Named Entity Recognition (NER) Dataset for Medieval Latin Charters from Monasterium.net

This dataset contains a collection of sentences extracted from medieval Latin charters, specifically curated from the Monasterium.net (MOM) portal, mainly from the AT-StiAH fond. The data is formatted in JSONL (JSON Lines), where each line represents a single sentence with its corresponding semantic annotations (spans).

The dataset is designed to support tasks related to Natural Language Processing (NLP) in the digital humanities, particularly Named Entity Recognition (NER) and Information Extraction (IE) from historical legal documents.

Data Format

Each entry in the .jsonl file follows this structure:

text: The original Latin sentence from a medieval charter.

spans: A list of annotated entities within the text, including their character-level start and end offsets and a semantic label.

Entity Labels

The annotations include, but are not limited to, the following labels:

PER: Personal names.

ACTOR: Individuals or groups acting in a legal capacity.

TITLE: Ecclesiastical or secular titles (e.g., abbas, iudex, episcopus).

LOC: Geographic locations, towns, or regions.

INS: Institutions (e.g., monasteries like Sancte Marie de Cripta).

LEG: Legal formulas and references to legal procedures.

DAT / TIM: Dates and time-related expressions.

MON / NUM: Monetary units and numerical values.

TRANS / PROP: Transaction types and property-related terms.

REL: Family or social relationships.

Potential Use Cases

Training NER Models: Developing machine learning models specifically for medieval Latin.

Historical Research: Automatically identifying key actors and locations in large-scale charter collections.

Linguistic Analysis: Studying the formulaic language of medieval legal documents.

Technical Specifications

File Name: 458_mom_sentences.jsonl

Language: Medieval Latin

Encoding: UTF-8

Format: JSONL (one JSON object per line)