Translating PDFs is easier and quicker with TransPDF

Try it FREE for your next PDF project.

One talk I listened to at last week's -- very successful! -- virtual ATA conference was by my good friend Jay Marciano. Jay has been working with machine translation from the very early beginnings (well, maybe not the very early beginnings in the late 1950s -- though he sure looks like it!) and has worked for a number of machine translation providers and service providers that use machine translation. For the last few years he has been the most consistent voice from within the machine translation community to reach out to translators, especially in the US, to inform us of machine translation's current and future impact on the industry from his particular vantage point.

This time, however, he gave a different kind of talk in which he suggested a new kind of pricing model. I'm probably not going to do it complete justice in this short article, but let me try to rephrase his main points. He argued that since there is no discernible and reliable differentiator between fuzzy translation memory matches and machine translation suggestions (assuming that you are using a specifically trained machine translation engine for a text type that is suited for that kind of translation), we should stop differentiating them in their pricing. Instead, they should all be paid by edit distance. ("Edit distance" is the now widely used approach to evaluating the number of changes the editor or translator had to make to an MT suggestion before delivering it.) Doing this, according to Jay, will protect the translator from poor-quality machine translation (because the edit distance -- or rewrite from scratch --will in that case be large enough for 100% payment) as well as from bad translation memories (same reason). Also, he suggests payment for MT suggestions with no edit distance, i.e., suggestions where no edits were deemed necessary (20% of the word price) at a rate twice as high as a 100% TM match (10%) to compensate for the effort to evaluate their accuracy. He also suggests a 110% rate for an edit distance of 91-100%, taking into account the larger effort needed to "correct" something that was rather useless in the first place.

If this is not clear, I would be glad to ask Jay to write a little something about it in the next edition of the Tool Box Journal or maybe transcribe a discussion with him about that.

My immediate reaction to this proposal is that it seems to be an attempt at a fair -- but hard-to-predict -- compensation scheme. On the other hand, I think we should still be looking for an hourly-based compensation. I have not yet encountered discussions on how a mechanic, electrician, accountant, or lawyer takes too long to do a job as a reason not to hire them, and I see no reason why we should be different. What I could see, however, is that we would use something like this system as a foundation of sorts to come to agreements about hourly compensation when negotiating an hourly rate with clients for whom we work TM+MT-based jobs.

Trick or treat? We choose to treat you with awesome software for translators! Leap ahead of your competitors with smart project management and word count.

Surprise for Tool Box Journal readers from AIT at the link below:

Stay vigilant, everyday superheroes!

As I was considering how to tackle this piece about XTM's latest feature, an off-hand comment from XTM's CEO Andrzej Zydron came to mind: XTM has 160 people working full-time for them at this point, with 40 of those writing code. They do serve some large clients -- including Expedia, SAP, Toyota, Twitter, and LinkedIn -- but this is still a large operation. I realized once again how this age of big data and narrow artificial intelligence (the kind of AI that we have access to right now) has made it difficult for very small vendors to compete. You can still develop a great tool with a one- or two-person outfit (for a small and targeted [and devoted] user group -- see for instance CafeTran Espresso), but if you want to equip it with data-driven abilities, you will struggle.

And, as we've stated many times before, data-driven processes does not only include machine translation but also so much more -- from user-generated data on how to make your tool more efficient, to ways to allow your user to collect data on how to streamline processes, to equipping your tool with linguistic knowledge that goes beyond static spell-checking dictionaries and links to language-specific termbases, translation memories, and machine translation engines.

XTM has recently invested in a technology that was first developed by Google in 2013 (see this paper) and then taken up by Facebook (see this paper from 2018 which contributed to this very recent announcement). XTM has now continued to develop one strand of this technology, calling it "Inter-Language Vector Space." The idea is essentially this: Since languages all have to express more or less the same reality, it should be possible to map the overlay of these realities by how languages express them. By looking at and comparing large amounts of monolingual data, you should find similar "shapes" (called "vectors") occupied by terms and words, and you might be able to conclude an overlap of use (see this paper from Babylon Health showing how this can be done).

XTM has done just that, with data for 157 languages.

Clearly this technology was originally thought of in connection with machine translation, but this is not how XTM approaches it (at least not directly). Instead, it looks at common annoying problems for anyone who deals with translation and then tries to find answers in its data for how to approach them.

What comes to mind when you think of common annoying problems? Tags!! You know, those inline codes in your translation environment tool that function as placeholders for anything from hyperlinks to formatting to footnotes or sometimes just completely unnecessary coding. (Also: The ones that can make it really hard to find translation memory matches or that royally confuse a machine translation engine.)

In its latest version, XTM automatically places the tags in the target segment once you leave the segment by looking at the data in its massive trove of inter-language vector space. It's a great concept, and while not foolproof, it is easy to work with and immediately fix it if necessary because you are still right with the next segment as you continue to work on your text.

Two other features using the data derived from the Inter-Language Vector Space are term extraction (monolingual and bilingual) and alignment (XTM's alignment has been using dictionary data for a long time and now includes the ILVS data as well), both features that are heavily used by XTM's LSP customers.

Some of the next features XTM hopes to use with this (and other technology) include auto-deciding on what workflow steps are necessary for a given project, comparing and evaluating different MT engines, storing post-edits so they can be automatically applied again, and fixing of fuzzy and MT matches.

Andrzej is a very euphoric man who tends to get very excited about his technology -- some of which pans out and makes a difference and some not -- but I was really glad to see his excitement here because most of these features are immediately relevant to a translator's job. I like that.

New Book on Translation Quality in the Age of Digital Transformation

The collective volume represents a strong link between theory and practice.

Find out more: bdue-fachverlag.de/detail_book/150

What do you do when you come across an article in a language you don’t speak? Run it through Google Translate. When you need to have a quick chat over video with someone and don’t have a common language? Skype Translator can help. Or that exotic menu written in a foreign language while on vacation? Just open the handy translator app on your smartphone, point it at the text, and let the magic begin.

For many basic tasks, these automatic tools work well enough. They’re certainly not perfect, but do help us grasp the gist of the message and make ourselves understood.

First, big data. Strip away the jargon, and you’re left with a key idea: that software can analyze huge amounts of data and detect useful patterns far faster and better than a human ever could. This applies to plenty of things other than language, of course - from real time traffic reports to targeted advertising or Netflix’s recommendations for what you should watch next.

In language technology, companies like Google and DeepL draw on huge bilingual and multilingual corpora. Their databases include millions of sentences from original documents along with their translations produced by humans. These treasure troves of well-translated content form the backbone of modern machine translation.

The new kid on the block, neural machine translation, uses a different approach: Instead of relying on probability for the right translation, artificial intelligence recognises patterns in the source and the target language and matches the two. These patterns go beyond single words to entire phrases and sentences. And recent studies have shown that this method yields better translations than before.

Nevertheless, technological developments are changing the face of the language industry. Many companies already save time and money by having their texts translated by a machine and polished by a professional linguist; the jury’s still out on quality and job satisfaction. Others use controlled language with a smaller subset of words and strict rules. After all, a simple text is easier for a machine to translate.

As for machine interpretation, it’s even harder than machine translation to get right. That’s because machine interpretation has traditionally included three stages: transcribing speech, using machine translation to convert that text into another language, and then using speech synthesis for the spoken output. An error in the first step can be compounded in later phases. And new machine interpreting models that leave out the machine translation “middleman” are still embryonic at best.

Most importantly, machine translation and interpretation are based on the fallacy that language professionals just “translate words”. But we do far more than that. As research shows, interpreters draw on body language, information on screens, word lists, and more to form meaning; we add information, explain cultural references, advise speakers, change language registers when needed, and detect and solve all sorts of problems.

PS. Questions or ideas about interpreting technology? Drop us a line at info@techforword.com! We do the research, so you don’t have to.

Download your free copy now right here.

If you have ever registered with Across' crossMarket marketplace…. No, let me be more specific: If you use the Across Translator Edition (ATE), you are registered with the crossMarket marketplace. That means you received an email about a month ago announcing that crossMarket was about to close. Through personal conversations over the last few years, I'd been aware that crossMarket had not been the success Across had hoped it would be. (To be honest, I had thought that Across' hope might actually become a reality since more than any other translation environment tool, Across has a more closed infrastructure and therefore a better possibility of building and maintaining a marketplace.)

Well, it sounds like both Across and I were wrong. As a result, they have decided to scratch everything they've done in terms of a marketplace and essentially start from scratch.

What does this mean?

Well, first of all, the annual fee that had to be paid to become part of crossMarket and use the ATE still has to be paid -- only without crossMarket. crossMarket itself will be replaced by a new kind of marketplace sometime next year. I spent some time talking with the person in charge -- Nicolas Stumpf -- about the new plans and how they differ from what was there in the first place.

The neo-German name of the new place is "HeyTranslator." It can already be found online but so far is not functional. What you can find there is a little impression of the look and feel, plus you can register yourself to get further updates.

Nicolas is not an industry insider, which could be either good or bad. However, it seems positive to me that he certainly is very eager to learn and listen. After Nicolas looked at other available web-based platforms to connect translators and clients, he became convinced that the time is right for another kind of marketplace that reflects both a modern design and a 2020's approach. According to him, ProZ is stuck where it was about 15 or so years ago, so the most logical place for uneducated clients to look for freelance translators would be platforms like Upwork or Fiverr -- sites that certainly have no experience to offer when it comes to translation. And up to this point, I tend to agree with him.

With HeyTranslator, Nicolas and his team (made up mostly of the same people who worked on crossMarket) are trying to build a translation tool-agnostic (we'll see how that goes), transparent place for clients to meet translators and vice versa. While it will be possible for LSPs to register, they are not the target group, and translators are encouraged to create comprehensive profiles (which will not be migrated from crossMarket), including a price range for services (rather than one fixed price). Two things that make this proposition a little different are that a) not only can clients rate translators but translators are also encouraged to rate the clients; and b) payment is done via an escrow account that is paid into by the client before the job starts and is paid out of immediately after the job delivery is confirmed by the client.

The fee that HeyTranslator envisions for each transaction is between 6 and 16%, depending on the size of the job. I feel that the upper end of that range is too high since there really is no project management offered by HeyTranslator, plus at this point they also don't offer any workflow process or document management -- but this may change at a later point.

We might have to wait a few months before we will see the first functioning version of the site go live, and this time I will keep my mouth shut about whether I think it is going to be successful.

You and I know that virtually any and every tool vendor has tried some kind of marketplace at some point, and so far all of them have more or less failed. I'm not sure why this one should work, though I do agree with a need for some kind of online platform for inexperienced clients to look for qualified translators. One way to help clients find the value they're looking for would be by emphasizing translators' degrees, accreditations, and certifications, something that Nicolas has promised to do in the tool.

FREE presentations from over 40 industry speakers in multiple languages

Don’t miss out this November! REGISTER NOW

Three years ago I wrote about the English terminology extraction tool Prospector (see edition 280 of the Tool Box Journal). Serge Gladkoff, the CEO of the tool's maker Logrus Global, contacted me again to show me some of the latest developments of the tool. Feel free to read what I wrote about it previously and then add the fact that it's now completely cloud-based, it now also supports text and HTML files for extraction purposes, and it has a feature that allows you to select your existing termbases to automatically exclude already existing entries from your new term collection. You now have a pretty good overview of what the tool does. I really like it, though it seems that the licensing terms are a little convoluted. It's free to use for Logrus Global's existing clients and also free for freelancer translators. Well, kind of, since Serge would like to negotiate with you about contributing your glossaries in exchange for the tool's use. Most of us would agree that glossaries are valuable -- in fact, so valuable that they often can be sold as an extra service to the client, so I'm not sure this will be an attractive offer for many. But then, Serge did say it's a negotiation -- so feel free to start talking with him and his team.

Another tool developer who reached out to me during the last couple of weeks was Eugene Kraben. Eugene used to work for the UN in its statistics department where he developed significant expertise in the use of MS Access. He used that expertise to develop a tool that is able to access SDL MultiTerm databases significantly easier and faster than the often sluggish MultiTerm itself. It's called Tb-Scout, and it allows you to search through your MultiTerm termbases and locate matches in mere seconds. Plus it facilitates the fast export of termbases into formats like PDF and Excel. If this is relevant to you, you'll need to keep a couple of limitations in mind: You'll need to have a 32-bit version of MS Office as well as a MultiTerm Desktop installed on your computer, and you can search through only one termbase at a time.

Do you remember the translator-specific URL shortener xl8.link? It's still out there, but we had to password-protect it to exclude nasty spammer and phishers. If you want to use it to shorten your own URLs in a way that also shows who you are, you can go here and unlock it with the login: xl8talk and the password: 20xl8talk18.

And lastly, I would like to point you to a very interesting article on Kirti Vashee's blog by Nico Herbig on a multi-modal interface for translation environments. While he is mostly focusing on post-editing, I think that much if not everything can be applied to virtually any kind of translation. You can find it here. (And I think that this is especially interesting for tool developers.)

Oh, and one more thing: I could not attend Lilt's Ascend 2020 conference last week (being a loyal ATA citizen and attending the ATA conference and all…) but the recording of that conference are available now. I would highly recommend Kyunghyun Cho's conversation with Lilt co-founder John DeNero on the "Future of Translation." Little technical at times but super interesting.

Watch the short video for more information on STAR CLM functionality and usage:

If you would like to promote this electronic 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.

If you are interested in reprinting one of the articles in this journal for promotional purposes, please contact me for information about pricing.

© 2020 International Writers' Group