|
ADVERTISEMENT
|
|
memoQ 8.1 is now available
New PDF import. New Find/Replace. Better preview. And many
more.
Download. Install. Enjoy working with memoQ.
www.memoq.com/downloads
|
|
1. Mushrooms After a Drought
|
|
Lately I've written quite a bit in the Tool Box Journal
about project management and invoicing tools for freelance translators,
tools that seem to have cropped up like mushrooms after a night of rain.
This is actually not a very good analogy because they've come more in
response to a drought by the market leader AIT with its rather outdated Translation Office 3000
(TO3000).
Well, that drought has come to an end with the brand-new TO3000
3D. (Of course, the problem for AIT now is that all those other
"mushrooms" won't just go away.)
Before I share some of the impressions I had after using the
new version for a few days, let me tell you a bit about a conversation I
had with Vladimir Pedchenko, AIT's CEO.
There were many reasons for the "drought," according
to him, including the move to a completely new development platform and
underlying database engine (MS SQL Express instead of the
open-source Firebird SQL), a complete redesign of the interface to
make it touch-friendly and modern-looking, and the forking off from the
development of AIT's product for translation agencies, Projetex
(previously TO3000 was essentially a limited version of Projetex).
In my opinion, though, an equally important reason for the
delay was a change of vision for AIT. There was a time when they released
several completely new products a year, not to mention upgrades to existing
ones. They were the ones who essentially invented project management
software for the freelance translator -- reflected in a proud 5000+
licensed copies of TO3000 -- and envisioned eventually bringing all
their products into one large combined product that would do everything,
including translation memory and terminology management, time and volume
tracking, and so on and so forth. While this sounded attractive, it didn't
happen for two reasons: it ignored the realities of a multi-faceted market
where freelance translators are only one, albeit important, constituency;
and their actual translation solution AnyTM was not truly innovative
(according to Vladimir, it was more of a simplified version of the old pre-Studio
Trados).
Add to that the turmoil in Ukraine (AIT is Kiev-based) as well
as within the company (I previously reported about that in connection with
Technolex), and things did not move as quickly as they could have.
With that said, the
new product looks good, seems to be stable after a round of alpha and beta
testing, and will delight existing users of TO3000 versions 11 and
earlier.
I already mentioned the more modern-looking interface, which
includes the ability to work in several windows at the same time (making it
a lot easier to compare data entry between clients, for instance), it's
optimized for those who like to work with Windows-based touch
screens (yes, only Windows is supported), and it now offers email
integration, file management, and FTP integration. I'm not sure whether
it's the new ribbon-based interface or the horizontal versus the previous
vertical layout, but it's a lot easier to work in -- and I imagine that
this is true not only for upgraders but also new users.
One thing that made earlier versions of TO3000 stand
out was the integration of several other AIT products, including the
sophisticated word counting tool AnyCount. This is still the case,
and I encouraged Vladimir to bring the time-measuring utility ExactSpent
into the mix as well -- I wouldn't be surprised to see that soon.
The upgrade process worked smoothly and quickly (though I wish
the "Data Import Utility" had started automatically to port data
over from previous versions). On the other hand, startup and shutdown speed
of the tool seems excruciatingly slow (I assume this has to do with the
need to start and shut down the SQL Server program), and I'm a
little frustrated that the rather simplistic and very static currency
conversion still doesn't dynamically adapt itself to current rates.
Still, I'm glad I've upgraded, and I will continue to use TO3000
as my invoicing and business reporting tool for the foreseeable future.
Vladimir asked me what he can do to get the word out --
especially to those who are not familiar with TO3000 yet -- and I
encouraged him to give a special offer to Tool Box Journal readers.
That he did with a generous 50% discount right here. And if you
as a Tool Box Journal subscriber have already purchased a new
license with a lesser discount, write to support@to3000.com and they will
make an adjustment.
|
|
ADVERTISEMENT
|
|
crossMarket Job Board Is
Coming Soon
Faster translator search, hassle-free access to exciting
translation jobs: you will soon be able to find suitable matches in the
crossMarket job board.
We are already more than 10,000 members worldwide. Join us
now: www.crossmarket.net
|
|
Dear Basic Subscribers . . .
|
|
I've had two related thoughts on my mind lately: 1) I'm tired
of cutting articles from the Basic edition of the Tool Box Journal,
and 2) your colleagues who own the Translator's Tool Box
ebook are hooked on it, proven by their willingness to purchase it
again and again each time a new update is available. How do those two
thoughts fit together? Well, you might notice that you're receiving a full
Tool Box Journal today (and a long one at that). Along with that gift, I'd
like to sweeten the pot with this offer: Continue to receive the full
Premium edition and purchase the
Translator's Tool Box ebook through the rest of June for only $30
(rather than $50 -- just enter 30 as the amount to be paid). That price
buys you a free annual subscription for the Journal, and if you look at the
ebook and don't like it, just send me a note and I'll reimburse you the $30
-- no questions asked. You can even keep the book (and the subscription).
Let's see whether I can get you hooked as well >:D
(My kids tell me that's supposed to be the evil laugh emoji.)
|
|
2. Microsoft's Communication
|
|
The communication team for translation at Microsoft does not
always act in translators' favor. Think, for instance, of then-VP Gurdeep
Singh Pall who called Skype's auto-interpreter "pre-beta of
magic" in 2014. Or when Director of Product Strategy and Marketing
Olivier Fontana called a new
voice-to-text translation feature of PowerPoint something "straight
out of 'Star Trek'" just a few weeks ago. Those are exactly the images
and concepts that don't work in our favor (nor in Microsoft's either, I
think).
On a different level but equally unhelpful is Microsoft's lack
of communication when it comes to the availability and status of its Microsoft
Bing Translator. While there was a lot of fanfare around MS's move
into neural machine translation when it came to voice-to-voice, there's
been little publicity around the fact that this neural machine translation
has also been -- optionally -- available for text-to-text through the API
(for translation environment tools) since last October, and that neural
machine translation has now become the standard with regular web-based
access for English <> Arabic, Chinese (Mandarin and Cantonese),
English, French, German, Italian, Japanese, Portuguese, Russian, and
Spanish as well as Japanese <> Chinese text translations since April
28 (aside
from this announcement about Asian languages).
That's why I'm grateful for Chris Wendt
of Microsoft's machine translation group. Chris has been very willing to
engage if prodded just a little to share what needs to be shared.
So: For any translation environment tool that supports the new
Azure-based Microsoft API (I reported about that in edition
273 of the Tool Box Journal) -- and, mind you, the majority of
tools are not equipped to handle the new API -- you'll have a choice to use
either the statistical machine translation engine that Microsoft provides
or the neural engine. You can do that by entering generalnn
as the "Category" in the provided field -- or not. Here is an
example of what this would look like to enable neural machine translation
in two tools that actively support this, Déjà Vu and Memsource:
If you leave the Category field empty, you'll access
the SMT engine (you might have to close your tool and open it up again to
see the change take place -- at least that's what I had to do). Of course,
this works only for the language combinations mentioned above. (If you're
unsure which of the two engines you are "getting," you can always
check right here.)
(Not that this makes much difference to you, but in my limited
tests I actually found the statistical machine translation engine more
reliable than the neural engine, even though the latter "reads"
much better.)
Oh, and in case you wondered: Unlike Google, which (assures us
that it) is not using your data if you use the translation API, Microsoft
(still) does, so for many of you this really might be a piece of
not-applicable-news.
|
|
ADVERTISEMENT
|
|
Need
a system to manage your translation agency? Try Protemos!
Clients, vendors, projects management, invoices and payments,
business reports and more...
Free for freelancers, 3 month trial for agencies, try
it now!
|
|
3. The Tech-Savvy Interpreter: Training Interpreters Online --
Taking Stock after One Year (Column by Barry Slaughter Olsen)
|
|
Be sure to watch this month's Tech-Savvy Interpreter
video.
I recently finished my first year of teaching interpreting
online for the Middlebury Institute of International Studies (MIIS). All told, I spent over 450
contact hours teaching both consecutive and simultaneous interpreting and
some sight translation remotely. It has been an amazing learning experience
for me and for my students. In September 2016, in the Tech-Savvy
Interpreter, I presented a number of technologies I use to teach
consecutive interpreting via video conference. In this edition, I'd like to
share some initial lessons learned -- three methodological and three
technological.
Methodology
Let me start with methodology because no amount of cool
technology will make a poorly designed class successful. In fact, if not
appropriately applied, technology becomes a distraction.
My first conclusion? Well, after a year of teaching online, I
can say without a doubt, interpreting can be taught and learned
successfully online, but successful online teaching requires discipline and
initially, a lot of additional preparation time to be done well. The need
to maximize every minute of online synchronous time with my students has
forced me to give greater order and structure to my teaching and to carefully
define what my learning objectives are for each class. Not that my classes
were unstructured in the past, but when teaching remotely, wasted seconds
can seem like minutes and minutes, an eternity. So, doing all you can to
squeeze every bit of usefulness out of your synchronous class time is
crucial.
Second, converting my usual in-class "lecture time"
to online videos for watching outside of class helped maximize face-to-face
or synchronous instruction time for live interpreting practice and
feedback. In eduspeak this is known as "flipping the classroom"
(See "7 Things You
Should Know about Flipped Classrooms").Providing short video
lectures is more demanding and time consuming that I initially thought, but
the dividends for student learning in interpreting are clear. Producing the
videos, even if they are going to be simple and of a relatively low
production value takes a lot of time up front. As any educator knows,
curriculum design is a demanding and time-consuming process. This month's Tech-Savvy
Interpreter Video includes one of my short video lectures from an
introductory consecutive interpreting course. For these videos I use an
enterprise-class video management platform called Panopto provided by MIIS
(Thanks!). But this is not required, you could easily use a consumer-level
video platform like YouTube
or Vimeo to much the same effect.
My last observation on online teaching methodology is that
ironically some things are easier when working remotely. As I'll explain
next in the Technology section, software-defined videoconferencing
in the cloud has made some huge leaps forward in the last year. Video and
audio recording (including multiple files from one videoconference session)
is now as easy as pressing a button. This means that remote interpretation
testing is a snap. The videoconferencing platform I switched to in January,
Zoom Video, allows me to share
videos on screen along with my computer audio for testing. Or I can bring
in a live speaker from anywhere in the world to give a speech. Individual
student exam recordings (both video and audio only) are available for
download from the Zoom Video platform or I can choose to record them
directly to my computer's hard drive. The simple result is a lot of time
and headache saved.
Technology
And now for my three technological observations.
First, as alluded to in my last point, there is a revolution
going on right now in videoconferencing. Cloud-based videoconferencing
platforms are making it easier and much less expensive. Until January, I
had been using an enterprise-class videoconferencing solution by Polycom to
teach my consecutive interpreting classes. I switched from Polycom to Zoom
Video for the spring semester, which improved the experience for me and for
my students. Connecting is easier, and I now have an entire suite of
collaboration tools that I can use to teach (e.g. screen share, whiteboard,
video and audio sharing) all on one platform. One unexpected benefit from
using a cloud-based platform was that I was able to offer some much-needed
flexibility for a student who needed to attend class from home a couple of
times during the semester.
Second, ease of use is a big factor for any technology you
want to employ in online teaching. Getting on and off a platform needs to
be as simple as possible. In order to teach interpreting (both consecutive
and simultaneous) I have had to employ multiple conferencing platforms and
technologies. Each new platform or technological solution introduces
complexity into the teaching equation. So, the technology must have a clear
pedagogical purpose. In addition, you must factor in the time at the very
beginning to get every student up and running and comfortable with each
technology solution you will be using. If you don't, this can eat into your
synchronous class time during the semester -- something you definitely want
to avoid.
Third, not everything is perfect with the remote teaching
model. I'll be the first to admit that there are some things that would
just be easier if I were physically in the same space as my students and
fellow interpreting professors. Some students were flustered when the technology
did not perform perfectly, which is understandable. Ironically, however, I
interacted more with my students using videoconferencing, WhatsApp and
online journaling this academic year than I have face to face in past years
when on campus. Even so, I think it is imperative to establish a good
student-professor relationship from the beginning. This is why I still
travel Monterey at the beginning of each academic year to meet with my
students and teach the first two classes face to face. I then return to
teach two or three more times throughout the academic year. So, my teaching
model is actually blended (about 80% online and 20% face to face).
Conclusion
These are only a handful of the many insights I have gained
over the last academic year as I have taught interpreting online. In my
mind, the advantages clearly outweigh the disadvantages. The trend toward
more online interpreter and translator training is clear. The number of
online course offerings is growing quickly, and well-prepared interpreter trainers
are now offering individualized and group instruction online. I expect this
trend to grow given the huge need for interpreter training all over the
world and the inability of current academic models to fill that need
completely.
Do 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 inquiry@interpretamerica.com.
|
|
ADVERTISEMENT
|
|
Linguali is the next generation mobile-based
interpretation solution
Interpreters use a PC laptop as a console while delegates
listen to the interpreters using their own smartphone. Linguali
works stand alone or integrates with a sound desk, with high quality
full-band sound for both interpreters and participants. Linguali
is mobile, scalable and smart. It deploys in minutes for up to 60
participants.
Learn more and download your free trial at linguali.com.
|
|
4. Localization Essentials
|
|
Google has released a free six-part online course called Localization
Essentials. Those of you who have been around the block would not
particularly benefit from it, but if you're new to the field, you might
find it interesting (plus they're YouTube videos so you can play them at
double speed). Naturally you'll find some places where you'll disagree --
when it comes to pricing, for instance, or the exclusive focus on Google
Translator Toolkit as a translation environment tool (not sure why they
still use the term "CAT tool" in the course . . .) -- but if
you're able to filter out those shortcomings, you might benefit from it.
The six modules are Key Concepts, Content Types, Profiles
& Skills, Processes, Tools, and
Glossary & Resources.
|
|
ADVERTISEMENT
|
|
Whether you're new to translation and localization technology,
or experienced in both, SDL Trados has a huge library of free
resources to help you develop your careers and businesses.
Resource topics include:
See
a full list of all educational resources.
|
|
5. Confessions
|
|
No, not the ones from Augustine, but very heart-wrenching all
the same and likely to cause mixed feelings for some of you. A translator
friend who translates between two European languages with typically average
results for (statistical) machine translation (better than, say,
EN<>DE but worse than ES<>FR) and who wants to stay anonymous
(which really is part of the story but which I will otherwise leave
uncommented) shared this with me after the last Tool Box Journal:
"I am getting old, I am and
always have been a lousy typist, I am a slow, deliberate translator, I hate
Trados, on which I have spent way too much money and grief
throughout the years. So, I started using Google Translate some
years ago, and in the type of stuff that I do, usually medical journal
articles and medical reports, I discovered that in some cases it did better
than I could, plus I did not have to type it from scratch. Now, several
years later I am still using this method, but I have noticed that it has
gotten increasingly better. It now transposes phrases and handles syntax a
lot better -- at times even perfectly. So, in a nutshell, MT has made me a
better and a faster translator. I now mess around in my garden during
the day and do my translations after supper and I can do 2 to 3K words by
midnight.
"So, I am a believer and you
won't have to convince me, but this is still not something I can tell my
colleagues about, as it feels a little bit like cheating. But then again, I
am the one who decides what is correct, and that is because of my knowledge
of the two languages."
|
|
ADVERTISEMENT
|
|
Leave the office 20
minutes earlier today!
Using MindReader for Outlook, you can
write e-mails more quickly and more consistently.
Watch the short video for more information on MindReader
for Outlook functionality and usage: https://www.youtube.com/watch?v=YAPLHSvVrBc
|
|
6. My Journey into "Neural Land" (by Terence
Lewis) (Ⱦ)
|
|
A little more than a year ago, in the 260th
edition of the Tool Box Journal, I published an article about
Terence Lewis, a Dutch-into-English translator and autodidact who took it
upon himself to see what machine translation could do for him beyond the
generic possibilities out there. He taught himself the necessary
programming from scratch, once for rules-based machine translation, again
when statistical machine translation became en vogue, and, you guessed it,
once again for neural machine translation. I have been and still am
impressed with his achievement, so I asked him to give us a retelling of
that last leg of his journey. (And by the way, the Ⱦ in the title stands
for "technical").
It all started with a phone call from Bill. "B***dy hell,
Terence", he shouted, "have you been on Google Translate
recently?" He was, of course, referring to Google's much publicized
shift from phrase-based statistical machine translation to neural machine
translation which got under way late last autumn. Bill, an inveterate
mocker of lousy machine translation, had popped a piece of German into
Google Translate and, to his amazement, found little to mock in the output.
German, it seems, was the first language pair for which Google introduced
neural machine translation. I put down the phone, clicked my way over to
Google Translate and pasted in a piece of German. To say that what I saw
changed my life would be a naïve and overdramatic reaction to what was
essentially a somewhat more fluent arrangement of the correctly translated
words in my test paragraph than I would have expected in a machine
translation. But this was early days and things could only get better is
what I told myself.
Around that time the PR and marketing people at Google,
Microsoft and Systran had gone into top gear and put out ambitious claims
for neural machine translation and the future of translation. Systran's
website claimed NMT could "produce a translation overachieving the
current state of the art and better than a non-native speaker". Even
in a scientific paper the Google NMT team wrote that "additional
experiments suggest the quality of the resulting translation system gets
closer to that of average human translators", while a Microsoft
blogger wrote that "neural networks better capture the context of full
sentences before translating them, providing much higher quality and more
human-sounding output".
Even allowing for the hype factor, I could not doubt the evidence
of my own eyes. Being a translator who taught himself to code I was proud
of my rule-based Dutch-English MT system, which subsequently became a
hybrid system incorporating some of the approaches of phrase-based
statistical machine translation. However, I sensed -- and I say
"sensed" because I had no foundation of knowledge then -- that
neural machine translation had the potential to become a significant
breakthrough in MT. I decided to "go neural" and dropped
everything else I was doing.
What is this neural machine translation all about? According
to Wikipedia, "Neural machine translation (NMT) is an approach
to machine translation that uses a large neural network". So, what's a
neural network? In simple terms, a neural network is a system of hardware
and software patterned after the operation of neurons in the human brain.
Typically, a neural network is initially trained, or fed large amounts of
data. Training consists of providing input and telling the network what the
output should be. In machine translation, the input is the source data and
the expected output is the parallel target data. The network tries to
predict the output (i.e. translate the input) and keeps adjusting its
parameters (weights) until it gets a result that matches the target data.
Of course, it's all far more complex than that, but that's the idea.
Not knowing anything about NMT, I joined the OpenNMT
group which is led by the Harvard NLP Group and Systran. According to the OpenNMT website, OpenNMT
is an industrial-strength, open-source (MIT) neural machine translation
system utilizing the Torch/PyTorch mathematical toolkit. The
last two words are key here -- in essence, NMT is math. OpenNMT is
written in both the Lua and Python programming languages, but
the scripts that make up the toolkit, which are typically 50-100 lines
long, are in essence connectors to Torch where all the mathematical
magic really happens. Another NMT toolkit is Nematus, developed in Python
by Rico Sennrich et al., and this is based on the Theano
mathematical framework.
If you're thinking of delving into NMT and don't have any Linux
skills, get them first. It's theoretically possible to run OpenNMT
on Windows either directly or through a virtual machine, but most of
the tutorials you'll need to get up and running just assume you're running Ubuntu
14.4 and nobody will want to give you a lesson in basic Linux.
While in theory you can train on any machine, in practice for all but
trivially small data sets you will need a GPU (Graphical Processing Unit)
that supports CUDA if you want training to finish in a reasonable amount of
time. For medium-size models you will need at least a 4GB GPU; for
full-size state-of-the-art models 8-12GB is recommended. My first neural MT
training from the sample data (around 250,000 sentences) provided on the OpenNMT
website took 8 hours on an Intel Xeon X3470, S1156, 2.93 GHz Quad
Core with 32GB RAM. The helpful people on the OpenNMT forum recommended
me to install a GPU if I wanted to process large volumes of data. I
installed the Nvidia GTX 1070 with an onboard RAM of 8GB. This
enabled me to train a model from 3.2 million sentences in 25 hours.
However, I'm getting ahead of myself here. Setting up and
running an NMT experiment/operation is -- on the surface -- a simple
process involving three steps: preprocessing (data preparation in the form
of cleaning and tokenization), training and translation (referred to as
inference or prediction by academics!). Those who have tried their hand at Moses
will be familiar with the need for parallel source and target data
containing one sentence per line with tokens separated by a space. In
the OpenNMT toolkit the preprocessing step generates a dictionary of
source vocabulary to index mappings, a dictionary of target vocabulary to
index mappings and a serialized Torch file -- a data package
containing vocabulary, training and validation data. Internally the system
will use the indices, not the words themselves. The goal of any machine
translation practitioner is to design a model that successfully converts a
sequence of words in a source language into a sequence of words in a target
language. There is no shortage of views on what that "success"
actually is. Whatever it is, the success of the inference or prediction
(read "translation") will depend on the knowledge and skill
deployed in the training process. The training is where the clever stuff
happens, and the task of training a neural machine translation engine is in
some ways no different from the task of training a statistical machine
translation system. We have to give the system the knowledge to infer the
probability of a target sentence E, given the source sentence F (the
letters "F" and "E" being conventionally used to refer
to source and target respectively in the field of machine translation). The
way in which we give the neural machine translation system that knowledge
is what differs.
Confession time -- having failed math at school, I've never
found anything but the simplest of equations easy reading. When I worked my
way through Philipp Koehn's excellent "Statistical Machine
Translation" I skipped the most complex equations. Papers on neural
machine translation are crammed with such equations. So, instead of
spending weeks staring at what could just as well have been hieroglyphs, I
took the plunge and set about training my first neural MT engine -- they
say the best way to learn is by doing! This was accomplished by typing
"th train.lua -data data/demo-train.t7 -save_model demo-model".
This command applied the training script to the prepared source and target
data (saved in the file "demo-train.t7") with the aim of
generating my model (or engine). Looks simple, doesn't it, but under the
hood a lot of sophisticated mathematical operations got under way. These
come down to learning by trial and error. As already mentioned, we give our
neural network a batch of source sentences from the training data, and
these are related word by word to words in the target data. The system
keeps adjusting various parameters (weights) assigned to the words in the
source sentence until it can correctly predict the corresponding target
sentence. This is how it learns.
My first model was a Dutch-English engine, which was
appropriate, as I had spent the previous 15 years building and refining a
rule-based machine translation system for that language pair. I was
delighted to see that the model had by itself learned basic rules of Dutch
grammar and word re-ordering rules which had taken me very many hours of
coding. It knew when to translate the Dutch word "snel" as
"quick" or as "quickly" -- something that my rule-based
system could still get wrong in a busy sentence of some length. "De
paard die door mijn vader is gekocht" is rendered as "The horse
bought by my father" and not "The horse which was bought by my
father," reflecting an editorial change in the direction of greater
fluency. Another rule the system had usefully learned was to generate the
English genitive form so that "de paard van mijn vader" is
translated as "my father's horse" and not "the horse of my
father," although it did fail on "De hond van de vriend van mijn
vader" which came out as "My father's dog" instead of
"My father's friend's dog", so I assume some more refined
training is needed there.
These initial experiments involved a corpus of some 5 million
segments drawn from Europarl, a proprietary TM, the JRC-Acquis,
movie subtitles, Wikipedia extracts, various TED talks and Ubuntu
user texts. Training the Dutch-English engine took around 5 days. I used
the same corpus to train an English-Dutch engine. Again, the neural network
did not have any difficulty with the re-ordering of words to comply with
the different word order rules in Dutch. The sentence "I want to
develop the new system by taking the best parts of the old system and
improving them" became "Ik wil het nieuwe systeem ontwikkelen
door de beste delen van het oude systeem te nemen en deze te
verbeteren". Those who read any Germanic language will notice that the
verbal form "taking" has moved seven words to the right and is
now preceded by the particle "te". This is a rule which the
system has learned from the data.
So far, so good. But -- and there are a few buts -- neural
machine translation does seem to have some problems which perhaps were not
first and foremost in the minds of the academics who developed the first
NMT models. The biggest is how to handle OOVs (Out of Vocabulary Words,
words not seen during training) which can prove to be numerous if you try
to use an engine trained on generalist material to translate even
semi-specialist texts. In rule-based MT you can simply add the unknown
words either to the general dictionary or to some kind of user dictionary
but in NMT you can't add to the source and target vocabularies once the model
has been built -- the source and target tokens are the building blocks of
the mathematical model.
Various approaches have been tried to handle OOVs in
statistical machine translation. In neural machine translation, the current
best practice seems to be to split words into subword units or, as a last
resort, to use a backoff dictionary which is not part of the model. For
translations out of Dutch I have introduced my own Word Splitter module
which I had applied in my old rule-based system. Applied to the input prior
to submission to the NMT engine, this ensures that compound nouns not seen
in the training data will usually be broken down into smaller units so
that, for example, the unseen "fabriekstoezichthouder" will break
down into fabriek|s|toezichthouder and be correctly translated as
"factory supervisor". With translations out of English I have
found that compound numbers like "twenty-three" are not getting
translated even though these are listed in the backoff dictionary. This isn't
just an issue with the engines I have built. Try asking Systran's Pure Neural
Machine Translation demonstrator to translate "Two hundred and
forty-three thousand fine young men" into any of its range of
languages and you'll see some strange results -- in fact only the
English-French engine gets it right! The reason is that individual
numerical entities are not seen enough (or not seen at all) in the training
data, and something that's so easy for a rule based system becomes an
embarrassing challenge. These issues are being discussed in the OpenNMT
forum (and I guess in other NMT forums as well) as researchers become aware
of the problems that arise once you try to apply successful research projects
to real-world translation practice. I've joined others in making
suggestions to solve this challenge and I'm sure the eventual solution will
be more than a workaround. Combining or fusing statistical machine
translation and neural machine translation has already been the subject of
several research papers.
Has it all been worth it? Well, customers who use the services
provided by my translation servers have (without knowing it) been receiving
the output of neural machine translation for the past month and to date
nobody has complained about a decline in quality! I have learned something
about the strengths and weaknesses of NMT, and some of the latter
definitely present a challenge from the viewpoint of implementation in the
translation industry -- a translation engine that can't handle numbers
properly would be utterly useless in some fields. I have built trial MT
engines for Malay-English, Turkish-English and Lithuanian-English from a
variety of bilingual resources. The Malay-English engine was built entirely
from the OPUS collection of movie subtitles -- some of its translations
have been amazingly good and others hilarious. I have conducted systematic
tests and demonstrated to myself that the neural network can learn and its
inferences involve more than merely retrieving strings contained in the
training data. I'll stick with NMT.
Are my NMT engines accessible to the wider world? Yes, a
client allowing the translation of TMX files and plain text can be downloaded from our website,
and my colleague Jon Olds has just informed me that plug-ins to connect memoQ
and Trados Studio (2015 & 2017) to our
Dutch-English/English-Dutch NMT servers will be ready by the end of this
month. Engines for other language pairs can be built to order with a cloud
or in-premise solution.
|
|
7. New Password for the Tool Box Archive
|
|
As a subscriber to the Premium version of this journal you
have access to an archive of Premium journals going back to 2007.
You can access the archive right
here. This month the user name is toolbox and the password is
dingodominance.
New user names and passwords will be announced in future
journals.
|
|
The Last Word on the Tool Box Journal
|
|
If 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.
If 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.
Should 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
|
|
