Multilingual.com - AI's role in project management Tips for building an internal AI team Deep learning, a chip off the old block
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
NOV/DEC 2018
AI’s role in project management
Intelligence
Tips for building an internal AI team
Artificial
Deep learning, a chip off the old block
multilingual.com
Translation
shouldn’t be this hard.
We provide powerful tools to translate your content and products.
Stop wasting time and start growing your business.
Get started today at
www.smartling.com
One platform, All the resources you need. In your hand. For free. Linguistic resources, machine translation, project management, workflow control. Matecat is a centralised platform that enhances your performance, whatever it is. For translators, agencies, large enterprises. For teams of any size. www.matecat.com Open, free, online.
Pos t Ed i t i n g
#180 Volume 29 Issue 8 November/December 2018
Editor-in-Chief, Publisher: Donna Parrish
Managing Editor: Katie Botkin
Proofreaders: Bonnie Hagan, Bernie Nova
News, Calendar: Kendra Gray
Production: Darlene Dibble, Doug Jones
Cover Photo: Doug Jones
Technical Analyst: Curtis Booker
Assistant: Gayle Hallock
The idea that AI is coming for us all sort of makes Circulation: Terri Jadick
me happy. Special Projects: Bernie Nova
Advertising Director: Kevin Watson
I mean, it’s not like we’re doing such a bang-up Marketing Director: Marjolein Groot Nibbelink
job ourselves. We’re blithely hurtling toward our Finance: Leah Thoreson
own demise, feeding the machine everything it
Editorial Board
might eventually need to completely control us. The
Games: Miguel Á. Bernal-Merino
technology may not be there, but our attitude is: if Standards: David Filip
there were a universally intelligent entity that could Business: Aki Ito
control all the world’s tech, and by extension almost Marketing: Nataly Kelly
everything in the world, it wouldn’t even need to User Experience: Ultan Ó Broin
wage war on the human race to get what it wanted. Technology: Jost Zetzsche
It would just need to offer us small rewards, free Advertising
in-game purchases. We do this anyway; for a free advertising@multilingual.com
cheeseburger, we’d offer up all our secrets. https://multilingual.com/magazine-ads
+1 208-263-8178
In a way, we already are half-controlled by AI.
What else could you call the massive voter fraud Subscriptions, back issues, customer service
made possible by creating fake news stories and subscriptions@multilingual.com
propagating them via social media on individuals https://multilingual.com/subscription-information
likely to accept them as real?
Submissions, letters
So maybe by the time it occurs to AI that it can do editor@multilingual.com
the same thing to accomplish its own goals — which, Editorial guidelines are available at
who knows, maybe it already has — we’ll probably https://multilingual.com/editorial-submissions
need it as a species. We’re not there yet: all the Reprints
AI articles in this issue, with its focus on language reprints@multilingual.com
processing and data analysis, are limited in scope. As MultiLingual Computing, Inc.
CSA Research’s Arle Lommel once said, machines can 319 North First Avenue, Suite 2
translate and they can also play chess, but we don’t Sandpoint, Idaho 83864-1495 USA
have a machine that can do both. And we don’t even
info@multilingual.com
have a roadmap on how a machine would bridge that
https://multilingual.com
gap without human input.
© MultiLingual Computing, Inc. November/December 2018. All rights reserved. Repro-
So perhaps AI is coming for us all, and perhaps it duction without permission is prohibited. For reprints and eprints, please email
reprints@multilingual.com or call +1 208-263-8178. MultiLingual (ISSN 1523-0309) is
isn’t. And perhaps the ways that it is can be harnessed published monthly except May/Jun, Aug/Sep, Nov/Dec for US $58, international $85
per year by MultiLingual Computing, Inc., 319 North First Avenue, Suite 2, Sandpoint, ID
and controlled, with the examples laid out in this 83864-1495. Periodicals postage paid at Sandpoint, ID and additional mailing offices.
POSTMASTER: Send address changes to MultiLingual, 319 North First Avenue,
issue. Suite 2, Sandpoint, ID 83864-1495.
This NewPage paper has been chain-of-custody certified
by three independent third-party certification systems.
MultiLingual is printed on 30% post-consumer recycled paper.
C o n te n t s
Focus:
Artificial
Intelligence
28 32 36
AI’s role in project How AI is changing Tips for building an
management the future of translation internal AI team
Hélène Pielmeier management Myra Goldie
Rob Vandenberg
39
39 46
Profiling giraffes and Deep learning, a
reindeer chip off the old
Jim Compton block
Aidan Collins
28
Education Business 53
50 53
Creating a translation The changing scope
program for the 21st of enterprise localization
century management
Filip Šanca Edmund Blogg
The path to expanded human knowledge attained
through higher education is depicted in this painted
roadside mural on St. Thomas, US Virgin Islands.
6 November/December 2018
Con tent s
I social m e d i a I
Recent Facebook posts
Up Front 32
5 Post Editing
News
8 Recaps 8
12 News
14 Calendar
Columns
16 Client Talk
Terena Bell
18 Terminology Glosses
Laura Di Tullio
20 Community Lives 11
Jeannette Stewart
24 Perspectives
Devaki Datar-Kunte
66 Takeaway
Ally Gilboa
16
see more on our Facebook page at
www.facebook.com/
58 Buyer’s Guide multilingualmagazine
or visit our website at
https://multilingual.com
65 Advertiser Index
November/December 2018 7
Re cap s
LT Industry Summit focuses on AI, more
The 7th Language Technology
(LT) Industry Summit (www.lt-sum-
mit.com) took place May 28-29, 2018,
in Brussels, Belgium. The LT Industry
Summit is Europe’s event showcasing
the latest developments in the three
technology stacks driving language
intelligence: speech interaction; deep
language and meaning processing;
and multilingual communication and
cognition.
The summit presents the most
recent technological developments
through practical use case presenta-
tions. The mix of participants allows
for fruitful networking and paves
the way for future collaborations. interviews. “Language is the number- clients add value.” He summarized
As Andrzej Zydron of XTM Inter- one issue that prevents human beings this as “digitize or die!” David Canek
national said: “As a company we are from delegating repetitive service of Memsource presented recent AI
regular participants and have ben- tasks to collaborative robots,” said developments empowering machine
efited tremendously from participat- serial entrepreneur John Rauscher, translation and Justin Wyman of
ing both in terms of learning about pointing toward main opportunities Socialgist uncovered the hidden trea-
latest developments in the language and challenges in the field of natural sures in social media data.
field, networking and meeting busi- language generation. Arthur Wetzel LT-Innovate’s next event takes
ness partners.” of 24translate underlined the impor- place November 29-30, 2018, in
Four keynote speakers provided tance for “Language Service Providers Vienna, Austria: “Fake News and
challenging insights in their presen- to build up their technology expertise Other AI Challenges for the News
tations and further reading through and leverage technology to help Media in the 21st Century.”
8 November/December 2018
Recaps
2018 user conferences focus on automation and user feedback
Wordfast Forward 2018 WFF is never complete without a interesting tasks that are uniquely
Wordfast organized its fourth user sports outing. On Sunday morning, dependent upon human experience.
conference June 1-2, 2018, in Cascais, participants learned how to surf at One product announcement was
Portugal, bringing together nearly 80 the legendary Guincho Beach. Next the introduction of “Hey memoQ.”
users, developers, trainers, hotline year the event celebrates Worfast’s This new feature supports dictation
support, spouses and even children 20-year anniversary March 21-23 in directly into memoQ via an Apple
for the occasion. Wordfast Forward Martinique. device. Although memoQ did previ-
(WFF) gives stakeholders a chance ously support dictation through a
to take stock of new features and im- memoQfest 2018 third-party application, the breadth
provements, learn why and how us- At memoQfest 2018 May 30-June of language support was limited. Hey
ers apply them, and seek solutions to 1, 2018, there were a variety of pre- memoQ piggybacks on the Apple
problems by sitting down and listen- sentations about topics such as the dictation services for iPhones and
ing to each others’ stories. emerging role of AI in the language iPads, and therefore vastly expands
The conference kicked off on Thurs- industry and what that will mean for the languages and dialects to over 60.
day with a trainers’ symposium followed the role of translators in the future. Enhanced dictation support opens
by an informal, beachfront dinner Other presentations involved real- new doors to translators whose
and poolside cocktails sponsored by world case studies about how memoQ previous productivity was negatively
partners WordFinder and IntelliWeb- practitioners had solved thorny prob- impacted by being constrained to
Search. Friday morning started with lems with unfriendly file formats. The keyboard input or a limited language
keynote speaker Filipe Carrera explain- preconference day featured 25 train- set. The first release of Hey memoQ
ing how to translate networking into ing sessions that ranged from intro- can be expected soon.
business, and the power of weak links. ductory topics for memoQ newbies One of the most popular sessions
Participants later learned how to lever- to geeky, challenging subjects such as was “The User Strikes Back.” This is
age the power of server-based transla- regular expressions or automated file a formalized complaint session, in
tion memories and glossaries for free, conversion through scripting. which attendees can air their griev-
and enjoyed an end-of-the day tasting With over 250 attendees from 30 ances in front of other attendees who
session from our Chablis wine expert countries, the community that gath- frequently nod their heads in agree-
translator and user Ruth Simpson. ers for memoQfest is diverse and ment. The grievances are fielded by a
The first day ended with a white party cohesive. With the exception of some panel of memoQ employees and the
dinner overlooking the world-famous guest speakers or representatives of management team. The preconfer-
Boca do Inferno sea cliffs. partner companies, attendees are uni- ence day concluded with a networking
On Saturday, attendees heard from formly memoQ users. Some have been event and dinner at the historic Kiscell
terminology expert Clarisa Moraña memoQ aficionados from the begin- Museum, complete with acrobatic
and about advanced tool features from ning. memoQfest itself dates back ten flame-jugglers. The gala conference
expert users and hotliners Reginaldo years, and memoQ as a company was dinner was served at the Castle
Francisco, Jamie Lucero and Domi- born four years prior to that. The com- Garden. After aperitifs in the garden
nique Pivard. Wordfast intern Ziqi pany has rebranded itself to match the overlooking the Danube, attendees
Zhou then explained how she and her name of its leading product, memoQ. descended into a decoratively-lit sub-
team localized Wordfast Pro 5 into Therefore, the former company Kilgray terranean space. At the dinner, three
Chinese. Product managers later took is now known as memoQ. stalwart memoQ users were honored:
the floor to explain how they choose Nataly Kelly presented the keynote Juliet Macan, Stéphan Gervais and
what goes in the next release and how address on the growing influence of Roberto Savelli, who have attended
important user feedback is in influenc- AI. She expressed optimism about all ten memoQfests. Their loyalty was
ing their decisions. Before the closing the future of translation profession- rewarded with a free pass to attend
dinner in downtown Cascais, each par- als. Kelly’s premise was that AI will the 2019 event next summer.
ticipant was given 300 Wordfast dollars help remove the boring, repetitive Dinner was followed by entertain-
to “buy” the features they wanted most stuff that translators don’t really ment, including a 1960s dance and
during a mock auction and help devel- want to do anyway. This will facilitate hairstyle retrospective performed by
opers prioritize for the coming year. future translators digging into more memoQ staff.
November/December 2018 9
Re cap s
F e a t u re d R e a d e r at present. Whether it’s prin-
cess parties, dance lessons or
Would you introduce yourself? swimming, there is often need
Tom Olliver, localization pro- for “Dad's taxi," and I wouldn’t
gram manager at Canon Europe change it for the world. She is
Ltd. currently determined to be a
Where do you live? palaeontologist and I was un-
I live in Binfield, a rural village der strict instructions to bring
in the Royal County of Berkshire, back a dinosaur from LocWorld
with my wife, seven-year-old Warsaw. The closest I could get
daughter and three cats. was a photo from the Warsaw
How did you get started in this National History museum.
industry? Sadly/fortunately my pockets
I got into this industry via a weren’t big enough to bring
happy accident — ironically by anything back.
a poor keyword search online. I I’m also an avid science fiction
finished University in 2000 with fan and enjoy the odd game
a degree in aeronautical engi- dustry only speaking my native of tennis, as well as joining the
neering, large debts and no real language, English. Along the weekly quiz nights at the local
idea of what to do next. When way I’ve picked up bits from the pub.
searching for engineering jobs, I broad spread of languages we Why do you read MultiLingual?
found an advert for a localization localize, although these often Within a large, global organi-
engineer. Despite not having have limited relevance in the zation and the numerous func-
heard of it previously or, if I’m outside world. You would be tions involved in releasing and
being honest, fully understand- surprised how seldom “printer supporting products through
ing what it was either, I applied. driver” or “document feeder” their life cycle, localization can
At the time I remember thinking come up in casual conversation. sometimes feel like an invisible,
I “might stick around a year” but I also have a grounding in underappreciated function. Feel-
I’ve been in the industry, and Japanese, after taking lessons ing connected to the industry
with Canon, ever since — working in the language, which I find is important, and MultiLingual
my way through project man- has helped understanding and is a key part of that for me. It’s a
agement until my current role communication with our parent consolidated snapshot of the in-
as program manager, managing company. dustry and will often send me off
a brilliant team of localization What do you like to do in your in many interesting and produc-
project managers. spare time? tive directions, after seeing an
What language(s) do you speak? Most of my spare time re- interesting article or surprising
Unusually, I entered this in- volves around my daughter personnel move.
10 November/December 2018Unleash the Potential of
Your Digital Impression
The launch of our new vertical
platform in 2019 will further
increase the value of your ad
campaign in MultiLingual.
Enhance your ad
in MultiLingual
digital to increase
engagement.
Use animation,
audio. video or a
call-to-action button.
Become an
exclusive digital
issue host.
learn more at
magazine.multilingual.com/media-kitN ews
Business Women in Localization 501(c)(3) certified
Women in Localization, a professional organization for
Moravia rebrands women in the localization industry, has announced its
The Moravia IT Group has rebranded per its acquisi- official certification as a 501(c)(3) tax-exempt nonprofit
tion by RWS, a provider of intellectual property support organization.
services. RWS Moravia will continue to operate as an au- Women in Localization www.womeninlocalization.com
tonomous division of RWS.
RWS Moravia www.rws.com/what-we-do/rws-moravia People
Venga acquires TI Recent industry hires
Venga Global, a provider of localization services, has Eriksen Translations Inc. has hired Will Lach as director
acquired Translators International (TI), an information of sales, Kevin Hudson as director of client services, Yasmin
and communication technology security language services Menon as director of production, Ryan Hansen as multilin-
company. gual quality manager, Shelby Greenwood as vendor manager
Venga Global http://vengaglobal.com and Aisha Moritz as project management team lead.
Eriksen Translations Inc. www.eriksen.com
Questel acquires MultiLing ICD Translation has hired Steven Perhach as head of the
Questel, an online intellectual property services pro- company’s new video remote interpreting product.
vider, has acquired MultiLing, a provider of patent transla- ICD Translation www.icdtranslation.com
tions. Questel will partner with MultiLing’s management TripleInk has hired Cristina Segar as an account supervisor.
team. Cofounder and managing director, Christa Tiefenbacher-
MultiLing www.multiling.com Hudson, is retiring. Stephanie Cooper and Uta Moncur will
jointly lead the agency.
Translata buys Localization Guru TriplInk www.tripleink.com
Translation agency Translata has bought translation
startup Localization Guru. Localization Guru will remain Resources
a distinct entity with services, management and adminis-
tration delivered under the Translata brand. Reports on translation proxies, intelligent
Translata www.translata.eu content and defining depth of localization
Common Sense Advisory, Inc., an independent market
research firm specializing in the language service industry,
12 November/December 2018News
has published several reports. “Re-assessing Translation new features and functionality to its apps for vendor
Proxy Solutions” defines translation proxy, revisits the management, linguistic quality evaluation and multilin-
myths, and identifies scenarios where proxies should al- gual business intelligence.
ways be considered or should generally be avoided. Lingotek www.lingotek.com
“Four Futures for Global Intelligent Content” defines
what intelligent content is and why it matters, and ana- SYSTRAN Pure Neural Server
lyzes four content trends. SYSTRAN International, a provider of translation ser-
“How Deeply Should You Localize” addresses how to vices and technology solutions, has launched SYSTRAN
define depth of localization and how deeply organizations Pure Neural Server. It supports unlimited user access and
localize, with an emphasis on the role of site size and num- integrates with any business application and document
ber of languages in this decision and why organizations treat workflow.
languages differently. It closes with a set of recommenda- SYSTRAN International www.systransoft.com
tions for how to determine appropriate depth for sites.
Common Sense Advisory, Inc. www.commonsenseadvisory.com Language Insight adds WPML plugin
Language Insight, a provider of language services, has
Products and Services added the WordPress multilingual plugin (WPML) to its
online service offering. WPML includes advanced features
XTM v11.4 for translation management and an interface for profes-
XTM International, developers of a translation manage- sional content translation.
ment system and computer-aided translation tool, has Language Insight www.languageinsight.com
released XTM v11.4. The latest version introduces the new
neural machine translation engine Amazon Translate, en- Clients and Partners
hancements for XTM Connect and the option to provide
different plural forms of a noun during translation of iOS enLabel integrates Argos TMS
source files. enLabel Global Services, Inc., provider of integrated
XTM International https://xtm.cloud packaging management (IPM) solutions for medical de-
vice and life sciences organizations, has partnered with
Lingotek system updates and enhancements Argos Multilingual, a language solutions provider, to inte-
Lingotek, a developer of collaborative translation tech- grate its enLabel IPM platform with the Argos translation
nology, has further enhanced its translation management management system (TMS).
system and content management connectors, and added Argos Multilingual www.argosmultilingual.com
Give The Gift Of
Knowledge
— for free! Thanks
If you find information in any issue
of MultiLingual that you know would
benefit a friend or colleague, let us
send them a copy.
Simply email freecopy@multilingual.com with the person’s name, postal
address and the date of the issue you’d like to share.
They will receive a complimentary copy of the magazine along with
a note, letting them know you were thinking of them.
November/December 2018 13C al e n d ar
October EXPOLINGUA Berlin
November 16-17, 2018, Berlin, Germany
Meet Central Europe Conference ICWE, www.expolingua.com
October 30-31, 2018, Budapest, Hungary
AATC, ACTA, ATCSK, Proford, www.meetcentraleurope.com TAC Conference 2018
November 19-20, 2018, Beijing, China
EMNLP 2018 Translators Association of China
October 31-Nov 4, 2018, Brussels, Belgium www.tac-online.org.cn/annualmeetingen
Association for Computational Linguistics, http://emnlp2018.org
Nordic Translation Industry Forum
November November 21-23, 2018, Oslo, Norway
Anne-Marie Colliander Lind, Cecilia Enbäck, http://ntif.se
Best Practices for Communicators
in Highly Regulated Industries Tools and Methods for
November 8-9, 2018, New York, New York USA Corpus-Based Translation Science
Ragan Communications, Inc., http://ow.ly/NnGb30m7t0R November 22-23, 2018, Innsbruck, Austria
TransBank, https://transbank.info/workshop
tcworld 2018 - tekom
November 13-15, 2018, Stuttgart, Germany Information Development World
tekom, http://conferences.tekom.de/tcworld18/registration November 27-29, 2018, Menlo Park, California USA
The Content Wrangler, Content Rules
Translating and the Computer 40 (TC40) http://informationdevelopmentworld.com
November 15-16, 2018, London, UK
AsLing, www.asling.org/tc40 Translation and Cultural Sustainability
November 28-30, 2018, Salamanca, Spain
TAPICC: Translation API Class and Cases University of Salamanca
November 15, 2018, San Jose, California USA http://traduccionysostenibilidad.fundacionusal.es
International Multilingual User Group, https://bit.ly/2tjGYJJ
Cologne Conference on Translation,
Interpreting and Technical Documentation
November 29-30, 2018, Cologne, Germany
TH Köln – University of Applied Sciences, https://bit.ly/2AQfpxv
December
ND Focus – Elia’s focus on
Sales & Marketing
December 6-7, 2018, Malaga, Spain
Elia, http://events.elia-association.org/nd-focus-sales-2018
January 2019
How Technology Is Redefining
When and Where We Meet
January 17, 2019, San Jose, California USA
International Multilingual User Group, http://ow.ly/a28g30lLmnj
ALC UNConference
January 17-19, 2019, Huntington Beach, California USA
Association of Language Companies
www.alcus.org/page/UNConferenceOverall
14 November/December 2018Calendar
Translation Technology Round Table May
January 24-25, 2019, Boulder, Colorado USA
The Localization Institute, http://ow.ly/A6WS30lJprA ALC Annual Conference
May 1-4, 2019, Washington, D.C. USA
Lenguas 2019 Association of Language Companies
January 24-26, 2019, Mexico City, Mexico www.alcus.org/page/AnnualConfOverall
Italia Morayta Foundation, InterpretAmerica, www.lenguas2019.com
2019 STC Technical Communication
February Summit & Expo
May 5-8, 2019, Denver, Colorado USA
Together Society for Technical Communication, https://summit.stc.org
February 21-22, 2019, Barcelona, Spain
Elia, http://events.elia-association.org/together-2019
June
LocWorld39 Kuala Lumpur CIUTI 2019
February 27-28, 2019, Kuala Lumpur, Malaysia June 3-5, 2019, Melbourne, Australia
Localization World, Ltd., https://locworld.com Monash University, http://ow.ly/5OSo30lLlOL
tcworld India Content Connections 2019
February 28-March 1, 2019, Bangalore, India June 10-12, 2019, Boston, Massachusetts USA
teckom, http://tcworld-india.com Acrolinx, https://acrolinxcc.com
March LocWorld40 Portugal
June 11-13, 2019, Portugal
Wordfast Forward 2019 Localization World, Ltd., https://locworld.com
March 21-23, 2019, Sainte-Luce, Martinique
Wordfast, http://wordfast.com/conference
GALA 2019
March 24-27, 2019, Munich, Germany
Globalization and Localization Association
www.gala-global.org/all-events/gala-2019-munich
LATA 2019 connecting your visions,
March 25-29, 2019, Saint Petersburg, Russia
Saint Petersburg State University, IRDTA, http://lata2019.irdta.eu technologies and customers
Tenth International Translation Conference
March 26-27, 2019, Doha, Qatar
Translation and Interpreting Institute
www.tii.qa/en/tenth-annual-international-translation-conference-call-papers
April
MadWorld 2019 Information Creation · Translation/Localization
April 14-17, 2019, San Diego, California USA Desktop Publishing · Illustration/Animation
Automatic Publication · Information Retrieval
MadCap Software, Inc., www.madcapsoftware.com/conference
Process Automation · IT Services · Training · Consulting ...
14th International EUATC Conference
April 25-26, 2019, Tallinn, Estonia
European Union of Associations of Translation Companies www.star-group.net
http://euatc.org/conference
STAR_ad-ML.indd 1 November/December 2018 29.06.2018 10:40:11 15Column
Client Talk
BuzzFeed
Terena Bell
Terena Bell is an independent journalist writing for
The Atlantic, Washington Post, Fast Company and
others. She is former CEO of In Every Language
and was on the GALA and ALC boards.
Welcome to Client Talk, where we chat with the people who
buy (or should buy) language services. When do they say profes- Bilingual adaptation editors in
sional translation is worth it? BuzzFeed’s major international
By talking with clients outside the sales environment, we hope to markets make up the team.
uncover what really drives purchasing. Each month offers a different D’Addario says they curate and
profile to learn from. The challenge is to find patterns from one issue to edit all translation for the country
the next. Together, what do these interviews tell us about how clients see where they’re located and “approach
our industry as a whole? curation through a combination
Gabi D’Addario is based in Los Angeles, of in-market editorial expertise
California, but spent time in Brazil growing up. and analytical models that suggest
She “learned Portuguese as a result” and, thanks translations based on previous
to a year spent living in Italy, speaks Italian as well. successes and what is trending on
Russian, Chinese and Arabic are on her hope-to- domestic channels.”
learn list. During a December 2017 presen-
tation at AI Summit in New York,
The client vice president of data science Gilad
As senior manager of international Lotan broke the model down: it’s
publishing, D'Addario oversees translation a machine learning algorithm that
for BuzzFeed, a cross-platform news and Gabi D'Addario uses social network, site traffic and
entertainment network. Based in the United other data points to choose which
States, the company markets to readers in Canada, Australia, the United news articles to localize. “We’re
Kingdom, India, Brazil, Mexico, Spain, France, Germany, Russia, Japan obsessed with giving every piece of
and the Philippines. “The content published by these editions is in the content a fair opportunity to reach
primary language or dialect of the market,” she says. “Internal communi- the right audience,” he said.
cations within these offices takes place in the local language.”
So what gets translated?
The client’s need and how they meet it Once the algorithm gods
According to D’Addario, “BuzzFeed has an in-house global adaptation shine down on certain content,
team that translates content across all verticals and formats across the BuzzFeed sends it to an outside
company: entertainment and news, long and short-form video, articles, provider, has bilingual staff
illustrations and memes.” perform the translation or starts
16 November/December 2018Colum n
from scratch with subject translation
matter in-language. D’Addario difficulties
says it’s usually the latter: “The with existing
majority of content published to industry tech
our non-English, international — like transla-
editions is created by in-house tion memory
teams in their primary language or translation
or dialect.” She does add that for management
certain types of content, BuzzFeed systems (TMS).
outsources “an initial translation But BuzzFeed
from a vendor or freelance trans- presents a new first:
lator,” but wouldn’t answer when the company is aware
we asked which type. of industry tech, but
Once translation is complete, decided to develop
D’Addario says, “An in-market, its own. Lotan’s
adaptation editor edits it for tone, algorithm is not a
style and accuracy. In other cases make the best BuzzFeed content TMS, but Buzzfeed did consider at
the translation is completed from available around the world. It's least one existing industry solution
scratch by an in-house adaptation a fast and cost effective way for before his team designed it. And
editor. Our adaptation editors us to expand our reach and take since the program predicts a piece
review any translation that we advantage of all content created by of content’s ROI — an offering
outsource before publication.” BuzzFeed teams, no matter what a top TMS would include — the
language it's in. BuzzFeed's business company has essentially built
So 1-5, how important team also relies on translation to something it could have bought
does BuzzFeed think pro- globally distribute branded content off the shelf. Maybe these system
fessional translation is? for certain advertisers.” vendors should ask themselves
A full 5. Despite outsourcing why BuzzFeed chose to invest time
relatively little work to pros, Is there an emerging pattern? and money into building some-
D’Addario claims translation is Over recent months, Client Talk thing that data-driven solutions
critical to BuzzFeed’s international has seen its share of buyers who could give them for a low monthly
branding. “Translation allows us to could solve the vast majority of their fee. [M]
November/December 2018 17Column
Terminology Glosses
AI and paradigm shifts
Laura Di Tullio
Laura Di Tullio is a terminology management
consultant who has developed termbases and man-
aged enterprise terminology for large multinationals.
She has been in the localization industry for over 20
years, holds an MA in terminology management and
a degree in translation studies.
Last century’s young people were faced with solid ideolo- [[ transhumanism: a cultural move-
gies that had liquefied by the turn of the millennium, only to ment that promotes the use of
vaporize into the intangibility of a virtual world in the next few science and technology to improve
decades. On several occasions, the idea of a paradigm shift has human beings with a particular
surfaced, more or less explicitly, between the lines of this col- emphasis on physical and cognitive
umn. We have always observed such shift through the lens of transcendence in addition to estheti-
cal improvement.
language, with awareness of the dual role played by language. [[ posthumanism: a philosophical
Language always reflects and, at the same time, always explains approach that studies “the ethical
change. implications of expanding the circle
Yet when it comes to artificial intelligence, the change is so striking and of moral concern and extending
extraordinary that the language is not entirely codified yet and even less so subjectivities beyond the human spe-
are the new concepts developing around such change. For terminologists, cies” according to Wikipedia.
this is a very intriguing moment. Change — and this time change involves Languages typically grow follow-
the very essence of human beings — is accompanied by new terms and ing concrete to abstract patterns: the
concepts that define new scenarios. Strictly linked to artificial intelligence first to appear are usually words that
are terms like: denote objects, then abstractions
[[ cyborg and The Universal Declaration of Cyborg Rights. and so on. Word formation may
[[ biohacking (or grinding): in one of its nuances, the practice of altering happen through prefixation and suf-
one’s body by implanting cybernetic devices. fixation. If a new noun derives from
[[ germinal improvement: improvements humans want to add to their an adjective, this is called a deadjec-
selves not only for themselves, but also for all the future generations. tival noun. The more of these traits
[[ genomic editing: a technology that uses molecular tools to modify the we find in the new words, the closer
DNA inside a cell. we are to real neologism. At the
[[ designer baby: a baby created based on the taste and needs of the parents. other end of the spectrum, we find
18 November/December 2018Colum n
new combinations of existing words. source language. According to the crowd sourcing and to use the data
The lexical universe of artificial intel- article, the results obtained with collected and interpreted to “enhance
ligence gives us an important sign of the studies were not very competi- the machine-learning system.” This is
how deep the semantic change is by tive, but they were still better than what it provides: real time translation
means of words like transhumanism word for word translation. service, initially in four languages:
and posthumanism: neologisms In the domain of education, German, English, Arabic and Persian;
representing abstractions or, in other language learning and assistance legal review of documents on the
words, new notions and thoughts as with languages, artificial intelligence crowd; and other services.
opposed to new things. is starting to be used productively As for our ideal termbase, this
Consider posthumanism, for in a variety of domains. Among the time we would have to pick from
example, a deadjectival noun formed others, Glossika, a linguistic company, a large selection of new words and
by adding the prefix post- and the in its webpages states that their smart specialized terms, but, as always,
suffix -ism to the adjective human. algorithm automatically adjusts to terminologists need to be cautious
This word by itself testifies that the the learner’s level, learning speed and with neologisms. In short, my pick
field is already beyond the level schedule by using a learning-method for today are the terms artificial
of naming objects: -ism really is based on repetitions. Capiche, on the intelligence (noun, full form) and its
the suffix of philosophy. When other end, is a “new artificial intel- acronym AI defined as: “An area of
appended to a stem, it can mean ligence and crowd-based innovation… computer science emphasizing the
doctrine, system or method, and it that aims to support refugees with a creation of intelligent machines that
is used in terms like realism, prag- mobile translation and search service.” work and react like humans,” based
matism, rationalism, postmodernism It is the first AI-system to integrate on Techopedia. [M]
and so forth.
However, is artificial intelligence
already applicable to the language
industry? At the end of 2017,
ScienceMag published “Artificial
intelligence goes bilingual—with-
out a dictionary,” an article that
looked at machine translation
from the angle of unsupervised
machine learning. Two studies
that had been presented, but not
peer reviewed at the time, showed
that “neural networks can learn
to translate with no parallel text.”
They could also use unsupervised
machine learning to build bilingual
dictionaries and translations at
the sentence level. The training
strategies were slightly different:
the first study used back transla-
tion, which we are all more or less
familiar with, whereby a sentence
in language X is translated in
language Y and then retranslated in
language X to see the differences.
The other study used a strategy
called denoising. In this case the
sentence was translated, then some
noise was added to it by removing
or rearranging words and then the
sentence was translated back in the
November/December 2018 19Column
Community Lives
AI and the language
community
Jeannette Stewart
Jeannette Stewart is the former CEO of CommuniCare, a
translation company for life sciences. An advocate for the
language industry, she founded Translation Commons, a
nonprofit online platform facilitating community collaboration.
How does AI affect the language community? Natural language A disclosure: I’ve been married to
processing (NLP) is a significant AI activity. Specifically, machine an AI researcher for many years. My
translation (MT), speech recognition and language understanding husband, it so happens, is a refugee
are all the focus of massive research and development programs from the so-called AI winter of the
distributed throughout the enterprise and academic worlds. There mid 1980s and early 1990s. This was
are increases in productivity due to new technologies that have a period in the UK when Sir James
Lighthill published a scathing report
revolutionized our working lives. AI’s achievements are already on basic AI research methodologies.
colossal and are driving global communications to dizzying heights. Coupled with a general pessimism
Our community has been enriched by entirely new disciplines, spear- amid the AI pantheon of big-name
headed by engineers without whose work localization, internationalization academics who were survivors of an
and globalization would remain pipe dreams. Of course, as with all change, even earlier wave of disenchantment
there has been some negative impact. Automation has altered the econom- with AI’s accomplishments, funding
ics of working as language professionals. And there is the contentious shrank and research hit the wall.
issue of human translators feeding the machine: the very machine that Although new life would grow from
threatens to make professional language skills redundant. On the one hand, fallow ground, the original impetus
AI seems to have a well-earned place in our community, yet on the other, within the so-called AI community,
its presence is perceived as a menacing threat. How exactly, then, is the effectively in abeyance, was halted
language community dealing with AI? and many highly talented people took
AI, it seems, is everywhere. Everybody’s doing it. their skills elsewhere.
chat version It therefore seems fair John MacLeod, my husband’s
to ask, is there an AI fellow Glaswegian and fellow AI
AI is simultaneously seen as having a community and are we winter victim, puts it succinctly in his
place in the language community, and part of it? The answer, in thick Glaswegian accent, “We flew
a word, is yes. But such too close to the sun and ended up in
as being a threat. a simple answer is quite the drink wi’ Icarus and a’ the other
misleading. punters wi’ big ideas.”
20 November/December 2018Colum n
MacLeod was a huge contribu- In simple terms, our enterprises, aka a time when data is king. Machine
tor to the AI community but has corporate businesses, are leverag- learning (ML) has propelled us
changed his professional path since ing technology like the Internet of forward by astonishing leaps and
those early difficult days. He does, Things (IoT) and vacuuming up data bounds. But the troubled voices of
however, point out that Daedalus in volumes that even leave cosmolo- AI researchers warn us that while
went on to complete his task of gists mind-boggled. The numbers vast corpora yield fascinating and
building a labyrinth for King Minos are indeed beyond astronomical. actionable insights from data pro-
of Crete. Mission accomplished? For example, Google Translate has cessing, insights are not necessarily
Perhaps, but whether on time and 500+ million users a day, translating knowledge! Knowledge, the fruit of
on budget, the mythologists do not some 150+ billion words a day. intelligent thinking, is what we strive
divulge. MacLeod further notes that Between Google, Facebook, Amazon to endow our brains with from birth.
many other ancient cultures have and Microsoft some 1200 petabytes As MacLeod and Stewart ask, “How
parallel myths relating the dangers of data are currently being stored. much thought goes into a transla-
of over-ambitious ideas. “AI is,” he That’s only four of the biggest! Our tor’s daily work? Is that reflected in
opines, “the biggest idea humanity friends at Cisco maintain that we are MT outputs?” There are metrics, of
has come up with. But if you think now in the Zettabyte Era of storage course, to measure quality, but these
the ultimate goals are achievable on — a zettabyte is 1021 or 1 sextillion are not hard and fast. Will our com-
a von Neumann machine crunch- bytes. Even if we only work with a puters achieve parity with human
ing squillions of bits a parsecond, small fraction of that total in our translated texts, as has recently been
prepare for your feathers to be work as language translation provid- claimed? The jury is still out.
plucked.” ers, I’m thinking we need to take a Of course, there are as many
When the conversation broached cold, hard look at what lies ahead. potential AI apps as there are smart
the idea of AI and science fiction If today’s numbers fall off the edge human activities and then some. AI
and the AI apocalypse, MacLeod just of the language universe, where on is excellent for diagnostics, predic-
laughs. “Where’s the engineering earth will they be in five years’ time? tion, problem-solving and so on. It
evidence for the robot rebellion?” he I’d say, do the math, but you need a can be used in countless different
asks dismissively. He agrees that we supercomputer for that! fields of endeavor from life sciences
rigorously need to watch our step Common Sense Advisory’s to finance to NLP. AI has as diverse a
ethically and must not play with (CSA Research) founder and chief range of applications as our intel-
fire. He is more optimistic about strategist, Don DePalma, recently ligence can devise.
our human/computer future than addressed AI in the language indus- Stewart helped me understand the
ever, but he does caution that we try in an article entitled, “Planning kind of problems AI seeks to tackle
can expect our ideas about what for the Onslaught of Artificial with a simple analogy. He’s a Sudoku
computation can do to do somer- Intelligence.” In it, DePalma does not freak, so he pointed to a fresh puzzle
saults. Homing in on the language simply address the fears and woes on his desk. “Here we have a regular
community and our rapidly-evolving of language workers; he offers great space containing a few clues,” he tells
automated world, he believes that advice to C-Suite bigwigs. It is critical me as he points to the given starting
when we move beyond conventional to plan for an inevitable future where numbers in the grid. “Our task is to
computing platforms and develop automation is the first resort. Is this a use these numbers to work out what
other tech based on, say, biological Brave New World we face? Well, ask numbers go in the blank spaces. We
architectures, we will possess syn- anyone in the food industry, brick do this by applying thought and use
thetic communication abilities that and mortar retail, or even a Tesla car a process of elimination to decide
will truly master languages. plant about robots and automation, with certainty what the solution is.
If there is still life in the AI of the and you quickly discover we are play- ML can be trained to do tasks like
past, as Stewart, MacLeod and many ing catch up in the language industry. this. In fact, it can achieve much
others staunchly maintain, where This is easily explained by the fact more complex tasks these days. But
does that put us in the present day? that natural language processing is it’s not thinking that does the trick.
I learned that there is a prevailing a very hard nut to crack with logic- It’s number crunching.” He mentions
notion among many thinking techies driven computers. the anguish of chess master Gary
that our present endeavors amount Language processing has been Kasparov and Go world champion
to more of an alliance of different central to the AI mission since its Lee Sedol as IBM’s Deep Blue and
disciplines than a coherent whole. earliest days. We currently live at Google’s AlphaGo respectively
November/December 2018 21Column
triumphed over them. These truly costs have mushroomed. The with their vision well focused on the
deep thinkers were stymied by the project was launched before all the future have already understood that
tech juggernaut. DePalma’s view technology needed even existed! In currently data is the enabler of many
is that we better get ready for a other words, the need for invention AI applications. AI does not replace
lot more Deep Blue and AlphaGo was built into it from the get-go. but rather it enhances all human
moments. We can’t put the genie Was this a wise move? Given that activities. The more we use AI in
back in the bottle, but if we use our innovation does not always hit the our daily lives the more we accept
wishes wisely, we will adapt and bulls eye the first time, the cost our partnership with the machine.
survive. There may be plenty of of fixing problems and the time As we move forward, applying AI to
AI pessimists out there, but those involved has badly impacted its our activities will become a skill that
who spend their lives working on achievable goals. will enhance our lives and make us
developing the big ideas are genu- So, is there an AI community considerably more efficient across
inely optimistic about our future in 2018? It seems that saying yes the board.
as “post-sapiens,” as Stewart and involves a genuine attempt to link All communities face change and
MacLeod describe us. academia with business, justifying sometimes those changes can be
But let’s pause for a moment expensive work on some very far-reaching and have transform-
before we get too carried away. AI futuristic-sounding ideas. ing effects. Research analysts like
can learn a salutary lesson from However, I had a recent conversa- DePalma effectively warn us not to
NASA and the James Webb Space tion with Francis Tsang, head of be caught out like lumbering dino-
Telescope (JWST). The NASA international engineering at LinkedIn, saurs. I strongly advocate that we in
website declares that, “opportuni- who has a solid grasp on AI’s current the language community embrace
ties for collaboration will highlight status in our industry. AI and make a strenuous effort to
our common interests and provide Tsang believes that perhaps there bring members of what passes at
a global sense of community.” They is no AI community because there present for the AI community and
certainly have made awesome is no clearly marked AI industry. AI bring them on board. As the pace of
contributions to progress with in its full capacity is a way of life that technological change gathers more
the Apollo program, the Mars we will all need to embrace in order and more momentum, we need to
missions, the ISS and countless to fulfill its massive potential. AI has innovate processes, services and
other projects. But the JWST has been creeping into our everyday lives abilities that will give multilingualism
been something of a nightmare as human beings and into our work. its rightful place in our globally-
as its schedule has slipped and When it comes to companies, those networked world. [M]
A thousand different workflows. One Solution.
The Business Management Solution for the Translation Industry
www.plunet.com
22 November/December 2018Join the conversation on
Go Global, Be Global
Discover LocWorld in 2019
LocWorld39 LocWorld40 LocWorld41
Kuala Lumpur Portugal United States
February 27-28, 2019 11-13 June 2019 Fall, 2019
The world’s #1 Localization Conference & Exhibition Series
Find out all you need to know at
www.locworld.comColumn
Perspectives
The future of localization in India
and other emerging markets
Devaki Datar-Kunte
Devaki Datar-Kunte has run the family business, Languages
Services Bureau, established in 1979 by Madhuri Datar, for
the last 16 years.
The advances in information and communication technology In the case of Hindi, only half of the
have contributed to the formulation of a global business con- Indian population can understand it.
text, though one of the most fundamental trade barriers still A recent survey by KPMG about
Indian language internet users has
exists: language. That’s why it is important to not only localize
brought to light eye-opening data
your products and services into the target market’s language, about the usages and preferences
but also understand the customs, culture and sensitivities to of these new users when it comes
ensure the product is a success in that market. to language support. By 2021, it is
“Content is king,” they say…but of what use is it if it’s in a language one predicted that the number of Indian
does not understand? Businesses have long ignored the language impact of language internauts will be thrice
their communication! as compared to English language
Africa is something of a last frontier in this regard — there is high internet users, and will constitute
language diversity here, though Arabic, French, Portuguese and English 75% of India’s internet users.
are supposedly widely known due to past colonization. It is home to one- It is therefore a myth that English
third of the world’s 6,000 languages. Swahili is the most spoken language, will be all pervasive, or Hindi will help
followed by Amharic, Yoruba, Oromo and Hausa. What’s more, seven of you communicate with your target
the ten fastest growing economies are in Africa, beating even the Asian customers. A consumer is more likely
tigers in growth rates — so this is a market a business can’t afford to miss! to buy a product or service when com-
Asia, especially India, will be the global growth engine for various municated to in his or her language,
industries, thanks to its growing consumer power and internet penetra- as proven by many surveys the world
tion in the remotest corners. India is a densely multilingual country. over. In a survey of App Annie, it
Despite having 30 languages spoken by more than a million native speak- was found that users in most Asian
ers each, and over 1,600 dialects, it is still considered an English-speaking countries prefer to download and buy
country. However, roughly 10-13% of the population understands English. apps localized in their language. On
24 November/December 2018Colum n
the same lines, according to the to be an essential corporate strategy insights on cultural sensitization,
KPMG survey, 88% of Indian lan- to succeed in today’s market. content optimization, geo-political
guage internauts are more likely to End-to-end localization enablement sensitivities, various scripts of
respond to an Indian language ad. is a critical requirement. Take, for the same language, locale-specific
After having saturated markets in example, a user who clicks on a terminology and so on. During our
the Tier 1 cities of the country, busi- local language ad and lands on a 40 years of business experience
nesses are already looking at the Tier page in a different language. Or in the industry, we have seen
2 and Tier 3 cities, as well as the vil- perhaps the user finds that some businesses facing entry barriers
lages of India, for untapped markets. images on the website are in a due to lack of localization of
The rural internet users, in fact, have different language. This internaut product or service information.
a higher engagement level than their is bound to be disgruntled! Sometimes companies have
urban counterparts. However, most Every business stage, right from tight budgets for localization,
of them state limited language sup- discovery R payment R fulfillment whereas marketing enjoys wealthy
port as a major challenge for access- R after sales service, needs thorough budgets. They fail to understand
ing various online apps and sites. localization, not only in terms of that localization is critical to
Local language enablement content, but also in terms of images, marketing, in an international and
is therefore imperative for audio and video, currencies, timing, local frame of reference. Unfor-
the following categories: measurement units, color schemes, tunately, such businesses don’t
[[ E-tailing political correctness and geographi- even have their marketing and
[[ Online government services cal boundaries, just to name a few. commercial material or website
(fastest growing category) Though machine translation is ready for the target market!
[[ Digital payments the current buzz for the language On the other hand, in an actual
[[ Digital news industry, it is still in a nascent stage business case, one of our clients saw
[[ Digital classifieds for all Indian languages. Partnering a 40% rise in enquiries after multi-
[[ Digital entertainment with a professional language com- lingual localization of their website.
[[ Chat apps pany is therefore imperative to be Only those businesses that convert
[[ Social media platforms able to speak locally to your clients. this tremendous opportunity in
[[ Apps When companies write their India and other such roaring mar-
[[ Video content corporate content or localize it, it kets with the help of localization can
Translation and localization as is crucial to have a language service hope to stand their ground in times
an integral business practice proves provider to guide them with their of disruptive business trends. [M]
November/December 2018 25W h i te Pape r
Applying AI to NT, MT and Beyond
By Aleš Tamchyna, Memsource AI Engineer segments can be left without any changes. Some of them are
In the translation industry, the disruptive effect of AI is not very simple (numbers, email addresses) but others can be
yet apparent. Machine learning/AI has traditionally been quite difficult to detect — consider the case of “Madrid” not
associated with one feature: machine translation (MT). It is requiring a translation into Czech (where the same spelling
true that with the recent advancements in neural MT, the is used), or the snippets of software documentation.
output quality is inching closer to human translation.
Our customers report significant savings thanks to this
However, neural MT still makes serious mistakes and its feature (read Semantix’s findings on our blog). Being
quality can be upset by more complex sentences. More able to offer a competitive price up front and saving
importantly, professional translation has different stan- overall costs is important. But the feature is beneficial
dards than simply passing for human translation; transla- for translators as well: segments that do not require their
tions might require a specific style, consistent terminology, attention do not waste their time, thereby allowing them
coherence across sentences and paragraphs, etc. to focus on the more interesting cases.
But a translator’s goal is to convey the original meaning as This approach fits into our overall goal of automating
closely as possible. They have to carefully navigate ambigu- tasks that are tedious and repetitive. If a task can be fully
ity and craft wording that best reflects the emphasis in the automated by AI, it is probably not enjoyable or interest-
original text, ensuring that there can be no confusion about ing for humans to work on in the first place — their time
the meaning. Consider the severity of mistakes within legal and effort is better spent focusing on the hard cases and
texts or medical records; in marketing, a good translation creative solutions.
can be the difference between a successful campaign and
an international embarrassment. It’s clear that MT is not Detection of non-translatables works with a high pre-
going to replace human translation anytime soon, if ever. cision and is designed to minimize the number of false
positives (marking a regular sentence as a non-trans-
With this in mind, we have focused on applying AI to latable is a major problem). Still, user adoption of such
enhance the translation process for all involved. a feature can be a complicated process and we expect
this to be the case for most AI-based systems. Any error
Non-translatables made by the AI has a very negative effect on users’ trust.
We first took a step back and looked at the process as a The fact that models based on deep neural networks
whole. Our goal is to find equally novel and efficient ways are hard to interpret makes matters worse. When a
of applying AI. user asks why the AI makes a seemingly random error,
even the AI developers may not be able to explain the
Of course, in translation, the exact workflow varies
behavior. Yet deep learning often outperforms previous
greatly from company to company and each role is very
approaches by such a large margin that even the model
different. Despite the complexity, we can identify some
opacity can be a reasonable trade-off.
common subtasks that are rather repetitive and could be
automated or at least simplified. Machine translation
While translation memory greatly helps to minimize the
Our first AI-powered feature can serve as an exam- monotony of translation by allowing translators to reuse
ple. We developed a system for automatic detection of translations of similar segments from the past, not every
non-translatable segments. project has a suitable translation memory to start with.
Moreover, even segments with a good TM match can be
Since the feature is powered by AI, it is not limited to a dull and time-consuming to translate.
fixed set of “types” which can be covered by rules or regular
expressions. The system learns from data to recognize which We believe that in such situations, MT can greatly help.
sponsored information multilingual.com/white-papers
multilingual.com/whitepaper
26 November/December 2018You can also read
