A Structured Dataset of Payments, Places and Persons in the Cartulary of Santo Toribio de Liébana (790–1626)

Repository location

https://zenodo.org/records/22847548 (Last accessed 2026-09-24).

Context

The archive of Santo Toribio de Liébana yields a continuous charter series for a single Cantabrian valley from the late eighth century onward. Sánchez Belda’s (1948) critical edition prints 566 documents, of which 249 carry a full transcription; coverage by century runs from two charters in the eighth to 57 in the tenth and 80 in the thirteenth. The edition also supplies editorial regestas, a toponymic index and an index of persons. Of the 566 charters, 317 survive only as regestas, while 249 are transcribed in full.

Cartularies of this kind are used in qualitative economic and social history (Davies, 2003; Portass, 2023), but their contents have rarely been converted into machine-readable form (Torres Aguilar et al., 2022). Payment elements are a case in point. The language in which they are recorded is formulaic but orthographically unstable and distributed unevenly across the dispositive and penal sections. Retrieving it requires diplomatic segmentation rather than plain text search.

Steps

1. Diplomatic segmentation: Each charter was separated into critical apparatus, dispositive section, and penal clause. Penal clauses specify nominal sums that never changed hands (pectet in cocto regi mille solidos) and are legally distinct from prices (Carvajal Castro & Escalona, 2020); the audit below reports what treating the two alike costs. Segmentation used incipit and penal-clause markers covering Latin and Romance formulae.

2. Lexicon-based recognition: Goods were identified with an inflected lexicon covering Latin and Romance endings, in seven categories: coin, grain, livestock, wine, foodstuffs, manufactured goods, and real property. Quantities were parsed from Latin and Romance cardinals, Roman numerals and Arabic digits.

3. Payment-element identification: Extraction operated in two passes:

   • A window following an explicit counter-prestation marker (in adherato precio, accepimus de uos, in roboratione).

   • The direct syntagm pro/por/per + numeral + unit, the dominant thirteenth-century price formula.

4. Contextual enrichment: Place names were assigned from the edition’s toponymic index, which links each place to the charters in which it occurs, and person data from the index of persons. The monastery’s role in each transaction (acquirer, alienator, cited, absent) was derived from the editorial regestas.

5. Manual adjudication: The 49 event records flagged “needs_review” share a single review reason (payment not located). These cases occurred in transaction types where parties do not normally expect a counter-prestation: 28 inventory/record, 6 litigation/judgement, 5 ecclesiastical act, 3 donation, 3 renunciation/cession, 2 confirmation/privilege, and 2 undetermined. The pipeline allows for manual verification that the automated extraction had not overlooked a genuine payment. Both authors independently reviewed all 49 cases and reached consensus on each.

Sampling strategy

The dataset covers all 566 charters in the edition. Two conditions limit coverage of the payment variables. First, 317 charters survive only as editorial regestas; for these, payment information was sought in the regesta and flagged with a distinct provenance value. Second, 88 of the 325 processed charters record no counter-prestation, being donations pro remedio animae, judicial sentences or inventories. These cases are retained as structural zeros because their proportion is informative. The total of 325 reflects two disjoint subsets: 236 of the 249 fully transcribed charters yielded at least one payment element from the charter text, and 89 of the 317 regesta-only charters provided payment details from the editorial summary (236 + 89 = 325). An additional 241 were excluded from the released payment dataset after joint review by both authors. These excluded charters date predominantly from the fourteenth to sixteenth centuries and have minimal surviving text (median of 66 characters, compared to 1,516 for the released payment dataset), meaning that any price information would depend entirely on the editorial summary.

Quality control

• Numbering verification: The edition numbers of charters differently from the manuscript cartulary. Alignment was tested by checking whether place names listed in the toponymic index occur in the corresponding charter text under each candidate numbering. Publication numbering matched in 57.2% of assignments, compared to 1.5% for cartulary numbering and 2.5% for random assignment. An independent reconstruction of this test using the released gazetteer, concordance, and charter texts yielded the same qualitative pattern (37.6% under publication numbering, 3.2% under cartulary numbering, and 1.4% under random assignment). Although the reconstruction necessarily uses an alternate matching methodology, these findings corroborate the direction of the effect.

• Penal-clause contamination: An audit of an earlier, unsegmented extraction of the same corpus located nine mentions inside penal clauses. Removing them lowered the mean of the monetary series from 55.4 to 13.8 solidos and the maximum from 1,000 to 200: the contaminating values sat at the upper tail (Carvajal Castro & Escalona, 2020). This audit concerned an earlier, unsegmented extraction of the corpus. In the revised payment files, the charter-level has_penal_clause flag is retained for 81 charters to preserve this context for users; it does not indicate remaining contamination of the payment series.

• Systematic (Optical Character Recognition) OCR defects: Some characters render u as a throughout (aobis for uobis, qae for que). A repair routine, triggered only where the pattern is detected, recovered payments the lexicon would otherwise have missed. This routine covers only the u-for-a substitution; three further corruption patterns identified during revision (u misread as n, as g, and as q) are not yet covered and are noted as a limitation below.

Limitations

The 57.2% verification rate is not an error rate for the index. Failures concentrate where the modern form differs from the medieval one, as in Baró against Barao or Oña against Onia. The rate measures the distance between spellings, not the reliability of the editor’s identifications.

Coordinates were established for 94 of the 603 place names. Of these, 74 enter the ring-distance calculation, while 20 are excluded because they refer to episcopal sees or kingdoms that appear solely in royal confirmation clauses. Twenty-two names could not be matched against modern gazetteers.

The supplementary geolocation file (liebana_places_geocoded.xlsx) covers 400 entries for which the editor supplied a modern identification, but the main gazetteer did not originally provide coordinates. Coordinates were added for 348 entries, drawing on GeoNames and OpenStreetMap or, where digital records fail, local informants’ testimony; the remaining 52 entries are explicitly marked as unidentifiable with current sources.

Homonyms present an additional limitation. The 1,348 released person entries represent unique name strings from the printed index. Consequently, no accidental name-collision merging occurs, but the dataset does not attempt to disambiguate true namesakes across charters.

Regarding charter dates, 47 of 566 charters (8.3%) lack a recorded year. The current dataset does not distinguish between securely dated charters and those whose dates the editor notes as disputed.

The 49 event records retain the needs_review flag as an audit marker, although all 49 cases were manually reviewed and adjudicated by both authors.

Files

The files included in the released dataset are described in Table 1.