Ep01 – Managing the Machine
– Welcome to ROAR: Deep Dive—a podcast produced by the Middlebury Institute’s Translation and Localization Management Program, bringing together global voices from the Localization industry. I’m your host for this episode, Rebecca Guttentag.
Artificial Intelligence is a buzzword that has been around for decades, and lurking under the surface has always been a lingering sense of unease. As far back as 1927, with the German film, Metropolis, media industries have been portraying intelligent machines as a threat to mankind. And Metropolis wasn’t the only example. Terminator, 2001: A Space Odyssey, Bladerunner—the list goes on and on. After all, people are only human, and humans are fallible. Machines? Not so much. Like Frankenstein & his monster, if we make our own creations too intelligent, sooner or later, they’re bound to surpass us.
But as many of us in the localization industry know, that fear mongering is only part of the story. Machine learning has been supporting the linguistic side of localization for a long time, so much so that when most people even outside of the industry think of machine learning, their immediate thought is translation engines. Google Translate took the internet by storm in 2006, and while it brought its own fears of replacement, now more than a decade later, it’s obvious that the translation industry is doing just fine. If anything, intelligent machines like these have helped translators. Many CAT tools are able to integrate machine translation APIs, and as companies move to systems like Microsoft Azure in order to create Machine Translation engines of their own, they’ve been able to streamline the work that goes to the human translation. By letting the machines focus on the repetitive stuff, linguists can focus on bringing the human side to translation.
But these strategies don’t just live within the linguistic side of the localization industry. In order to shed some light on this topic, I sat down with Max Troyer, the Chair of the TLM program, to talk about how he sees these technologies supporting the industry going forward, as well as how he’s helping lead students towards future opportunities.
– Hi Max, thank you so much for coming to join us for this podcast today!
– My pleasure, thanks for inviting me!
– Like I’ve said before, this is our first episode, of course since we’re being a little bit transparent that we’re students, we’re using this first episode to reach out to a professor. But it’s not just the fact that you’re a professor, that’s not the only reason, you’re also someone who has taken this program from, not the ground up, but you’ve really helped it evolve over time. So, I really was looking to ask, straight from the horse’s mouth, tell us a little bit about yourself, what is the Max Troyer elevator pitch?
– As time has gone on, I have worn many hats. A background in Computer Science, a degree in Translation, which I think we’ll probably talk maybe more about. I’ve been doing freelance desktop publishing; freelance localization engineering and it was kind of doing all these different roles that got me where I am today which is thinking about solutions for clients. And taking kind of the big picture and one of my favorite things to do is to help clients evaluate different Translation Management Systems and see what they can do, especially when they’re trying to switch from another TMS to a different TMS, all the challenges that has. I guess that my biggest complain about what I’m doing now is that I don’t get to get my hands as dirty as I used to and I’ve found that this is true as you move up in any career. When you start out, you’re getting your hands very dirty. We just had someone from Welocalize onsite who now is a manager and she was lamenting that… what she said to students, you know, get your hands as dirty as possible because, as you move up, then you’re gonna be thinking about it a lot more, talking about it a lot more, so…
– Yeah, I mean that’s something that I think a lot of us have heard, where… at the beginning, when you’re starting out you really get to stick your fingers in a lot of different pies, you get to do a lot of different things, but, as you move up, you really need to kind of focus on what it is you’re doing…
– … narrow that down. To jump back a little bit, you’ve mentioned you have a CS degree, you’ve mentioned that you also have the Translation degree. How did you transition from CS into the world of language services? How was it that that really got onto the map for you?
– When I was younger I was really black and white, I had trouble seeing nuances and understanding that there are many ways to doing things and, when I became a Software Engineer I basically thought that I was abandoning my passion for language and I didn’t see a way to reconcile it in any way. No guidance counselor ever said, you know, there are other ways to combine technology and language, so I became a Software Engineer, and just went to France on vacation and read French literature, but, when that job fell through, Enron, auditing, corruption, long story, not on my behalf, but on the company’s behalf, I decided to get a degree on French Translation from what was known as the Monterey Institute and I was gonna go 180, I was gonna become a translator or interpreter and just leave software engineering behind, it was gonna be a career shift for me. But then I learned about the localization industry, we had… at that time we had CAT courses, Computer-Assisted Translation courses, even a Software Localization course, so we did have some courses in localization and I had no idea… it was just the big epiphany like… whoa! It takes really language people to do localization project management [unintelligible], so that’s what I did when I graduated with my degree in Translation, I became a Project Manager and I think that just… I’m so jealous of current students that have realized this a lot earlier that I did. I don’t know where I would be… I don’t know, I’m pretty happy where I am, so, I don’t have any complaints there.
– Yeah, well, that’s a good **** because my next question was: where you are right now, you mentioned before that you’ve really enjoyed taking this TMS tools in consulting companies that are looking to go from one to another or really jump into TMSs to begin with. So, what do you like most about having that position, being able to do that kind of consultation and what are you taking from what you’ve learned… like you said it was just CAT tools, to now where there’s this whole plethora of tools that you can advise companies on. What is it that you’re bringing to the table now?
-First of all, it’s super challenging in the sense that whenever I work with a company of organization. In many ways it’s kind of like the software development that I used to do when I was a Software Engineer in the sense that in the whole consulting process you really have to talk to every primary person involved in that process, and it might be people who are creating original content and finding out what tools they use because they’re doing different content types and if you interview someone who is doing some weird thing and there’s no way to get that content into the TMS they’re considering then you kind of have to rethink everything and so, that’s what’s Solutions Architecture is, looking at everything that needs to come from this company, getting it to the Translation Management System, and also look at them, the people who are doing the requests. The requesters: who is in charge of that company, of implementing or creating a request and working with that content creator and, how does it all fit together it’s kind of like, we talk about workflows, but this is a really big workflow of going from person to person, and can it be automated in any way? Or, how much are we just emailing everything to everyone? It’s just really interesting to go into a company and see how they do things and… yeah, it reminds me of the movie, I think Office Space where I think the guys job is basically to take papers from the first floor to the second floor and they can’t really find anything else he does in the entire process. It’s sometimes just uncovering really wasteful things where literally a person is just wasting their time in processes that could be totally automated. So, the human component is a little bit unfortunate sometimes, but I think most companies rather than fire people tend to train them in other ways, at least that’s what I’d like to think.
– Right, so you brought up automation and that’s really a big talking point in the industry right now, because we see, like you said there’s a lot of change that’s happening, you know, even just over the past ten years we’ve seen new tools come into place, we’ve seen companies adding tools to their repertoire, which is both a blessing and a curse in some cases, because on the one hand, it allows companies using these services to really hone in on the tools that they’re using and the integration of those tools, but, on the other hand it means that they’ll kind of be entrenched in the tools that you have. Finding a way to lead out of those tools can be difficult. But, the other thing, when you talk about automation, and how localization professionals find their way in the industry right now, we see this talk of, are we going to lose jobs, are people going to be suffering as they’re trying to enter the industry now because so many jobs are being automated, or do you see this more as a vantage point for these newer generations of people coming into it to really expand the skills that we’re able to use?
– I think what we’re seeing is the evolution of the typical TLM student, basically, I think when I started teaching I could almost take anyone and turn them into a Localization Project Manager, I don’t know if that’s true or not, but it felt like I could basically take anyone and they could do Localization Project Management as it’s been done for many years. But, as automation rolls out in companies, and people aren’t emailing files to each other, to some degrees project assignment is taking place automatically as well, then, what is left for the Project Manager to do is really the human component and, as time goes on, I’m realizing not everyone is fantastic at that human component. I think what we’re gonna see is the TLM program is gonna need to evolve and we’re really gonna need to work on the soft skills a little bit more, especially with what we’re doing with intercultural competence. You can be really smart but be a jerk about it and no one is gonna wanna work with you and people aren’t gonna wanna complete your translation projects on time. To some degree is not just about being nice but knowing how to interact with people to get them to want to work with you and I don’t know if that requires everyone being kind of a leader. I’m sure that there’s two types of people those who believe that leadership is either something you have or you don’t. Can you cultivate leadership skills? When I ask people in the industry what skills they think TLM students need and many times they… there’s this one woman I won’t even name, but she always tells me that they need strategy and I love that because, what am I gonna do, create a strategy class? I always kind of feel strategy is something you have or you don’t, but it’s definitely something that we’re thinking about in the program and, basically, trying to create more courses that cover leadership roles and getting TLM students above the trenches and kind of managing what goes on in the trenches. What would be in the trenches? The machines, basically. To take the analogy with full force.
-Something I know that I’ve felt is we’ve seen this shift not just for project managers, but everyone needs to have these tech skills; all these different people within the company need to have some sort of ability to do tech. And so I feel like I’ve seen what used to be a clearly defined difference between a Localization Engineer and a Project Manager, and a Vendor Manager, you see this kind of overlapping because certain parts of these jobs are being lost to automation. What do you feel about evolution, this kind of weakening of the lines that used to separate them?
– If I could talk to someone who knew a lot about AI or, like, hire an AI team, one of the things I would like to experiment on is project assignment. And I think, like Smith Yewell, the CEO of Welocalize, in several talks he’s mentioned these micro-translations where they basically send a few strings at a time out to translators to translate them. In some ways it seems like death by a thousand cuts, because the translators are just getting paid, you know 30 cents, 50 cents, a dollar here and there and it’s like micro-translations or micro-transactions downstream. But I was wondering if we could apply some AI and have the AI do some experimenting with batching up some micro-translations into little bundles and experimenting with the size of the bundle to see what improves translation quality, or even experimenting with the time of the day or the day of the week that the bundle is sent and we could ask a human Project Manager to do experiments like that, but it would be so tedious, but we could easily, well, somewhat easily, program a machine that could start doing these experiments and then track that quality rating of that string way down in the workflow and be able to make some conclusions about how big are the batches we should do, when we should send them… That’s just zeroing in in something that I think that we can apply AI to because, when you said you wanted to interview me for this podcast and you wanted to talk about AI, I really started thinking: well if I could use AI and had a budget for what I do, and that’s one of the things I thought of that I would like to experiment with.
– It’s so funny that you bring up that example because I had that exact example in my own history. I was working for an internship over the summer where I was leading a client that was sending these single, maybe at max, 200 words a day, at minimum, something like 50, because it was just a constant, updateable website, you know, every time a new string was added in it just got automatically registered into the TMS. We were having to take from the client-side TMS into our CMS, take it from our CMS, put it into a Trados package, send that Trados package out, have it come back, send it out to the reviewers, have it come back, put it back into our CMS System, then put it back into the client TMS and I was doing that by hand. Every single day. It was really, like you said, the death by a thousand cuts, it felt like that…
– …so I think that that’s definitely a really exciting way that we could… Obviously I want to be interacting with that client, but, not only is it a problem with the tediousness of it, but it’s also the human error side of things where… There were times where I misquoted the exact number of words going in. I’m having problems setting up the Trados package to send it out, so, there are really… I think you’re right, there’s a lot of great ways that we can streamline processes and be able to not completely bleed out the project management side of things, but support it from… Yeah, take out the nitty gritty that we don’t need to be wasting time with.
– And, just to dive a little deeper into that, I don’t think that humans are really that innately good at remembering all the details that are involved in a project. I think humans have a short-term memory of maybe seven working items that you can keep in your head. If you don’t give a Project Manager automated technology to remind them when they need to check, then you’re setting them up for failure. And an example of that: back when I was working at an agency that had somewhat automated systems, it would let you send out a project using the automated system, but it didn’t make any checks to make sure that the translator had started, so you could get to the deadline and the Project Manager would receive a notification that was supposed to be delivered and the deadline would go by and you would reach out to the translator and they’d say: -Oh, I never started- so there was no follow up. A machine, you know, if you’re using a TMS, a machine could just see if any segments were touched, and then one of the Project Managers would know well in advance that this was sent to the translator but they haven’t started yet. It’s little things like that where we could really improve accuracy and quality.
– Yeah, and I think that that’s really the core of this argument, right? Where, if humans bring the human aspect of it, and that’s a very personal, subjective thing, machine support through automation brings the quality. And that’s something that… humans are fallible and we’re always going to make some sort of mistakes, and that’s not exactly a bad thing if everyone is going to do it at some point, but the way that we use automation to improve quality are extremely helpful. I guess another thing that I wanna ask though, is we’ve talked about automation as it helps the Project Manager. Do you see other avenues that automation can help with, because, like we said technology is really good at doing the repetitive things that you have the same task every single time, or you have a metric that you’re testing against every single time and the machine is really good at telling you: -Oh, we’re needing that or not.- There are a lot of different sides for the localization industry that are less repetitive and more focused on those soft skills. We see sales, we see marketing, especially marketing. There’s a lot of talk about not just automation in terms of setting projects up on the PM side, but automation in terms of, you know, training machine engines to be able to support some of the translation processes, support some of the marketing processes. What do you see is technology’s role in those sides of things?
– Well I think that’s one place where the machines aren’t really doing much damage in the sense that… a classic marketing campaign, or an advertising campaign that’s pretty much still managed by humans, we’re talking about transcreation, something the machines are not that good at, luckily for now. But, you know, just the other day I read about AI that’s been programmed to write original content based on some seed keywords and it can just basically go out and browse a database of thousands and thousands of articles and it can go out and recreate… pull from those articles and recreate kind of coherently written text about a subject and that really… It excites me, it’s interesting, but it also makes me a little bit nervous that, maybe someday, the marketing campaigns will be handled by machines and will be able to go out and look into their vast databases of content and create something cohesive. Now, when we’re talking about automation of marketing campaigns, I also just read an article someone posted on LinkedIn about how much time she spent translating seven words. I mean, it was basically a transcreation story, I can’t remember who it was and all the details are a little fuzzy to me already, but it was basically how she translated these seven words in the source language into ten sample translations, she slept on it, in the morning, she was down to five, she sent those out to all of her translation colleagues, and they voted on them, suggested some different ones, and she tracked her time and it was like ten hours or whatever to translate seven words. I just… I have a hard time with our current AI frameworks and, even as magical as deep fakes are, I don’t think that they can do quite that just yet. I think project management can be automated, maybe AI can be applied. In translation, AI can definitely be applied, we’re using Machine Translation, Neural Machine Translation, which is essentially AI, we’re still waiting on the document-level fluency; we have fluency on the segment level, but not on the document level, as most people are fairly aware of, so, one day we’ll have fluency on the document level and, when we do, what will translators have left?… I’ve told this to a few people, I’m not very loud about it but I think out Translation program will kind of evolve into a trancreation program, especially if the machines are really good at really boring translation, but, as someone who studied translation, I’m perfectly fine with the machine handling some boring, corporate translation, like, I wanna work on videogames, storytelling, and kind of back to literature translation in many ways, which I don’t think anyone in T&I would be happy to hear about, we’re definitely not a literature translation program, although there are more and more electives in literature translation.
– Yeah, it’s actually funny that you say that because there’s a lot of talk in the T&I programs here. And really, in terms of professional translation there’s a lot of, you know… Translators have to, really hone in on a genre or a subset of the industry to really make a name for themselves. You know, the people who do legal, the people who do financial. Those translators end up having a lot of success, being able to make a name for themselves in those industries, but a lot of those industries… The reason why w se those translators end up being do highly coveted is the fact that it’s very high-level translation.
– It’s not always a high-level of skill, although, if we talk about legal and medical, it does require specific subject matter expertise, but, for me, the real issue for legal and medial is the liability. In medical, if you translate something incorrectly people could die, in legal, if you translate something wrong people could go to jail, so there are huge ramifications and, I think that’s why… If you’re gonna say you’re a medical translator or a medical interpreter, you’d better be an SME in the medical field, and the same with legal. And I’m sure there are other cases like that too. If you’re a scientist and you wanna help NASA translate their content; if you don’t know what you’re saying rockets could explode and there may be people on them someday headed to another world and…
– …and that would be very bad…
– Well that’s something…
– …all because of a translation mistake
– Right, well that’s something that, in a lot of ways we see through training machine engines. In a lot of these domains, you get a lot of repetitive speak, everyone talks about how there’s the legalese. So, in some ways, people may look at these machine engines and be like: -Oh it would be so easy to train them to be able to translate this information and because you’re repeating the same kinds of phrases over and over again, you’d be providing the engine with a lot of different training materials, so, of course someday it would be able to take over.- But you’re right, you know, there’s that element of, if there is any single problem within the translation, it has extreme ramifications. I thinks that’s a really good example of why, you know, there’s always that panic within the industry of: – the machines are going to come and take our jobs-, well, there’s that balance of, of course the machines are better at, sometimes, delivering the same translation, the same wording every single time. There’s always going to be that element of you still need that human check, you still need something to be able to make sure that, even in that perfect translation, it fits that exact scenario, that exact situation.
– Yeah, definitely
– So, I think that that’s… We can all wipe this one off our brows, in terms of worrying about the machines taking all of our jobs. At the beginning you talked about how we need to focus on the soft skills. When you’re talking about doing all these creative translation projects like games and literature… That’s not a soft skill, but it really is, again, that human side of translation where you’re building some story. You know, we’ve talked about the automation of the PM role; we’ve talked about the automation of the Localization Engineering role. What do you think about DTP, ‘cause I know you teach the Desktop Publishing course here and you’ve done a lot of work with Desktop Publishing in the past? Do you think that there are avenues to be able to automate that when a lot of it comes down to a single human person having to go through and make edits by hand?
– Yeah, I do think about DTP, Desktop Publishing, a lot. The first time I saw the Google Translate app where you can point your camera at text and it would replace the text with the translation, basically doing a content-aware fill and getting rid of existing text and it replaces the text with the translation, obviously. It matches the color of the text, it doesn’t choose a different font, it doesn’t know how to chose sans-serif or a serif font, but the first time I saw the Google Translate functionality work, it was basically real-time DTP, it scared me a little bit. So, I’ve been thinking, you know, how, as content-aware evolves… Now we have content-aware in After Effects, not super easy to apply, but it will be soon; in Photoshop it’s pretty good, for the most part. I don’t know if we’ll get to a future where we can just open up a JPEG or a PNG directly in a CAT tool and it will use Machine Learning to hide text and replace it and even do font substitution. Like, I think maybe that’s a really good place where AI could be applied to DTP. You give it a font treatment and it comes up with choice A, B, and C with target fonts in that language and then you kind of choose, but it uses the power of AI from a vast library of fonts and it gives you kind of a preview. And you could choose to send this context sheet, or whatever you wanna call it to your client and say: – which of these treatments do you like for your Chinese text? – and it was developed by a machine, and maybe that’s automatic, maybe the client requests or uploads a JPEG and it looks at the fonts and realized the client hasn’t chosen a font for that treatment and it automatically kicks the process back to the client and emails: – click on the font you like best – and no human is involved at any time in that font selection process. But that’s something that used to be tedious, the Project Manager would have to go to the Desktop Publishing person: – Give me some fonts! – and the DTP would be like: – Well, in the past I’ve used this font for Chinese, I’ll go get these three fonts and…- you know, there’s an hour and 60 bucks down the drain, so…
– Yeah, I think that there is another aspect of that as well, which is how being able to take this kind of data from different places on the Internet and being able to apply it there as well. I know something that I’ve recently noticed on Google Fonts is it will tell you, if you choose a font, all of the fonts that people have paired with it…
– …you know, to be able to match the styling and that’s sort of crowdsourced by the people who are, you know, using Google Fonts and then downloading the fonts alongside them and using them together and Google Fonts is compiling all that data. Something similar to that could be used for these different languages. You know, if someone’s using Times New Roman in English, like you said, what is the CCJK font? What have people used to mimic that in the past? What has been successful in the past? And I think, like you said, that really cuts down time and that could really help streamline the industry because I know, for me, myself, having to go to the slug of finding the exact, right font has been a pain
– And, this example, we really zoomed in on something, but you could talk about dubbing and voice actor selection where the machine has a database of actors and sends three of them to the client based on properties of the people’s voices rather than some producer curating the samples you’re gonna send. So, I think…
– … there are opportunities.
– I went to a talk where the conversation about automating dubbing was less so focused on that and more so on creating AI voices and I remember looking at that and cringing because, well I think that there’s a lot to be said about that. Being able to, you know, streamline the process of hiring actors and cutting the time of finding the exact right voice. It’s amazing how there are so many different people trying to use so many different things that AI can help support and automation can help support to tackle these issues in so many different ways. Because, what you just said, that excites me so much more because, again, it’s that melding of the machine aspect of things and the human aspect of things.
– I mean, I probably wouldn’t go to school to become a voice actor because, already, I’ve seen demos of famous people’s voices being hijacked and saying things that they didn’t say, and it sounds perfect. And so, if you’re a voice actor, be careful that you don’t give out all of your phonemes because, as soon as someone gets all your phonemes, they can take your voiceprint and say anything; they don’t need you anymore. And you get royalties for someone using your voice? What if they modified it a little bit and it doesn’t sound like you anymore? I don’t know, it’s a scary world for voice actors… It’s a scary world for everyone, you know, bankers, let’s just list the careers at risk
– That’s really the gist of it. We’re seeing everyone so excited about what technology can help with and so completely terrified of how it’s going to replace us. So, for you, as you’re looking at this future, considering the past that brought you here and all the different changes that have occurred just within the past ten years, if you look forward five years, ten years into the future of this program, What do you see for the program supporting the students coming through it and where do you see that leading the industry?
– Well I can give you a good example. In my Website Localization class I just published an article, or updated an article that I’ve used explaining that I don’t want everyone necessarily to become a coder or programmer, I think that’s kind of a low-level skill at this point, what we need more of is people who can conceptualize and design software applications and think about what they should do and features they should have and then the minimum-wage coders can make those exist. Really, it’s the designer that it’s gonna be paid the most. And, in terms of localization, I think that, as we move forward, we are going to need people who have gotten their hands dirty, and who know how to do things via email. And I’ll tell you, as this is totally a parenthesis right now that I’m opening, but, you know, I’ve worked with a lot of agencies who are fully automated and then, when automation breaks, knowing what the Project Managers are supposed to do to keep projects going… you know, now their running back to emailing everyone because the TMS is down and try to keep things going. So, when you look at the future of the Translation and Localization Management program, I think that we’re going to be designing these automated systems and working with the people who understand Machine Learning to apply Machine Learning to what we do. We’re the subject matter experts in localization. Now, is that in our interest? This is… you know, I joke about the TAUS conference, their whole point is to make it so that humans don’t have to be involved anymore. And so, I wonder, if we become experts at Machine Learning or designing Machine Learning systems, is that in out interest? But I think that where we’re gonna go whether or not we want that. I think it’s a good move for the program and we’re gonna start doing a lot more with data analysis, not so much visualization because I now kind of see visualization as kind of a trenched skill like, making an infographic is cute, but it’s the person who can look at the data, and analyze the data, and make conclusions from the data that it’s gonna be paid the big bucks. And, so, if you’ve got a future TLM student five, ten years out, I want him to be really good at thinking about how these automated systems work, how we can make changes to them to make them work better, how we can apply AI in a very niche area to do something amazing, to take away the pain that someone is doing in their role, and then also, basically design these systems. And there’s also Computational Linguistics and Natural Language Processing that is kind of, also something that may or may not continue to be an on-demand skill or in-demand skill. So, there are more dimensions where we could take the TLM program, but definitely more leadership, more strategy, even though we probably won’t have a strategy course, but just, you know, thinking in terms of what we’re doing, why we’re doing it, and how AI can help
– That’s great, well thank you so much for your time!
– Thank you, Rebecca! This has been my pleasure, and I look forward to hearing future podcasts, I’m very excited to see what you’ll do with this!
– Yeah, thank you so much for assisting us as our first interviewee!
– So maybe machines aren’t the enemy to localization, but its ally going forward. As we embrace technology in our workplaces, we don’t have to think of it so much as a force to rival human capabilities, but one that can help to, in essence, get rid of all the nitty-gritty, repetitive processes that can drive human management mad. In this sense we won’t see technology replace human efforts but support them by weeding out processes that can fall victim to one-off human error or processes that are just a pain to go through repetitively day in and day out. At its best, artificial intelligence has the potential to streamline production, and when we do that, hopefully we’ll see human roles at all levels of the industry do what they do best—thrive by bringing that human touch.
– Thanks for listening to this Deep Dive episode of Roar—a MIIS Podcast bringing together global voices from the Localization industry. This episode was made possible by the help of the faculty and students here at the Middlebury Institute and listeners like you. For future episodes, be sure to check out our website at sites.miis.edu/roar/.
Also, if you liked what you heard, don’t forget to check out our companion series to Deep Dive, Roar: Speed Bumps. These episodes will focus on particular pain points of the Localization industry, and how loc professionals are finding ways to move the industry forward beyond them.
You can find both series of the Roar podcast on Apple Podcasts and Spotify. Thanks for listening, and we’ll see you in the next episode!