📄 social_science

Computational History and Heritage: New Enquiries Through Synthesis

This paper argues that computational methods, including modeling, simulation, and machine learning, are essential for synthesizing vast multidimensional datasets to transcend traditional historical analysis, thereby enabling scholars to systematically reconstruct past environments, test hypotheses, and explore complex socio-cultural dynamics to unlock a more holistic narrative of human history.

Original authors: Eugene Ch'ng, Simon See, Ka Chun Cheung, Ping Shu Ho

Published 2026-07-20

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Eugene Ch'ng, Simon See, Ka Chun Cheung, Ping Shu Ho

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery that happened thousands of years ago. In the past, detectives had to read every single diary, letter, and stone tablet they could find, one by one, using their own brains to piece together the story. This is how history and heritage studies used to work: careful, slow, and limited by how much a human could read or remember. But then, we got computers. First, we just used computers to make digital copies of these old things, like scanning a book so you can read it on a screen. This was the "Digital" era—it was great for saving things and letting more people see them, but it was mostly just a fancy library.

Now, we are entering a new phase called "Computational History and Heritage." Think of this not as a library, but as a giant, super-smart simulator. Instead of just reading a story, this new approach uses powerful math and artificial intelligence to build models of the past. It can take millions of tiny clues—like pollen from ancient soil, the shape of a broken pot, and the words in a forgotten letter—and smash them together to see patterns that no human eye could ever spot. It's like taking a million scattered puzzle pieces and having a robot instantly snap them together to reveal a picture you didn't even know existed. This matters because it lets us ask "what if" questions about our ancestors, testing ideas about why civilizations rose or fell in ways that were previously impossible.

The Paper: A New Way to Look at the Past

This paper, written by Eugene Ch'ng and his team, is basically a map of this new territory. They wanted to see exactly where the field of history and heritage is going. To do this, they didn't just guess; they went on a massive data hunt. They grabbed the titles of over 22,000 research articles from Google Scholar and used a special kind of AI to group them by what they were actually talking about. They treated these article titles like a giant jigsaw puzzle, looking for clusters of words that showed up together to see which topics were growing fast and which were staying the same.

The Big Discovery: Two Speeds of Growth

The authors found something really interesting: the field is moving at two different speeds.

On one side, you have Archaeology and Heritage. These are the "anchors" of the field. They are sprinting ahead! The paper shows that research in these areas is growing at a rate of 5.2% to 5.6% per year. This is just as fast as, or even faster than, the rest of the "hard" sciences like physics or chemistry. Why? Because these fields are great at using computers to build 3D models, scan ancient ruins, and simulate how people moved around in the past. They are turning physical objects into digital twins that can be tested and played with.

On the other side, there is a "stagnant" area: Historiography (which is basically the study of how we write and understand history). The paper found that while we are getting amazing at reconstructing the stuff of the past (like buildings and tools), we are not getting any better at computationally understanding the stories and ideas behind them. The growth rate for these theoretical topics is flat or even slightly negative. The authors suggest this is because it's much harder to teach a computer to understand human emotions, complex causes, and the messy "why" of history than it is to teach it to recognize a brick wall.

What This Approach Is (and What It Isn't)

The paper is very clear about what "Computational History" is not. It's not just a fancy search engine that finds facts for you. It's not just digging through data to find a hidden number. And it's definitely not just reading a bunch of old books faster.

Instead, the authors define it as synthesis. Imagine you have a box of millions of tiny, colorful tiles (tesserae). By themselves, they are just random bits of color. But if you use a computer to arrange them all at once, a beautiful mosaic picture appears. That picture is the "new knowledge." The computer takes different types of data—like climate records, DNA from ancient bones, and old texts—and mixes them to reveal hidden rules and connections that were invisible before.

The Tools of the Trade

The paper highlights that the magic is happening because of new tools like Deep Learning and 3D reconstruction.

3D Reconstruction: They talk about how we used to use cameras and math to build 3D models of ruins. Now, new AI tools (like something called "Gaussian Splatting") can build these models even faster and with better detail, even on shiny or smooth surfaces that used to confuse computers.
Deep Learning: This is the brainpower behind the scenes. It allows the computer to "learn" from massive amounts of data to spot patterns. For example, it can look at thousands of ancient texts and figure out how language changed over time, or simulate how a whole city might have reacted to a drought.

The Gap We Need to Fill

The most important takeaway from the paper is that we have a gap to fill. We are currently excellent at using computers to reconstruct the evidence of the past (the physical stuff). But we are still struggling to use computers to write the history (the stories and explanations).

The authors suggest that the future isn't just about making better 3D models or finding more data. It's about synthesis. We need to combine the physical data (like ruins) with the abstract data (like ideas and social rules) to create a complete picture. They believe that once we can do this, we won't just be able to answer old questions; we will be able to ask brand new ones that we never even thought of before.

In short, the paper argues that we have successfully built the engine (the digital tools and data), but now we need to build the driver (the new ways of thinking) to take us to the next level of understanding our human story. We are moving from just looking at the past to simulating and understanding it.

Technical Summary: Computational History and Heritage: New Enquiries Through Synthesis

1. Problem Statement

The paper identifies a fundamental limitation in traditional humanities scholarship: the inability of human cognition to process vast, multidimensional datasets to detect high-dimensional patterns spanning millennia. While the "Digital Humanities" (DH) and "Digital Heritage" eras successfully established infrastructure for digitization, access, and representation (e.g., 3D scanning, online archives, "distant reading"), they largely remain focused on retrieval and static analysis.

The authors argue that the field is currently experiencing a schism. On one side, applied disciplines like archaeology and heritage have successfully integrated computational methods to reconstruct physical evidence (sites, artifacts). On the other side, the theoretical core of history—historiography, which deals with narrative construction, causality, and human agency—remains largely untouched by computational reasoning. The paper posits that the field must transition from "digital" (representation and access) to "computational" (algorithmic modeling, simulation, and synthesis) to move beyond simple fact retrieval and address complex socio-cultural dynamics.

2. Methodology

To empirically map the trajectory of this field and validate the shift from digital to computational, the authors conducted a large-scale bibliometric analysis using the following technical pipeline:

Data Collection: Instead of relying on curated indices like Scopus or Web of Science (which the authors argue miss critical interdisciplinary and "grey literature" in computer science and heritage), they utilized the Google Scholar API. They queried keyword pairs (e.g., "computational history," "computational heritage," "computing archaeology") across 100 result pages per query, retrieving approximately 22,449 unique article titles published since 1904.
Semantic Embedding and Clustering:
- Article titles were embedded using the pre-trained Transformer model all-MiniLM-L6-v2 (Sentence-BERT) to generate 384-dimensional dense vectors, capturing semantic similarity beyond surface-level keyword matching.
- K-Means clustering was performed for $k \in \{2, \dots, 100\}$ . The optimal number of clusters ( $k=56$ ) was selected based on the elbow method (inertia stabilization) and silhouette scores.
- TF-IDF and KeyBERT were used to extract representative unigrams and bigrams for human-interpretable cluster labeling.
Dimensionality Reduction: UMAP was used for primary 3D visualization to preserve local and global structure, with PCA serving as a deterministic baseline.
Growth and Influence Metrics:
- Linear Regression slopes were calculated to determine publication velocity.
- Compound Annual Growth Rate (CAGR) was used to normalize growth across topics.
- Mann-Kendall Trend Test assessed the statistical significance of monotonic trends ( $p < .05$ ).
- Zipf's Law analysis was applied to citation distributions to distinguish between established foundations and emerging frontiers.

3. Key Contributions

The paper makes several distinct technical and conceptual contributions:

Definition of Computational History and Heritage: The authors define the field not as simple data mining or fact retrieval, but as the use of computational methods to synthesize vast, diverse datasets (textual, material, environmental, genomic) to model complex systems. This involves discovering complex rules and structures that reveal deep patterns invisible to traditional scholarship.
Empirical Mapping of the Field: Through the analysis of 22,000+ titles, the paper provides the first quantitative evidence of a structural transformation in the discipline, identifying specific "anchor" disciplines and "stagnant" theoretical areas.
Identification of the "Historiographical Gap": The study quantitatively demonstrates that while applied fields (Archaeology, Heritage) are growing at rates matching or exceeding the global computational baseline, the theoretical core (Historiography) is statistically stagnant.
Synthesis as a Core Mechanism: The paper frames "synthesis" as the primary epistemic shift. It argues that computational methods (specifically deep learning and simulation) allow for the combination of dissimilar elements (e.g., climate data, artifacts, texts) into coherent models that generate new questions and reveal hidden relationships, analogous to assembling a mosaic from individual tiles.

4. Results

The quantitative analysis yielded the following specific findings:

Growth Rates:
- The global "Computational/Computing" baseline (Cluster 11) shows a growth rate of 5.3%.
- Archaeology/Archaeological (Cluster 4) aligns with this baseline at 5.2%.
- Heritage/Cultural Heritage (Cluster 27) exceeds the baseline at 5.6%, identifying it as a growth outlier.
- Digital Humanities (Cluster 21) grows at 4.8%, lagging behind the computational baseline.
- Historiography/History (Clusters 28 and 29) are marked as "Stagnant" with negative or negligible growth slopes (-0.01 to -0.05).
Temporal Stratification: Heatmap analysis (1988–2025) reveals a "wall" of high intensity for Archaeology and Heritage starting around 2010 and accelerating post-2015. In contrast, Historiography and Literary studies show diffuse, non-accelerating bands, indicating persistence rather than exponential growth.
Citation Distribution (Zipf's Law): The top-cited keywords are dominated by hard sciences (e.g., "history shelx" referring to crystallography software, "quantum information," "cloud computing"). Humanities keywords (e.g., "Digital History," "Museums") appear in the middle segment, indicating the field is currently borrowing frameworks from other sciences rather than having indigenous, highly-cited computational theories.
Technological Drivers: The growth in Heritage and Archaeology is structurally linked to the rise of 3D technologies (NeRFs, Gaussian Splats) and Machine Learning/Deep Learning (Transformers, LLMs).

5. Significance and Claims

The paper claims that Computational History and Heritage represents a decisive epistemic shift.

From Representation to Simulation: The field is moving beyond the "digital" era of static representation and preservation toward a "computational" era of dynamic modeling, simulation, and algorithmic reasoning.
New Capabilities: By synthesizing multi-modal data (objects, monuments, climate records, genomes), these methods enable scholars to:
- Reconstruct past environments and behaviors.
- Simulate alternative scenarios and test competing hypotheses.
- Explore the interplay between culture, environment, and agency.
- Ask entirely new questions that were previously impossible to formulate.
The Challenge Ahead: The authors conclude that while the field has successfully built its foundations in physical reconstruction (Archaeology/Heritage), the next great challenge is to develop a robust computational historiography. The goal is to move from computing the evidence of the past to computationally synthesizing the narratives and theories that explain it, bridging the gap between quantitative data and humanistic inquiry.

The paper does not propose specific new algorithms or experimental applications but rather provides a meta-analytical framework and a call to action for the field to mature from data collection and visualization toward complex system modeling and theoretical synthesis.