especially if a contract agreement is signed in which
the translator is requested to give away their rights. The seemingly involuntary
transfer of ownership and user rights deserves a wakeup call among translators.They
could lay claim on the industry’s mantra and take ownership of translation quality,
efficiency, and profit by informing the client and LSP what they can deliver within
the given parameters.

Project-based assignment

Objective:
Planning, implementation, and management of your own web-based trans-
lation project through teamwork and contracted translators with new
cloud-based TEnTs

Assessment and outcome:

Team- and peer-assessment after completion of the project. Peer evaluation can
be arranged on a platform where team assessments are posted. Responses
can be social media style, such as threaded discussion, informal language.
The primary team assessment is based on some or all web-based activities
suggested below.

Method:
The assignment is designed for a project management team but collaboration
within a project team or between individuals is possible

Tools:
Server or TMS; CAT tool with internet access; ASR
150 Web-based translation environment tools

Suggested resources:
HTML source text for localisation

Language pairs and direction:

As required

Assignment brief
In this assignment you must

• apply all your management skills, design your own team translation project
and run it on a TMS or server. Alternatively, use a CAT tool but access digital
platforms, or create your own (marketplace? crowdsourcing? portal?) platform.
Your choice depends on what is available to you
• apply any new TEnTs available to you: TMS, CAT + server, plugins, APIs,
speech recognition
• recruit translators and revisers in the cloud on existing platforms or on a self-
created platform
• apply localisation skills (SEO) to the text through recruited translators, through
localisation tools, or in your team.

Do not forget a plan and workflow at the beginning and the quality assessment of
the translation at the end of this assignment.
Make sure that you can tick the following activities in your evaluation of your
project management skills:

• Design and implementation of workflow in TMS (see 1.7.2)

• Assessment of shared TM/Tmdb benefits
• Assessment of integrated (adaptive) MT quality
• QA and third-party revision
• Use of digital platforms
• Crowdsourcing of (untrained) translators (from undergraduate language degree
courses?)
• Quality control of recruited translators via platforms
• Digital ethics
• Administration and accounting
• Other TEnTs (ASR).

Possible pitfalls
• Impact on quality by unqualified translators
• Impact on quality as a result of incompetent usage of TEnTs or platforms
• Impact on product (translation) as a result of project manager incompetence
 Web-based translation environment tools 151

• Stagnation in the workflow if project management team members are not sure
about their roles

Concluding remarks
How does the industry’s mantra sit among freelance translators and LSPs? Data
for this assessment are based on interviews in 2018 with translators and LSPs
conducted by Lucas Nunes Vieira and Elisa Alonso (University of Bristol UK)
in collaboration with the Institute of Translation and Interpreting (UK) and the
Western Regional Group of the Institute of Translation (UK). The interview
questions related to MT output. Bearing in mind that the translation management
systems discussed in this chapter use MT as their primary tool, the answers in the
survey are pertinent.
The mantra of quality was considered most important and yet also most prob-
lematic: the concept of quality needed defining and quality needed negotiating.
LSPs would ask translators to post-edit within a given time. It was not a problem
if MT output was good but detrimental to final quality if the output was poor.
Translators felt that their sense of quality and their pride in delivering quality was
under attack.We have seen that adaptive MT in TMS can resolve this problem if the
translator can select and modify proposed MT matches on the fly rather than per-
form a post-edit. None of the interviewees stated that MT provided better quality.
The mantra of productivity saw an imbalance. LSPs did not see initial gain
in productivity when the MT technology was introduced, but improvement after
the arrival of NMT increased productivity by more than 100%, from 2000 words
a day to 4000 to 5000 words. Translators commented that they saw little gain if
MT output was poor. In this respect adaptive MT in TMS may not be a temporal
improvement if the translator is faced with having to make too many match choices
and too many edits. They are time-consuming.
The mantra of profit is closely related to the experiences described by LSPs
and translators about MT quality. The LSPs appreciated the increased translated
word rate, and the translators felt that post-edits were more profitable than trans-
lation if the quality was good. It was agreed that the pricing structure of word
rates did not suit hybrid translation when MT was used. Translators were not
happy with alternatives such as pricing according to edit distance because they
felt it did not account for effort. Time spent on thinking was not part of the
calculation.
A pertinent conclusion in the survey was that in large collaborative projects
translators felt like a small cog in the wheel with little say, especially when they were
expected to work within a given structure or procedure, such as TMS or platform/
server. Translators stressed the human side of translation, which meant that they
considered feedback important and that communication between all parties was
of prime importance. They would like the industry to listen and respond to their
concerns about TEnTs they want to use or are expected to use.
152 Web-based translation environment tools

Further reading
Cronin, Michael (2003). Translation and Globalization. London and New York: Routledge.
GALA (Globalization and Localization Association) www.gala-global.org/what-translation-
management-system
Jiménez-Crespo, Miguel A. (2013). Translation and Web Localization, pp. 193–7. London and
New York: Routledge.
Nunes Vieira, Lucas and Elisa Alonso (2018). ‘The use of machine translation in human
translation workflows. Practices, perceptions and knowledge exchange’. Milton Keynes:
Institute of Translation and Interpreting.
O’Hagan (2017). ‘Crowdsourcing and the Facebook initiative’. In: D. Kenny (ed.), Human
Issues in Translation Technology, pp. 25–44. London, UK: Routledge.
Pym, Anthony (2010b). ‘The Translation Crowd’. Tradumàtica 8. www.fti.uab.cat/
tradumatica/revista/num8/sumari.htm# [accessed November 2019].
Shuttleworth, Mark (2015). ‘Translation management systems’. In: Chan Sin-Wai (ed.),
The Routledge Encyclopedia of Translation Technology, pp. 687–91. London and New York:
Routledge.
BIBLIOGRAPHY

Anderson, Lorin W. and David R. Krathwohl (eds) (2001). A Taxonomy for Learning, Teaching,
and Assessing: A Revision of Bloom’s Taxonomy of Educational Objectives. New York: Addison
Wesley Longman.
Austermühl, Frank (2001). Electronic Tools for Translators. Manchester: St Jerome Publishing.
Baker, Mona (2018). In Other Words: A Coursebook on Translation. London: Routledge.
Baker, Mona and Gabriela Saldanha (eds) (2009). Routledge Encyclopedia of Translation Studies.
London: Routledge.
Baroni, Marco and Sylvia Bernardini (2004). BootCaT: Bootstrapping corpora and terms
from the web. Proceedings of LREC 2004.
Berthaud, Sarah (2019). ‘Ethical issues surrounding the use of technologies in the translation
and interpreting market in the Republic of Ireland’. In: Ethics and Machines in an Era of
New Technologies. ITI research e-book.
Bowker, Lynne (2001). ‘Towards a methodology for a corpus-based approach to translation
evaluation’, Meta, 64: 345–64.
Bowker, Lynne (2002). Computer-aided Translation Technology: A Practical Introduction.
Ottawa: University of Ottawa Press.
Bowker, Lynne (2005). Productivity vs quality? A pilot study on the impact of translation
memory systems. Localisation Focus, 4(1): 13–20.
Bowker, Lynne (2015). ‘Computer-aided translation. Translator training’. In: Chan Sin-
Wai (ed.) The Routledge Encyclopedia of Translation Technology, pp. 88–119. London and
New York: Routledge.
Byrne, Jody (2014). Scientific and Technical Translation Explained: A Nuts and Bolts Guide for
Beginners. Hoboken: Taylor & Francis.
Chan, Sin-Wai (2015). The Routledge Encyclopedia of Translation Technology. London and
New York: Routledge.
Chan, Sin-Wai (2017). The Future of Translation Technology. Towards a World without Babel.
London and New York: Routledge.
Chaume, Frederic (2012). Audiovisual Translation: Dubbing. Translation Practices Explained.
Manchester: St Jerome Publishing.
Chesterman, Andrew (2001). ‘Proposal for a hieronymic oath’. In Anthony Pym (ed.), The
Return to Ethics, special issue of The Translator, 7(2): 139–54.
154 Bibliography

Chesterman, Andrew (2018). ‘Translation ethics’. In: Lieven d’Hulst and Yves Gambier
(eds), A History of Modern Translation Knowledge. Sources, Concepts, Effects, pp. 443–8.
Amsterdam: Benjamins.
Ciobanu, Dragoş (2019). ‘Speech technologies: the latest word in AI-driven game-changing
language technologies’. In: Ethics and Machines in an Era of New Technologies. ITI research
e-book.
Cooper, Alan (2004). The Inmates are Running the Asylum: Why Hi-Tech Products Drive Us
Crazy and How to Restore the Sanity. Indianapolis: Sams Publishing.
Cronin, Michael (2003). Translation and Globalization. London and New York: Routledge.
Cronin, Michael (2010). ‘The translation crowd’. Revista.Tradumàtica 08. Catalonia: UAB.
Cronin, Michael (2012). Translation in the Digital Age. London: Routledge.
Daems, Joke, Lieve Macken, and Sonia Vandepitte (2013). ‘Quality as the sum of its parts: a
two-step approach for the identification of translation problems and translation quality
assessment for HT and MT+PE’. In: Sharon O’Brien, Michel Simard, and Lucia
Specia (eds), MT Summit XIV Workshop on Post-editing Technology and Practice, Proceedings,
pp. 63–71. European Association for Machine Translation.
Doherty, Stephen (2017). ‘Issues in human and automatic translation quality assessment’.
In: D. Kenny (ed.), Human Issues in Translation Technology, pp. 131–48. London: Routledge.
Doherty, Stephen and Joss Moorkens (2013). ‘Investigating the experience of translation
technology labs: pedagogical implications’. JoSTrans 19. www.jostrans.org/issue19/art_
doherty.pdf
Dragsted, Barbara (2005). ‘Segmentation in translation. Differences across levels of expertise
and difficulty’. Target, 17(1): 49–70. John Benjamins. https://doi.org/10.1075/
target.17.1.04dra.
Drugan, Jo (2009). ‘Intervention through computer-assisted translation: the case of the
EU’. In: J. Munday (ed.), Translation as Intervention, pp. 118–37. London & New York:
Continuum International Publishing Group.
Drugan, Jo (2013). Quality in Professional Translation: Assessment and Improvements. London/
New York: Bloomsbury Academic.
Drugan, Jo and Bogdan Babych (2010). ‘Shared resources, shared values? Ethical implications
of sharing translation resources’. In Ventsislav Zhechev (ed.), Proceedings of the second
joint EM+/CNGL workshop. Bringing MT to the USER: Research on Integrating MT in the
Translation Industry. Available from: https://pdfs.semanticscholar.org/4acd/2c229ef9dfa3f
a903911ed7447e62f726edc.pdf.
Dunne, Keiran J. and Elena S. Dunne (eds) (2011). Translation and Localization Project
Management. Amsterdam and Philadelphia: John Benjamins.
EAGLES-EWG (1996). Eagles Evaluation of Natural Language Processing Systems,
Final Report EAG-EWG-PR.2, Project LRE-61-100, Center for Sprogteknologi,
Copenhagen, Denmark. [available at: www.issco.unige.ch/projects/ewg96/]
Ehrensberger-Dow, Maureen and Gary Massey (2014). ‘Cognitive issues in professional
translation’. In: John W. Schwieter and Aline Ferreira (eds), The Development of Translation
Competence;Theories and Methodologies from Psycholinguistics and Cognitive Science, pp. 58–86.
Cambridge: Cambridge Scholars Publishing.
Ehrensberger-Dow, Maureen and Sharon O’Brien (2014). ‘Ergonomics of the translation
workplace: potential for cognitive friction’. In: Deborah A. Folaron, Gregory M. Shreve,
and Ricardo Muñoz Martin (eds), Translation Spaces, 4(1): 98–118. Amsterdam and
Philadelphia: John Benjamins.
European Commission. Directorate-General for Translation (2015). DGT Translation Quality
Guidelines. Brussels/Luxembourg. [Online] https://ec.europa.eu/translation/maltese/
guidelines/documents/dgt_translation_quality_guidelines_en.pdf [accessed October 2019].
 Bibliography 155

Farajian, M. Amin, Marco Turchi, Matteo Negri, and Marcello Federico (2017). ‘Multi-
domain neural machine translation through unsupervised adaptation’. Proceedings of
the second conference on machine translation. Denmark, Copenhagen. DOI: 10.18653/v1/
W17-4713.
Flanagan, Kevin (2014). ‘Subsegment recall in translation memory — perceptions,
expectations and reality’. JoSTrans 23.
Forcada, Mikel L. (2015). ‘Open-source machine translation technology’. In: Chan Sin-
Wai (ed.), The Routledge Encyclopedia of Translation Technology, pp. 152–66. London and
New York: Routledge.
Garcia, Ignacio (2015). ‘Computer-aided translation’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 68–87. London and New York: Routledge.
Gintrowicz, Jacek and Jassem Krzysztof Jassem (2007). ‘Using regular expressions in trans-
lation memories’. Proceedings of the International Multiconference on Computer Science and
Information Technology, pp. 87–92.
Gouadec, Daniel (2007). Translation as a Profession. Amsterdam and Philadelphia: John
Benjamins.
Guerberof, Ana (2017). ‘Quality is in the eyes of the reviewer. A report on post-editing
quality evaluation’. In: Arnt Lykke Jakobsen and Bartolomé Mesa-Lao (eds), Translation
in Transition. Between Cognition, Computing and Technology, pp. 187–206. Amsterdam/
Philadelphia: John Benjamins Publishing Company.
Hernandez-Morin, Katell, Frenck Barbin, Fabienne Moreau, Daniel Toudic, and Gaëlle
Phuez-Favris (2017). ‘Translation technology and learner performance: professionally-
oriented translation quality assessment with three translation technologies’. In: Arnt
Lykke Jakobsen and Bartolomé Mesa-Lao (eds), Translation in Transition. Between Cognition,
Computing and Technology, pp. 207–34. Amsterdam/Philadelphia: John Benjamins
Publishing Company.
Holmes, James S. (1988a). Translated! Papers on Literary Translation and Translation Studies.
Amsterdam: Rodopi.
Holmes, James S. (1988b/2004). ‘The name and nature of translation studies’. In: Lawrence
Venuti (ed.), The Translation Studies Reader, pp. 180–92. London and New York:
Routledge.
Hutchins & Somers (1992). An Introduction to Machine Translation. London: Academic Press.
ISO 704:987 Principles and Methods of Terminology – http://lipas.uwasa.fi/termino/
library.html [accessed 2019].
Jiménez-Crespo, Miguel A. (2013). Translation and Web Localization. London and New York:
Routledge.
Jiménez-Crespo, Miguel A. (2015). ‘Translation quality, use and dissemination in an internet
era: using single-translation and multi-translation parallel corpora to research translation
quality on the web’. JoSTrans 23.
Kenny, Dorothy (2010). ‘The Ethics of Machine Translation’. Proceedings XI NZSTI
National Conference.
Kenny, Dorothy (2011). ‘The effect of translation memory tools in translated web texts: evi-
dence from a comparative product-based study’. In: Walter Daelemans and Veronique
Hoste (eds), Evaluation of Translation Technology, series 8/2009 Linguistica antverpiensia.
New Series – Themes in Translation Studies, pp. 213–34. Antwerpen: Artesis Hogeschool
Antwerpen. Departement Vertalers en Tolken.
Kenny, Dorothy (ed.) (2017). Human Issue in Translation Technology. London and New York:
Routledge.
Kiraly, Donald C. (1995). Pathways to Translation: Pedagogy and Process. Kent, Ohio: Kent State
University Press.
156 Bibliography

Kiraly, Donald C. (St. Jerome 2000/Routledge 2014). A Social Constructivist Approach

to Translator Education: Empowerment from Theory to Practice. Manchester, UK and
Northampton, MA.
Kiraly, Donald C. (2005). ‘Project-based learning: a case for situated translation’, Meta:
Translators’ Journal, 50/4:1098–1111.
Kiraly, Donald C. (2012). ‘Growing a project-based translation pedagogy: a fractal perspec-
tive’, Meta: Translators’ Journal, 57/1: 82–95.
Kit Chunyu and Billy Wong Tak-ming (2015). ‘Evaluation in machine translation and
computer-aided translation’. In: Chan Sin-Wai (ed.), The Routledge Encyclopedia of
Translation Technology, pp. 213–36. London and New York: Routledge.
Lammers, Mark (2011). ‘Risk management in localization’. In: Keiran J. Dunne and Elena S.
Dunne (eds), Translation and Localization Project Management, pp. 211–32. Amsterdam and
Philadelphia: John Benjamins.
Land, Ray, and Meyer, Jan (eds) (2006). Overcoming Barriers to Student Understanding: Threshold
Concepts and Troublesome Knowledge. London and New York: Routledge.
Lawrence, Oliver (2018). ‘The finishing line’, ITI Bulletin. The Journal of the Institute of
Translation and Interpreting, Nov/Dec.
Martin, J.R. and P. White (2005). The Language of Evaluation: Appraisal in English.
New York: Palgrave Macmillan.
Melby, Alan K. (2012). ‘Terminology in the age of multilingual corpora’, The Journal of
Specialised Translation, 18. [Online] www.jostrans.org/issue18/art_melby.php [accessed
December 2019].
Melby, Alan K. and Sue Ellen Wright (2015). ‘Translation memory’. In: Chan Sin-Wai
(ed.), The Routledge Encyclopedia of Translation Technology, pp. 662–77. London and
New York: Routledge.
Mileto, Fiorenza and Muzii, Luigi (2010). ‘Teaching computer-assisted translation and
localisation: a project based approach’. In: Łukasz Bogucki (ed.), Teaching Translation and
Interpreting, pp. 13–14. Newcastle: Cambridge Scholars Publishing.
Mitchell-Schuitevoerder, Rosemary (2013). ‘A project-based methodology in trans-
lator training’. In: Catherine Way, Sonia Vandepitte, Reine Meylaerts, and Magdalena
Bartłomiejczuk (eds), Tracks and Treks in Translation Studies, pp. 127–42. Amsterdam and
Philadelphia: John Benjamins.
Mitchell-Schuitevoerder, Rosemary (2014). ‘A project-based syllabus design’ – Innovative
Pedagogy in Translation Studies’. PhD thesis. Durham University.
Mitchell-Schuitevoerder, Rosemary (2015). ‘A reconsideration of translation quality and
standards’. Journal of Siberian Federal University. Humanities & Social Sciences 8(12): 2908–19.
Moorkens, Joss (2012). ‘Measuring consistency in translation memories: a mixed-method
case study’. Thesis. Dublin City University.
Moorkens, Joss (2019). ‘Uses and limits of machine translation’. In: Ethics and Machines in an
Era of New Technologies. ITI research e-book.
Moorkens, Joss, Stephen Doherty, Dorothy Kenny, and Sharon O’Brien (2013). ‘A vir-
tuous circle: laundering translation memory data using statistical machine translation’.
In: Perspective Studies in Translatology. DOI: 10.1080/0907676X.2013.811275
Mossop, Brian (2014, 2019). Revising and Editing for Translators. Manchester, UK: St Jerome
Publishing.
Mossop, Brian with Jungmin Hong and Carlos Texeira (2019). Revising and Editing for
Translators. London and New York: Routledge.
Müller-Spitzer, Carolin and Alexander Koplenig (2014). ‘Online dictionaries: expectations
and demands’. In: Ebook Package Linguistics, pp. 144–88, Berlin: De Gruyter,
ZDB-23-DSP.
 Bibliography 157

Munday, Jeremy (2012). Evaluation in Translation. Critical Points of Translator Decision-Making.

London and New York: Routledge.
Munday, Jeremy (2012; 2016). Introducing Translation Studies: Theories and Applications. London
and New York: Routledge.
Muzii, Luigi (2012). A Contrarian’s View on Translation Standards. [e-book], www.lulu.com.
Nord, Christiane (1991). Text Analysis in Translation. Amsterdam-Atlanta: Rodopi.
Nunes Vieira, Lucas (2017). ‘From process to product: links between post-editing effort and
post-edited quality’. In: Arnt Lykke Jakobsen and Bartolomé Mesa-Lao (eds), Translation
in Transition. Between Cognition, Computing and Technology, pp. 161–86. Amsterdam/
Philadelphia: John Benjamins Publishing Company.
Nunes Vieira, Lucas (2018). ‘Automation anxiety and translators’, Translation Studies. London
and New York: Routledge. DOI: 10.1080/14781700.2018.1543613
Nunes Vieira, Lucas and Elisa Alonso (2018). ‘The use of machine translation in human
translation workflows. Practices, perceptions and knowledge exchange’. Milton
Keynes: Institute of Translation and Interpreting.
O’Brien, Sharon (2012). ‘Towards a dynamic quality evaluation model for translation’. The
Journal of Specialised Translation, 17. [Online] www.jostrans.org/issue17/art_obrien.pdf
[accessed May 2014].
O’Hagan, Minako (2012). ‘The impact of new technologies on translation studies. A techno-
logical turn?’ In: Carmen Millán and Francésca Bartrina (eds), The Routledge Handbook of
Translation Studies, pp. 503–18. London and New York: Routledge.
O’Hagan, Minako (2017). ‘Crowdsourcing and the Facebook initiative’. In: D. Kenny (ed.),
Human Issues in Translation Technology, pp. 25–44. London, UK: Routledge.
Olohan, Maeve (2004). Introducing Corpora in Translation Studies. London and New York:
Routledge.
Olohan, Maeve and Elena Davitti (2017). ‘Dynamics of trusting in translation project man-
agement: leaps of faith and balancing acts’. Journal of Contemporary Ethnography, 46(4):
391–416.
Papineni, Kishore, Salim Roukos, Todd Ward, and Wei-Jing Zhu (2001). BLEU: A Method
for Automatic Evaluation of Machine Translation, IBM Research Report R C 22176
(W0109-022).
Pavel Terminology Tutorial, https://termcoord.eu/wp-content/uploads/2018/04/Pavel_
Silvia_and_Nolet_Diane_Handbook_of_Terminology-002.pdf [accessed 2019].
Project Management Institute (2004). A Guide to the Project Management Body of Knowledge.
(PMBOK® Guide). 3rd ed. Newtown Square, PA: PMI Press.
Pym,Anthony (1993). Principles for the Teaching of Translation, pp. 100–16.Tarragona: Universitat
Rovira i Virgili.
Pym, Anthony (2008). ‘Professional corpora: teaching strategies for work with online docu-
mentation, translation memories and content management’. Chinese Translator’s Journal,
29(2): 41–5.
Pym, Anthony (2010a). ‘Text and risk in translation’. Intercultural Studies Group. Universitat
Rovira i Virgili. www.researchgate.net/publication/283363313_Text_and_risk_in_trans-
lation [accessed 2019].
Pym,Anthony (2010b).‘The Translation Crowd’. Tradumàtica 8. www.fti.uab.cat/tradumatica/
revista/num8/sumari.htm# [accessed 2019].
Quah, Chiew K. (2006). Translation and Technology. New York: Palgrave Macmillan.
Risku, Hanna (2016). ‘Situated learning in translation research training: academic research as
a reflection of practice’, The Interpreter and Translator Trainer, 10/1:12–28.
Robinson, Bryan J., Clara I. López Rodríguez, and María I. Tercedor Sánchez (2008).
‘Neither born nor made, but socially-constructed: promoting interactive learning in
158 Bibliography

an online environment’. TTR: Traduction, Terminologie, Rédaction,Vol. XXI (2): 95–129,

Montreal.
Rohlfing, Katharina J., Matthias Rehm, and Karl Ulrich Goecke (2003). ‘Situatedness: the
interplay between context(s) and situation’. Journal of Cognition and Culture, 3(2), Leiden:
Brill.
Samuelsson-Brown, Geoffrey (2010). A Practical Guide for Translators. Bristol, Buffalo,Toronto:
Multilingual Matters.
Sánchez-Gijón, Pilar, Olga Torres-Hostench, Bartolomé Mesa-Lao (eds) (2013) Conducting
Research in Translation Technologies. Oxford: Peter Lang.
Shuttleworth, Mark (2015). ‘Translation management systems’. In: Chan Sin-Wai (ed.),
The Routledge Encyclopedia of Translation Technology, pp. 687–91. London and New York:
Routledge.
Shuttleworth, Mark (2017). ‘Cutting teeth on translation technology: how students at
University College London are being trained to become tomorrow’s translators’. Tradução
em Revista, 22: 18–38. 10.17771/PUCRio.TradRev.30595
Suojanen, Tytti, Kaisa Koskinen, and Tiina Tuominen (2015), User-Centered Translation.
Translation Practices Explained. London and New York: Routledge.
TAUS Predictions Webinar, 21 January 2010. www.taus.net/events/user-calls/taus-predicts.
TAUS (2019). DQF BI Bulletin – Q1.
Toury, Gideon (1995). Descriptive Translation Studies and Beyond. 1st revised edition.Amsterdam
and Philadelphia: John Benjamins.
Wagner, Emma, Svend Bech, and Jesús M. Martínez (2014). Translating for the European Union
Institutions. London and New York: Routledge.
Wang, Vincent X. and Lily Lim (2017). ‘How do translators use web resources?’
In: Dorothy Kenny (ed.), Human Issue in Translation Technology, pp. 63–77. London and
New York: Routledge.
Warburton, Kara (2015). ‘Terminology management’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 644–61. London and New York: Routledge.
Wilks,Yorick (2008). Machine Translation. Its Scope and Limits. New York: Springer.
Wong, Cecilia Shuk Man (2015.) ‘Machine translation in the Chinese University of
Hong Kong’. In: Chan Sin-Wai (ed.), The Routledge Encyclopedia of Translation Technology,
pp 237–51. London and New York: Routledge.
Zanettin, Federico (2015)’. Concordancing’. In: Chan Sin-Wai (ed.), The Routledge
Encyclopedia of Translation Technology, pp. 437–49. London and New York: Routledge.
Zetzsche, Jost (2003–2017). A Translator’s Tool Box for the 21st Century. A Computer Primer for
Translators. Winchester Bay: International Writer’s Group.
Zetzsche, Jost (2019). ‘The 297th Tool Box Journal’. The International Writers’ Group.
Zetzsche, Jost (2019). ‘The 298th Tool Box Journal’. The International Writers’ Group.
INDEX

Note: Entries given in bold refer to tables; entries in italics refer to figures.

acceptability 53, 96 bilingual reference 38, 100–1

accounting 2, 13, 136, 152 bilingual revision see revision
accuracy 53–5, 64–5, 82, 84, 99, 104 BLEU (bilingual evaluation understudy) 60,
adequacy 53, 108 63, 65, 67, 96
administration 2, 109, 136; public 71–2 BootCaT tool 83
administrative function 144 bootstrapped corpora 82
administrative skill 1 Bowker, L. 96, 98–9
administrative tool 142; see also portal browser 2, 9; browser-based 144
AEM (automatic evaluation metrics) see business model see model
metrics
AI (artificial intelligence) 52, 58, 86; strong CAT tool 15–21, 24–30, 34–8, 51–9, 85–8;
and weak 135 manufacturer 37, 41–2, 133; see also
algorithm 53 TEnT
alignment 38–41, 81; automated 39 cause-driven crowdsourcing 141
alphabetic language 12; see also non- check see QA; revision
whitespace language cloud-based tool 119; CAT tool 3, 13;
API (application programming interface) see also SaaS (software as a service)
12, 56–8, 63, 80; connector 133–4, 143 CMS (content management system) 143–4;
authoring see copyright see also TMS (translation management
automated speech recognition system)
(ASR) 134–5 code 7, 10, 11; of ethics 121; key 56;
autosuggest 27, 38 see also predictive writing shortcut 69; source 98; writing 47;
see also regex; tag
benchmark 64, 96, 101: human translation cognitive ergonomics 14, 21
64, 93, 96; cognitive friction 21
standard 96 collaboration 3, 5, 13, 16, 79, 114
Berthaud, S. 119 compatibility 2–3; 13–14; 39; see hardware
bilingual CAT tool format 28 and software
bilingual file 2, 6, 100 completeness 77, 99
bilingual format see format concordance: CAT 30–1, 83–5; external 85;
bilingual glossary see glossary parallel 85; web 85; see also consistency
160 Index

confidential disclosure agreement see digital Ehrenberger-Dow, M. and G. Massey 21

ethics/non-disclosure agreement error: management 63, 65; repair 52;
connector see API spotting 13; warning 42, 95, 100
consistency 26–7, 30–3, 64, 71, 77 error typology 64–5, 96, 105–6
consumer-oriented crowdsourcing 142 extensible markup language (XML) see
content management system see CMS format; tag
context see match
conversion tool 2, 139; see also pdf Facebook 141
copyright 114, 116–23; authoring 122, 138, false positives, 27, 45, 48, 73, 86
143; digital material 98, 113, 122, 130; file: clean 2; dirty 2; see also format; TBX;
intellectual property 113–14, 117, 122–3; TMX; XLIFF
ownership 113–14, 116, 120; see also filter 3, 45–6; 95, 134 see also regex
digital ethics fluency 53, 104, 107–8, 135
corpora: bilingual 39; digital 79; folder: directory 2, 8; parent 7, 8; see also
multilingual data 80–1 subfolder
corpus 39, 60, 73, 79, 81–3; British National Forcada, M.L. 56, 63
Corpus 85 format 41–6: bilingual 5; docx. 2–13;
corpus-based machine translation 53 editable 2, 14, 30, 44; HTML 42; MS
criteria: assessment 104; evaluation 62, Excel 41; rtf 80, 100; source 30; target 47;
93–4, 104; search 83; testing 64 see also interchangeable format; tag
Cronin, M. 141, 146 formatting see tag
crowdsourced material 142 freelancer 20; freelance translator 94, 103,
crowdsourcing 141–2; see also platform 132, 138; see also human resources
customisation 99 freeware 56
customised TM 27, 37–8, 41, 50, 72, 137 fuzzy see match
fuzzy match threshold 26
data: centre 122; protection 119–20; sharing
119–21; see also metadata GALA (globalization and localization
database: external 25–7, 38, 86; internal 26, association) 12
69; see also file glossary: bilingual 73; online 71
deep learning 52, 54; see also machine Google 12, 52, 63, 81, 83; Analytics 147;
translation Cloud Translation 121; Translate 120
desktop-based CAT tool 136, 144 Gouadec, D. 96, 103, 122
dialog box 10–11, 14, 29, 30–3 Guerberof, A. 105–6
dictionary 71, 78–82, 85; digital 79–102;
sample-based 79; see also glossary hardware and software 3, 123–5; see also in/
digital ethics 113; non-disclosure compatibility
agreement (NDA) 115–16; see also human resources 2, 18–20, 138; in-house
copyright translator 18, 103; project manager
digital material see copyright 16–20, 122, 144; see also freelancer;
digital platform 137–9; see also platform reviser; terminologist
directory see folder human translation 55, 62, 64, 78, 96
Doherty, S. 64
domain 38, 58; knowledge 18; specific 37, image file 3, 40–1, 44; see also hardware and
53, 63; suffix 78–9 software; incompatability
DQF (data quality framework) see metrics incompatibility 124, 126
drop-down menu 10 indemnity insurance 94, 115
Dunne, K.J. and E.S. Dunne 19 in-house translator see human resources
input: consistency 57; data 65, MT 57–8
EAGLES 4, 59–60 translator 57; user 12; see also output
edit distance 38, 55–6, 64 interchangeable format 13, 40, 88, 124;
editor: term extraction 73–4, 85, 88; see also see also XLIFF; TBX; TMX
translation/grid interface 30, 41, 58, 138
efficiency 5, 55–6 internationalisation (i18n) 98
 Index 161

invoice 113, 134, 142, 144 morphological change 69, 75, 84, 86
ISO (International Organisation for Mossop, B. 99, 105
Standardisation) 60, 77, 92–4, 101–4, 120 MQF (Multidimensional Quality
Framework) see TAUS
Jiménez-Crespo, M.A. 97–8, 111, 141 MS Word: competence 1; shortcut code see
joined and split segments 27, 29, 39, 42 code; status bar 10; toolbar 10–11
Müller-Spitzer C. and A. Koplenig 79
Kenny, D. 1 Munday, J. 94
knowledge platform 141
KWIC (Key Word in Context) 81, 85 natural language processing 52
NDA see non-disclosure agreement
l10n (localisation) 98 neural machine translation (NMT) see
language: code 7; direction 27, 34; pair 11, machine translation
18, 44, 60, 63, 138; quality assurance no match 26–7, 30, 34, 36, 105–6
92; tag 7 non-disclosure agreement (NDA) see
language service provider 18, 80, 102 digital ethics
Levenshtein distance 108 nonlinear 97, 142 see also syntagmatic
leverage 5, 37, 54, 56 non-translatable 40, 92
lexicographer 79, 82 non-whitespace language 28; see also
licence 114; software 13, 134, 136–7, 144 alphabetic language
linear 77–8, 100; see also nonlinear Nunes Vieira L. 106, 108
linguist: contracted 34, 103; contractee Nunes Vieira L. and E. Alonso 151
120–3, 127, 132, 136, 138; see also
freelancer; freelance translator; reviser; O’Brien, S. 96, 106
subcontracted 121 offline editor 137
LISA (Localization Industry Standard O'Hagan, M. 142
Association): model 64 Olohan, M. 85
localisation 95, 97–9, 140–6 on-demand software see SaaS
lookup 12, 26–7, 87, 134–5 open-source software 3, 56
LSP see Language Service Provider optimisation see search engine optimisation
output quality 65
machine translation: adaptive 54–60; outsource 15, 143
integrated 12, 21, 26, 48, 51; neural 52–3, outsourced material 141; see also
60, 96, 107, 122; see also deep learning crowdsourcing
management: business 2, 145; content 144; ownership see digital ethics
file 6, 24, 144; quality 60, 65, 98; risk
113–4, 122, 124, 128 package 39–40; project 136; return 39, 136
mantra 14–5, 147, 149, 151 paid crowdsourcing platform 141
marketplace see platform paradigmatic 77, 107
match: accuracy 55; context 26, 34, 87; parallel corpora see corpora
fuzzy 26, 34, 37, 52–8, 86; perfect 26–8, parent folder 7–8; see also subfolder
33, 37, 54; reversed 27; statistics 13; see pattern matching see regex
also no match pdf 40, 43–4; see also format
metadata 33, 36–7, 53–5, 73–4, 120, 147; platform 116, 125; marketplace 139,
see also data 142; translation 58, 138, 140, 143,
metrics 46, 52, 62–5, 96, 98; automatic 145; translator 4, 138; see also
evaluation (AEM) 63; data crowdsourcing; digital knowledge base;
quality framework (DQF) 64–5; localisation; portal
multidimensional quality (MQM) 64–5 PO see purchase order
Microsoft (MS) Office 2–3, 13, 24; MS portal 127, 134, 137, 142–4; see also
Windows 1–3; operating system 3 platform
model: business 137, 139; encoder 53 see post edit machine translation (PEMT) 21,
also BLEU; LISA; metrics; TAUS 61–2, 107–9, 141
monolingual see revision post-editing (PE) 37, 54–5, 62, 102–3, 106
162 Index

post-editor 52, 92, 107–9 score: edit distance 55; error 52, 59, 63, 95,
precision 7, 30, 64; see also recall 102; MT 59, 63–4; underscore 7, 95
predictive feature 41, 53; typing 13; search: local 134; phrase-based 95, 146;
writing 38 tool 12
pre-translate 27–8, 94, 140 search engine 56, 62, 69, 84; optimisation
preview 12, 29 143, 146–7; MT 56; specialised 77
process: conversion 3, 125; file 13, security: breach 121; breach of
42; revision 99; segmentation 28, confidentiality 21, 113–15, 121–2;
42; translation 13–17, 71–3, 92–4, digital 115–18
136–7, 144 seed 146; list 147–8
professional organisation 4, 95, 116, 121 segment: source 4, 12, 26; split 27, 29; target
project management 17–19, 124–5; team 12, 26, 29, 80; subsegment 4; see also
13, 19–21; translation 15, 125–6; see also join; split
project manager segmentation 28, 42, 84, 135; rule 28–9, 34,
project manager 13, 16–20, 33, 115, 122 46–7; see also subsegmentation
propagate 27, 33, 58 server 13–14, 42, 125, 132–3, 136–7, 144
pseudo-translation 42–4 sight translation 135; see also text to speech
purchase order (PO) 17, 34, 113, 144 smartphone 14, 38, 134, 143
push and pull approach 87, 140 software see hardware
Pym, A. 29–30, 78, 103, 122, 142 source: file 6, 18, 98, 125–6; language
28–9, 32, 55, 104; text 42, 57, 73,
QA see quality 99, 115–16
quality: assessment 65, 92–6; assurance speech recognition see automated speech
(QA) 46, 64, 77, 85–6, 92–5; check 20, recognition
60, 92; control (QC) 74, 88, 93, 98, 104; speech technology see automated speech
estimation 55, 62–3; specification 93;TQA recognition
(translation quality assurance) 63, 92, 96 speech to text 134–5; see also automated
speech recognition; text to speech
ranking: algorithm 146; order 4 spellcheck 12–13, 41, 93, 99, 134
recall 37, 45, 64, 69, 84; see also precision split see joined and split segments
reference: file 38–41, 85, 105, 122, 126; standalone 2–3, 13–14, 52, 87–8
material 70–2, 80, 83, 85, 137; translation standard see ISO; quality
55–6, 64, 96, 99; see also alignment standardisation
regex 44, 46–7, 95; pattern matching 46; statistical machine translation 53, 121
see also filter see also neural machine translation
repetition 5, 34, 36, 95 statistics 13, 34–6, 53
repetitive strain injury 11 status bar see MS Word
review: bilingual 9, 29–30; monolingual storage system 6, 25
52, 104; preview 12, 29; tab 11, 31; string 27–30, 37–8, 53, 55, 84,
translation 17, 34, 53, 64, 102, 107; 98–99
see also revision subfolder 7–9
reviser 17–18, 94, 104–5, 112, 136 subsegment: matching 8, 36–7, 70; see also
revision: bilingual 100–1; CAT tool 99–102, segment
105, 107, 136; model 95; monolingual 29; subsegmentation 38, 90
self- 17–18, 94, 100, 105; third-party 5, syllabic language 12; see also alphabetic
17, 94, 101–2, 104, 150 language
ribbon 10–11, 44–5; CAT tool 11, 30–1, 34, syntagmatic 77–8, 107; see also nonlinear
45, 69; MS Word 10, 45
risk: breakdown structure 124–5; factors tablet 14, 119, 135, 143
124–5, 127; management 114, 124, 127; tag; language 7, see also code; formatting 14,
response 124, 127 41–2, 99
target: file 9, 39, 44, 95, 100; language (TL)
SaaS (software as a service) 136, 143–4 32, 42, 55, 60, 96, 147; text (TT) 34, 42,
sampling 79 53, 61, 96, 99–100
 Index 163

TAUS (Translation Automation User Translation Management System (TMS)

Society) 4, 37, 55–8, 65, 81–2, 96; see also 56–7, 68, 136, 143; see also content
DQF (Data Quality Framework) management system (CMS)
taxonomy 65 translation memory (TM) database 3, 12,
TBX (TermBase eXchange) 42, 79, 114, 25; edit distance 38; edit time 74; editor
124, 136, 139; see also interchangeable 27, 45; interchangeable 25–6, 38, 58;
format; terminology database repair 45
teamwork 19–20 translation project see project
TEnT (translation environment tool) 1–2, Translation Results window 26–7, 38, 40,
132; ASR see also see automated speech 69–70, 74, 84
recognition; CAT tool; concordance; translation unit (TU) 3, 12, 25–6, 39
conversion tool; dictionary; translation Translators without Borders 138
management system TU see translation unit
term: complex 73; definition 74, 77; extract
73; extraction 13, 72–3, 88; lookup icon usability 96
87; pair 27, 37, 73–4, 84, 86, 88; search
13, 78; simple 73 vendor 18, 21, 132, 139
TermBase eXchange (TBX) see TBX vocabulary 71, 83; see also glossary
terminologist 112 volunteer translation see crowdsourcing
terminology database (Tmdb) 77–80,
83–8, 133–7 web: corpora 82; localisation 97–9, 139–40;
terms of business 119, 121, 130 search 77–8; web browser see browser;
terms and conditions 113–17, 121; web/cloud-based tool 21, 116, 119,
translation services 117 132–3, 137, 149
text genre 56 web content management system see
text to speech 134, 136 content management system
threshold 54–5; match 26; see also fuzzy weighting 55, 64
TM see translation memory whitespace languages see alphabetic
TM editor 27, 45 language
TMS see translation management system word count 2, 17, 33–4, 41; costing 26, 34;
TMX (translation memory eXchange) fees 34, 109, 123; pricing 15, 19, 21, 34,
38, 41, 42, 58, 114; see also file; 139, 151; rates 14, 21, 26, 34
interchangeable format workflow 15–16, 140, 142–4; chart 16–7;
transcreation 147 LSP 18, 125, 136–7; translation 130, 132–
transfer 42, 116, 119–20; file 116, 3; translator 17, 58, 88
119–25; of ownership 149; rights 148; World Wide Web (WWW) 42, 78, 85, 146
rule 46
Translation Automation User Society XLIFF 2–3, 6, 13, 42, 93, 137; see also file
see TAUS XML 3, 13, 80; see also formatting
translation editor/grid 36, 39, 41, 74,
99, 133 Zanettin, F. 84
Translation Environment Tool see TEnT Zetzsche, J. 37, 53, 73, 81, 86, 88, 107
translation grid see translation editor ZIP archive 39