3104: Amplifying Voices: AI, Language, and Women in Technology

[00:00:04] What does it take to lead innovation in language technology, while also championing diversity and inclusion in one of the fastest growing sectors in technology?

[00:00:14] Well today I'm going to be exploring this inspiring story and groundbreaking work of my guest Olga.

[00:00:20] She's the VP of AI and Machine Translation at a company called Smartling.

[00:00:24] And with more than 25 years of experience in language technology, she has been at the forefront of advancements in NLP, Machine Translation and AI.

[00:00:35] Not to mention driving transformation across multiple industries. But her story goes beyond her professional achievements.

[00:00:42] She's a trailblazer for women in STEM, a mentor to aspiring technologists, an advocate for creating more inclusive tech ecosystems.

[00:00:51] So today Olga is going to share insights into how Smartling is leveraging AI to revolutionize translation processes.

[00:00:59] The challenges and opportunities of deploying AI in localization.

[00:01:04] And how hyperlocalization is shaping the future of global content.

[00:01:09] And she's also got an inspiring backstory to boot.

[00:01:13] So whether it be that backstory or how AI is reshaping language and localization.

[00:01:19] And what lessons we can learn from Olga's journey.

[00:01:22] We've got it all today.

[00:01:23] But enough rambling and scene setting for me.

[00:01:26] Let's get Olga onto the podcast right now.

[00:01:30] So a massive warm welcome to the show.

[00:01:33] Can you tell everyone listening a little about who you are and what you do?

[00:01:38] Okay, so name is Olga Berakavaya.

[00:01:41] Olga Berakavaya.

[00:01:42] And I've been in language technology and language technology industry for, well, I like to say 20 years, over 20 years.

[00:01:48] But now, I mean, I need to be honest with myself.

[00:01:50] It's over 25 years.

[00:01:52] I'm getting close to 30.

[00:01:55] I've held leadership positions in language technology, natural language processing, machine translation throughout my career.

[00:02:04] I've actually started in an executive role at 25.

[00:02:08] So that's what I do.

[00:02:09] That's the only thing I know how to do, which is language technology and been doing it my whole life.

[00:02:13] I absolutely love that.

[00:02:15] And I love how you've been doing it your whole life.

[00:02:16] The whole world went crazy over AI, what, two years ago almost to the day.

[00:02:21] And you've been doing this for 20, 25 years.

[00:02:23] I've got to ask, if we go way back to the beginning, can you share what initially drew you to this field and how you've seen everything evolve, especially with the rise of large language models and all that stuff?

[00:02:35] I feel there's got to be a big story there.

[00:02:38] Actually, the story is fairly simple.

[00:02:41] I went to, I mean, obviously born and raised in what was Soviet Union at the time.

[00:02:46] I went to a school that specialized in languages.

[00:02:50] I went to the university studying structural linguistics, continued with my master's at UC Berkeley, again, continuing with languages and language theory.

[00:03:00] So the path was pretty much laid out for me.

[00:03:02] As I said, that's the only thing I know how to do.

[00:03:04] And that's the only thing I've been doing my whole life.

[00:03:07] Now you graduate with a degree in humanities.

[00:03:09] You obviously graduate with a lot of question marks, right?

[00:03:12] What is it I'm going to do next?

[00:03:14] The path is not as obvious as for somebody with a degree in applied physics.

[00:03:19] And that was just like most things in our lives.

[00:03:22] That was an absolute coincidence.

[00:03:24] I happened to be talking to somebody like, hey, you know, structural linguistics.

[00:03:28] Why don't you try and build out parsers and dictionaries for what used to be a rule based machine translation at the time?

[00:03:36] Right.

[00:03:36] And rule based machine translation was basically you parse out one language and then you either use interlingua or like another methodology to transpose this to the other language.

[00:03:46] So I started doing that.

[00:03:48] It was kind of fun.

[00:03:48] I was probably about 24 at the time.

[00:03:51] And then basically my career and what I was doing in the field of AI was evolving alongside the evolution of natural language processing and potentially what would be from NLP more graduating into the AI field.

[00:04:06] So that was my first answer as to what drove my interest.

[00:04:10] It was super exciting.

[00:04:11] It was super interesting.

[00:04:12] I don't know if you remember like AltaVista Babelfish and Yahoo Trader, right?

[00:04:18] I mean, they were not the shiniest examples of what you can achieve with automated translation, but they were there and they paved the way to now is an explosion in language technology.

[00:04:30] So good company there.

[00:04:32] I remember all of those, Matt.

[00:04:34] Well, then I'm back again to how many years I've been in the industry.

[00:04:37] And of course, fast forward to present day then.

[00:04:41] I mean, as VP of AI and machine translation at SmartLink, what are some of the key innovations that you're currently working on?

[00:04:49] How is SmartLink leveraging AI and NLP to enhance language translation and localization for global audiences?

[00:04:57] Because it seems to be on such an incredible journey, but the exciting stuff's happening right now, right?

[00:05:01] Yeah, the exciting stuff has started happening.

[00:05:04] I would say that the most exciting stuff started happening.

[00:05:09] I mean, first, when you went from rule-based approaches to statistical approaches, that was one.

[00:05:14] And I would say what would have been probably 2000, maybe 16, when statistical approaches actually prevailed over rule-based.

[00:05:23] And that's when we actually started applying algorithms to processing natural language.

[00:05:29] And that was a major breakthrough in quality because you were fetching directly from the training corpus of language.

[00:05:36] And then if we follow the trajectory, then we have the famous paper, right?

[00:05:41] Attention to what you need.

[00:05:42] And the transformer models were introduced, maybe earlier generation transformer models.

[00:05:48] And we still live in the world of transformer models.

[00:05:51] It's just that the models have evolved and the training data completely exploded, right?

[00:05:57] How much, and you look at the trillion parameters models, right?

[00:06:01] That basically scraped all over the world's knowledge.

[00:06:03] And this is probably four years ago, three years ago, when large language models actually emerged and then grew exponentially in terms of the volumes of training data, the phenomena covered, and the number of parameters they're trained on.

[00:06:19] So obviously, large language models are somewhat taking over the world.

[00:06:24] But I think we need to go into it with open-eyed.

[00:06:27] And that's exactly how we're doing it at Smartlane.

[00:06:30] I would say that 2023, and I don't know if you'd agree with me, was probably a year of like a little bit of a frenzy and a little bit of wild west, right?

[00:06:39] Like, oh, let's grab GPT and let's see what it can be.

[00:06:41] Let's grab this, let's grab that.

[00:06:42] Let's plug it in and pray.

[00:06:44] And I think we all saw the world saw a lot of plug and play deployments.

[00:06:49] And I have to admit that with Smartlane, we're also like, what do we do?

[00:06:52] We work with neural machine translation.

[00:06:54] Where do we place large language models in our process or generative AI, or even smaller language models?

[00:07:01] And I think what happened and where we at Smartlane, and maybe even the world landed at 24, and will progress into 25 and onwards is what I like to call measured deployment.

[00:07:13] We learned a lot about the advantages, capabilities and shortcomings in the past year.

[00:07:20] And in 24, and I'll be talking about Smartlane here, I think we've landed at a very good place where we actually trademarked language AI.

[00:07:30] It's our trademark, and I would say that's exactly what we do.

[00:07:33] We landed at a very good place where we know what each of the natural language processing approaches are good at, and where they don't necessarily shine.

[00:07:43] So, at Smartlane, all of our translation process is completely AI powered.

[00:07:48] Our development processes, our marketing, our sales processes are also backed by AI.

[00:07:54] But again, I think we've graduated to we know what we're doing, and we know where we want to be from.

[00:08:01] Like, hey, what is it? Let's experiment with it.

[00:08:04] I completely agree with what you're saying a moment ago.

[00:08:07] I think that frenzy that you mentioned, it felt like 2017 when businesses added blockchain to their tile and their share prices rocketed.

[00:08:16] It was exactly the same in 2023, but you stick AI on the end, you know.

[00:08:20] But of course, outside of this and everything that's happening here, something I wanted to shine a light on is that you've been an advocate for women in STEM

[00:08:29] and have also mentored the next generation of female tech leaders for more than 15 years.

[00:08:35] So, kudos to you here. This is something that I'm passionate about as well.

[00:08:39] So, in your view, what would you say are the most pressing challenges that women face in AI today?

[00:08:45] And what steps could the industry take to create more inclusive opportunities?

[00:08:50] Because there's some great work been achieved over the last few years, but there's still so much more that needs to be done, right?

[00:08:56] You know what, I'll take you back to maybe neural machine translation and large language models.

[00:09:02] And I'll explain why.

[00:09:03] If you still punch in into large language models, and I know I was actually following a journalist

[00:09:09] who was playing with a multi-model as in visual and textual model.

[00:09:15] And she was trying to squeeze females of color as programmers out of the model without explicitly prompting for it.

[00:09:24] So, she was using different vocabulary to like, how do you land not a room of Caucasian white males?

[00:09:33] How exactly do you land that?

[00:09:35] What needs to happen?

[00:09:36] What do you need to do to the model to actually populate the room of programmers and developers with women and even more so women of color?

[00:09:44] And she was playing with adjectives and eventually she landed where she wanted to be.

[00:09:48] But that was like a two-hour journey.

[00:09:51] So, and the reason why I'm bringing it back to large language models, they,

[00:09:54] Jane Radevair reflects the anthropological phenomena, right?

[00:09:58] The phenomena of the world.

[00:09:59] And the phenomena of the world has traditionally been that STEM has traditionally been a male profession, right?

[00:10:06] Like you wake up and I was just hosting a couple of panels on the topic.

[00:10:10] And especially I come from Soviet Union, basically the path for you would be like, okay, what do you want to be a teacher or a nurse?

[00:10:16] I mean, the options were very limited.

[00:10:18] And I think unfortunately the trajectory still carries on to the modern day.

[00:10:23] I just looked up actually before our conversation, I looked up the statistics.

[00:10:27] Now we're talking about 29% of STEM students being female, about 27% of workforce, STEM workforce being females.

[00:10:39] And then when you go into executives in STEM, actually it drops like probably to roughly 25.

[00:10:46] And this is fantastic news because when I was looking up pulling the same statistics in 22, we were in the teens equally for STEM female students, STEM female workers and STEM executives.

[00:10:58] So that's very promising.

[00:10:59] That's very encouraging.

[00:11:00] The world is recognizing that women actually can and are contributing to the world of STEM.

[00:11:07] And I would hate to sound bitter or scorned, but having been in that industry, I know, and I think it's just a very fair and honest answer.

[00:11:16] I know that the journey of a woman in the STEM field, when you're the only female VP in the room of 25 males, it's a harder journey.

[00:11:25] But I think, I really think that we're landing at a very good place.

[00:11:29] There are a lot of opportunities.

[00:11:31] There are a lot of nonprofits.

[00:11:33] There are a lot of like women in technology.

[00:11:36] As you said, I'm on the board of women in localization.

[00:11:38] There are amazing things that our friends and peers for women in localization are doing in like Uzbekistan, Kazakhstan, Africa, just going there and helping girls, younger girls build up their career and make their choices around STEM field.

[00:11:56] So I think it's extremely encouraging and extremely hopeful.

[00:11:59] I would say hiring practices have changed dramatically.

[00:12:03] And there is a driver on the recruitment and HR side to make sure that the balance between male and female STEM workers is more even than it is now.

[00:12:15] And I think also the stereotypes and societal cliches are changing, right?

[00:12:21] So I think the stereotype like, okay, you wake up a female and next thing you know, you're going to be an occupation X.

[00:12:27] I don't think it's the case anymore.

[00:12:29] And I can proudly say that even on my team, the chief data scientist is a woman, director of strategy is a woman, a couple of junior developers are women.

[00:12:42] So I think we're moving in the right direction.

[00:12:44] And mentorship, you also mentioned that I've been mentoring for 15 years.

[00:12:49] I think it's also that those of us who've been down that path and those of us who are at a latest stage in life and in our journey, I think it's our responsibility.

[00:12:58] It's not even our choice to make sure that we guide younger ladies towards their path in technology field.

[00:13:07] And I can proudly say that out of my mentees, I was thrilled.

[00:13:11] One landed a job at Facebook just recently.

[00:13:14] One landed a job at Apple.

[00:13:16] One landed a job at Netflix.

[00:13:18] So obviously, we're as a society, are doing something right.

[00:13:21] You really are.

[00:13:22] And keep doing it.

[00:13:23] Absolutely love it.

[00:13:24] And there's also so much excitement right now around LLMs that we've talked about.

[00:13:29] And something that often gets missed is language is so much more important than just words.

[00:13:34] It's also about culture.

[00:13:36] It's about nuance and nuance and context of things that are sadly lacking very often in the media, for example.

[00:13:43] But how do advancements in AI and machine translation handle some of these complexities?

[00:13:49] And what other challenges still remain in achieving that truly accurate and culturally sensitive translation too?

[00:13:58] Okay.

[00:13:58] Well, I mean, maybe first I would say that machine translation is nothing but a form of AI, right?

[00:14:04] It's just maybe a slightly earlier generation of AI.

[00:14:07] But at the end of the day, it is.

[00:14:08] It is what it is.

[00:14:09] And then if we go into language AI, I will go back to what I said earlier, that language, as you rightfully said, is not just words, right?

[00:14:18] It's not just sentences, syntax, and morphology.

[00:14:21] Language is also about meaning and cultural phenomenon, local nuances.

[00:14:25] And I think that actually in our industry, now there is a term that I love, which is called hyperlocalization, where you can actually adapt your multilingual content to local phenomenon.

[00:14:37] And there are multiple ways of achieving it.

[00:14:40] I mean, it's the way you build the algorithms.

[00:14:42] It's all about the training data as well.

[00:14:44] It's all about attention.

[00:14:48] And what was happening, for instance, a year ago, which I think was fascinating, I read an article by an Italian translator who was saying,

[00:14:56] Okay, good, you got my Italian words in my translation, in my translation by AI, right.

[00:15:03] But guess what?

[00:15:04] You're only reflecting English-centric phenomena.

[00:15:07] You know absolutely nothing about my culture.

[00:15:10] So I thought it was really funny that in Italian, in proper, well-rounded Italian, everything that was described had absolutely nothing to do with Italian culture.

[00:15:20] And I think what's happening now, I think this realization had been made.

[00:15:24] I mean, we need to be honest with ourselves, right?

[00:15:26] The majority of generative AI and large language models, it's coming from the English-speaking world, right?

[00:15:33] And it's reflecting English phenomena.

[00:15:35] So I think it's a combination of adding more training data and, again, tweaking the way that context, right?

[00:15:43] The more you open the context window of a large language model, the more context you get, the more contextually accurate translation or contextually accurate generated content you can produce.

[00:15:55] So I think the mitigation is happening.

[00:15:57] And there is another thing that I find extremely interesting.

[00:16:01] And that's what we're very attentive to here at Smartlane.

[00:16:04] That's the more diverse your development team is.

[00:16:09] And I don't know if you've come across this research.

[00:16:11] The more diverse, culturally diverse and people coming from different countries.

[00:16:16] If your data science and AI development team comes from different backgrounds, they actually bring different perspectives into the development process and in the way that models interact with cultural phenomena.

[00:16:30] Just because they come from this background.

[00:16:32] Like one of the ladies on the team would be Chinese.

[00:16:34] And she brings her perspective even into the development process and even into the product design just because she's from that background.

[00:16:42] So, and also the other thing that we do at Smartlane, and I see it a lot happening in the industry a lot, is also prompt engineering, right?

[00:16:50] Which is now a field of its own.

[00:16:52] It is essential that you engage linguists from different cultures and linguists from obviously speaking different languages in the prompt engineering process.

[00:17:02] Because they have the cultural knowledge and the sensitivity to what resonates with the audience in a particular country.

[00:17:10] Like I don't speak Hebrew and Swahili, right?

[00:17:14] I speak a couple of languages.

[00:17:15] But I would never be able to reflect the cultural phenomena of those geos in those countries.

[00:17:20] So, I hope I answered your question here.

[00:17:22] You really did.

[00:17:23] And I'm also glad that you mentioned localization because I also have a unique perspective on the industry around this too.

[00:17:30] So, is there anything else you can share around your mission and some of the initiatives that you've helped support underrepresented voices in localization and the language technology space?

[00:17:40] Because you must be seeing a lot here.

[00:17:42] It's on the front line here.

[00:17:44] What are you noticing here?

[00:17:45] So, what we see in localization is, as I said, it's really how do we reinvent ourselves?

[00:17:51] That's the biggest thing in localization now.

[00:17:54] How do we, the industry has been doing things the same way for many years.

[00:17:58] And if you look again at the trajectory of technology industry, there was computer assistant translation.

[00:18:04] And then now I think we're more talking about language co-pilot or AI co-pilot.

[00:18:10] And I think this is the biggest thing that's happening in the industry now.

[00:18:14] How do you reinvent the world of technology, the role of technology in your global content transformation processes?

[00:18:22] And how do you reinvent the workforce?

[00:18:26] We want to make sure that people have, not only people have jobs, but people have exciting and stimulating jobs.

[00:18:32] And these are probably, if I were to sum up the trends is, how do you reinvent the processes with a new tech and everything that's now available?

[00:18:42] And how do you mitigate the shortcomings of the tech that's available?

[00:18:48] Models still hallucinate.

[00:18:50] Like one of my colleagues punched in something from a user manual and he received back a personal ad in Italian.

[00:18:57] And there was absolutely nothing in the source that suggested the personal ad.

[00:19:01] So I would say the main themes are evolution of generative AI, mitigating hallucinations and making the best of being able to mitigate those hallucinations.

[00:19:13] Regulatory and compliance measures.

[00:19:17] Regulatory and compliance measures.

[00:19:18] I've never filled out this many infosec questionnaires probably in my entire life after having worked with regulated industries and government entities.

[00:19:29] Regulatory and compliance.

[00:19:35] What's going to happen to my data?

[00:19:37] Are they going to scrape my data?

[00:19:38] Are they going to use my proprietary data to train their models?

[00:19:42] Who even owns the IP when everything is available on the internet?

[00:19:46] How do I interact with global governing bodies?

[00:19:51] Again, I live in the localization space, right?

[00:19:53] And SmartLink as a company, we live in localization space.

[00:19:56] What does it mean?

[00:19:57] We deliver content to 200 plus countries.

[00:20:00] And that again means that you need to know everything about GDPR and you need to carry

[00:20:05] quite a few ISO certifications.

[00:20:07] So that's another one.

[00:20:09] Governance, data governance and regulations and risk mitigation.

[00:20:13] And very, very important.

[00:20:16] And again, it probably spills a little bit over into the regulatory space is the workforce.

[00:20:21] And how does the workforce in multilingual space reinvent themselves?

[00:20:27] And there are multiple avenues there.

[00:20:30] Like I mean, not everybody is born a data scientist, but even linguists or project managers

[00:20:35] program managers can master the basics of data science and help AI help them with their work.

[00:20:44] Like as I said, for instance, we engage linguists in prompt engineering.

[00:20:48] There are still models still produce factually inaccurate results, even more so for foreign

[00:20:54] languages and for under-resourced languages.

[00:20:56] So you do need linguists to validate and correct the model output.

[00:21:03] And I'm pretty sure you have, I mean, I'm pretty sure you're familiar with the term human

[00:21:06] in the loop.

[00:21:07] And it has never been more important now because we want to build the most accurate and the most

[00:21:13] useful AI models.

[00:21:15] And for that, human input is of essence.

[00:21:17] I think also looking back at your personal journey, which is also incredibly inspiring for landing

[00:21:23] your first B-B role at 25.

[00:21:26] I'm curious, if you put all your experiences together in your career, how have they shaped

[00:21:32] your leadership style and maybe even approach to mentoring others in the tech industry?

[00:21:37] I came to the US as a single mother of a two-year-old at 21.

[00:21:41] And I think just because I landed like with absolutely empty pockets and no idea what's going

[00:21:49] to happen here in a foreign land.

[00:21:51] And thank God my mom came with me and she was extremely supportive.

[00:21:54] And so was my family.

[00:21:56] I think just that drive towards, yes, you can do it.

[00:22:00] Yes, not only can you survive, but actually potentially, hopefully you can thrive.

[00:22:04] I got admitted to UC Berkeley, which I never thought was even possible.

[00:22:09] And just going forward, I mean, there were obviously coincidences that helped me shape

[00:22:14] my career.

[00:22:15] But there was also this drive towards, yes, I can do it.

[00:22:19] And I think my leadership style is also the same.

[00:22:22] Don't take no for an answer.

[00:22:24] Always be able to pull rabbit out of the hat.

[00:22:26] Always have a plan, whether it's plan A, B, and C.

[00:22:30] And I think that's what I've been doing.

[00:22:32] And I hope my teams, the teams that have been working with me over years would agree.

[00:22:36] It is always goal-oriented.

[00:22:38] And there is always a way of achieving that goal.

[00:22:43] People on my team, I don't know if they were to listen to this, they'll probably have a good

[00:22:47] chuckle.

[00:22:48] Things like could have, should have, and we may, and we can, they should not be in your

[00:22:53] work vernacular.

[00:22:55] There is only we did and we will.

[00:22:58] So I think that's what shaped it up.

[00:22:59] I had to do it myself.

[00:23:01] And I'm happy to coach people in that direction.

[00:23:04] No matter what, you can always get it done.

[00:23:08] Absolutely love that.

[00:23:08] A big hello to everybody listening that's from your team.

[00:23:12] And of course, you've spoken at so many prominent conferences, including AI Summit and Localization

[00:23:18] World.

[00:23:19] And we're almost on the precipice of entering 2025.

[00:23:22] So from all the conversations that you're having, what trends do you see emerging in language

[00:23:28] technology over the next few years?

[00:23:30] And equally, any big conversations you're hearing around AI and localization space?

[00:23:37] Is there anything that you're hearing that you want to share about, that you want to

[00:23:41] share about, and even things that we don't talk about enough?

[00:23:45] How do you see it all taking shape over the next few years?

[00:23:47] I think what's happening and what's going to happen is, I mean, first of all, the language

[00:23:52] models that deal with text are going to become more and more accurate.

[00:23:58] And I already mentioned things like factual inaccuracy and hallucinations.

[00:24:02] I think those are going to go away very soon.

[00:24:05] The other thing is, we do know that large language models are extremely computationally

[00:24:10] expensive.

[00:24:11] So I think, and I know that bigger conversations of those industries are, how do we compress

[00:24:17] models or the new term had emerged, which is small language models, not large language

[00:24:22] models.

[00:24:23] And they are smaller models that are trained specifically for specific tasks.

[00:24:27] And in our world, those would be translation.

[00:24:31] And the industry sees a lot of successes with smaller models that are trained specifically for

[00:24:36] quality estimation or translation tasks, actually outperforming general purpose foundational

[00:24:42] models.

[00:24:43] And I think that's another direction.

[00:24:45] How do you actually train a fit for purpose model, as opposed to using the model that can

[00:24:50] do anything from calculating to summarizing court records?

[00:24:54] Data governance, again, that's a huge one.

[00:24:56] Who owns the data?

[00:24:58] What data is needed?

[00:25:00] How to clean the data, right?

[00:25:01] In language industry, the term garbage in, garbage out has been around forever.

[00:25:05] How do you optimize your training data to get the best outcome?

[00:25:10] That's probably another big conversation.

[00:25:12] And overall, again, overall data governance.

[00:25:15] Unsupervised learning is probably another one.

[00:25:17] Supervised learning is expensive.

[00:25:19] It requires a lot of labeling.

[00:25:21] Models actually being able to generate synthetic data to further train the models.

[00:25:27] And how do you actually make the best?

[00:25:29] How do you save most while getting the best results?

[00:25:32] That's probably another one.

[00:25:33] Another big question is neural machine translation is still great.

[00:25:37] It's factually accurate.

[00:25:39] It produces give or take fluent translations.

[00:25:42] And another big, like what's been happening in the break rooms at conferences is, so which

[00:25:47] one is it?

[00:25:48] Is neural machine translation here to stay?

[00:25:51] Or is it going to be completely overwritten by LLMs?

[00:25:54] And the consensus right now is that both have their good home in the industry.

[00:26:00] NMC for certain content types.

[00:26:03] And then LLMs for other content types.

[00:26:05] And specifically user-generated content and noisy content.

[00:26:09] So what else is there?

[00:26:10] Oh, multimodality, of course.

[00:26:12] Multimodality, because now you can generate multilingual images.

[00:26:15] You can, like the whole lip syncing and automated speech recognition and multilingual digital humans.

[00:26:22] That's obviously another big theme.

[00:26:24] So let me see if there is anything I'm missing.

[00:26:28] No, there would probably be.

[00:26:29] These are like, this is what's top of people's mind.

[00:26:32] And maybe another thing, quality estimation.

[00:26:35] Right.

[00:26:35] You can produce something, but you also need to know how good it is.

[00:26:38] And again, this is where AI can actually validate AI output.

[00:26:43] Wow.

[00:26:44] We've packed so much into 30 minutes today.

[00:26:46] And I cannot thank you enough for showing your insights learned throughout your career

[00:26:51] and also to the present day and the conversations we are having with your teams,

[00:26:56] with your customers, and the tech chauffeurs as well that you frequent

[00:27:00] and all the people that you speak with there.

[00:27:02] But before I let you go, I'm going to ask you to leave one final gift.

[00:27:06] I always ask my guests to leave either a book that means something to them

[00:27:10] that we can add to our Amazon wishlist that they can check out,

[00:27:12] or a song for our Spotify playlist that they can check out.

[00:27:16] Guilty pleasures are allowed, but what would you like to leave everyone listening with?

[00:27:20] And I suspect your team members will be particularly interested in how you answer this one.

[00:27:26] Well, there are quite a few Russian songs.

[00:27:29] I have a side life of a music producer.

[00:27:32] But let's leave Russian songs out of the equation,

[00:27:36] because I'm not sure everybody would relate to those.

[00:27:39] Interestingly, the song that influenced me most,

[00:27:42] or the first song that I was listening to when graduating from high school,

[00:27:47] it was Suzanne Vigatom's Diner.

[00:27:49] And the reason being, I grew up in the Soviet Union, right?

[00:27:54] So suddenly, the borders opened for Western music.

[00:27:59] And somehow, Tom Waits and Suzanne Vega were on the front line of that.

[00:28:03] I come from St. Petersburg.

[00:28:04] We were listening to a lot of that.

[00:28:05] So I was always fascinated by Tom's Diner,

[00:28:08] just because how does it feel to just sit there and observe, right?

[00:28:12] Being a third-party observer.

[00:28:13] So that'd be my song.

[00:28:14] And the book, do we want a professional book or just any random book?

[00:28:18] Any book that means something to you.

[00:28:20] Okay.

[00:28:20] Any book that would be actually Vile Bodies by Eveline Wall.

[00:28:25] And that being, I think, again, going back to the Soviet Union,

[00:28:31] we were a little bit at a loss, right?

[00:28:33] You grow up in one system, and then everything falls apart under your feet.

[00:28:37] And it fell apart from my parents, who had no idea what to do going forward.

[00:28:40] And I think Vile Bodies, right, about the lost generation

[00:28:44] and people whose world was completely destroyed by World War I,

[00:28:48] and the world fell apart under their feet.

[00:28:50] I think that's one of the books that I could relate to most.

[00:28:54] Well, I'll add that book to our Amazon wishlist for listeners to check out.

[00:28:58] And for your team members listening, next time you walk past them,

[00:29:02] I encourage them all just to go,

[00:29:04] do-do-do-do-do-do-do-do.

[00:29:07] Maybe they could do something like that to raise a smile.

[00:29:10] But for anyone listening that would just like to find out more information

[00:29:14] about anything we talked about today and learn more about Smartling

[00:29:18] and connect with you or your team.

[00:29:19] Any way you'd like to point everyone listening?

[00:29:21] I think Smartling website would actually be the best place

[00:29:24] and Smartling LinkedIn page because we post a lot of blogs.

[00:29:27] We do have regular webinars.

[00:29:30] We write white papers.

[00:29:32] So I think just simply just follow Smartling on LinkedIn,

[00:29:36] and that would be the easiest way of learning about all the work

[00:29:39] that we're doing and we're very excited about.

[00:29:41] Excellent.

[00:29:42] Well, I will add links to that so people can find you nice and easily.

[00:29:45] But just love chatting with you today,

[00:29:47] especially learning how you've dedicated yourself

[00:29:49] to mentoring the next generation of women in technology.

[00:29:53] Really inspiring work there.

[00:29:54] And also sharing so many valuable insights on AI, language,

[00:29:59] and so many different topics there.

[00:30:01] We packed a lot into 30 minutes.

[00:30:02] I'd love to hear more from people listening, what they thought.

[00:30:05] But more than anything, just thank you for joining me today, Olga,

[00:30:08] and also leaving us with a great tune too.

[00:30:11] Thanks so much, and thanks a lot for having me.

[00:30:14] I think Olga's story is a testament to the power of resilience,

[00:30:18] innovation, and a commitment to making technology work for everyone.

[00:30:22] And from her pioneering work in AI-driven translation at Smartling

[00:30:27] to her advocacy for women in STEM,

[00:30:29] I think she's shown today that technology can be both transformative and inclusive.

[00:30:36] And her insights into hyper-localization, language model advancements,

[00:30:41] and the importance of diverse development teams,

[00:30:44] I think offers a roadmap for a more connected and thoughtful global digital landscape.

[00:30:49] But over to you.

[00:30:50] Did today's conversation spark new ideas or questions?

[00:30:53] I'd love to hear from you.

[00:30:55] Please email me now, techblogwriter.com,

[00:30:58] LinkedIn, X, Instagram, just at Neil C. Hughes.

[00:31:03] Let me know, what role do you see AI playing in bridging cultural linguistic divides?

[00:31:08] Lots to think about.

[00:31:09] I'll be back again tomorrow with another guest,

[00:31:12] but thank you for listening today as always,

[00:31:14] and hopefully you can join me again tomorrow.

[00:31:16] But bye for now.

[00:31:17] Bye.