EP 136: Jenalea Rajab,Championing Linguistic Diversity and Inclusion in AI

In today's episode, our guest Jenalea Rajab from Lelapa AI talks about her work in reshaping the landscape of natural language processing but and championing the cause of underrepresented languages. Jenalea's transition from the field of electrical engineering to AI innovation is as intriguing as it is inspiring, offering invaluable lessons in how diverse experiences can culminate in a career that's both dynamic and impactful. 

LinkedIn: Jenalea Rajab

Transcript

Anne Chisa: 0:21

Hello everyone and welcome back to another episode of the Root of the Science podcast with your girl, Anne With an E. If you are new here, welcome to the show. It is such a pleasure to have you on. Remember that you can follow the show on all the various social media platforms and you can also tell us if you would like to be on the show or if you think there's somebody we should have on the show. Reminder that every Friday we have a newsletter that goes live on our LinkedIn pages as well as on email, so make sure you subscribe. The newsletter is information and stories by other African Science communicators not only me, african Science Communicators, not only me and if you would like to subscribe or contribute to that also, do send us an email at info at rootofthesciencemediacom.

Anne Chisa: 1:16

Now that we've got all of that out of the way, this is another episode of the Women in AI series. It's been such an amazing series and it's lovely just getting to know some of these incredible women who are doing some phenomenal work in this field of AI, particularly in Africa, and today is no exception. Today we have Jenalea Rajab on the show. She is from South Africa and works at Masakana, at Lelapa. In this episode we will talk about her AI journey, the winding road of it and how she got there, some of the amazing work that she's part of and the projects that she's part of at Lelapa with language processing models, and also we talk about some of the ideas of how to have more diversity and inclusion in this field of AI and also the future that she envisions for this field. So come on, let's take a listen, let's go. Good morning Jenilee.

Jenalea Rajab: 2:20

Welcome to the show thank you, it's really nice to be here.

Anne Chisa: 2:24

It is such a pleasure to have you on today. Could you just quickly brief us? Who is Ginalee, where are you from and, just briefly, what do you do?

Jenalea Rajab: 2:35

Sure, so my name is Jenalea Rajab. I am from South Africa. I am an AI research scientist at Lelapa AI and part of my role, I lead the product research team and we build natural language processing models for low-resource languages.

Anne Chisa: 2:58

Awesome. Thanks for that intro, ginalee. It's a career that is pretty big at this time and a lot of people want to get in it, and AI there's a lot of overlaps into so many other different professions. But for you, how did you get into this field like, from the beginning, what were you doing? Did you always know that this is something that you want to do?

Jenalea Rajab: 3:21

No, definitely, definitely not. I certainly didn't take a very linear path as a researcher. I have had a lot of different roles in my career. So actually I started as an electrical engineer that's what I have my degree in and then, after I'd finished studying I decided that I was really overstudying and I didn't want to continue into Masters etc. I just wanted to work. So after that I did a graduate program and it was in information systems. So with that I got to work overseas and travel a lot and explore. So that was really fun. But I ended up coming back to South Africa because I love it here and it's always going to be my home.

Anne Chisa: 4:12

Yeah.

Jenalea Rajab: 4:15

And then after that I actually became a project engineer. So I worked on the substations in sub-Saharan Africa for a while. I really enjoyed it. It was kind of back to my roots as an engineer and I also had an amazing team that I worked with. But obviously it's not always easy to be a female in these rural substations in Africa, so you don't really get a lot of autonomy in a role like that Africa, so you don't really get a lot of autonomy in a role like that, yeah, so then it's going to be quite a winding story.

Jenalea Rajab: 4:54

No, let it wind, let it wind, okay, yeah. So then following that, I decided to rather do system analytics. So I joined a small mining company in South Africa and I helped with the system analysis so everything from the inventory to the asset management to the HR systems. So that was quite a learning experience and I got to touch so many different parts of how a business is run and how it works and stuff. And then, through that, actually that's how I decided to go into AI, because after doing all those systems, we had this mountain of data that was all digitized and I thought you know, actually you could do something with this data. And AI was obviously all over the news, especially data science.

Anne Chisa: 5:45

Yeah.

Jenalea Rajab: 5:46

And so that's when I decided to work part-time and go and do my master's at WITS, yeah, and then, following that, I then got involved with Masakana and Jade Abbott and then eventually made my way to being a researcher at Dalapa. So it's not a straight path and I actually am very grateful for the fact that it wasn't, because I've had so many exposures to how businesses work in different aspects and you can really pull from that. You know, I think it gives you a lot of different perspectives.

Anne Chisa: 6:26

um, recently actually, I've been reading this book it's called range by david epstein and um, I think it just shows that you can kind of start something wherever you are in any stage of your career, and it might not be considered the natural progression, but it can always work out no, this is amazing, genevieve, and I'm so glad that you've already touched on that, because that's what I wanted to touch on, I think for many of us okay, well, rather, let me not speak on other people's behalf, but this is something that I've also just been thinking of but the idea of pivoting because I think maybe you sometimes feel like, if this doesn't work out, oh my goodness, I'll have to, like, start again and learn something new. So how? I mean, how was that like the actual processing? Was it easier or was it hard in the beginning? And then it got easier to go from electrical engineering and then to pivot to here, and then to pivot to here.

Anne Chisa: 7:26

Did it get better over time? Because I think that's what keeps many of us stuck in specific roles, because we think, oh my goodness, I don't know how I'm going to go from here to here to here. But I think, just listening to your story, um, it's quite amazing how you started from one particular place. You went somewhere else, you kind of got back to it, then went somewhere else, then you know, and you're now here. So how is that like the process?

Jenalea Rajab: 7:53

um, in your, in the, in the road, in the road ship um, yeah, I agree, it's always very daunting to try something new, um, and often I felt, you know, when I was doing my master's and stuff, that I I got in there so late you know, I was like one of the oldest people in my master's class and um, and also just seeing how much experience people had over me. You know, some some people have obviously been in the AI field for like five years and even even more, and I was just starting, um, but you know, I think sometimes you just have to go with the flow a little bit and seize the opportunities that come your way. So one of the biggest benefits that I had was just leaning on the people around me, you know, just accepting that, yes, ok, I don't know as much as them, that's fine. It doesn't mean that I won't someday know as much as them. So, while I can, let me just absorb as much knowledge as I can from the people around me and often people are very receptive to that.

Jenalea Rajab: 9:05

If you admit oh sorry, I actually have no idea what that concept is. Would you mind explaining it to me? It's very rare that somebody will say no, but yeah, I think that it is very hard. I'm not going to sugarcoat it and say that it's easy to just change your career direction. But if you find something that you're really passionate about and you would really like to try, I mean, even if it doesn't work out, at least you know right. I mean, even if it doesn't work out, at least you know right, at least you know. Okay, I wasn't really meant to be a research scientist and maybe I should just be an engineer. But then you have more range and you can use all the stuff you learned trying to be a better engineer.

Anne Chisa: 9:48

So yeah, they say it's never failure, it's just lessons upon lessons upon lessons yeah, exactly oh, awesome, okay, so now let's talk about some of the work that you're currently doing. Um, could you tell us about some of the projects that you're currently involved in and, um, something, the things that you can share, of course, and, um, yeah, in terms of the language processing, um field that you're share, of course?

Jenalea Rajab: 10:20

and, yeah, in terms of the language processing field that you're a part of, yeah, sure, so at Lapa AI, we focus on low-resource African languages. So currently we are working on our Vula Vula product. It comprises of transcribes, excuse me, transcribes, so text to speech. I mean, sorry, let me just reverse it, okay.

Jenalea Rajab: 10:47

So we're focusing on our Vula Vula product. It has four different components. So one of the components is transcribes, so where we do speech to text components. So one of the components is transcribe, so where we do speech to text, and this is obviously one of the major focus points for call centers and customer facing interfaces, where people obviously phone in to get advice and then these calls need to be analyzed. The second one we're focusing on is speech, which is text to speech, so the reverse of transcribe. We also do analyze which is involved in named entity recognition and sentiment analysis and all the analytics that you can put on top of text. And then, lastly, we're focusing on converse, which is creating products that enable intelligent chat spots in these languages that we care about. So our focus for now is predominantly on South African languages. Obviously that's where we have the most knowledge, because the founders are all from here, and also, then, some of the major African languages like Swahili Hausa, yoruba. Those are also some of our focus points.

Anne Chisa: 12:01

That's so fascinating and obviously very needed work. I've always had this question, actually, generally, when we're speaking of the work that you do in terms of language processing Many of you are data scientists, etc is there an overlap with other disciplines, like working with people who've done like languages, like I don't know, like literature, etc. In terms of helping you develop these types of language processing models, or is it just purely the coding?

Jenalea Rajab: 12:43

Yeah, so at Lelapa we draw heavily on data experts, or data experts, as well as language experts.

Anne Chisa: 12:50

Yeah.

Jenalea Rajab: 12:51

So we don't pretend that a computer scientist can do everything. Certainly you can't, and especially when you're working on languages that you personally don't pretend that a computer scientist can do everything. Certainly you can't, and especially when you're working on languages that you personally don't speak.

Jenalea Rajab: 13:01

You know it's very important that you involve linguists. So we have a lot of linguists that we contract out to evaluate our models, to tell us if they're firstly, culturally relevant, if there are any harmful outcomes from our models. To tell us if they're firstly culturally relevant, if there are any harmful outcomes from our models, and then also to guide it from a more linguistic perspective. You know they have studied this. They are experts in this language. They can tell us oh, for example, your system is not working very well on verbs or it's not capturing this particular diacritic that you know is part of the language. So certainly we draw on them quite heavily and we never put a model out into production until it's been evaluated by a linguist or a cultural expert to make sure that it is on track. I know that a lot of cases don't do that and I think that's why you see a lot of cases don't do that.

Jenalea Rajab: 13:56

Yeah, and I think that's why you see a lot of like harm in AI, especially with these large language models. But it's very important for us at Lelapa that we create models that like fit the people they are made for right. So if you, for instance, speak Istizulu and you want to use this product, it will be a seamless experience for you and you won't feel at some point, oh, that's not really part of my culture or that's quite offensive, you know.

Anne Chisa: 14:26

Fascinating, fascinating, and yeah, I mean those are the conversations that, as much as AI is very helpful, it does have some, you know, issues. And apart from that, and apart from just the language aspect, what other ethical considerations do you have to consider, or maybe other challenges that you have to consider when working with AI, particularly, as you know, as there's a huge increase of the integration of AI into various aspects of society?

Jenalea Rajab: 15:02

Yeah, that is a huge cornerstone of AI at the moment around index, especially when these huge corporates release these models into the world. And I think one of the things that we particularly do is obviously, aside from making sure there's no bias, that it's the model is representative and not harmful yeah, is. We also try and make sure that our models are explainable, that we understand why our models are producing the outputs that they produce. So explainability is a huge thing in our team, and one of the things that is driven by the founders is to really make sure you understand why your model is doing what it's doing. So we go into the data ourselves and we see, okay, this output is happening because of this specific aspect in our data, but we need to get more data for that.

Jenalea Rajab: 15:56

And then I think another thing that's very important is to really define your scope of your model, so you can't release a model into the wild and say, oh, this will fit every use case. You have to specifically say this has been made for a call center, right? So if you now start using our transcription model for general domain WhatsApp transcriptions, it might not work as well. Model for general domain WhatsApp transcriptions it might not work as well. You might have things that you know creep in that were unexpected, so I think that's one of the ways that you can promote, like ethical AI, and safe AI is making sure that people understand the scope of your models, that you're transparent with the data that it's been trained on and also that you've taken the necessary steps to make sure that it's not a harmful model.

Anne Chisa: 16:47

No, for sure, for sure. So, for example, in the Vula Vula project that you're currently working on, you spoke about the idea that this can be used in core centers. So this is sort of the end use application of this in the real world. So, with the work that you do at Lilapa, is it companies let's say, for example, something that both of you and I know let's say Vodacom, which is a cellular network would come to you and say would like to use this type of model in our core center, in our core centers? Is that is that how evidently this all works?

Jenalea Rajab: 17:29

yes, yeah. So, um, we, we try and do um these products as a service, so, um, for, exactly as you just suggested, people who have call centers in Africa, who have a large client base, for example, who don't only speak English which is a lot of what the call centers cater for now, and they have no way of analyzing, for example, the conversations that the call center agents are having with the customers, and they would come to us and say could you please? You know, can we use your product, for example, to detect which language is being spoken and then do the transcription so that we can do our quality assurance checks on top of that? That's kind of the approach. We also get approached by companies to do chat interfaces. So we are not in ourselves, a chatbot company, but we can enable chatbot companies to service their customers in more languages, you know, detect intents and stuff in Isisulu, in Susutu, in Afrikaans, for example, and then make their platforms and their businesses more successful and reach a wider range of audience.

Anne Chisa: 18:42

Fascinating work, absolutely incredible work that's being done at Lelapa Kudos. To you and the whole entire team. This is something that's obviously very well needed and it gets me excited. It gets me so excited that there are people who are thinking that way to make um for inclusivity and just accessibility at a language level, because english is not the most common language in the world and there are many people who get left behind when that assumption is made.

Anne Chisa: 19:16

so congratulations to you and the team. I just want to go zoom out a little bit you touched on it earlier on the idea that in your journey you worked in some places where it was predominantly male dominated and there were some you know challenges and I know this is also a topic that you really see, particularly in the tech industry. Right, and you can talk maybe on your own perspective or just generally. I just wanted to ask you how do you believe we can encourage more women and people of you know unrepresented groups to pursue careers in ai and tech, given this ecosystem that is currently existing, which is not always so welcoming and particularly favorable?

Jenalea Rajab: 20:15

uh, yeah. So I think that's one of the biggest ways that we can do. That is just exposure. So one of the things that I help with with Pilonomi, who's the CEO of Lelapa and one of the co-founders, is we work at this organization or she founded it. It's called Code Comorso, and we go to high schools and do IT courses for particularly girls' schools. Often girls' schools don't have that as a subject, so that in and of itself is amazing exposure. I mean, you don't even know what you could do potentially if you've never seen it right. So, unless maybe you have somebody in your family who's in the tech field, often girls don't even think of computer science as a possibility. So that's one of the biggest things is exposure. And then also there are some amazing communities specifically devoted to women in AI, devoted to women in AI. So I would definitely encourage anybody who's thinking of getting into the field or has already been in the field but is not getting the support that they think they need is to find these communities.

Jenalea Rajab: 21:30

I myself am involved in quite a few and it's such a nice space. You get so much support, you get a shoulder to cry on. You know everybody's kind of going through the same things so you don't feel so alone. Um, just recently, actually, with Masakane, we had a all like a female only research paper that we republished. So, um, you just have to find your people to be.

Jenalea Rajab: 21:57

To be honest, I mean, it is quite hard, especially if you find it quite intimidating to be in a room full of men. Um, I've had that since undergrad so I think I'm a bit used to it. Now maybe it's easy advice, but, um, but honestly, there's such amazing communities out there, um, and they are really, really supportive and I think you know women supporting women is really really often in my career, some of the the harshest comments I've received are actually from women and I've always thought that, you know, there's so few of us here like we should be pushing each other up as much as possible yeah so if, for example, you are already in the field of ai, I think, if you can and if you have capacity, you should try and help uh, you know people coming up in the fields and add them to your communities, and add them to like initiatives that you're doing, for example, and just make them feel a bit more included.

Anne Chisa: 22:58

No, so true. Yeah, so so true. It does require, I think, from the inside as well, as much as we do say that the men need to take some accountability. But I think women helping women to make them feel safe, to make them feel, you know, valued as well, and there's so much space. I fully agree. Thank you so much for that input as we wrap up, last question what do you envision? Because I mean, this field is obviously moving at such an accelerated pace, so maybe even what you envision could happen, like next year. But I'm just thinking, like looking ahead into the future of AI, what do you envision for it as a whole and what role do you hope it plays in shaping the future?

Jenalea Rajab: 23:50

Yeah, that is a very hard question. Not the role I hope it plays, but just the next exciting breakthrough.

Jenalea Rajab: 23:58

I mean like you say, this field is just accelerating at such a pace that there's something released every day. I think the most exciting thing about particularly the NLP field is the amount of open source models that are released. It's quite mind-blowing that you can have access to these models that Facebook and Google, et cetera, have spent millions on creating and you can just build whatever you want on top of it. That makes me very excited because it means that people have the agency. You can take it, fit it to a use case that solves a problem that you have or your community has, and you can use it. So my ideal future would be that AI becomes more inclusive because of it. I mean there's lots of people putting in a lot of work to make sure that these LLMs are fit for more cultures, more languages are not just represented of the Western world, but also represented of our lives in Africa. So ideally, it would become more inclusive and that people can start solving their own problems.

Jenalea Rajab: 25:14

You know, I always well, one of the things that Lallapa always says is that nobody's going to solve our problems like we can, because we're the only ones who know the problems. Yeah, and even yesterday I was talking to a friend of mine and they were saying how one of the biggest problems with ambulances at the moment, especially in South Africa, is obviously in some of the townships, a lot of the streets are not named or they're not shown on Google Maps, for example. And that's something that we could fix. You know, it's a problem we have and we could fix, but the West is definitely not going to fix that problem for us because they don't have that problem right.

Jenalea Rajab: 25:52

So I think that just the idea that there's so many people growing in this field, particularly on the African continent, we have so much talent here, and so we can just take these models that are being released by these big companies and we can just solve our problems.

Anne Chisa: 26:09

Yeah, what an exciting, exciting space that's so exciting. African solutions for African problems. Yeah, I love that. I love that. Generally, it's been amazing chatting with you. Thank you so much for taking the the path for so many other women um to to come into this space and to to see you guys as a beacon of hope.

Jenalea Rajab: 26:48

so, thank you yeah, thank you so much and thanks for having me.

Anne Chisa: 26:52

It's it's been a nice conversation it's been a pleasure and to everybody else who's tuned in, thank you for listening to another episode of the root of the scienceot of the Sands podcast with your girl and with an E. Until next time, goodbye.