Subscribe
Share
Share
Embed
A podcast from the ETEN Innovation Lab exploring acceleration in Bible translation. Tune in for experimentation, updates, and conversations about new methods and technologies advancing Scripture accessibility.
Isabella Scarinzi 0:07
Hi friends, and welcome back to the Bible translation innovation podcast. I am Isabella currency, your facilitator, and I'm also joined here today by clappy. Hi, clappy. Hello, Isabella and Joel. Hi, Joel, Isabella. We are part of the E 10 Innovation Lab, and in this podcast, we chat through emerging technologies and methodologies in Bible translation. So for today's episode, we have a great topic ahead to discuss. You have probably heard or used the popular messaging app called WhatsApp before. So we're going to talk about how WhatsApp can be used for Bible translation. Whatsapp was officially launched in November of 2009 as a chat app service, and it was bought out by meta or Facebook in 2014 it's estimated to have more than 1.5 billion users, and it's ranked as the most used mobile messenger app in the world. So to Jill and clavi, we can begin with this. What makes WhatsApp so appealing, or maybe useful in our Bible translation context?
Klappy 1:19
Well, the quick answer is, there are actually teams already using WhatsApp for Bible translation. It's kind of mind blowing that without any aid or helpers, they're using it as their primary mechanism for sharing their recordings, audio recordings for OBT oral Bible translation. So there are teams around the world already using it without any extra functionality, just because they can record voice notes and share those audio recordings back and forth between teams without any extra technology, without any training. Just out of the box, you have instant sharing capabilities that they're able to use. So I'd say, fundamentally, the reason why it's appealing to me to use for Bible translation is there are teams already using it.
Joel Matthews 2:09
Yeah, that makes a lot of sense. And I'll just to add to that, I think it's also the fact that it's already installed in probably a translator's phone. We don't need to market it, we don't need to build it, we don't need to maintain it, and it's there. And if we can find a way to leverage that entry point for Bible translation also to go to, going back to clappies point, I mean, it's very natural for people to continue a conversation on WhatsApp, even though they may be using another application for doing their translation, primary translation work, it's common for them to ask, ask a friend or a colleague over WhatsApp Some questions, and it's end to end encrypted. And also multimodal, so audio, video and tech support, those are, I think, great reasons to explore the possibility of using WhatsApp for Bible translation. One interesting aspect of WhatsApp, because, well, it's meta, and meta paid a lot of money to buy WhatsApp, and so they are attempting to monetize insane ways the whole of the platform. And one of their approaches has been to have business accounts, and I think that opens up the door for us to programmatically have bots on their platform that we can leverage to interact with human users. And this is where it gets really interesting.
Klappy 3:59
Yeah, in even though people are already using it in in ways that are unaided by AI, I think you know you mentioning the the commercial or business accounts, which enable you to create LLM based bots that allows us to actually come in alongside of existing workflows using WhatsApp and augment their existing processes. In fact, the first time I had heard somebody asked about using WhatsApp for Bible translation was somebody saying, hey, my team is recording on Android recorder. They're sharing their files through WhatsApp. Can't we just use AI to help us organize our files and manage those audio recordings a little bit better. And that was the first time I was like, Huh? Why would anybody use WhatsApp for Bible translation? That's so weird. But then the more I thought about it, you know, that lines up with the things we've already been sharing. Like, oh. It's already on everybody's phones. There's no training involved. You just mainly have to help set people's expectations of what they can and what they cannot do through the existing tool that they already use. And if they were, if they were interacting with anybody as a mentor online, they're using some sort of chat platform. Either it's voice communication messages that they're sharing back and forth, or they're typing back and forth, depending on their preferred communication mechanism, they're still using something like WhatsApp to send messages back and forth with their team or with their consultant or mentor that's helping them. They're through their process, agreed. So if we start so, if we take that a step further, we start building in these large language model or bot helpers, they would just interact with them the same way they would with a human.
Joel Matthews 5:55
One other reason, I think of is especially in the developing countries, I have seen differential pricing for Facebook and meta services like WhatsApp. So you'll find people in the villages with a smartphone that has air quotes Internet, and what that means is they have access to the Facebook app and the WhatsApp app and maybe Instagram, but they don't have really access to a web browser, Internet through a web browser and and It just goes to show how deeply penetrated the ecosystem is and how the industry has been trying to push these sort of technologies which we can leverage to our advantage, and that's exciting. And like you're saying, Flappy, I don't think either of us saying that this can replace what what's already being done. It can be a really powerful support for making a big difference in the Bible translation, speed and quality.
Isabella Scarinzi 7:16
So I'd love to go more in depth into that. We talked about how it can be combined, maybe with other tools and technologies. What are some of the ways that it can be integrated with the tools that are already meeting the Bible translation needs?
Klappy 7:32
That's a great question. One of the ways that it can be used to augment existing tools and workflows and processes is the fact that most of the time, we already have a process using tools in one of the biggest gaps we have is community checking. So having a way to get your Bible translation checked by the community and distributed and get meaningful feedback back incorporate it into your process. There's usually a new tool that needs to be developed, and some tools have built their own custom tools that do community checking, but there's limitations to how well they can be distributed. Either they're not available on mobile phones or, you know, as Joel mentioned, people may not have access to the full internet. They only have access to WhatsApp, Facebook, Instagram, a few few select services. So imagine being able to take any Bible translation workflow or tool that you're already using and be able to somehow distribute the translation to the community for for community feedback. So get quantitative feedback from the community. Is it meeting the need? You know, there's, there's different ways of community checking that could be performed through WhatsApp that doesn't require an entire new tool to be built. It's really just kind of augmenting and tapping into existing processes and servers and just finding new ways to distribute the audio recordings or maybe even the text right.
Joel Matthews 9:13
That's right. And to add to community checking, I would also think WhatsApp is a really cool way for us to do some crowdsourcing, generally. So if, if we want to, say, collect some seed data, and what I mean by that is an initial set of curated data, data from humans, so so that we can then fine tune a AI model, I believe WhatsApp could be a very instrumental tool to do this quickly and efficiently, rather than building our own phone app for. Of a bespoke phone app, maintaining and and distributing, which itself is a huge challenge, even if you had the best phone app to do this, getting it on to the phones, installed in the phones of all the end users, even if it's a website, getting the links to them and opening getting them to use it is actually a challenge that is underappreciated. Just because it's available doesn't mean it's used, and that is where WhatsApp shines. It's already being used, and so we could technically start using WhatsApp to request for collecting text, audio, even video content that can then feed into an AI pipeline, which we could fine tune and use for specific Bible translation tasks like machine translation or say speech to text for oral Bible translation. Yeah.
Klappy 11:03
I think data collection is probably one of the biggest multiplying effects, right? So WhatsApp could be a force multiplier in many of the steps or processes of Bible translation, but the one that that it would multiply the most would be in the beginning stages, like data collection. If you think about the number of times a tool needs to be installed for quality checking versus data collection, it's probably an order of magnitude of 10 or 100 to one, right? That the number of people that would be needing to collect data from would be, hopefully hundreds of people per language, instead of like a select few to help curate and do translation work or to check the translation so at the very beginning, you know, if you were to look at a funnel, you want to have the widest amount of people being able to collect data from per language, and then multiply that out at the number of languages that we would like to engage as we get to these long tail languages, they're in the harder to reach areas in that it's those smaller languages are in these regions that are harder to deal With modern technology, and so having the broadest internet connection and using a nice laptop is not going to be practical. So if you're doing data collection on a laptop requiring good internet, that's a stark contrast from just using the phone you have in your hand and the app you already have installed and just having a conversation with it to collect data. So I think data collection is probably the biggest leverage point.
Joel Matthews 12:47
One thing that makes me also excited about this is it opens up the possibility of breaking away from the linear cycle of doing Bible translation. So we have this concept of doing drafting and then some sort of checking in different ways involving different people, and it's usually one after the other, and I'm oversimplifying, and there are ways to optimize this, but I think WhatsApp allows it to be fundamentally asynchronous and a little more just in time. So for example, for doing community checking like you mentioned clappy, what if an AI bot is looking at the current state of the translation as it's being done on your favorite platform, and it realizes that you are or you even ask it that you are fumbling over how to translate say the word baptism best that it communicates the right meaning to the end users. And at that point, you know, despite figuring out what it means and reading through all the notes and study material around the word baptism and how best to translate it, you would also value actual speakers of the language giving you some feedback and their thoughts. So rather than waiting for the end of the whole process of your drafting to then go and have a community checking process where you might even have to bring people physically together, which may be unsafe or not feasible, depending on where you are. What if the bot could turn around and ask people who have the bot as a friend on their phone, which could be trusted group? Of community members, it could ping on the on the community through WhatsApp and ask them specific questions. And the beauty with llms being available today is that now this can be so smooth and conversational, and it's so easy to make a bot that just is able to easily have a conversation with you while steering that conversation towards a goal that you would ultimately like to achieve through that conversation. So in this case, you you might ask the question for the goal of the bot is to clarify how this translation sounds and what this evokes in the mind of the trend of the speaker and how the bot might go about it. This might be a series of questions and answers through WhatsApp, and they might answer through text, through audio, and asking five people the same question, gathering the feedback, summarizing, getting kind of a consensus, and giving that information back to the translator in a synchronous manner, I think, unlocks certain powers that I don't think we have leveraged so far in Bible translation, where the process has been seen strictly, not always. I mean, I'm simplifying, but more linear than not. I don't know what you think about that.
Klappy 16:33
Yeah, as you were talking, what came clear to me is there are two key things that make WhatsApp very attractive. The first is, it's a natural way for us to engage more people without having to build more apps, right? So it's the common user interface that we don't have to go out and then build and rebuild and maintain. The second thing is, it's a natural way not just to engage with people, but to engage with AI, right? I mean, every time you use AI, you're using a chat bot. Like 90% of our interaction with AI is we log into chat. GPT, we log into our favorite. You know, whether it's grok or anthropic, you know, Claude, you're having a conversation with the LLM, so WhatsApp is bringing people in AI, together in a natural way, right? I think it's the common nexus in the beauty of WhatsApp, is it is probably perfect for all of that to come together.
Joel Matthews 17:37
Yeah, totally. And it's, it's awesome how like you're saying, people are already using it, even starting to think of doing this in some pockets.
Isabella Scarinzi 17:49
So this might be a little too far out there, but could WhatsApp and AI bot? Could that be the application used for doing Bible translation?
Joel Matthews 18:01
That is a bold question. And I think making, I don't want to be the person making predictions for what will happen in five years or 10 years about this, but I do see at the rate at which llms and conversational AI is being adopted and is able to produce value, I do see that it'll only be able To do more than what it can do now in the next two or five years. And so at that point, if we just think of the more traditional app, like clappy, you said, the UI is basically the interface of chatting with it, whether through text or audio or even video, then WhatsApp could very well be the place that Bible translation. A lot of the Bible translation parts happen because it's collaborative by default and already has the pieces needed for doing something like this with a more powerful LLM and AI system backing it?
Klappy 19:30
Yeah, I'm right there with you. Joel, like asking if it is going to be the Bible translation tool, right? It's, a that's a leap. Five years from now, we might look back and laugh and say, obviously, of course, it was. Or we might say, Well, that was a dumb idea. Why do we even try that? So I don't want to predict either way, but I think the natural thing for us to do either either way is to try the first. First thing we do is we build a companion app. The companion app starts off being a reference implementation for key augmented functionality that you can use with any tool, any process, whether you're using Paratext, scribe, Codex, or, I don't know, pick your tool. You can use it. It doesn't matter. This would be an augmenting tool that could potentially be a bridge between all of the tools, right? It doesn't matter what your process is. It doesn't matter what your primary tool is. We can share files. We can share conversations through WhatsApp, and it's just an augmenter and a companion app that helps us tap into the power of not just the community and more people to get involved with the process, but also tie in, tie in some AI bots to be able to help augment your process without having to change your tool.
Joel Matthews 20:54
That's right, yeah, and the use case you mentioned about seeing it as an assistant. I think the most relevant way we can see it being practically useful for us today is seeing it as a way to answer biblical questions, which is helpful for the translator. And I know clappy, you know a lot more about that, having done POCs and such for this. So maybe you can share a bit about that too, because I was mind blown seeing what all it can be if we just leverage it in that space.
Klappy 21:36
Yeah, I appreciate you bringing that up. You know, we're talking about building proof of concepts, and we're talking about using it for assistance. The first example for using it as a translator's assistant to understand scriptural principles is something called the BT servant, or Bible translation servant. And that's something that unfolding word is putting together. And the idea is, what if WhatsApp was your primary Bible translation? Go to assistant to learn about biblical principles or learn about a passage and the proof of concept that they brought to us uses a typical rag based system or retrieval augmented generation, to where you can ask it a question, it searches its database, and it comes back to you with with an answer. So instead of building a whole new tool, they built it into WhatsApp, because people already have WhatsApp in their pocket, and so they were already kind of leaning into the thoughts in the discussions we had earlier. You know, in this episode, focusing in on let's reach people with the tool they already have in their pocket, and let's not worry about building another tool or building it into a tool. Let's let them just ask questions and let the bot answer and help them learn about certain passages or concepts, and they went up a little bit deeper of how do we have some sort of stack ranking of biblical resources that the team trusts. Like, let's start and look in and this resource, and then if it doesn't have an answer there, we'll fall back into some other resources and ensuring that it properly quotes scripture and it doesn't reword it. You know, how do we sandbox it from from saying heretical things or hallucinating. So their their approaches, like the Bible translation servant, just like we are with any other chatbot for for scripture, you know, in the Bible translation process, we're trying to prevent those types of things, hallucinations, and the idea of it, making stuff up and not properly quoting our resources. So that's being solved in a way that can be reused across other tools.
Joel Matthews 23:48
Yeah, that's a that's a really cool example of how it's I think it's so easy to deliver value go from an idea to actually have it in people's hands that they can try in a short cycle by just implementing an LLM connected with WhatsApp. A couple thoughts that come to my mind, even though we are like praising and praising WhatsApp and all that we can do with it, there are a couple limitations that I think are worth highlighting too, that that is worth considering for our listeners, if, if they ever want to apply some of this in their own context. One, one obvious one is, WhatsApp is an online app. Ultimately, even though you know you can view your previous conversations offline, you would still need to be online to be able to have a conversation with somebody or chat with a bot, for example. So it does need internet, though, to counter the. But I think the internet, especially on the mobile phone, the coverage for that, that part of the demographics, is just ballooning. I think even even in the most remote, remote parts of the world, there would be probably a cell phone connection, and there is more possibility for a cell phone connection, at least, than a wired connection for the internet. So I think if anything, this is bet, the best bet we have for using an online app to do such a thing,
Klappy 25:41
I'd like to jump into your first point, right? Yeah, you're talking about the limitations of it and how it requires Internet, while that's becoming less and less of a factor as more people get internet, or, you know, sell access on their mobile phones or smartphones. I also believe that, if we're careful, we can architect the exact same stack, the exact same technology. And I don't want to get too technical, because I don't know how technical our audience is, but there are things we can do to ensure that all the hard work we're doing to build our WhatsApp bots, we're not painting ourselves into a corner that only works for WhatsApp. We're integrating the same things for maybe a web based solution or offline based solution. So there are a few proof of concepts that have been going around for offline, offline, large language models that work on a laptop to where you can ship a laptop to a remote region, and everything works on that laptop. So it's got a large language model that runs on this fancy laptop. It can take your voice input, transcribe it, use the multi, multilingual, multimodal, large language model that runs offline without an internet connection, and then generate a voice answer. So it's using these resources that it already has offline on this laptop, and then responding only from the resources on the laptop. So I believe that we can build technology that serves both WhatsApp and these offline laptops with the same stack. And I don't want to get technical on this episode or on this podcast, but there are ways we can do it, but we have to architect it to where we're not painting ourselves in a corner for one or the other.
Isabella Scarinzi 27:42
Yeah, yeah, that's great. So I do have a question for our listeners, our Bible translation friends out there that might be wondering, how can they get started with this? How can they use WhatsApp? What are some practical advice that you guys might have and clappy, I know you don't want to get too technical, but this might be the time to nerd out just a little bit. Yeah.
Klappy 28:06
Well, if you are technical and you want to build your own WhatsApp bot, you can ask chat, GPT, right? You can just say, help me build a whatsapp bot, and it'll walk you through the process. The first time I did it, within a couple of hours, I had my first WhatsApp, but like I took a week off to help my wife with her, with her business, right? My wife has a business selling curly hair products, and she's got this is following online, and she has a lot more people reaching out to her, about her, either her products that she sells, or trying to get an appointment with her to do for her to fix their hair, right? So she has lots of intake, like a lot of people are reaching out to her, and she doesn't have a lot of time to answer everybody. So I thought it would take me a week to build a WhatsApp, but to help her funnel those questions. You know, the field all the questions, create some sort of intake form and have it conversationally, ask the questions that she would normally ask manually. So it only took a couple of hours to get the first proof of concept. And I had, I went from having no clue how it was done to just asking chat GPT to walk me through it, and it helped me build my first one. Now, while it took a couple of hours to build a proof of concept, it wasn't integrated into WhatsApp yet that took that took more time than anything else. Basically, you do have to have a business account, like Joel mentioned at the beginning of the episode, you do have to have a business account to be able to create a whatsapp bot. So once you have that business account, you can create a whatsapp number. With that WhatsApp number, if you can prove your identity and prove that you are a legal entity that helps them prevent fraud, that helps them prevent abuse from happening on their system so they. Limit your ability to create the WhatsApp bot to you having a proven identity, so you're not doing this for scamming purposes. So if you can get through those hurdles, create your own business entity that you're going to prove you're actually somebody who has a real business use case, then then you're able to actually get started.
Joel Matthews 30:25
Yeah, and, and, I think going back to your point clappy about making a system that is not only going to be working for WhatsApp, but just generally outside of it, too. I think if I'm sure there are more people now looking into generally making bots or conversational ai, llms, chat, gpts, however you call that, call it, there is a way that we can just fine tune those systems to work well for our use cases, and think of WhatsApp as an integration of your system to suddenly distribute it to a larger audience. So it's, it's kind of a very, a big gain for a little effort sort of thing, where it's not taking away from what you would otherwise probably be already building, and it's allowing you to suddenly distribute it in a much to a much larger user base. Yeah, and there may be some criticism around if I move it to WhatsApp, what? What happens to my data? Is data, all of it is in WhatsApp, and I don't own it anymore. And that is, there is some truth to that, where not all your conversations are probably going to be in your in your data store. However, one of the ways you can design your bot is if you give it goals, like in our example of community review earlier, earlier that we gave, if the goal is to finally pull out, what is the user finally reviewing this piece of content as then the conversation between that for for for the between the bot and the human to get that get to that point is rather irrelevant. And so the the AI can then just get the final answer that actually matters to you, and then save it in your data store in the back end, so it can still save stuff outside of WhatsApp, if you design it like that, and it can call out to your own systems and pull things together. So it is, there is a bit of a trade off, but I think we can use it in a productive way still.
Klappy 32:57
Yeah, I think you made two really good points. The first point of this makes it easier for non technical people to get started in creating these. There's there's examples already of the OBT helper. A lot of people have heard about it. That's actually become quite popular for a non technical tool. It's not quite on WhatsApp yet, but the goal is to get there. But it started off as just gpts, which are open AI chat GPT, having custom system prompts to create your own little bots, right? But it only runs inside of chat GPT, but using the same system prompts, you can port those over to WhatsApp or WhatsApp bot and do the exact same thing over open AI's API to ensure you get the exact same functionality, but over WhatsApp instead of in chat, GPT interface. So first off that that allows non technical people. I always like to praise Marcia Suzuki. She's not a technical person. She does not know how to program she's a linguist, and she builds tools, AI tools, just by creating gpts. And then I've worked with her over the past year, like porting her gpts over to other systems, and one of her goals is to get that useful on WhatsApp. And so that should be a pretty straightforward migration use case, but to me, that's exciting. That means non technical people can go fine tune their AI bot. Now, without involving any technical people, Marcia did it on her own. She's just like, hey, I want to try this. Didn't ask anybody. She just did it. And awesome. Yeah. It's very exciting to see the progress she's made, and she's continuing to innovate and create new ones like i. Not only does she have like story helpers that will help tell stories in a language about a certain biblical principle or biblical concept, so instead of giving a definition, a dictionary definition, to a term for oral communities, they don't have a dictionary. They're not accustomed to learning about a term by looking it up in a book and just reading a definition. They need to hear stories about it. So she created one of her OBT helpers is a storyteller. Tell me a story about grace so I can understand the biblical concept of grace, and in that target language, it'll tell a story in a way that would be meaningful to that culture. So that's an example, just one example. Another is there's Bible translation projects that she wanted to check up on and see what was the health of their project. And so she created an AI assistant just to interview the teams, and based on a short interview, it would give a an overall report of the health of that project to let them know, are they on track? Are they behind? What areas are they the weak on, or what areas did they need extra support from, from the organizations? So there's a diverse type of tools that she's been able to create with no technical expertise. And that the second point you brought up, sorry, I probably went a little long winded on that first one, but the second point you brought up was not owning the data per se on the conversation. And so by default, there's an assumed privacy for messaging. So I don't know if this gets into ethics or if it gets into legal rights or both, but when you're dealing with conversations with AI, people are very sensitive, so you need to think through how much of that data you actually store and how you use that data. So you need to be clear and upfront with your users, how you use that data, and do you store it? Is there a permanence to it, and will people be reading that? So those are the types of things that when you're building a whatsapp bot, you're not just thinking through, how do we use this bot to help the user, but how do we store and use the data, the conversations that they're having with it? Because we could use that in so many different ways, whether to make the future versions better for humans to review it. And in fact, one of the major use cases they're talking about with the BT servant is maybe this could actually be a hybrid help desk support system. So what if you had people on the other side of the world that were ready to help answer the questions that the AI wasn't able to answer yet? Right? What if there were some uncertainties in the AI response for whatever the person was asking of the the bot, and then a human can review the questions it had a low confidence on, and actually help build a better knowledge base for future questions to be answered
Joel Matthews 38:11
from. That's a great idea. Wow,
Isabella Scarinzi 38:14
wow. Yeah, these, these are great discussions. So we've talked about WhatsApp, a very popular chat app worldwide and accessible and even harder to reach places and in Bible translation, WhatsApp AI bots could support tools that are already out there bringing people and AI together in a more natural way. So Jill and claffey. Do you guys have any last words to our listeners?
Joel Matthews 38:39
I would just encourage folks to start taking a look at all this, because it's low effort and potentially high yield. So it's better to try it out than to even think too much about it.
Klappy 38:59
Yeah, I think my last thought would be, don't be afraid to check it out. Like, I think the biggest innovations in this area are coming from outside of the technical experts in our field, right, like the non technical people are having the ideas of how it could help and actually that it's probably the best place for things to start. If you're actually a linguist like Marcia, or you're somebody on the field working with with teams and Bible translation, you're actually working on the translation team, you might be one of the best positioned people to start playing around with potential use cases. And I think that's what technical experts like Joel and the people we work with, we would love to get more of the the direct input from the field, from from those types of people.
Isabella Scarinzi 39:57
That's great. Yeah. To our listeners. Joe. So did recently write a blog post that went into a little bit of more detail. We're going to add the link to our show notes if you want to read more about WhatsApp and Bible translation and AI bots as well. Thank you for listening, and we will see you next time. Thank you. Thank you.
Theophany Media Media 40:26
The Bible translation innovation podcast is brought to you by the Eaton Innovation Lab. This episode is edited and produced by Jake dobrins With Theophany media. Your hosts were Joel Matthew and Christopher Clapp, with facilitation by Isabella scarenzi, please subscribe on your favorite podcast platform, and we'll be with you again next month.