Talking to Arabic audiences
Language is the cornerstone of communication and essential not only to communicate with but also in understanding data of audience sets.
In this episode, Paul discusses how there is a surprising lack of data analysis in the native languages and especially so in Arabic. He describes how he created Sila to solve this problem in his region, and why it is so important for brands to get it right.
Dr. James Piecowye: Welcome to know your audience, a marketing mini cast that explores how knowing an audience can unlock greater insight. In this episode, we talk more specifically about gleaning authentic insight from Arabic-speaking audiences. Paul walks us through the challenges and his solutions with Sila.
So let’s drill down a little bit here, Paul. And let’s take everything that we’ve been talking about now; audience first, how we’ve gotten there, scaling the benefit to our data interpretation, how this allows us to create better briefs for our creatives.
Now, let’s take this, and let’s be a little bit more specific and talk about what you’ve, in fact, alluded to in our last episode. Let’s talk about our region, let’s talk about the UAE, let’s talk about Saudi Arabia. And let’s talk about understanding these Arabic-speaking audiences, and how the audience first, in the context of this region, is ripe with insight.
Paul Kelly: I think the best way to talk about this is probably applying our own experience, at D/A with this region, and the Arabic speaking world, which is much bigger than the countries that we’ve talked about across North Africa, and also in other parts of the world as well, but is trying to understand, I guess, the audience a bit better here. And now even I’m not a native Arabic speaker, anything like that, but my team is and all that sort of thing. But we were very surprised when we were trying to do some research about how bad a lot of different tools, not just social listening other sorts of tools as well, we’re at understanding and interacting with the Arabic language. And it seems like a very underserved market because this one or this technology was basically taking Arabic text, translating it through a translation service, trying to understand something like sentiment, and then translating it back. And then getting the sentiment from the translation.
A bit like Spanish, Arabic has a lot of nuance to it in language, and you don’t necessarily; (something negative could actually mean something positive). yeah, English is also, every language is the same.
But if it’s not created for that audience, if that tool hasn’t been graded, with that audience in mind, particularly with Arabic being a right to left language, rather than left to right, then you miss a lot of stuff. And what happened, what happened was we were just getting bad or no information. Everything was neutral, for example, on sentiment.
So we knew that there must be a much better way of understanding this audience that seemed to be very underserved. And so we built our tool; Silal, which means ‘connections’ in Arabic. To really understand the audience and bring this audience’s first philosophy to life in a much deeper and richer way. And that’s and that’s using, I guess, machine learning and artificial intelligence and understanding natural language processing but without any translation. So it just happened in Arabic. Why is that important? Well, if you’re using tools to make decisions like we were, that wasn’t necessarily doing things in a native way, then you’re getting the wrong information. And so we can’t really apply an audience first philosophy, to information that’s being translated into another language, because we’re not thinking audience First, the audience wouldn’t speak English, to begin with.
James: So, let me get this straight. Sila is working natively in Arabic. And it’s working in a manner that not only is interpreting what’s being said in Arabic, but it’s interpreting what’s being said in Arabic, according to the dialects of the input.
Paul: Yes, yeah, we created dialect models and all this sort of thing to understand, you know, proper sentiment, predictive models, like all that sort of stuff through natural language processing.
And, also image processing things like regional, huge brands in our region, are not huge brands in a global context. I mean, their sales are probably getting them into bigger lists. But there are brands that are very unique to this region, or otherwise, brands whose name is in Arabic, particularly for instance, in the UAE, the United Arab Emirates, particularly Dubai – 99% of the stuff is in English. So it can sometimes tend to help you have our entire region, but most packaging is in Arabic, and English, but people if they’re shooting photos, or they’re sharing moments or things, it’s always the Arabic that you see, you don’t necessarily see the original English brand, and every brand has an Arabic equivalent. And so, you know, image recognition doesn’t pick up this stuff from Western sources.
James: So you’re losing the insight.
Paul: Yeah, and so and the sentiment is another extreme example, where, as I said, you translate things and it’s like, what the hell does that mean? You know, because it’s a bad translation, and just everything is neutral. So then you manually have to go and train the model. But if we just built the model like this, to begin with, to understand the language input, then you get the result and if you’re, you know, and then the mistake obviously, if you’re in the, if you’re in an English native speaking country, you know, the US or UK building these tools, you don’t necessarily understand that there are huge differences between dialect here, although there’s the unification of traditional Arabic, people don’t talk that way. You know, what might be said in Jeddah would be completely different to Najd, you know, the central region of Saudi Arabia and, or Kuwait, where it’s the Gulf dialect. So understanding all that stuff becomes really important, and if everything else in the market doesn’t necessarily do that, then we found it very difficult to apply this principle.
So in 2018/2019, we started building cellar 18. And then in the last year, it sort of came into the world and has been a great asset for us to better understand and better apply our audience first principle, which we’ve had inside our company for a very long time. But I think it’s been, it’s been hard for us to articulate it because we never had a way of understanding our audience. And this is where I keep going back, like, the audience first means being that audience first. So if that audience is Arabic speaking, then don’t for a minute, try as a nonnative speaker of that language to understand it, because it doesn’t work. Nothing works that way. Do you know what I mean? And then this is the biggest problem that I think measures, like not just measurement from perspective, but also audience first and applying these principles. I think a lot of marketers are sometimes skeptical in our region because they get these results that don’t make sense.
James: So, Paul, we’ve had conversations where you’ve talked about the fact that the results have been gleaned from sentiment, don’t necessarily jive with the ideas that people have. And then those people have gone and done focus group research and found out that, indeed, the sentiment research and the focus group research line right up, but what we would have thought was going on, because of the way we do our research in marketing and in business, doesn’t actually connect with what people really are saying, because they tend to lie.
Paul: Yeah. consumers will always lie. There is not a stronger truth that’s ever been said. And I think, yeah, exactly what you said. So I think there are two layers to sort of quickly unpack there. And the first layer is that with research, there are these biases that we’ve talked about. But also you have sometimes the issues like that I’ve talked about where, the thing, doing the research, the tool that’s helping you don’t necessarily understand the language. And that can really skew results. But if you take that out, and you look at the second layer, which is using a tool like Sila that does understand dialect and understand things, you’re able to better uncover what’s going on.
James: Trends and actual usage of the product.
Paul: Yeah, yeah, an example that we sort of found was that a particular product, I’d love to give more details, but I can’t, but a particular product being used in a certain way, that product is an everyday cooking product. And it wasn’t necessarily the way that we get the best results it was, in fact, there’s a substitute that people use, which I could, which I could talk about there, which is rice, this region is rice, everything’s rice, we always we all love the rice dishes of various Mandy and, and biryani and things like that. But this other product, which was adjacent wasn’t necessarily being or was being, cooked the same way as rice, but it shouldn’t be.
And so what was happening was that people were getting a really underperforming product, not very nice at all. And when you add in flavor enhancements and things like that, things like sauces, things like toppings, it doesn’t help the taste, it’s awful. And you get people disappointed so that they then use that particular product in the cheapest possible way. Because everyone, you know, I go out and have it and it’s really nice, but cook it at home and it’s not quite right. And, you know, that’s why you go to restaurants, I guess so, but I’m not going to spend extra money on making this nice, I’m just going to choose the basic absolute baseline tastes, you know, whether it’s a source or whether it’s a topping or the core ingredient itself, that wasn’t necessarily coming through the category data. So there was a big difference within the category about spend and how that was all working. But that behavior element was missed. And then when the research was out there, you know, of course, people never actually said that they weren’t sure how to cook it.
You’re not going to tell a researcher that, in your house, that you don’t know how to cook something, right? And this is where that social desirability bias comes in. You want that person to think that you’re, you know what you’re doing, you’re disappointed. No, just choose this because it’s price-based, you know, it’s performance-based. And then, you know, there was a bit of skepticism, I think among the client about what actually, you know, is there? ……. and then a focus group was conducted.
And then that really sort of drove home that was actually what was happening, you know that the research had found that insight and that insight would never be found, had it not been being able to understand dialect and understand the language you just wouldn’t have been found at all.
And this is one example of many that come up. And the sentiment is a really important one as well. Because when you understand the sentiment, you really understand what someone feels on a topic and it can be hit and miss. Granted, even tools that are in English for English can sometimes miss but it does give you that baseline of measurement that can understand feelings towards a certain topic at a really big scale really quickly. And unless you understand that in a dialect that someone talks about natively, then it makes it (really makes it) much more difficult to understand. You know, what’s happening at a real true scale in that audience.
James: The takeaway from this episode is that audience first means putting the audience first and the first principle of this is working and collecting the data in the language and dialects of the audience being looked at.
You can get in touch with me across the socials @thejamescast or [email protected]
Thanks for listening