style, your wording, your content. Utilize translations quickly and
throughout your company.
machine translation powered by STAR MT
the short video for more information on STAR MT functionality and
The Multiple Birth of Adaptive Neural MT
the rush to embrace neural machine translation (NMT), it has finally
arrived at our -- the translators' -- doorstep with adaptive models.
And at this point, not just one has been released but three, two just
in these past couple of weeks. They really are three different kinds of
models because they each use a very different approach to adapting the
MT output -- so different, in fact, that I wonder whether they should
all be called by the same "adaptive" moniker.
start with the first one, KantanMT's
neural system, which has been in use since March of this year.
I talked with KantanMT's CEO Tony O'Dowd about this, it became
increasingly clear to me how disadvantaged an MT provider is who does
not also offer a full-fledged translation environment. To put it
differently: There are equally important fields of responsibility on
the side of the translation environment provider and machine
translation provider to make machine translation work smoothly and
productively in a translator's workflow.
back up first, though. Here's how KantanMT deals with neural
machine translation. As for any neural machine translation system, KantanMT's
initial processing requirements to train neural engines are very high.
Tony mentioned that for the first training pass, the processing of a
100-million-word bilingual corpus takes three to four days. It's
overwhelming to imagine that this would have to be done again and again
for an adaptive system. The nice thing about neural machine
translation, however, is that the ongoing training for adapting the
translation model does not require a complete retraining (as it did in
many cases with statistical machine translation, with the notable
exception of Lilt -- see below).
has set up the incremental training of the neural MT engines by
collecting data in a TM of sorts (which first is just kept in the
temporary cache of the server but is converted into a file once the
user logs out). At certain intervals (but not every time a segment is
finalized), this TM is fed into the neural engine for adaptive
purposes. In the meantime (and after the passing of the data), the TM
(in KantanMT-speak "Total Recall") is used as a translation memory that
acts very much like most TMs in combination with MT: perfect matches
and any match with a fuzzy match rate of 85% or higher is preferred to
MT suggestions and entered automatically to be confirmed or edited by
the translator. Otherwise, the neural machine translation engine
suggests translations. Since the engine in most cases is solely trained
on the data that the client has provided and is being adapted and
improved as the projects are being processed, there is a relatively
good chance that the suggestions are of reasonably good quality (at
least as far as terminology is concerned).
might have the same question I had: How does this process using a
front-end TM differ from any other translation environment tool? It
doesn't really, if not for the fact that both the MT and the TM are
cloud-based and therefore available to everyone in a real-time
translation team. And while this is also available in other
server-based workflows offered by the various translation environment
tools, more often than not, this is not set up for projects that you
and I work on.
here is the hitch: While many translation environment tools support the
use of KantanMT, only a handful allow its use upstream (i.e.,
sending data back to the KantanMT's cloud so it can then be
used across translators). The ones that do include Memsource, Trados
Studio, and (in a limited fashion) Across. Tony's
(unsurprising) prediction for 2018? "Next year adaptive MT will become
the big thing that will be seen in CAT tools," meaning that tools like KantanMT
will have more complete access.
main clientele for KantanMT is clearly the translation buyer.
Early on there were plans to also offer products to other stakeholders,
but its easy access to large amounts of focused data made the
translation buyer the primary target. Still, additional features that
would be helpful for translators would be welcomed, such as more
interactivity between the machine translation engine and the "Total
Recall" TM engine to "fix" TM matches (with MT) or MT suggestions (with
other engine that has just gone live (within Translated's translation
environment tool MateCat
and the paid Pro edition of the MyMemory
app for SDL
Trados Studio) is ModernMT. ModernMT
is a three-year EU-funded project with a number of partners, including
Translated, much like MateCat itself was a few years ago. If
you remember, MateCat's original purpose was to "investigate
the integration of MT into the CAT workflow [with] self-tuning MT that
adapts MT to specific domains or translation projects and user-adaptive
MT that quickly adapts from user corrections and feedback" (Source: Proceedings
of the 17th Annual Conference of the European Association for Machine
Translation). While the adaptive system worked reasonably well,
that part was unceremoniously and frustratingly dropped from MateCat,
and the EU agreed to confer another three-year contract. This time the
adaptive MT is here to stay, according to Translated's Alessandro
Cattelan, whom I spoke to for this report.
gotta feel for all these MT developers working so hard on the cutting
edge of technology when, just like that, a new and improved technology
enters the field and, alas, all their prior work is ready for the trash
heaps of technology history. I'm not completely sure this is entirely
how it went with ModernMT, but I would guess it's pretty close,
since only this summer they turned their attention to neural machine
translation after having spent more than two years on statistical,
adaptive MT. Amazingly enough, they were still able to present a result
just a few months later.
adaptive part of the technology is fundamentally different than other
adaptive engines because there are actually no changes in the baseline
engine happening at any time. Instead, the system uses a technology
called "instance-based adaptive NMT." Similar to KantanMT but
for a different purpose, this consists of the translation request first
being sent to a TM layer (which can consist even of a relatively small
TM as long as it's highly-tuned). With similar segments found in that
TM layer, the NMT engine's "hyperparameters" are adapted on-the-fly so
that a more suitable suggestion is generated. This concept is based on this paper by the Fondazio
Bruno Kessler, which is part of the consortium working on ModernMT.
benefit -- in theory -- is that you don't ever need to actually train a
specific MT engine, but you can instead use a large generic engine
whose suggestions are specialized by having the query parameters
adapted as the translation is happening.
tried to work with the engine in MateCat after I talked with
Alessandro, but I wasn't able to get any suggestions other than the
ones from Google Translate (you can't actually choose the
engine because the system selects it for you).
have to be honest and say that I don't completely understand the
concept of this technology, but if it is indeed able to produce better
results, then all power to it. I am pretty sure, though, that
"adaptive" is a bit of a misnomer. Not because the machine translation
suggestion is not being adapted (it is!), but because the process is
different from what is typically understood as adaptive MT. Maybe a
term like "responsive," "reactive," or even "ad-hoc-tuned" might be
is completely open-source (in fact, you can download everything from
its website), but that does not include the data, of course. This is
where Translated's hyper-TM MyMemory comes in
and where the company sees a possibility for itself to market this
solution -- to translators and LSPs as a paid service through MateCat,
Trados, and possible other translation environments
and to large translation buyers as an in-house solution. (According to
Alessandro, at least one large Silicon Valley company has already done
some extensive testing and found the results better than those from Microsoft
Translator Hub, the trainable machine translation engine Microsoft
also asked Alessandro how the non-adapted baseline engine compares with
the Google Translate NMT engine, and he was honest enough to
say that it produces worse results -- unless the adaptive process is
taking place and then it's better.
SDL Trados Studio 2017 - the complete CAT tool for professional
a free trial of the leading CAT tool SDL
Trados Studio 2017 and enjoy delivering projects faster,
maximum reuse of previous work and accepting any type of job.
your free trial >
The Multiple Birth of Adaptive Neural MT (continued)
Lilt has also released a
neural MT engine (for now only between English and German and English
and Chinese and only for select users -- see more on this below). Since
the team at Lilt consists of remarkably young and energetic folks, they
not only introduced their new engine but also completely converted the
user interface, oh, and rebranded (see below as well).
yes, they also settled the silly lawsuit SDL had brought against them
273 of the Tool Box Journal).
(The very non-communicative and, it seems, legally prescribed official
statement from Lilt: "Lilt is pleased with the resolution of this
dispute. The settlement was mutually agreeable. We will continue to
focus on products and services that democratize access to information.")
I was really intrigued and -- I admit it -- at first puzzled about
Lilt's move toward neural machine translation. Here's why: Lilt
essentially broke down all the borders that existed between the
different assets translators use by creating one single database for
all the data used by the (at that point, statistical) machine
translation, the translation memory engine, and the automated term
extraction/termbase engine. This was made possible because the data was
sitting in an open text-based table that could be utilized for any of
these options. Even though I'm not super-technical, I also knew that
wasn't possible when it came to neural machine translation, where the
translation model really consists of numbers rather than language data.
So my confusion had to do with me wondering whether Lilt had thrown all
of its basic concepts overboard.
out it hasn't, at least not completely. As in the previous incarnation,
there is still a massive text-based table containing data that the TM
and the termbase features are based on and that is being used to train
the neural machine translation engine initially and continually. (The
difference involves a different level of immediacy of the data in the
data table and how it's used for machine translation purposes.)
NMT engine is trained on the same bilingual data as the previous SMT
engine (that's at least true for the four existing language
combinations) minus the monolingual data needed for SMT but not for NMT.
engine interacts with the user as he or she translates through the same
process as previously: Every time a word is entered, the machine
translation engine calculates a new suggestion for the rest of the
segment to match the entered data. (Lilt's Carmen Heger, whom I talked
to about this, mentioned a barely noticeable increased delay in the
response. She was right, it's hardly noticeable; in fact, I didn't
like before, every time you confirm the translation of a segment, the
machine translation data model (which now runs parallel to the data
table) is immediately adapted -- in fact, much more efficiently than
before. And just as in the case of KantanMT, while the initial
training and so-called batch updates (like when you upload a whole
translation memory) are very resource- and processing-intensive,
incrementally adding to the neural engine is not, allowing it to be
done on regular CPU servers (vs. the more powerful GPU servers).
what about the quality of the suggestions? One important consideration
when answering this question is that with Lilt it does not
primarily have to be the initial suggestion for the whole segment to be
useful as long as the constantly evolving suggestions for each segment
are useful. In my very limited and unscientific testing in EN>DE,
the initial suggestions seemed OK but tended to be less accurate than Google
Translate or DeepL (for test purposes I usually copied and
pasted the same segment in the other interfaces). But when I diverged
from the original suggestion they became ever more helpful, and tended
to be more helpful than in the earlier SMT-based Lilt.
is a bit more scientific data: Lilt ran a case study with one of their
corporate partners comparing the statistical vs. neural quality by
professional translators. The results: NMT was chosen 44% of the time,
29% of the time translators chose SMT, and in 27% of the time it was
we're talking about corporate clients, you might have seen that Lilt
has also done some rebranding with the slogan, "The New Engine for
Enterprise Translation Workflows." What does this mean for translation
professionals? Not much as far as the availability
of the tool, but it does reveal that we haven't been as eager to
switch to a new technology paradigm as some might have expected. And
that in turn may point to something more profound (forgive the
over-generalization): Traditionally translation professionals have been
very slow to accept technology. But once they
accepted translation technology as a normal part of their work
environment, they were (and are) just as slow to change their chosen
technology (which was so hard to accept in the first place). Maybe it's
just the human condition, but I think what we might not completely
understand is that being technologically adept by definition means a
constant willingness to change the technology that becomes available as
I'm not saying this because I think Lilt is the one and only
kind of technology out there. It isn't. There are plenty of interesting
new tools and technologies. And if you honestly look at yourself, are
you completely open to continuously moving on? (I have to admit that
I'm not, and I also know that I would be a better translator if I were.)
finish this up, as mentioned above, Lilt also has a complete new
interface as you can read in this
document. I like that there's no separate editing interface, it's
now very easy to switch between a vertical and a horizontal display,
and there is a better sense of accessibility -- even as far as
accessing available options and keyboard shortcuts.
did not like the "on-demand tag editor" which, while a good idea in
general, did not work as seamlessly as it should have.
and I forgot to mention that the neural mode is at this point available
only for corporate clients (and LSP clients on demand), but it will be
available for all sometime in the new year with all language
Friday Special: Save 50% on the Across Translator Edition!
your crossMarket Premium Subscription including the Across
Translator Premium Edition. Join now and save 50 percent for
the first year. Log in to crossMarket, click Upgrade,
and enter the discount code BLACK50 during the payment process.
only on November 24.
from LinkedIn, Twitter is my preferred
social network, and it can also be very productive -- productive for
networking with colleagues, for meeting clients, and for publicly
displaying who you are and what you stand for. In relation to other
forms of social media, it's also less unproductive because it doesn't
necessarily require you to read never-ending posts (though it might
lead you to some of those), and you're supposed to express yourself in
a relatively precise manner (though this brevity has been a little
diluted by the recent extension to 280 bytes, i.e., 280 single-byte and
characters per tweet).
looking at some colleagues' tweets, I've noticed a couple of tips that
might be worth repeating:
start a tweet that is supposed to be seen by everyone with @username
because this will be displayed on Twitter's homepage only for username plus
everyone who follows username and
you. Enter a period (or some other character) before the @ sign. The
same is true if you reply to someone.
- If you
feel like tweeting about private things as well as professional things,
set up two different accounts. Really! I immediately unfollow other
tweeters who start to regularly report on personal matters -- and while
it's not important what I do, many of those who you really want to
reach are doing the same. It's not enough to say that you don't promote
your Twitter account anywhere in your professional materials -- as soon
as you participate in a professional discussion, you are effectively
- If you
get frustrated with someone's barrage of tweets but you don't want to
frustrate them by unfollowing them, you can also "mute" that account by
clicking the down-arrow at the top of one of their tweets and selecting
credit where it's due. If you find something through someone else's
tweet or email, mention that in any related follow-up tweet you might
send yourself. If you post a link to an article that someone
interesting wrote, research his or her Twitter handle and mention that
in the tweet as well. And if you're retweeting someone but have changed
some of the content, precede it not with RT (retweet) but with MT
(modified tweet) or HT (hat tip).
#hashtags judiciously, not to #the point that #your #tweets are really
#hard to #read. It's probably a good idea to set a hashtag in front of
a central term (like #translation
or #xl8), but I even prefer not to do that. Instead, I really enjoy
using the hashtag as a way to explain or comment on my own tweet.
are some other helpful tips:
- It's a
good idea to download your archives of tweets every once in a while --
especially if you tweet a lot. This will help you to quickly locate a
tweet you might have sent a long time ago. You can do that under Settings>
Account> Request your archive.
- You can
also use helpful keyboard shortcuts in the non-mobile Twitter
interface. To see all of them listed at once, just press the question
mark key when you're in the Twitter web interface. (Cool, huh?)
the best way to (legitimately) gain more followers is to post
interesting and engaging tweets. If you want to speed it up a little
bit, here are a couple of ways to follow others so that you engage with
them and hope that they will follow you back:
there are translation-related conferences happening (and they're always
happening somewhere...), click on the conference hashtag to see who is
tweeting from the conference. Not only will you get a good (and cheap)
overview of what's happening at the conference, you can follow the
tweeters and maybe even engage them in a conversation while staying
under the hashtag umbrella.
tweets under translation-related hashtags such as #xl8 and consider
it's not possible to format text in tweets the conventional way, you
can perform a Unicode conversion. This
converts your text into formatted look-alike characters. The drawback
is that this text is not
not searchable but helpful for text that is too long to tweet: make a
screenshot of text and embed it as a graphic in a tweet.
additional benefit of an embedded graphic is that, unlike hyperlinks,
it doesn't count against the character limit. Plus you can tag up to 10
Twitter users in a graphic to involve them in a discussion.
- If your
tweet links to a source in a language other than the one you usually
tweet in, be sure to note it [SV] or [ZH]...
use services like "The Daily ... is Out" (paper.li) or scoop.it. This
is annoying for followers since it requires more clicks. Also, don't
have services like commun.it or SumAll.com tweet how many retweets you
had and how many followers you gained last week. No one is really
auto-tweet from LinkedIn, Xing, or Facebook and have part of that tweet
quoting from an article or blog post, be sure to <offset> your
tweeting a link to a news article, research the Twitter handle of the
article's reporter to include it. Do that especially when you're
critical of the article. This can lead to great discussions that might
help you and others within the world of translation.
tweeting a link to a news article, consider making the headline in the
tweet descriptive rather than using the original title, which is often
formulated as click-bait.
- To do
an advanced search in Twitter, you can go to twitter.com/search-advanced. Or
you can use the Twitter-specific syntax in Twitter's regular Search
field. Helpful search parameters include from:user and to:user to
search for sender and recipient (if applicable) respectively.
- If you
want to shorten a URL in a tweet, use xl8.link as a
URL shortener to show some pride in being a translation professional.
don't just gain colleagues as followers but also potential clients.
Think of Twitter as your interactive, ever-evolving business card.
You'll be glad you did.
all the world's best dictionaries at your fingertips!
a fixed monthly fee, on all your devices, integrated in your daily work
Unlimited -- One service, 5 applications and more than 260
here to find out 10 things you need to know about WordFinder Unlimited
The Tech-Savvy Interpreter: Interpreting and the Computer? Translating
and the Computer 39 (Column by Barry Slaughter Olsen)
recently returned from London where I attended the 39th
installment of Translating and the Computer (TC39), a yearly scientific
conference organized by the International Association for Advancement
in Language Technology, or ASLING for short. Translating and the
Computer has been around since 1978 and will celebrate its 40th
anniversary in 2018. The original organizers were prescient indeed and
clearly foresaw the potential of the computer to assist with
conference brings together researchers, academics, and translators (and
this year interpreters too) for two days of learning and discussion
about the latest developments in machine translation, corpus
linguistics, translation memories, and other technology topics
affecting translation. As one might expect, the big focus this year was
on neural machine translation (NMT)-how it works and how it might be
folded into the translation workflow.
interpreting and technology, however, TC39 was a watershed. The
organizers decided to include a track on interpreting and the computer
for the first time in the history of the conference. They even
dedicated the Friday keynote address to interpreting and the computer-a
speech by Dr. Alexander Waibel on the current state of speech-to-speech
translation and the use of deep neural networks for speech recognition
combined with machine translation. The interpreting track ended with a
panel discussion on "New Frontiers in Interpreting Technology"
moderated by Professor Danielle D'Hayer from London Metropolitan
University. Panelists included Dr. Anja Rütten, Alexander
Drechsel, Joshua Goldsmith, Marcin Feder and yours truly.
could go on and on about the presentations, from what kinds of tablets
are being used by professional interpreters in the field (Joshua
Goldsmith), to an overview of terminology management tools for
conference interpreters (Anja Rütten), to automatic speech
recognition (ASR) and how it is being combined with terminology
management software to provide automated terminology lookup in the
booth (Claudio Fantinuoli). But I won't. Instead, If you are interested
in the topics covered at TC39, I encourage you to read Alexander Drechsel's
November 21st blog post reporting on TC39. It's full of detailed
information about each of the presentations and even includes a pair of
videos we shot live from the conference.
to hoping this initial foray into interpreting territory by ASLING is
the beginning of a lasting effort to provide a forum for research and
discussion not only on translating, but also interpreting, and the
computer. If you interested, mark your calendars now. The 40th
anniversary edition of Translating and the Computer (TC40) will take
place on November 15-16, 2018 in London.
you have a question about a specific technology? Or would you like to
learn more about a specific interpreting platform, interpreter console
or supporting technology? Send us an email at firstname.lastname@example.org.
Interpreting at GALA 2018
Interpreting 5 will take place in Boston, USA on March 13-16, 2018, as
part of GALA's 10th Anniversary Conference.
This 'n' That
book that is most certainly going to be helpful for novices to the
world of translation as well as to the majority of more experienced
colleagues is Language of
Localization, edited by Kit Brown-Hoekstra. It gives a great,
two-page-each overview of 52 topics (by 52 different authors) that are
relevant to the modern translation and localization workflow and
might remember that I strongly encouraged tool vendors awhile back to
offer switchable access to either the statistical machine translation
or the neural machine translation engine of Google Translate. A
number of them obliged, with the latest being Okapi.
(If you don't know much about Okapi, and especially its
flagship tool Rainbow, there's going to be an excellent article
about it in the next ATA Chronicle. You don't even have to be
an ATA member to read it: Its online edition is available right here. While
you're there you can also read the interesting current article about OmegaT.)
tool that had this option much earlier than most and should have been
mentioned by me is Smartcat,
the translation environment tool that was originally developed by ABBYY
and is now independent. The way Smartcat handles machine
translation is a little different than other tools. In other tools --
and depending on which machine translation engine you use -- you either
have to enter your API key or information about the server and the
respective authorization to access it. Depending on your use, you then
have to pay the machine translation provider. Smartcat, on the
other hand, acts as a reseller between you and the MT provider, so you
pay a usage fee to Smartcat unless, and this is important to
realize, you agree to send your edited data back to Google, Microsoft,
or Yandex (the three providers that are supported). Clearly this is not
a good option for many translators, so you have to really understand
which option you choose and what that means if you are using Smartcat.
option in Smartcat starting next month will be Intento, a relatively
new Russian company that is primarily geared toward larger translation
buyers but will make its foray into our world starting with Smartcat
(other translation environments will follow soon). Intento is
somewhat similar to Fair Trade Translation
(see in issues 254 and 263 of the Tool Box Journal). It's a
tool that allows you (better: will allow you) to get an automated
estimate of which machine translation engine suits your present project
best and at what cost. The engines include engines by SAP, SDL, PROMT,
IBM, DeepL, Microsoft, Baidu, GTCOM Yeecloud, Google, Yandex, and
SYSTRAN. It remains to be seen whether this is usable from a
translator's perspective, especially when it comes to issues like
confidentiality. (I'll look into that more deeply once there are more
connectors to other tools.) Still, it was very interesting to talk with
Kontantin Savenkov, Intento's CEO, earlier this week. Here is one
insight that he shared. As long as English is either the source or
target language, there is a relatively great likelihood that Google
Translate (or DeepL for the languages that it covers) will
come out on top by Intento's automated assessment. As long as
that is not the case, all bets are off. Why? Because the system then
uses English as the "pivot language," so there are two machine
translation processes taking place (in, say, Chinese to Indonesian it
would be Chinese to English and then English to Indonesian).
more on that once it's more widely available.
Eby recently did an interesting survey on the use of monitors among
translators that some of you will find it interesting. You can find it right
tuned! memoQ 8.3 will be released soon
WordPress Connector. Enhanced Terminology Management. A set of new
3 December, at www.memoq.com/downloads.
Last Word on the Tool Box Journal
you would like to promote this journal by placing a link on your
website, I will in turn mention your website in a future edition of the
Tool Box Journal. Just paste the code you find here into
the HTML code of your webpage, and the little icon that is displayed on
that page with a link to my website will be displayed.
you are subscribed to this journal with
more than one email address, it would be great if you could unsubscribe
redundant addresses through the links Constant Contact offers below.
you be interested in reprinting one of the articles in this journal for
promotional purposes, please contact me for information about pricing.
2017 International Writers' Group