Nigerian Salafi Discourse Corpus (NSDC): Construction, Methods, and Use

In the study of how people use language, researchers have long relied on two different tools. One tool is the close reading of a few texts, where a scholar examines every word to understand the deeper meaning and social context. The other is the massive counting of words, where computers scan millions of pages to find hidden patterns that the human eye would miss. For decades, these two approaches often worked in separate worlds. However, a growing field of study now combines them, using the sheer volume of data to guide the close reading, and using the close reading to make sense of the numbers. This approach is particularly powerful when studying religion, where beliefs are not just private thoughts but public arguments fought out in the streets, in mosques, and increasingly, on social media. In Nigeria, a country with a vast and diverse Muslim population, the internet has become a primary arena where different groups debate their identity, authority, and place in society. To understand these debates, researchers need a way to capture the sheer scale of the conversation without losing the nuance of the individual voices.

This is exactly what a team of researchers from the University of Ibadan set out to do. They built a massive digital library of Facebook posts to track how Salafism, a specific and often controversial movement within Sunni Islam, is discussed in Nigeria. They did not just collect random posts; they constructed a carefully organized archive called the Nigerian Salafi Discourse Corpus. This collection contains nearly 5,800 posts published over ten years, from 2015 to 2025, by pages and profiles with large followings. The text within these posts amounts to nearly 2.5 million words. The researchers gathered this data using a formal research interface provided by Meta, the owner of Facebook, ensuring they only accessed public content from public figures and organizations. The goal was to create a clear, verifiable record of how Salafis are portrayed, how they portray themselves, and how they argue against other groups, such as Sufis or Shia Muslims, within the Nigerian context.

To build this library, the researchers faced a significant challenge: the raw data from Facebook is messy. It contains broken formatting, duplicate posts, emojis, foreign scripts, and links that can confuse a computer trying to count words. To solve this, they employed a specialized artificial intelligence assistant, which they named Salih. This tool did not just guess at the answers; it acted as a rigorous editor. It followed a strict set of rules to clean the text, removing noise like emojis and URLs while keeping the meaningful words. It checked its own work at every step, ensuring that no posts were accidentally deleted and that the final count of words was accurate. The result was a clean, organized collection of text that could be analyzed by other researchers. This process was designed to be completely transparent, meaning any other scientist could follow the same steps and get the exact same result, a crucial standard for scientific trust.

Once the text was clean, the researchers added three layers of detailed information to every single word. The first layer simply identified what kind of word each one was, such as a noun, a verb, or a proper name. The second layer grouped words into specific topics, such as "religious authority," "conflict," or "worship practices." The third layer, which is the most complex, looked at the purpose of entire sentences. It coded whether a sentence was trying to legitimize a group, attack an opponent, call for action, or report what someone else said. This coding was based on a method called the discourse-historical approach, which helps researchers understand not just what is being said, but why it is being said and what effect it is intended to have. The researchers were careful to note that while the computer did the heavy lifting, the system was not perfect. They found that the computer sometimes over-identified certain types of sentences, like calls to action, and they plan to refine these rules in the future.

The data they uncovered reveals a vibrant and often heated conversation. The researchers found that the volume of posts fluctuated over the decade, with a peak in activity around 2017 and another rise in 2024. A significant portion of the discussion involves naming and defining groups. The data shows that the term "Muslim" is used as a default identity in the vast majority of posts, but specific labels like "Salafi," "Sufi," "Shia," and "Wahhabi" are used to draw boundaries between groups. The analysis suggests that the label "Wahhabi" is often used as a negative term to criticize others, while "Salafi" is frequently used by the group to describe itself. The researchers also observed that the language used to discuss religious authority is intense, with a heavy reliance on quoting scholars and religious texts to support arguments. Interestingly, the data shows that while religious scripture is cited, it is not the only tool used; arguments about law, politics, and social order are just as common, indicating that the debate is deeply tied to the broader Nigerian state and society.

One of the most striking findings is how the length of these posts has changed over time. While the number of posts remained relatively steady in recent years, the amount of text in each post has decreased. This suggests that the nature of the discourse is shifting toward shorter, more compressed messages, perhaps reflecting the fast-paced nature of modern social media. The researchers also noted that the conversation is not just about theology; it is deeply entangled with security and conflict. The data captures how religious groups are linked to or separated from violence, particularly in the context of the insurgency in northern Nigeria. By keeping the term "Boko Haram" as a single unit in their analysis, the researchers ensured that the word "haram" (forbidden) within that name did not accidentally trigger codes related to religious rules, keeping the security discussion distinct from the theological one.

The value of this work lies in its ability to turn a chaotic stream of online arguments into a structured resource that anyone can study. It allows researchers to move beyond anecdotes and look at the actual patterns of how millions of words are used to shape religious identity in Africa. The corpus is now available to other scholars, who can use it to investigate everything from how stereotypes are formed to how political loyalty is expressed in religious language. The researchers emphasize that while the computer did the counting, the interpretation of these patterns requires human insight. They have provided a map of the territory, but the journey of understanding the deeper meanings of these debates is just beginning. This project demonstrates that with careful design and transparent methods, artificial intelligence can help us understand the complex ways in which communities define themselves and their place in the world.