https://colourquraan.netlify.app/

TLDR: This is an experimental Qur’aan reading app that visually encodes grammatical meaning within a reading experience. It also allows for a bilingual English-Arabic reading.

You can read about the basics and Version 1 below.

Reading the Qur’aan Through Structure: Introducing Colour Qur’aan

Most digital Qur’aan apps focus on presentation: beautiful scripts, recitation, translations. Colour Qur’aan takes a different approach. It focuses on how meaning is constructed through grammar, structure, and language, making those layers visible to you while you read.

Rather than separating reading from study, Colour Qur’aan blends the two, offering both an Arabic colour-coded (not tajweed but grammar!) reading view and a linguistically aware English view. Each mode supports a different stage of understanding, while pointing toward the same goal: deeper engagement with the Qur’aan’s language.

These updates are to test and promote two principles when reading and learning Quranic Arabic:

1. That breaking sentences into smaller semantic units supports understanding, making it easier to break a sentence apart into meaningful phrases that can be interpreted separately, step by step.

2. That gradually making reading more difficult but still feasible supports learning. That easiest → hardest can be proxied by only showing the deepest syntactic relationship in arabic → showing the full sentence, and we can create a spectrum of difficulty by selectively showing different syntactic relationships at different ‘depths’ of a sentence.

Syntactic Relations (i.e. Dependencies): The traditional Arabic grammar of iʿrāb (إعراب) assigns a syntactic role to each word in a sentence (i.e. what’s a subject (fa’il), an object (mafool bihi), etc.).

For more details, see Corpus Quran’s Descriptions.

V2 Updates

1. Adding Dependencies

Thanks to the work in “A complete, multi-layered quranic treebank dataset with hybrid syntactic annotations for classical arabic processing”, we can encode syntactic relationships into the reading experience.

Dependencies, however are complicated, because language is complicated. Clauses nested within clauses. So we introduce this concept of ‘Depth’: how far you want to go deep within the sentence, or at what depth level you want to view Arabic in a bilingual reading experience.

In a bilingual experience, it is easier to interpret words and grammar at the lowest depths (because you have a lot more english context from the rest of the sentence). As you go to higher depths, more and more grammar relationships in language become Arabic and more interpretive skill is needed.

In Arabic mode, at default depth (= zero), we see sentences bracketed. However, as we go into more ‘negative depth’, we split the sentence into deeper and deeper nested clauses, visualizing the relationship of subtrees in the sentence.

2. Learning Presets: Designing a Bilingual Reading Curriculum

Given the data of each word’s Part-of-speech (POS: nouns, verbs, etc.), and also their syntactic relationship (DEP: possessive, subject, object, etc.), we can create a curriculum between fully English to fully Arabic readings.

This allows educators or students to create a bilingual, code-meshing reading experience that fits one’s level. This tool gives custom ability to select the set of POS and DEP alongside the ‘depth’ of the sentence to view in Arabic (vs. the rest to leave in English). As student learns more concepts, they can convert more and more words to be viewed in Arabic.

As always, I welcome any critical feedback or suggestions on how we can use tooling to support various pedagogical approaches.

Say you’ve done some Arabic intensive and know some basic grammar and vocab. What’s left is a lot of practice. But you’re also already in contact with Arabic text. You might read Qur’an with a translation nearby. You might follow a lecture in Arabic and catch 1/3 of it.

You have an Arabic reading list, but you’re not quite there yet to understand it alone. This makes engaging with Arabic text daunting, and a pre-requisite of spending a year practicing intermediate-advanced text, more drills, complex grammar is.. not a preferable option.

For a lot of people in that spot, the usual options do not fit well. Graded readers simplify the content. Immersion asks you to parse everything cold. Drill apps keep you in isolated sentences. All of these can work, but many adults drop off because they are inconvenient (too much lookup, too much friction) or boring (too far from the texts they actually want to read).

LangBraid starts from a different assumption: you already have a mature first language, and you can use it strategically while you read Arabic, without leaving the passage you care about. As children, we pick up languages without any technical understanding of grammar. Presumably, enough practice with the ‘right’ material (where right = not too far away from what you know and not too easy either) allows someone to pick up even complex language patterns and structures in the background.

What LangBraid does

LangBraid braids Arabic and English phrase by phrase in one reading view. You read connected prose, not flashcards. Tap a phrase to flip between languages. Use a slider to show more Arabic and less English as you get comfortable.

In a way, this is the next iteration of an earlier experiment, Colour Qur’aan, that tried to make grammatical structure visible while reading scripture. LangBraid generalises that idea to any Arabic passage you paste in, with adjustable reading levels and interactive phrase lookup.

Published translations have done something like this for a long time. The Orthodox Jewish Bible weaves Hebrew and English on the same line. LangBraid applies that braid logic to Arabic and English, and you choose how much of each language stays visible.

Learning in-context

We can call this code-meshing / translingual: two language systems visible in the same stretch of text, woven together rather than kept in separate columns or separate mental modes.

The hypothesis is straightforward. You already know how arguments work in English. You know how sentences hang together, how reference works, how an idea develops across a paragraph. What you need is a way to map Arabic structures onto that existing competence while you stay inside a text worth reading.

That mapping happens in context. You notice how a clause opens in Arabic compared to how you would say it in English. You see verb agreement, definiteness, sequencing. You do this without stopping the read every line to open a dictionary. Research on translanguaging in reading suggests this kind of cross-lingual support can help comprehension and motivation rather than weaken it.

Second-language research also talks about comprehensible input: language that is mostly understandable but stretches you slightly.

Learning through a feasible challenge

Pure immersion overwhelms a lot of readers in this middle band. Too much unknown vocabulary, syntax you cannot parse yet, no way to recover the thread of an argument. Graded readers solve that by simplifying the content. LangBraid solves it by keeping the original text and adjusting the support.

At lower levels, harder phrases stay in English with Arabic available on tap. At higher levels, the same phrases appear in Arabic with English available on tap. The meaning stays constant. The linguistic load increases at a pace you choose. You challenge yourself inside a feasible envelope, on material you actually care about.

Give us Feedback

LangBraid automates the “braid”: a model segments Arabic into short phrases, attaches English glosses, and scores each phrase for complexity (simpler algorithm for now) so the level system can reveal easier phrases in English first.

This is a prototype. It works well enough to test the idea. It is not a polished product, and the braiding will not be perfect. I am less interested in whether the tooling is finished and more interested in whether the reading experience itself feels useful for people who are already trying to engage with Arabic text but need a better on-ramp than drills or raw immersion.

What I am looking for

If you are in that middle band level of Arabic knowledge:

• Does a translingual (braided) reading feel natural when you are working through a real passage, or too awkward?

• When you see an original Arabic phrase within an english translation that you understand, do you appreciate it and prefer seeing it in Arabic?

  • What if you don’t understand it and you have to see its translation, does that add too much resistance to your reading or is it bearable? How would you like to encounter Arabic words/phrases that you don’t understand?

• Would you keep using this tool alongside your current routine, or does it feel like too much of a novelty? Is there another routine for Arabic-practice that you prefer?

You can try sample passages on the site without an account. If you want to paste your own text and save passages (limit of three per user), you will need to sign in.

Try it @ LangBraid Samples

https://exegetical-edges.up.railway.app/

Example Input: The Heifer (2:130)

وَمَن يَرْغَبُ عَن مِّلَّةِ إِبْرَٰهِـۧمَ إِلَّا مَن سَفِهَ نَفْسَهُۥ ۚ وَلَقَدِ ٱصْطَفَيْنَـٰهُ فِى ٱلدُّنْيَا ۖ وَإِنَّهُۥ فِى ٱلْـَٔاخِرَةِ لَمِنَ ٱلصَّـٰلِحِينَ

Who but a fool would forsake the religion of Abraham? We have chosen him in this world and he will rank among the righteous in the Hereafter.

Intratextual Connection Results

This project utilizes data science and network analysis to examine the practice of tafsir al-Qur’an bi-l-Qur’an (explaining the Quran through the Quran) via a ‘distant reading’ approach. By aggregating over ~100K scholarly citation links across 30+ classical and modern exegetical works, this work creates a statistical model that maps intratextual relationships between verses. This approach allows for a bird’s-eye view of how exegetes agree, diverge, or identify idiosyncratic connections, ultimately providing a tool to discover patterns and potentially novel links that traditional, manual analysis might overlook.

• Data Aggregation: Compiled a massive dataset of over 100K citation links from more than 30 tafsir works.

• Modeling Intratextuality: Developed a machine learning model that generates “embeddings” (numeric fingerprints) for Quranic study units based on morphological data, literal translations, and exegetical commentary to predict and score intratextual relatedness between any 2 verses.

• Search and Visualization: Built an interactive web application featuring ranked semantic search, 2D landscape mapping of verse clusters, and filters to compare statistical similarity against historical citation frequency.

• Structural Analytics: Performed multi-level statistical analysis to quantify consensus, divergence, and “distinctiveness” at the levels of verse-pairs, individual verses, specific scholars, and thematic communities.

• Methodological Framework: Established a scalable, data-driven approach to exegetical research that highlights gaps between historical scholarly tradition and computational similarity metrics.

This project introduces a computational framework that maps the complex landscape of Quranic intratextuality, potentially offering scholars and students alike a new, scalable way to analyze historical exegetical trends and uncover previously unexamined connections between verses.

Table of Contents

1. Motivation: The inspiration behind the project and the goal of moving from manual cataloging to scalable data science.

2. The Data: An overview of the 100K citation links pooled from over 30 major exegetical works.

3. Technical Overview: How the project turns Quranic verses into study units and uses machine learning embeddings to learn patterns of relatedness.

4. Features, Analytics, & Visualization: A breakdown of the web-based tools, including semantic search, the Explore Map, and citation plotting.

5. Structural Analysis: Insights into how scholars align—or diverge—at the level of verse pairs, individual exegetes, and thematic communities.

6. Limitations & Future Extensions: Reflections on the current methodology and ideas for future development, including domain expert verification.

1. Motivation

Dr. Sohaib Saeed Bhutta (SOAS, University of London) did something wonderfully patient: he took one whole chapter, Surah al-An’am (165 verses), and manually catalogued every time a major exegete pointed from one verse to another elsewhere in the text (the practice known as tafsir al-Qur’an bi-l-Qur’an, “explaining the Quran through the Quran”). He did this across roughly eight major works, from the earliest complete commentary on record to modern ones, including scholars from different eras, schools of thought, and even different sects.

So, natural extension from a data science perspective: how can we automate and visualize such a procedure?

Intuitively, we can imagine that intratextual connection could be ‘predicted’ if we had the right signals between any 2 verses:

• morphological data for each word in the verse (Quran Corpus data)

• overall literal meaning(s) of a verse (proxied by translation text)

• expanded commentary from experts about a verse (exegesis text)

Gathering this data and combining it with scholarly citations of intratextual connections, we could technically build a model that could ‘score’ the intratextual connection between any 2 verses. This “score” would be interpreted as:

Dr. Saeed’s finding was that the overlap between these scholars’ chosen connections was surprisingly low — even between works you’d expect to line up closely because one scholar studied under, or was clearly influenced by, another. Two people looking at the exact same verse, both trained in the same tradition, would often point to entirely different “parallel” verses elsewhere in the text.

This project looks at the statistical results if we pool many of the exegesis connections together, and then try to discover the pattern behind their choices.

From there, we can compare citations in a variety of ways, measuring agreement, disagreement, statistical anomalies, connection strengths, gap analysis, etc. and even uncover potentially novel connections not seen in the dataset.

2. The data: exegesis citations, pooled across 30+ works

A “citation” here means something concrete: while writing about verse A, a classical scholar explicitly points to verse B — quoting it, referencing it in brackets, or explaining A by way of B. Do it for 30+ different tafsir works and you get something closer to a pooled, cross-checked map: 6,236 verses of the Quran, and roughly ~100K (unique) citation links between them, each one tagged with exactly which scholars made it.

3. Technical Overview

Before any modeling happens, verses that are almost always discussed together and read near-identically in most commentaries get merged into single “study units”. We turn the 6,236 raw verses into about ~2K units that better match how the text is actually studied.

The core idea: learn from what’s known to find patterns

Instead of treating the citation data we extracted as a fixed list to look up, the project uses it as training signal for a small model. The idea, in plain terms:

• Take pairs of verses that scholars did explicitly connect. Call these “yes” examples.

• Take pairs of verses picked at random (with some care to avoid trivial cases, like verses that are simply next to each other). Call these “no” examples.

• Show the model many of both, alongside three views of each verse — its Arabic root patterns, the language of its surrounding commentary, and its translated meaning — and have it learn to tell the “yes” pairs apart from the “no” pairs.

Do this enough times, over enough examples, and the model isn’t just memorizing “verse X and verse Y were cited together.” It’s picking up on the shape of what tends to make two verses feel connected — shared vocabulary, shared imagery, shared argument — well enough to make a judgment call about a pair of verses that, perhaps, no scholar in the dataset ever explicitly linked.

Ultimately, this model then gives us an ‘embedding’ for each quranic unit. We can then use this embedding to do a bunch of things very easily like finding ‘the top 10 closest units to this unit’, etc.

Quick technical detour - What is an embedding exactly? Since the word comes up a lot in this kind of work: an “embedding” is just a way of turning something — a verse, in this case — into a short list of numbers, chosen so that two things which are alike end up with similar-looking lists, and two things which are unalike end up with very different-looking lists. Once that’s done, “how related are these two verses?” becomes a fairly simple math question: how close are their two lists of numbers?

Once every verse (or merged study unit) across the whole Quran has one of these numeric fingerprints, comparing any two of them at scale gives an estimate of citation-style relatedness for the entire text at once.

The rest of the work goes towards visualizing, tooling, and surfacing these connections and research. This research now sits underneath a small, living search tool for exploring those connections. The tool has its own guided tour on the web app through which you can learn how to use it.

4. Features, Analytics, & Visualization

For the purposes of this section, some vocabulary to understand:

1. a unit = a set of consecutive verses

2. a unit-‘pair’ = represents 2 units being compared with one another for their intratextual connection

3. a citation = this is a unit-pair that has been specifically cited in tafsir literature by a scholar

4.1 Search: type in a verse, get ranked matches

Type in one or more verse references (a single verse, a range, or a scattered list) and the app returns the verses it considers most intratextually related, ranked highest to lowest. If you give it several verses at once, it ranks by what’s related to all of them on average — useful for exploring a whole passage or theme rather than a single line.

Reading View of Search Results

So you can see intratextual connections within context.

4.2 Explore map: the whole Quran as a landscape of clusters

A separate “Explore map” page projects every study unit’s fingerprint (embedding) down into a single 2D map you can pan and zoom, colored by which chapter it belongs to and sized by how often it’s cited in tafsir. Zooming around this map is a good way to build intuition for the shape of the whole embedding space at once (which chapters cluster together, and which verses stand out as unusually well-connected hubs).

4.3 Structural Analysis: Bird’s eye view of agreement and divergence

Complete Structural Analysis:

https://exegetical-edges.up.railway.app/structural

This is really the summary statistics of this data analysis. In this section, we consider all the pairs of citations across exegetes and calculate metrics like Jaccard similarities, Shannon’s entropy, and information gain. These help us determine how much each citation pair (verse X → Y) is in agreement across the pool of exegetes. We also then compare ground truth citation networks of the exegetes to the statistical model we’ve trained to find distinctiveness and cohesion of exegetes to the ‘statistically dominant patterns.

We can calculate things at 4 levels:

• Per unit-pair (X → Y): an agreement score measured by “of the exegetes who show any engagement with either endpoint (X or Y), what fraction actually cite this specific connection (X→Y)?

  • High = broad consensus, low = only one or two idiosyncratic voices.

• Per unit (X): a divergence score — Shannon entropy of the exegete-weighted distribution over its cited targets.

  • Low = exegetes converge on the same target (a consensus anchor verse), high = every exegete points somewhere different (a contested/ambiguous unit).

• Per exegete: Complementary distinctiveness scores — how different is this exegete’s set of drawn connections (not just which units they touch) from every other exegete’s or the model’s?

• Communities (Set of X units): Community detection directly on the unit-to-unit citation graph (weighted by the per-edge agreement score), with simple per-community size/cohesion stats.

Basically, at 4 levels, from a unique pair (edge) to a community of pairs, we get statistics regarding agreement and divergence between entities. This is one structural method to compare and contrast citations that gives us insight from a birds-eye-view instead of the detail and reasoning of particular connections.

In other words, once we have a network of citations and a model that can score every pair of verses for its ‘intratextuality’, we can do ‘math’ to calculate and answer subjective questions like “which 2 exegetes are similar in how they make connections?”, “which verses have a high rate of divergence in regards to how they are used to explain other verses?”, or “what are the set of verses that tend to be connected to each other across multiple scholars?”

The next section explores some of these questions.

5. Structural Observations and Analysis

Dr. Saeed’s finding, that different exegetes pick different connections, was based on close reading of one chapter. We structurally ask similar questions in this project at full scale: across all the tafsir works and every citation in the dataset, just how much do scholars agree or disagree, verse by verse and scholar by scholar — and does the model’s own sense of similarity line up with that pattern?

After some outlier removal, data cleanup, we do this over 35 medium-to-well represented exegetes and about 80,000 distinct cited connections to work with.

How much do scholars agree on any one connection?

For any specific connection between two verses, this asks: of the scholars who showed any engagement with either verse at all, what fraction actually drew this exact link? Nine independent scholars converging on the same connection out of a pool of ten who could have noticed it reads as near-consensus. One lone citation, buried in an otherwise-silent pool of many engaged scholars, reads as a single idiosyncratic voice — even if it’s the only data point available, the surrounding silence is itself informative.

Which verses are “settled,” and which are contested?

Flip the question around and ask it about a single verse rather than a single connection: when scholars do cite something else in relation to this verse, do they mostly converge on the same target, or does everyone point somewhere different? Some verses behave like a magnet — nearly every citing scholar lands on the same companion verse, a kind of consensus anchor. Others behave more like a crossroads, where equally serious scholars head off in many different, unrelated directions — a sign that the verse’s place in the wider text is genuinely open to interpretation rather than settled.

Do individual scholars have a citation “style” of their own?

Zooming out further, from single verses to whole careers: does a given exegete’s overall pattern of connections look like anyone else’s, or is it distinctive? This is checked two complementary ways — “how unusual is this scholar compared to a typical peer” and the stricter “is there even a single other scholar anywhere in the dataset who cites things similarly.” A scholar can look highly distinctive by the first measure yet still turn out to have one close counterpart by the second — the two questions aren’t quite the same, and both are worth asking.

Does the model’s sense of similarity match the tradition, or see something different?

For each scholar’s actual citations, does the model’s own learned similarity consider those connections obvious, or surprising? And separately, if you asked the model to guess a scholar’s citations from scratch, would its own best guesses actually include what that scholar really wrote?

Do citations cluster into neighborhoods?

Finally, treating every cited connection as a link in one giant web, a clustering pass checks whether verses naturally sort into tight-knit neighborhoods — groups of verses that reference each other far more than they reference anything else.

6. Limitations and Extensions

This research is not meant to be exhaustive or concretely answer one question. The purpose is to motivate a distant-reading data approach to certain exegetical research, particularly what can be framed through network science. The limitation of this current work is that it lacks domain expertise and verification in regards to its conclusion and research direction.

There are various other visualization views and extensions that are possible with this base data, depending on the research question and application. This work selected some preliminary and intuitive approaches to contrast exegesis networks across a few dimensions. With further domain expertise, this work can be extended to focus and derive more in depth relational insights and verify them against existing established conclusions.

Some ideas for areas of extension

1. Increased model sophistication (technical): feature engineering, model architecture, data collection

2. Data verification: domain expertise to verify data extraction and networks

3. Insight Extraction: domain expertise to spot outliers, gaps, or heavily suggested conclusions within the structure and explore further to see if any valuable research questions or answers can be derived. Furthermore, connecting this structural network data to domain knowledge of time periods, schools, political leanings, etc. can uncover more meaningful structural elements. Furthermore, there is potential to connect this work with cross-domain scripture data (law, history, etc.) as well as extend it beyond the closed corpus of Quran (ex. include biblical networks!)

4. Integrate User Data: We currently compare scholarly patterns with statistical patterns. But to really make research like this operational, there is potential to connect it with mass-user patterns to see how intratextual linking amongst lay encounters with the quran match up to scholars and statistics.

5. AI comparison: We have looked at ground truth data (tafsir) and compared it to a machine-learning modeling approach. The natural next step is to test it against the intratextual connections coming from various LLM models: are AI models making the intratextual comparisons that you find in tradition or not?

6. and more…

7. Closing Remarks

I plan to make this data and model available for further exploration. I would once again caution against considering patterns within here conclusive or exhaustive as there was a lack of domain expertise available to verify data extraction. God Willing, I hope to connect this work with more domain experts to add novel research value in Quran Exegesis studies.

References:

1. Dr. Sohaib Saeed Bhutta: Intraquranic Hermeneutics Theories and Methods in Tafsir of the Quran Through the Quran

2. Corpus Quran Morph Data

3. Quranic Universal Library (QUL): Translation, Exegesis Data

Share

Subscribe now

See Demo: https://ashurqbq.netlify.app/

Note: Visualization only available on Desktop! Give it a little bit of time to load, and then click ‘?’ on the top right to understand how to use it.

Qur’an bi’l Qur’an

There’s a long-standing Exegesis principle that the Qur’an explains itself. Verses clarify other verses. Language, themes, and concepts recur across different surahs, sometimes closely and sometimes at a distance. Classical scholars have always paid attention to these internal connections — often explicitly, sometimes implicitly — when interpreting the text.

Often times, this is mixed in with a whole lot else and it can be particularly daunting for an individual to engage with this particular kind of study of revelation.

Personally, as I participate in Revelation readings or study groups, there is always that desire when encountering a verse to see all the other intertextual connections and how they support a particular interpretation. This project came out of that itch. It’s an attempt to make that way of reading more accessible, using data science and generative AI as supporting tools.

The idea is simple: instead of encountering Qur’an bi’l Qur’an only through extended study or commentary, you can explore those internal relationships directly and get a birds-eye-view of which connections a Tafsir makes. Verses are linked to other verses through particular methodology to aid their interpretations.

Examples

Selected Verse (1:1)

Connected Verse (2:128)

Outgoing Reference (1:1 Exegesis references 2:128)

Structuring Data

What’s interesting here is that we have many instances (most people think of Isnads in Hadith, but there’s more!) where editors or writers have provided a form of structured data or at the very least, data that can be easily structured.

When you treat the Qur’an as structured data — without stripping it of meaning — you can surface those relationships at scale. Patterns that traditionally require years of familiarity become easier to notice, especially for students, teachers, and readers who are still building that internal map of the text.

Generative AI plays a supporting role here, allowing for 2 points of increased accessibility. Firstly, there is the ability to break language barriers. Secondly, we can use LLMs to automatically extract and process for relevance. It would be difficult to take Exegesis excerpts and determine through rules or traditional ML algorithms to know when a writer starts and stops regarding a connection they are making between verse. With LLMs we can do the following process:

1. gather all instances of verse Y mentioned in exegesis of verse X

2. get context around mentions of verse Y (ex. 500 characters before and after)

3. Prompt LLM to use this context to get the essential connection being made to aid interpretation of verse X

The goal of the use of Generative AI is not to replace or to summarize knowledge - it’s closer to orientation than interpretation: helping readers navigate and visualize the connections a writer has already made.

Used alongside a body of authoritative text, this kind of tool can:

• make thematic study more intuitive

• support comparative reading across chapters

• help learners develop a sense of Qur’anic coherence

• and lower the barrier to engaging seriously with exegesis commentaries

At its core, this is a digital humanities project rooted in a traditional method. Data and AI aren’t being used to redefine meaning, but to expose existing structure in a way that’s searchable, visual, and interactive.

Further Development

The obvious next step that’s intriguing me is that considering Ibn Ashur, as detailed and comprehensive his exegesis might be, is a human and did he catch or choose to reference all possible connections?

Given this dataset (20K connections between verses), can we train machines to use similar patterns and provide us with more candidates and potential connections?

Dataset

I plan to make this dataset of verse-to-verse connections and explanations available as well for further analysis or works.

Edit: For anyone who is interested in working with this data, take a look at quran-bil-quran-connections.

Sermons are, arguably, one of the most consistent and widespread (“educational”) institutions that Muslims have available to them with regular attendance. People of all demographics, social statuses, and ‘interest levels’ come out to fulfill, excitedly or grudgingly, their ‘sermon ritual’.

Misunderstood or not, there is a keen sense of sermon attendance as a ritualistic practice that only the lowest of the low would abandon willingly. It’s a stamp of being part of the social muslim polity.

Mosques, here in Canada, reaching capacity, may even hold 2 or 3 sermons a Friday. This results in a shortage of speakers, which results in a demand that is then difficult to meet unless compensated by lower standards, unqualified charisma, and/or repetitiveness.

All this to say, speakers often have a free buffet as to what they can do or say, constraints mostly by public opinion or hard organization deal breakers. Even when there have been attempts to govern the Sermon at a national level, or to ‘train speakers specifically for standardization of sermons’, it’s been abandoned quickly - mainly because of an incapacity to.. do anything about it. Speakers can express what their longing hearts desire and there isn’t much to do about it. For organizations with vision or mosques with agendas (I use this term positively!), this is likely their largest educational institution (for better or for worse) and there is minimal, anecdotal optimization that takes place, if at all.

In industry, a situation like this likely calls for a data driven approach to optimize messaging, branding, etc., generally for profit and alignment to values. Here, I don’t think it’s too different.

For a large mass, sermons are probably the only informative, education material they get about God’s Message in their year. We should prioritize their optimization. However, before that can take place, a diagnosis is important - we need to know what we’re dealing with.

Can Data help us Diagnose our Sermons?

What is the ‘Data’?

Attempts at ‘preparing’ sermons have often failed - it doesn’t seem feasible that our speakers provide pre-scripted sermons and we can distribute them, organize them, or plan them over the course of a year. So we really only have Real Time Data to work with - whatever the speaker decides to say.

With an increasing culture of live-streaming and virtual accessibility, more and more organizations present their messages on Youtube or the likes. This is a treasure trove of data.

Before I continue, we can learn from ChurchTech, which has often used technologies to optimize their sermons and bible studies. However, most large scale research shows the trend of those willing to invest in data analysis leans towards politicians seeking to gain leads and strategical insights for campaigns. Large scale studies of sermon topics across ethnicities, denominations, and area tend to extract political insights and gaps. This is also where the money is!

But the technology is available to do so much more. And likely, my hunch is this is already being done behind closed doors for applications like political campaigns. Why not get ahead and use this for our own sakes?

So the Data is the sermon itself - in Video, Audio, and transcribed, in Text.

The use of Generative AI to Enhance Signal

Processing such a variability of topics from a variety of speakers can be difficult and daunting. But our current technology provides us 2 strong advantages that makes this a ripe time to focus in this field:

1. Audio → Text Transcription is at a 90%+ Accuracy, even for multilingual speech

2. LLMs can help us enhance and clean text signals for downstream analysis

Sample Analysis

Audio → Text → Processing → Analysis → Application

Processing Unit

We can take Text and break it down into Summaries, or standalone Claims/highlights.

Sample of a Sermon:

Summary
This sermon emphasizes the inherent honor and dignity bestowed upon all of humanity by Allah, irrespective of any physical or intellectual challenges they may face. It highlights the Prophet Muhammad's (peace be upon him) exemplary treatment of individuals with disabilities, demonstrating compassion, respect, and empowerment. The discourse reframes disability not as a deficiency, but as a potential source of divine reward and mercy. It contrasts Islamic teachings with pre-Islamic ignorant customs that ostracized the disabled, underscoring Islam's call for inclusion, care, and the active participation of all members of society, drawing upon historical examples of leaders who prioritized the well-being of those with challenging circumstances.

Claims
1
All human beings are honored by Allah regardless of their physical or intellectual condition, and no one has the right to dishonor another.
2
Disability should not diminish a person's worth; instead, the challenges faced due to disability are a source of divine reward and blessing from Allah. The Prophet Muhammad's ﷺ example shows that individuals with disabilities deserve honor, dignity, and thoughtful attention to their concerns.
3
If an individual demonstrates contentment and patience with a disability or challenging circumstance, accepting it as a test from Allah, this submission itself is a means of divine mercy and a path to paradise.
4
Islam teaches that disability is not a sign of misfortune or evil influence, but rather individuals with disabilities deserve honor and priority, as exemplified by the Prophet's interaction with Abdullah ibn Umm Maktum.
5
Islam teaches that people with disabilities should be welcomed and included, as they bring blessings, not misfortune. The religion's true message emphasizes compassion and inclusivity over ignorant customs.
6
Muslims are called to actively seek out, care for, and include individuals with disabilities and challenging circumstances within their communities, following the example of early Islamic leaders who established systems of support and care.
7
The sermon claims that it is a community's responsibility to care for, include, and empower individuals with disabilities, as these individuals are capable of significant contributions to society, mirroring the actions of the Prophet Muhammad and historical Islamic scholars.
8
The Prophet Muhammad (peace be upon him) demonstrated that individuals with disabilities are valued members of the community who can contribute positively and should be empowered, as shown by his actions of visiting and praying in the home of a companion with a disability.
9
Those with genuine disabilities or challenging circumstances are excused from certain religious obligations and will receive the same reward as those who perform them, provided they have the sincere intention to do so.
10
Allah judges us not by our physical appearance or possessions, but by the intentions of our hearts and our good deeds, meaning we can all make a positive difference regardless of our circumstances.

Analysis

With fundamental building blocks like summaries and claims for a collection of sermons, we can then leverage the full suite of content analysis and modeling available from a variety of industries. We could do things like:

Topic Modelling

Topic presence over Time

Using Sermons as Study Data

What adds power to all this is once we are able to take this from a mosque or organization level to a grander study of sermons, correlated with things like census data. This would provide the kind sociological data that could help strategize and empower the spread of God’s Message, through the institute of a Sermon that still receives a widely diverse audience.

I’m planning to continue exploring this possibility and God Willing, create a case study of the use of sermon analytics. If any of this resonates with you, feel free to reach out and let’s talk.

Demo: See Interactive Topic Evolution Horizon Chart
(See sections below on how to interpret this plot)

This project explores that question by combining large-scale Arabic text corpora, dynamic topic modeling, and LLM-based signal enhancement to trace how the term “Ummah” has been used across Islamic intellectual history.

Motivation: Why “Ummah”?

“Ummah” is one of the most central — and semantically rich — terms in Islamic thought. Depending on historical context, it can refer to:

• A religious community

• A political body

• A legal or moral collective

• A cultural or emotional identity

• A global or even universal humanity

Traditional historical methods excel at deep reading, but they struggle to provide a birds-eye, diachronic view of how such a concept shifts across centuries and genres. This project asks:

How has the usage and conceptual framing of “Ummah” evolved over time in a large-scale Arabic Islamic corpus?

The Core Challenge

From a data science perspective, this problem has three major difficulties:

1. Scale
   The corpus consists of ~8,000 Arabic texts spanning centuries.

2. Language Constraints
   Arabic NLP resources — especially for historical text — lag behind English. Translation introduces interpretive noise.

3. Subtle Semantic Change
   Standard embeddings and topic models often fail to capture nuanced conceptual shifts, especially when contexts are implicit or rhetorically dense.

High-Level Approach

To address these challenges, we built a multi-stage analytical pipeline that combines classical NLP with generative AI:

Pipeline Overview

1. Corpus ingestion

2. Term extraction (“Ummah” contexts)

3. Disambiguation and filtering

4. LLM-based signal enhancement

5. Dynamic topic modeling

6. Temporal and hierarchical visualization

The Corpus

The foundation of this work is a large, metadata-enriched collection of classical Arabic Islamic texts, aggregated from sources such as:

• Shāmilah

• al-Jāmiʿ al-Kabīr

• Maktabat al-Shīʿa

• Ganjoor

Each text includes author metadata and approximate historical dating (using author death dates as temporal anchors).

Corpus Scale

• ~8,000 books

• Centuries of coverage

• Multiple genres (theology, law, history, ethics)

Step 1–2: Extracting “Ummah” in Context

We extracted 200,000+ sentence-level instances where the word Ummah appears. Because “Ummah” can appear in multiple senses (not all relevant), we applied:

• Arabic BERT-based embeddings

• Pre-clustering and filtering

• Rule-based disambiguation heuristics

This step reduced noise while preserving interpretive breadth.

Step 3: The Key Innovation — LLM Signal Enhancement

This project’s central contribution is what we call LLM Signal Enhancement.

The Problem

Raw historical Arabic sentences often encode meaning implicitly. Topic models struggle when:

• Key concepts are assumed rather than stated

• The rhetorical focus is indirect

• Multiple ideas are compressed into a short sentence

The Solution

Each filtered Arabic context is passed through a specialized LLM-based interpretation layer, which generates a concise English analytical summary answering:

• What does “Ummah” refer to here?

• What discourse is taking place?

• Is the focus legal, political, spiritual, emotional, or social?

Example

Original Arabic Context

واجتماع الأمة على ضلالة أمر غير ممكن، فإن عصمة هذه الأمة ثابتة

Enhanced Signal (Generated)

The author is discussing the idea of the Ummah — the Muslim community — as a collective body that is protected from falling into complete misguidance. The focus is on the reliability of the community’s unified moral and religious judgment.

By using the word Ummah, the author is asserting that the Muslim community, when it is united, cannot collectively agree on a serious error or false belief. This is because the community as a whole is considered to have a form of divinely granted protection from unified misguidance.

Rather than replacing the original text, the LLM amplifies latent semantic signals, making downstream modeling more robust and interpretable.

Step 4: Embeddings and Topic Modeling

Each enhanced signal is encoded using sentence-level embeddings, placing semantically similar usages of Ummah near one another in vector space.

We then apply topic modeling to discover recurring themes

Click here to see Interactive Map

Step 5: Visualizing Semantic Change Over Time

The final output is a temporal topic visualization that encodes:

• Topic prevalence over centuries

• Periods of rapid increase or decline

• Relationships between topics

• Relative dominance within each era

These visualizations allow researchers to see conceptual change rather than infer it manually.

Click here to see Interactive Topic Evolution Horizon Chart

What this Visualization Encodes

• What are the different Topics? (Y-axis Subplots)

• Relative % of Topic in Corpus? (Height of Topic Subplot)

• When does TopicX gain/lose prominence? (Bumps along X axis)

• Which Topics are related/close to each other? (Y Axis Hierarchy Graph to the Right)

• In which time intervals is a Topic in its top 20% or lowest 20% presence (Green/Red intervals along X Axis)

What This Enables

This framework allows scholars to:

• Identify historical inflection points in discourse

• Cross-check computational trends with known events

• Generate new research questions at scale

• Compare legal, political, and emotional framings over time

For example:

• Does rhetoric about Ummah unity spike after periods of fragmentation?

• When does political nationhood begin to overshadow religious consensus?

• How does the “Other” shift from threat to partner in modern discourse?

Limitations and Biases

No computational historical method is neutral. Key limitations include:

• Corpus bias (what survives, what was digitized)

• Disambiguation noise

• LLM prompting and model bias

• Topic modeling parameter sensitivity

• Temporal approximation via author death dates

These are not flaws to be ignored — but dimensions to be made explicit and iteratively improved.

Conclusion

This project demonstrates how generative AI can be used not to replace scholarship, but to enhance it — bridging linguistic complexity, scale, and interpretability in historical analysis.

By combining:

• Classical NLP

• Dynamic topic modeling

• LLM-based semantic amplification

we gain a powerful new lens for studying conceptual change across centuries.

Appendix: Auto-Interpreting Visualization (through Image model)

Below are some (auto-generated) trends visable in Figure 3 (Topic Evolution Map).

1. Does the rise of ethno-nationalism displace religious consensus as the primary unifier?

• Topic 14: Arab Nationhood

• Topic 6: Nationhood & Rights

• Topic 15: Consensus (Ijma)

Visual Inverse: Topic 15 is dominant in the early centuries (150–400 AH) but fades significantly. Conversely, Topics 14 and 6 are nearly invisible for 1000 years, only surging dramatically after 1250 AH (late 19th century).

2. Is the “Prophet’s Family” discourse distinct from the “Caliphate” discourse, signaling sectarian divergence?

• Topic 18: Bakr, Abu, Umar, Ali (Caliphs)

• Topic 31: Familial Bond/Believers

• Topic 4: Prophetic Lineage

Visual Timing: Topic 18 (Sunni-focused Caliphs) is heavily concentrated in the first 300 years. Topics 31 and 4 (Lineage/Family) appear to have different usage patterns, potentially spiking during periods of Shia influence (e.g., Fatimid or Safavid eras).

3. Is the call for “Ummah Unity” a reactive mechanism to internal cultural fractionation?

• Topic 12: Ummah & Muslim Unity

• Topic 0: Cultural Groupings & Divisions

Visual Correlation: Topic 0 (Divisions/Groupings) remains thick and steady in the middle era. A research study could overlay these two to see if spikes in “Unity” rhetoric (Topic 12) immediately follow spikes in “Groupings/Divisions” (Topic 0).

4. Has the “Other” shifted from a “Threat” to a “Partner”?

• Topic 16: Muslims & Threats

• Topic 9: Int. Organizations & Cooperation

• Topic 1: Global Humanity

Visual Shift: Topic 16 (Threats) is present throughout history, often associated with defense. Topic 9 and Topic 1 (Cooperation) are exclusively modern phenomena (right side of plot), suggesting a fundamental change in how the Ummah interacts with the outside world.

5. Did the Mongol Sack of Baghdad (656 AH) decouple Religious Authority from Political Rule?

• Topic 25: Imam & Imamate

• Topic 39: Legislation & Ruler

Visual Crossover: Topic 25 (Religious Leadership) appears distinct from Topic 39 (Political Rule). Investigating if Topic 39 becomes more prominent than Topic 25 after major political collapses (like 656 AH) would trace the secularization of power.

6. Is “Sectarianism” historically recorded as “Theological Minority” or “Deviance”?

• Topic 44: Orthodoxy, Division, Minority

• Topic 23: Muslims Following Islam

Visual Spikes: Topic 44 explicitly deals with “division” and “minority” views. Mapping the spikes of this topic against known historical sectarian conflicts (e.g., the Mihna) could validate if this topic serves as a proxy for sectarian tension.

7. Does the definition of “Belonging” shift from Legal Adherence to Emotional Identity?

• Topic 15: Consensus (Legal)

• Topic 37: Emotional Belonging

Visual Trend: Topic 15 (Legal Consensus) characterizes the early period. Topic 37 (Emotional Belonging) appears more fluidly. A comparison would determine if the bond of the Ummah transformed from a legal contract (Fiqh) to an emotional/identity-based construct over time.

https://colourquraan.netlify.app/

TLDR: This is an experimental Qur’aan reading app that visually encodes grammatical meaning within a reading experience. It also allows for a bilingual English-Arabic reading.

Reading the Qur’aan Through Structure: Introducing Colour Qur’aan

Most digital Qur’aan apps focus on presentation: beautiful scripts, recitation, translations. Colour Qur’aan takes a different approach. It focuses on how meaning is constructed through grammar, structure, and language, making those layers visible to you while you read.

Rather than separating reading from study, Colour Qur’aan blends the two, offering both an Arabic colour-coded (not tajweed but grammar!) reading view and a linguistically aware English view. Each mode supports a different stage of understanding, while pointing toward the same goal: deeper engagement with the Qur’aan’s language.

The Arabic Reading View: Seeing Structure, Not Just Script

In the Arabic reading view, the Qur’an is presented with colour-coded linguistic elements. Words are visually distinguished based on their grammatical or functional role, allowing readers to see patterns as they read.

This view can also display Punctuation (English-translated guided) and Word-by-Word Translations.

Furthermore, display options encode Classical Arabic Grammatical features as visual cues:

• Verb Aspect (Past/Command)

• Verb Form

• Conjugation

• Case Class (▲Nasb / ▽Jarr Or Jazm)

• Verb Voice (↑Passive)

• Derived Noun (⊤Fa’il / ⊥Maf’ool / ∅Masdar)

This helps with:

For Arabic readers, this reveals layers that are often mentally processed but rarely visualized. For learners, it offers a gentle, non-intimidating entry into Qur’anic Arabic without turning the text into a grammar chart.

The English View: Meaning First, Grammar Made Visible

The English View is not a standard translation. It is designed to encode Qur’aanic grammatical information directly into the English reading experience. Translations alongside the Arabic Phrases can be toggled on/off.

Readers can select a part of speech (POS) (e.g. nouns, verbs, etc.) and for that selected POS:

• The original Arabic word is shown

• All other words remain in English

This creates a layered experience where the reader stays grounded in meaning, while gradually encountering the Arabic language in context.

What This Helps With

• Explicitly converting Classical Arabic grammar into English to aid English readers understanding of how sentences are constructed.

• Maintaining readability: Unlike interlinear translations, the verse still reads smoothly and coherently.

• Building Arabic awareness naturally: Readers begin to recognize Arabic words in grammatical context

• Bridging translation and original text: Arabic appears intentionally, not all at once, making the transition feel accessible.

One Goal, Two Complementary Paths

Together, the two views support different readers — and different moments of reading:

Arabic View → for pattern recognition, structure, and linguistic awareness

English View → for meaning-first reading with guided exposure to Arabic

Colour Qur’aan is about making the latent grammatical structure of the Qur’an visible, so that reading becomes more intentional, reflective, and informed — whether you start with Arabic, English, or somewhere in between.

Contributions

1. Data is mainly from https://corpus.quran.com