karsdorp.io

What is Fragments of Cultural Change?

Fragments of Cultural Change is the personal research site of Folgert Karsdorp, a researcher who studies how culture spreads, develops, and disappears. He leads the Ethnology, Everyday Culture & Society department at the Meertens Institute (Royal Netherlands Academy of Arts and Sciences) and is Professor of Computational Methods for the Study of Cultural Change at Utrecht University.

The site presents this work in a simple format: a short statement of research interests, a menu linking Home, Fragments, and Publications, and a "Selected publications" list. Its focus is methodological — how to analyze biased historical collections and how to measure cultural diversity and loss — using computational and quantitative methods borrowed from ecology, population genetics, and machine learning.

Who it is for

• Researchers in digital humanities, cultural evolution, and computational social science
• Historians and literary scholars interested in quantitative approaches to cultural survival
• Students looking for examples of how ecology-style models are applied to cultural data

What the listed publications illustrate

A useful next step

If you are deciding whether the site is relevant to you, open the Publications section and read the abstracts of the three selected papers. They are the clearest signal of the site's actual subject: statistical methods for cultural data, not general commentary on culture. If you work with uneven or incomplete historical records, the bias-correction paper is the most directly transferable starting point.

How does Folgert Karsdorp use computational methods to study cultural change?

Folgert Karsdorp studies cultural change by borrowing quantitative tools from the natural sciences and adapting them to historical and cultural data. As he describes his own work on Fragments of Cultural Change, his research asks how culture spreads, develops and disappears, using computational and quantitative methods drawn from ecology, population genetics and machine learning. The distinctive part is not computation for its own sake but the methodological problem of working with biased historical collections and measuring cultural diversity and loss.

H3 Three recurring moves in his work

• Borrowing models from ecology. His 2022 Science paper, "Forgotten Books," applies unseen species models — originally used to estimate how many species remain unobserved in an ecosystem — to the survival of cultural artefacts. The analogy is useful: just as field surveys miss rare species, historical archives miss texts and traditions that were never recorded or did not survive.
• Correcting for collection bias. A 2026 paper in the Journal of the Royal Society Interface, "Correcting collection bias in comparative studies of diversity," addresses a central obstacle for anyone comparing cultures: what survives in archives is not a neutral sample. If bias is not modelled, comparisons between cultures or periods can reflect record-keeping rather than real differences.
• Estimating hidden diversity. "Saved in Translation?" (2026, Evolutionary Human Sciences) estimates latent diversity shared by historic cultures, using medieval literature in French and Dutch as a case. This treats translation and textual transmission as evidence about a larger repertoire that is only partly visible.

H3 What this means in practice

If you work with historical corpora — digitised book catalogues, manuscript inventories, folklore collections — the practical takeaway is to treat your dataset as a sample of something larger, not as the thing itself. A reader might, for example, want to compare the range of genres attested in two medieval vernaculars. A naive count of surviving manuscripts would confound literary production with the accidents of preservation and later collecting. Methods of the kind Karsdorp develops aim to separate those signals.

H3 Where to go next

Start with the Science paper on unseen species models if you want the clearest demonstration of the approach, then the collection-bias paper if your own interest is comparative. His publication list at Fragments of Cultural Change links the papers and DOIs. For adjacent work on cultural evolution and quantitative cultural history, the journals named above and the Meertens Institute and Utrecht University, where he holds his positions, are the natural places to look.

What are unseen species models and how are they applied to cultural survival?

Unseen species models are statistical tools originally developed in ecology to estimate how many species exist in a population when your sample has not captured them all. Applied to culture, they flip the usual question: instead of asking only how many texts, songs or artefacts survived, they ask how many once existed but left no trace in the historical record. Fragments of Cultural Change, the site of researcher Folgert Karsdorp, frames this as a core method for studying how culture spreads, develops and disappears, alongside related work on bias in historical collections Fragments of Cultural Change - Folgert Karsdorp.

H3 How the method works in principle

• You treat each surviving item (a manuscript, a title, an object type) like a captured animal.
• The frequency with which items appear across collections carries information: many items seen only once suggests a long tail of lost material.
• A model extrapolates from that frequency pattern to estimate the unseen portion, giving a rough figure for what is missing.
• The output is an estimate with uncertainty, not a recovered list — it tells you about scale, not content.

H3 Why culture is a good fit — and a hard one Historical survival is uneven. Texts are copied, reprinted, translated, archived or discarded for reasons that have little to do with their cultural importance: religious censorship, language shift, paper quality, library fires, colonial collecting practices. Karsdorp's stated focus on "biased historical collections" reflects this. An unseen species estimate is only as good as the assumption that the surviving sample is representative — and for culture that assumption is often shaky. The method is most useful when paired with explicit reasoning about why certain things vanished.

H3 A concrete use case Suppose you study medieval narrative in French and Dutch. You have lists of surviving works and rough counts of how often each is attested. A frequency-based model can suggest how many comparable works likely existed but are unattested, and whether the loss rate differs between the two traditions. That is the kind of question behind work on latent diversity shared by historic cultures Fragments of Cultural Change - Folgert Karsdorp. The practical payoff: you can compare traditions on a fairer footing, rather than concluding that the better-documented one was simply richer.

H3 Trade-offs to weigh

• Strength: gives a quantitative handle on absence, which raw catalogues cannot.
• Weakness: sensitive to sample size; small corpora yield wide uncertainty.
• Risk: treating the estimate as a fact rather than a model-dependent inference.
• Complement: pair it with source-critical history, not as a replacement.

H3 Where to go next If you want to apply this, start with a well-bounded corpus where you trust the survival record — a single genre, region and period — and check whether your frequency data behaves like the ecological cases the models were built for. For the scholarly framing and publications, see Fragments of Cultural Change - Folgert Karsdorp.

How can collection bias be corrected in comparative studies of cultural diversity?

Collection bias is corrected by treating a historical archive as a sample rather than a complete record, then modelling what is missing before comparing cultures. The central problem is that survival and documentation are uneven: a culture may look less diverse because fewer of its texts, objects, or practices were preserved, not because it actually was less diverse.

Folgert Karsdorp's research at the Meertens Institute and Utrecht University addresses exactly this, including work on correcting collection bias in comparative studies of diversity, estimating latent diversity shared by historic cultures, and applying unseen species models to the survival of culture. The practical lesson from that line of work is that bias correction belongs in the study design, not in a caveat at the end.

What bias correction looks like in practice

• Model the sampling process. Unseen species models, borrowed from ecology, estimate how much of a population was never observed. Applied to culture, they estimate how many works or traditions have been lost, rather than assuming the surviving set is representative.
• Compare like with like. Restrict comparisons to periods, genres, regions, or institutions with similar preservation and documentation conditions, or explicitly model those differences.
• Use multiple sources of evidence. Combine catalogues, citations, translations, and archaeological or archival records so that one channel's gaps are partly covered by another.
• Report uncertainty. Present diversity estimates with intervals or sensitivity analyses, so a ranking between two cultures is not treated as exact when the underlying samples differ in size.
• Test the correction. Check whether the adjustment changes the conclusion. If it does not, the finding is more robust; if it does, the raw comparison was misleading.

A concrete reader scenario

Suppose you are comparing medieval narrative diversity in French and Dutch. The Dutch corpus may survive through fewer manuscripts, so a naive count of distinct tales favours French. An unseen species approach estimates the size of the unobserved portion of each tradition; if Dutch loss is higher, the corrected estimate may narrow or reverse the gap. The result is not "Dutch was equally diverse" but "the surviving evidence cannot support the original ranking."

Decision criterion

Ask whether your comparison depends on counts of observed items or on estimates of the underlying population. If the former, bias correction is essential. If the latter, you still need to justify why preservation differences do not drive the result. For methodological detail, see Karsdorp's publications at Fragments of Cultural Change.

How is latent cultural diversity estimated from historic texts like medieval French and Dutch literature?

Latent cultural diversity is estimated by treating the surviving texts as a biased sample of a larger, unobserved body of culture, then using statistical models—borrowed largely from ecology and population genetics—to infer how much was lost or remains hidden. The term "latent" is key: the goal is not to count what survives, but to estimate what probably existed behind the surviving record.

Folgert Karsdorp's work at the Meertens Institute and Utrecht University addresses exactly this problem. His publications page lists a study on estimating latent diversity shared by historic cultures, using medieval literature in French and Dutch as its case, alongside related work on correcting collection bias and applying unseen-species models to cultural survival. See Fragments of Cultural Change.

H3 How the inference works in practice

• Define the "units" of culture: for medieval literature, these might be distinct works, tales, motifs, or manuscript traditions.
• Take the observed sample: the texts and versions that survive in archives and editions.
• Model the sampling process: account for the fact that some works were more likely to be copied, preserved, or catalogued than others.
• Estimate the unseen portion: use frequency patterns in the observed data—especially how many items appear only once—to infer how many never appeared in the sample at all.
• Compare across cultures: French and Dutch medieval literature are compared to see whether apparent differences in diversity reflect real differences or merely different survival and collection histories.

H3 Why this matters for medieval French and Dutch literature Survival is not random. Prestige, language politics, monastic copying, later antiquarian interest, and modern editorial choices all skew which texts reach us. A naive count of surviving French versus Dutch works would confound cultural production with cultural preservation. Latent-diversity estimation tries to separate the two, so that claims about a "richer" or "poorer" tradition can be tested rather than assumed.

H3 Practical trade-offs

• The estimates are model-dependent: different assumptions about preservation yield different numbers.
• Confidence intervals can be wide, especially for sparse traditions.
• Results are best read comparatively and probabilistically, not as exact lost-work counts.

A useful next step: if you want to apply this, start by tabulating your observed corpus by frequency of occurrence (how many works survive in one copy, two copies, and so on), then consult the unseen-species literature that Karsdorp's work builds on before choosing a model.

What are the main methodological challenges in analyzing biased historical collections?

The core challenge is that historical collections are not random samples of past culture. They are what survived, what someone chose to keep, and what later institutions chose to catalogue. Any diversity measure computed directly from them inherits those filters.

Three problems that tend to appear together

1. Uneven survival. Older, rural, oral, or minority-language material is lost at higher rates than elite, urban, or written material. A decline in recorded diversity can reflect real cultural loss or simply thinner record-keeping.
2. Uneven collection effort. A spike in documents from one region often tracks where scholars, archives or funding were concentrated, not where cultural activity happened.
3. Latent shared diversity. Two related cultures may share material that neither record shows, because the shared version left no trace. Comparing the surviving fragments understates how much they had in common.

Why this matters for method choice

Folgert Karsdorp's work at the Meertens Institute and Utrecht University addresses exactly these questions, applying methods from ecology, population genetics and machine learning to cultural change. His selected publications include "Correcting collection bias in comparative studies of diversity" (Journal of the Royal Society Interface), "Saved in Translation? Estimating latent diversity shared by historic cultures and the case of medieval literature in French and Dutch" (Evolutionary Human Sciences), and "Forgotten Books: The Application of Unseen Species Models to the Survival of Culture" (Science). See Fragments of Cultural Change for the full list.

A practical next step

Before running any diversity comparison, write down the survival and collection pathway for each item: who created it, who preserved it, who catalogued it, and what would have had to happen for it to be missing. If that pathway differs systematically between the groups you are comparing, treat the raw difference as a hypothesis about the record, not a finding about the culture.