Picture this: a room buzzing with energy, pizza boxes stacked high, soft drinks at every table, rapid Wi-Fi connectivity, and scholars engrossed in their laptops, training AI models. It might not be the first image that springs to mind when you think of theologians and humanities scholars. Yet, this was the vibrant scene on day 2 of the Network Institute hackathon, “Live-mapping Religious Difference Online”.
On the first day, we grappled with first-order issues of digital humanities scholarship, tackling how to source the data central for our research. Not an easy question to answer in times of X’s (former Twitter) API shutdown and increasing data collection restrictions due to the overuse of large language model training. Today, we delved into the second-order problems: analyzing this data. More precisely, analyzing religious language on social media was the task at hand. But before we dive into the details, let’s take a step back and understand why it’s not just a complex (and, therefore, fun) issue to wrestle with but why it’s crucial to track what’s going on in the realms of digital religion.
Prologue: The Complexity of Religious Language
Religious language is a fascinating realm. For instance, all three Abrahamic religions possess sacred texts strongly advising against using God’s name “in vain”. It is connected with a notion of divinity, possessing a central place in prayer and religious rituals. Yet, throughout the centuries, we have texts commenting how God’s name is actually used very much in vain: Muslim, Jewish, and Christian authors all address swearing during bargaining at the marketplace – an issue which seemingly was of grave importance to believers of the first millennium AD. While today, our marketplace might rather be online, proclamations such as “Jesus, this price is insane!” are equally common and do certainly not impose a divine connotation upon the reader. This begs the question: when is something truly religious?
Defining what is ‘religious’ is a complex challenge. As McNamara aptly put it in 1984, “Try to define religion and you invite an argument.” With Wittgenstein, the difference can be pointed out more specifically. Traditionally, a text’s topic was categorized via the topic its words refer to. Since Wittgenstein famously pointed out that the meaning of a word is defined by its usage, this poses a serious question for us as researchers: How can we be certain that the topics we assign to terms used in a given text are actually the right topics? Assigning a topic always refers to a concept outside the text. Yet, to not dissolve any meaning beyond a single text itself, we still have to assign topics and reflect on relations between different texts from different authors in different settings with different … it gets complicated. This complexity immediately translates into a digital humanities problem: How do we write algorithms that discern religious language in a sea of digital data if it’s not even that easy to define what religious language is?
Hackathon Sessions: A Deep Dive
The hackathon participants approached this challenge in various ways. Based on a set of problems to choose from, three groups were formed. Each decided on an issue to focus on and then started hacking away. For about six hours, concentrated coding and intense discussions about methodological issues or first findings set the atmosphere. At the end of the day, everybody came together and presented the initial findings, which we dive into in the following section:
Session 1: The Palestine-Israel Conflict
Right at the pulse of current events, the first group sought to understand how the Palestine-Israel conflict is represented in terms of political and/or religious terminologies. Applying a combination of qualitative analysis (backed by quantitative statistical checks) and two different inductive data visualizations, an intriguing finding was made: Although the conflict is often portrayed as a religious conflict in Western media, communication on X speaks a different language. There was minimal use of traditionally religious language. Instead, the discourse was dominated by various political aspects. Especially, there was no direct connection between the terms ‘Muslim’ and ‘Jew’, indicating that the conflict wasn’t being commented on as a religious war between Jews and Muslims.
Session 2: The Philosophical Discussion
The second group chose to focus less on current events and more on the ground-laying classification task of how to discern religious and political communication. Initially embarking on a philosophical journey, questioning the very concepts of religion and politics, they pondered whether these concepts should be identified via semantics or if semantics should be used to search for intent. In a practical session utilizing Tweets of the famous saying “Seek and You Shall find”, they employed different models with a named entity recognition approach. While some models, like ChatGPT, didn’t yield satisfactory results, others, like Bard, did a commendable job, albeit with a Christian-centric focus.
Session 3: Classifying the Palestine-Israel Conflict
The third group closed the round of presentations. And in doing so, they wonderfully combined the prior two group’s topics. The third group had set out to also design a Tweet classification pipeline using Tweets from the Palestine-Israel Conflict. Starting with a subset of Tweets, the group created a manually annotated training data set. From there on, their methodological approach involved using GPT-4 and OpenAI’s API to fine-tune a text classification algorithm specifically trained on the conflict. Their results strongly reinforced the findings of the first group while even going beyond them: the discourse was predominantly political, with minimal religious undertones, but was alarmingly violent.
Conclusion: The Intersection of Religion and Technology
The second day of our hackathon was a testament to the intricate interplay between theology, religious studies, and technology. As we delved into the vast world of social media data, it became evident that while technology equips us with powerful analytical tools, truly understanding religious language demands a profound grasp of theology and religious studies.
To all our readers, we leave you with a thought: In an age where tweets and posts shape perceptions, how can we ensure that the essence of religious dialogue is preserved and understood? The answer, perhaps, lies at the intersection of faith and code.
See also the hachakthons announcement. A report of the first day of the hackathons will be published soon.