Lanfrica Talks, PODCAST, 19 September 2023
There are diverse communities in the landscape of language technologies, including corporate entities driven by profit, grassroots passion-driven groups, and a middle ground that seeks to bridge the gap between the two. The goal is to provide data through data creations that supports the work of data scientists, academics, researchers, and grassroots organisations, fostering language inclusivity for the benefit of all.
Graham Morrissey is the Chief Operating Officer for. Way With Words, an Audio-to-Text services and solutions company.
This episode is part of Lanfrica Talks. Lanfrica provides a platform to showcase efforts in language technologies across Africa.
SUMMARY TRANSCRIPT
Good day everyone and welcome again to another exciting episode of the Lanfrica Talks. I am your host, Chris, and for those who are new to this show, the Lanfrica Talks is a place to discover inspiring stories like the one we’re gonna hear today, discover pioneering projects, research, businesses, and much more. Subscribe to our podcasts and YouTube channels for more exciting talk shows. Today, we host Graham Morrissey. Graham Morrissey is the Chief Operating Officer at Way with Words, a global provider of speech to text services and solutions dedicated to addressing the discrepancy between data scarcity and the potential presented by incorporating African languages into emerging technologies. I am very happy to have you, Graham, and the floor is yours.
Thank you, Chris. It’s a real pleasure to be here today and to represent Way with Words. We appreciate the opportunity and are big fans of Lanfrica Records and Lanfrica Talks. Thank you again. We’re really glad to be kicking off and having this time with you and the community to chat about the topic at hand, which we’ve set as the value of community in data creation sets, and that’s a really important word that will come up throughout the talk today. Just to sort of frame the discussion before we get started, you know it’s a time right now, I’m sure we can all appreciate rapid technological advancement. It seems every other week there’s a breakthrough that represents an exponential jump in what was before possible rather than incremental. And in the age of large language models, I think there are so many opportunities but also so many challenges. And perhaps the biggest challenge, despite popular fear-mongering around AI will take all of our jobs one day in the future, is more the way in which people are shaping technology rather than how technology is impacting people. I think that there needs to be some strong consideration there and how do we ensure that we create, deploy, and use technology in a way that best serves all end users and not just a select few. So it’s a bit of a story on our involvement in the landscape that has come into creation over the last while with regard to technologies, language, and how those merge. And looking forward to getting into it.
So, I think to get started, I’ll provide a brief introduction to Way with Words. Way with Words has been around for 20 plus years, and I’ve been with the company for the past eight years. We have very humble origins as a business process outsourcing unit that started in Cape Town to service clients in London. And we actually had guys riding around on little motorbikes delivering cassette tapes, if any of you out there can remember those, and dedicated transcribers using foot pedals and the equipment that was hard at the time to produce reams and reams of documents that we sent across the point if you will to clients that needed them for legal services primarily. And over the years, Way with Words has grown more into a global language service provider mainly focused on transcription, captioning, and particularly in the last four to five years, activities supporting the development of technology. So, you know, large volumes of machine transcription polishing to improve automated speech recognition speech collection annotation services and you know other tasks such as large language model prompt completion tasks and anything else really where the project management of people comes into play, so it’s a really interesting time.
In the language service space with rapid adoption of technology and primarily as a human transcription provider, we have our niche in the market and our space, but it’s a very interesting time for sure as automation improves efficiencies and brings costs down, so a lot to think about there for us, and that’s the journey we’re on at the moment but very exciting for us as a business to be involved in the project side of where everything’s headed, and we’re quite thoughtful and conscious of the role we play there and the role we might be able to play forward with a particular focus on South Africa and in Africa as well.
For the listeners out there, Way With Words is a South African and a United Kingdom registered company. We’re also an Australian registered company, but a lot of this talk will be focused on our efforts in a local context, local to South Africa, and our thought process around how this might apply to African languages as well.
So for context sake, quite a bit of the discussion will be around opportunities and challenges that exist currently in South Africa.
To focus the talk, I’d like to frame the question of what can be achieved considering developed systems of language Technologies and AI can do a lot of things for many people, but what about those that currently don’t reap its benefits?
I mean it’s very impressive to think that we’re nearing human levels of intelligence or human-like levels of intelligence and reasoning, thinking, you know, assisting us with tasks, but to be honest, that is really only true for a handful of languages and possibly even one to three. So the question is what can be achieved if that level of access is spread across the board and made available to those who need it and those who currently don’t have it?
When you think about Africa, it’s probably got about a third of the world’s linguistic diversity and a lot of languages. If the world has around 7,000 languages, you know, Africa has said to have about 2,000 of those, if not more, and with that in mind, there’s a ton of work to be done if we’re going to address digital equity and broach those questions of what about the rest of the globe when it comes to established Technologies and all the benefits that come with those.
Definitely a long way to go, but I think that’s exactly why platforms like yours exist, and that’s what we’d like to talk about today. So if we think about some of the things that can be achieved if we had some parity in terms of inclusion of digital Technologies and access thereof through data creation, one of the first things that comes to mind is education, of course.
In a context like South Africa, we’ve got a great disparity between the access to education that’s available for those within the country. It’s a worldwide scenario where it’s not equal, and there is a lot of information available and a lot of services and access when it comes to English and the dominant languages out there that are accounted for in terms of technological development. But certainly, in a local sense, there’s a lot that can be done if we can get to a place where language is represented in a digital sense in the multitude of languages we do have here.
Things like literacy rates, educational outcomes, and language barriers in schools can be addressed if that’s one of the areas that’s targeted when it comes to technological development with a focus on language. Particularly Healthcare is another big one. Just imagine that there’s a lot of people out there that struggle just to get to a clinic if they may be in an area that’s remote and hard for them to do things. If they do have some kind of mobile phone or technology at home, it might be a lot easier not having to travel and being able to use technology to get a diagnosis, to speak to a practitioner, or even to a chatbot and have some kind of tools to enable them to benefit from what others can do within the same space in the more represented part of the world.
Certainly, there are areas that can benefit significantly from improved access to technology and digital solutions, and one of these is business and e-commerce. In South Africa, many people still physically visit banks to conduct transactions or check their account statuses. If they had access to these services in their native language, they could engage with their banking applications more effectively, reducing the need for physical visits and enabling communication in their preferred language.
This shift toward language inclusivity isn’t just a matter of providing opportunities; it’s also a sound business practice. When institutions and organisations can communicate with their customers in their mother tongue, it enhances customer engagement and, ultimately, contributes to good business. This approach aligns with both ethical considerations and inclusivity while also serving as a strategic business agenda.
Expanding digital equality and inclusion can have a broader impact on social inclusion, supporting active civic participation and facilitating access to digital services that streamline various aspects of a country’s operations. While the mentioned areas showcase the potential for technology, it’s essential to acknowledge that numerous other economic sectors and developmental areas can benefit from technological advancements.
These include agriculture and rural development, economic growth, job creation, legal access and justice, disaster response and safety measures, as well as the preservation of cultural heritage and linguistic diversity. Additionally, technology can enhance media and entertainment accessibility in various languages, although it’s important to prioritise pressing issues before addressing this aspect.
Now, turning our attention to the main topic at hand, which is the value of community, particularly in the context of data set creation, let’s delve into Way With Words’ experience. As a small company that has been providing services to a select group of clients, we have observed the convergence of interests and expertise within a growing landscape that appears ready for progress. However, this landscape may benefit from more direction and focus.
It’s crucial to note that I’m not an expert in coding, natural language processing, or the languages familiar to most of our viewers. Instead, I can offer insights based on our experience in creating data sets for modelling algorithms and improving speech recognition systems. While I may not have all the answers, I hope to contribute to meaningful discussions that help us navigate these complex issues.
Within the realm of AI, machine learning, and natural language processing, we can broadly categorise these burgeoning communities into two groups: those driven by passion and those driven by profit. For the passion-driven community, we see notable and commendable organisations like Masakane, dedicated to expanding the benefits of natural language processing for social good. Lanfrica Records and Lanfrica Talks play a pivotal role in fostering community, sharing expertise, and making resources accessible to data practitioners. The Mozilla Common Voice project, with its vast collection of recorded hours across multiple languages, exemplifies the power of volunteers contributing to open-source resources.
These communities are driven by a fire in their members’ hearts and play a critical role in advancing language technology. They represent the collective passion of individuals and organisations striving for social good and public interest.
On the other side of the spectrum, we have communities that operate within the realm of profit. These typically include large corporations, venture capital investors, tech platforms, and consultancies. Way With Words, while a small business with humble beginnings, has evolved into a digital language service provider. We have ventured into speech collection and data set creation, driven by the exposure gained from providing speech-to-text services and solutions.
These communities often focus on profit generation and may have different motivations compared to passion-driven groups. They contribute to the development of language technology from a business perspective, seeking opportunities for growth and innovation.
In summary, both passion-driven and profit-driven communities contribute to the advancement of language technology, each with its unique motivations and goals. The collective efforts of these communities hold the potential to shape the future of technology and language inclusion.
It’s important to clarify that having a group of communities contributing to the broader goal of inclusivity within language technology is not inherently bad. In fact, it can be quite beneficial. For instance, Way with Words has worked with numerous individuals over the past year, and in our two decades of operation, we’ve provided contractual opportunities that have not only expanded professional development but also offered financial incentives to participants involved in our projects and jobs.
Another notable technology platform, Zindi, boasts a community of over 70,000 data scientists across Africa. While many of these scientists focus on solving corporate challenges, Zindi also dedicates itself to social good. The platform hosts competitions that address real-world problems, some of which are posed by corporations seeking solutions. Data scientists within the community collaborate to solve these challenges, receiving financial incentives and prizes for their contributions. This exemplifies a fantastic community dedicated to improving skills and addressing societal issues.
These different communities share common goals of making positive contributions, but they don’t always align perfectly. There isn’t always a common thread that unites them all. This disconnect, especially between public and private interests, is something I aim to explore further in today’s discussion. We need to examine the intersection of public versus private data, purpose, vision, and the future of language technologies and the people they serve.
To delve into our experience with data collection, it has been an enriching journey for Way with Words. Starting from our humble origins, we weren’t initially equipped for the opportunities that came our way. As a transcription company, we received a request for support in data collection efforts, specifically targeting South African English and Afrikaans. This was the first time we encountered a project that made sense for us to work on locally.
Initially, our approach was basic, with Microsoft Word documents and timestamps that didn’t align perfectly with the recordings. Fortunately, the technologically sophisticated client was able to re-time everything to make it work. However, this marked the beginning of our learning curve in data collection. Over time, we developed our own workflows to handle such projects more effectively.
In our early endeavours, we focused on creating proprietary data sets tailored to specific client needs, often in domains such as retail, banking, travel, insurance, and finance. These data sets were exclusive to the client and not re-licensable. Once created, they remained solely for the benefit of the commissioning client.
In contrast, there are off-the-shelf (OTS) data sets, which follow standard conventions and are not tied to one specific client’s requirements. OTS data sets are licensable, reusable, and customisable. The key distinction is that proprietary data sets offer a competitive advantage to the client, while OTS data sets are more affordable and versatile, adaptable for various uses.
It’s important to emphasise that the availability of data sets differs significantly between these two approaches. We’ve gained valuable experience working on a variety of languages and projects, all exclusively for a single client. However, as our strategic viewpoint evolved, we began to see increased interest in South African languages, particularly South African English and Afrikaans. This interest was evident in our machine transcription polishing work, where call centres in North America sought improvements in South African language models for more accurate results. This shift highlighted a growing demand for solutions tailored to the linguistic needs of South Africa and the broader African continent.
Certainly, it’s an area we are eager to progress in because it aligns with our expertise and our vision as a company. We aspire to be a company that supports individuals with valuable skills and talents to do meaningful work, especially in the realm of language services like transcription. While we currently have our niche, we aim to expand upon it. To discuss some of the work we’ve undertaken thus far, whether involving proprietary data or off-the-shelf data, certain considerations come into play during data collection. I’ll address these considerations now.
Firstly, data governance is of paramount importance. When embarking on a data collection project, it is crucial to be precise about the intended usage of the collected data. Contracts with participants must be established to safeguard and anonymise their information. Additionally, ensuring the data set is free from bias is essential. It should be fully representative, encompassing a wide range of participants to reflect the linguistic and demographic diversity relevant to the language in focus. These elements are fundamental in any data collection project.
One key aspect we discovered later on that proved to be pivotal in the success of our projects is the sense of community. Throughout numerous projects, we encountered challenges ranging from meeting deadlines to addressing socio-economic issues, technological hurdles, and linguistic diversity. Over time, we realised that establishing a strong sense of community among participants was the key to overcoming these challenges. When participants have a clear understanding of their contractual agreements, including how the data will be used and the benefits it will provide, it greatly enhances the project’s effectiveness.
Moreover, when working with participants who speak different languages, it is imperative to create instructional materials in their native languages. This ensures clarity and resonance with the participants. Designing simulated conversations for individuals who have never met can be deceptively complex. For instance, instructing person A to play the role of an upset customer needing to explain why a product should be returned, while person B takes on the role of the company agent explaining the return policy, requires clear and accessible instructions. Additionally, recruiting participants with diverse backgrounds is essential to the project’s success.
What we discovered was that fostering a sense of community was crucial. Engaging in group meetings, being accountable to project participants, getting to know them personally through video calls, and allowing them to become comfortable with each other and the process significantly improved project outcomes. Building this communal engagement and interaction among participants made it easier for them to contribute meaningfully to data collection. Often, by the end of these projects, we were pleasantly surprised to see how participants had developed bonds, engaged in conversations on various topics of interest, and, ultimately, enjoyed the process. This sense of community greatly facilitated the success of our data collection projects, making them more than just tasks but rather collaborative endeavours with shared goals.
The participants often expressed sadness at the conclusion of our projects. This sadness stemmed from the enjoyment they derived from being part of a group with a common goal and actively contributing to it. This experience fascinated us and highlighted the effectiveness of bringing people together for a shared purpose.
One notable revelation was the divide between public and private, and licensable versus non-shareable data. We believe it’s crucial to explore and understand this dichotomy as we move forward.
Another important consideration in data collection is understanding the motivations of participants. These motivations often align with either public or social good, or private and restricted interests. Altruistic purposes, such as enhancing professional networks, improving professional skills, preserving language and culture, and contributing to charitable causes, are some of the driving forces behind participation. However, in regions with a high quality of life and decent remuneration, financial incentives may be necessary to attract participants.
In summary, embarking on a data collection project necessitates careful consideration of data governance and ethics, as well as sustained community engagement to maintain motivation.
Looking at opportunities and challenges for digital language technologies in Africa, there is immense potential in preserving linguistic diversity and using language technology to provide inclusive digital services. Collaboration across languages and regions presents a significant opportunity.
However, challenges such as data scarcity, limited resources, and sustainability issues must be addressed. Data collection remains a painstaking process, and determining the sources for linguistic data can be challenging.
Advocacy and awareness are vital in prioritising language needs and committing to long-term solutions for real-world problems, whether in healthcare, education, or agriculture.
In conclusion, there are diverse communities in the landscape of language technologies, including corporate entities driven by profit, grassroots passion-driven groups, and a middle ground that seeks to bridge the gap between the two. The goal is to provide data that supports the work of data scientists, academics, researchers, and grassroots organisations, fostering language inclusivity for the benefit of all.
During the presentation, Graham expressed deep gratitude for the opportunity to share their experiences and insights and highlighted the importance of discussing these topics in the upcoming discussion session.