10thingsbook v4

10 THINGS YOU
PROBABLY DIDN’T
KNOW ABOUT MACHINE
TRANSLATION
What the Global 2000 Know
About Machine Translation
that You Can Leverage

1 10 THINGS YOU PROBABLY DIDN’T KNOW ABOUT MACHINE TRANSLATION

// INTRODUCTION
LexWorks provides industrial-strength machine translation for companies who
need a high volume of content localized. We use machine translation to provide
our customers with a speed and cost advantage without compromising on quality.

With our parent company based in Paris Through our work with the Global 2000, we
and offices on 3 continents, Lexworks have helped some of the most successful
counts many of the largest and best run companies in the world to leverage
companies in the Global 2000 among our advances in technology – particularly
top customers. machine translation – to leapfrog their
competition into international markets.


// WHAT THE GLOBAL 2000 KNOW
The Global 2000 are under constant pressure to bring
their products to new markets with all due speed.

The average multinational company routinely translates
content into 8-16 languages, depending on the markets they
serve. The amount of content is increasing as well, thanks to
user-generated content.

To build sustainable translation
processes, the Global 2000 made an
early investment in machine translation
and are farther down the experience
curve with this technology compared with
other companies. Over time, the Global
2000 have been able to reduce spending
on translation and speed up the release
of content across international markets.


// WHAT YOU NEED TO KNOW
01 This Isn’t Your Father’s Machine Translation

02 Don’t Even Think About Doing It Yourself

03 There Isn’t One Kind Of Machine

04 Your Data Security Matters

05 Look For “T–shaped” Companies

06 Start With A Pilot Program

07 Bring In The Experts

08 Invest In Training

09 Reap Savings With Translation Memory

10 Keeps The Humans For Quality Control


01 This Isn’t Your Father’s
Machine Translation.
“MT is often associated with a compromise of quality in favor of cost and turnaround
improvements. Contrary to all expectations, using machine translation has improved our
quality. In blind tests with each major language, reviewers consistently evaluated the content
as equal or superior to our traditional translation process.”

In 1954, in the New York head office of Now, a few advances in artificial intelligence
IBM, the first demonstration of a machine and machine learning later, machine
translation system took place – based on a translation is better – and we’re smarter
vocabulary of just 250 words. But despite about the processes to optimize it. While
the early fanfare that accompanied this, it may once have been an overhyped
automatic translation came to be identified technology, today Global 2000 companies
with translations such as this one, which are using machine translation to reach new
in 2007 nearly provoked an international customers and communicate with their
incident when sent by a journalist to a visiting international staff. Like Apple’s progress from
Middle Eastern diplomat: “Helloh bud, the Newton to the iPad, machine translation
enclosed five of the questions in honor of is at a point where it can deliver on its
the foreign minister: The mother your visit original promise.
in Israel is a sleep to the favor or to the bed
your mind on the conflict.”

Today’s machine translation has moved on
from this inauspicious beginning.


02 Don’t Even Think About Doing
It Yourself.
“For security reasons, we use an internal engine which we trained
ourselves. In order to reduce the time spent reworking translations we
turned to LexWorks to train for our 6 languages and 5 business units.”

It’s tempting to go to a machine translation vendor, purchase an engine,
set it up, and train it on a set of documents. Tempting like that extra slice of
chocolate cake is tempting

The reality is that these kinds of experiments almost always end up costing
you a lot in time, money, and most importantly, missed opportunity.

Instead focus on what you do best. Setting
up and customizing a machine translation
engine is not for the faint of heart and takes
expertise and skills that most enterprises
don’t have readily available.


03 There Isn’t One Kind
Of Machine.
“We use a rules-based engine for our documentation, but
for real-time translation for our online customer support,
we decided on a statistical tool with 24/7 availability.”

Rules based. Statistical. Hybrid. These are the three types
of engines available from the vendors who sell machine
translation technology.
They are all good choices depending
on the type of content you need to
translate, the source language, the
target language(s), and also the
level of quality desired.

Best practice companies know that
the single best way to cut through
the complexity and get to the ROI
they need is to call on outside firms –
firms who understand the technology
inside and out and who know how
to select the approach that will
perform the best.


04 Your Data
Security Matters.
“...a series of security problems because emails not
intended to go out behind our firewall were being
uploaded onto an open translation server.”

One localization manager we interviewed at a multinational
said it this way.

“We tried working with a new company that offered
translation based on a software-as-a-service model at a
low price per word. This seemed like it would work. Until
it didn’t. We had a myriad of problems not the least of
which was a series of security problems because emails not
intended to go out behind our firewall were being uploaded
onto an open translation server.”

Protect your IP by looking for service providers with
experience placing machine translation engines both
on premises and off with enterprise-class security.


05 Look for “T-shaped”
Companies.
McKinsey – the eminent management consulting
firm – defines two types of consultants. I-shaped
and T-shaped.

People who are I-shaped have deep expertise in one or LexWorks is one-such T-shaped company with a
more subject. T-shaped people – on the other hand – background in both linguistics and technology. LexWorks
demonstrate both deep expertise in a single subject plus a has been providing traditional translation services
broader based strategic understanding of business issues. for more than a quarter-century and has a decade of
To truly get the most from today’s machine translation, experience in cutting-edge machine translation.
look for an outside firm that is not all technology (“I”), not
all linguistics (“I”), but demonstrates depth of knowledge
in both these areas along with the ability to collaborate
with your business decision makers.


06 Start With a
Pilot Program.
“After a successful pilot with the French and German, the
process has been tested and is now expanding to the
remaining European languages.”

Is your content suitable for machine translation? What will A busy litigation department that is under the gun with
your Return on Investment be? To find out, ask a potential discovery may need less quality and a faster turnaround while
service provider to set up the pilot for a single department a manufacturing department responding to international
with a high volume of translation needs, say 1000 pages of RFPs will need higher quality. The purpose of your pilot is to
content across two or more languages. determine whether machine translation is a methodology that
will work in the real world and deliver real world results.
This enables the service provider to customize the machine
translation engine to a specific set of content and to meet a
specific quality goal. Processes can be fine-tuned to meet
your turnaround needs.


07 Bring In
The Experts.
“Our partnership with LexWorks has resulted in an average of
20% cost savings and 30% improvement in time to market.”

While there have been many advances in machine translation, it On the front end, people are needed to
remains true that to deliver quality at the same level of POTBH customize and train the engine on your
(plain-old-translation-by-humans), machine translation needs to rely company’s content. On the back end, linguists
on a repeatable process that uses people in small doses on both the and subject-matter experts may be required to
front end and the back end. post-edit the results, so as to enhance the quality
of the raw output. We say “may” because it
really depends on the quality level you need.
(See point #10 below.)

To have a sustainable process in place, round
up a village of experts to set up and run your
project for success. These should include an
experienced project manager, linguists and
experts in the subject matter(s) specific to your
company or industry sector. Or, better still, hire a
company that already has those experts in place.


08 Invest in
Training.
Working with LexWorks to automatically translate our international
RFPs with an engine that is trained for our industry (transport)
means that we gain an average of 5 days...”

Some early investments in Great post-editors are hard
machine translation did not work to find. They’re real experts in
out. The problem? Almost always linguistics and your exact subject
it’s not spending enough time matter. Work with professionals
who know how to train your
training and customizing the
machine translation engines
machine on enterprise-specific
properly so as to avoid this kind
content. This results in raw output
of burn-out.
from the machine translation
process that is low quality. Post-
editing content from poorly
trained MT is very time consuming
and will burn out your team.


09 Reap Savings With
Translation Memory.
“In addition to the 42-49% savings from using Translation Memory, we
benefit from an additional 10-30% from using MT with TM, for a total
average savings of 48-64% from translation technology alone.”

If you are unfamiliar with translation Translation memory works best when
memory, here’s how it works. Most the content domain is well established,
content includes phrases you will say with an existing product. But
repeat multiple times. With translation companies get the majority of
memory you save those translated their revenue and profits from new
phrases in a database. This gives you products. Also, the way you wrote
a set of phrases in pairs that you can about your products and services five
use as building blocks when translating years ago is probably not the way
additional content. you write about them today: things
change, and your translation memories
Translation memory cuts your may well be out of date.
translation costs because you end
up translating 1x phrases that come And this is where machine translation
up again and again. really shines. With your company’s
translation memories as training data,
It turns out that translation memory is LexWorks builds highly customized
a great strategy to reduce translation machine translation engines. And, you
costs. However, unless you have only can count on these engines to improve
just started working with translation over time.
memory, most likely you have already
realized most of the savings it can
offer; for additional savings, your
company needs to look at using
machine translation.


10 Keep The Humans
For Quality Control.
“We were surprised to see an improvement in our quality. For the
French files, the reviewer reported that the translation rated 8.25/10
for MT, while the traditional TM translation was at 8/10. For German
the MT translation scored 9/10 and our reviewer said: “It was the best
translation of courseware I ever read.”

The raw output of machine translation typically delivers This turns out to matter a lot because not all content
quality that is at the 60-70% level. Lots of errors, some needs to be translated to the same quality level. Some
of which get in the way of understanding. Apply subject- of it will need to be close to perfect. And some can pass
matter expertise to the post-editing process and you can muster at a lower level of quality.
turn the crank on quality and get it up to the same level as
plain-old-translation-by-humans. Machine translation enables you to take content you
normally could not afford to translate – for example
user-generated content – and make it available to your
customers. You can do this in a wholesale fashion or take
more of a self-service approach, by making a widget
available on your support site so customers can translate
on demand.

Either way, what you are doing is using machine
translation to cut the cost of translation while upping the
amount of content you make available in your customers’
languages. Which can also dramatically cut your
customer support costs.


Share This eBook

// NEXT STEPS
Global 2000 companies capitalize on advances in machine We’ll help you develop an ROI pilot for machine translation
translation to translate more of their content into more with a full feasibility report. We’ll work with you to help you:
languages.
• grow international revenues
Their goal? • decrease time to global markets
• increase customer satisfaction
A sustainable process that costs less and delivers more.
• reduce costs and
We’ve talked about ten ways you can leverage what they’ve
• improve profitability.
learned.

To get started on the journey to a more sustainable
translation process, email us at info@lexworks.com.

// ABOUT US // CONTACT US
LexWorks provides industrial-strength machine translation for Call: 1-604-303-9100
companies who need a high volume of content localized. We use Email: info@lexworks.com
SMART TM machine translation to provide our customers with a speed Visit our website: www.lexworks.com
and cost advantage without compromising on quality. LexWorks
has the people, processes, and technology to turn today’s most Follow LexWorks
advanced translation tools to your competitive advantage. Twitter: @LexWorksMT
Facebook: facebook.com/LexWorks
LinkedIn: linkedin.com/company/lexworks

United Kingdom | France | Venezuela | Canada
© 2013 LexWorks, Inc.


10thingsbook v4

More Related Content

Similar to 10thingsbook v4 (20)

More from Marcia Kadanoff (8)

Recently uploaded (20)

10thingsbook v4