DISSILEX: A machine-operable lexico-semantic network and valency lexicon for medieval Latin

DISSILEX is a controlled vocabulary in the form of a manually built lexico-semantic network of medieval Latin verbs and verbal expressions, featuring a detailed valency lexicon and connections to a large set of Latin and modern-English concepts, presented here as a single SQLite file (dissilex.db) that is readable by the sqlite3 command-line tool, any SQLite browser, or Python's built-in sqlite3 module. Rooted in the domain of inquisitorial records, DISSILEX covers general as well as more subject-specific meanings, with both standard (synonym, hypernym, etc.) and less canonical relations. DISSILEX is a product of Computer-Assisted Semantic Text Modelling (CASTEMO; Zbíral et al. 2026 - see README.md for full references), an approach to modeling statements as a four-slot structure of subject(s), predicate(s) and two objects, creating a thickly connected network of data points. Coverage is richest for human-interaction verbs (testimony, accusation, belief, religious practice) and the legal vocabulary of heresy trials. We distinguish two entry types: Actions (verbs and verbal expressions, each associated with a three-slot valency frame specifying entity type, morphosyntactic, and semantic valencies) and Concepts (single- and multi-word expressions for other parts of speech). The network is connected through a set of 11 relation types, including superclass (hypernym) membership, synonymy, antonymy, verb-to-noun mappings, and valency-specific relations. Each relation connects two entities, and can be unidirectional or bidirectional. As part of an ongoing effort to position DISSILEX within the Linguistic Linked Open Data (LLOD) cloud, many entries contain IDs to external sources stored in the database, specifically to the LiLa (Linking Latin) Lemma Bank and Princeton WordNet (PWN) 3.0 and 3.1 synsets. We applied the Collaborative Inter-Lingual Index (CILI) to map between the two versions of the PWN for entries where only one of the IDs has been added. We also indicate cases where no equivalent for a DISSILEX lemma exists ("NA"). Via the LiLa SPARQL endpoint, it is possible to use the linked LiLa lemmas, which feature as the central unit of linking sources in the Latin LLOD cloud, to retrieve data from several resources including dictionaries, corpora, treebanks, and various NLP tools. We have made use of this opportunity to enrich the database file with lemmas from the LiLa Lemma Bank, while also supplying lemmas from the LatinCy lemmatizer (model: la_core_web_lg). This release contains: dissilex.db: SQLite database, which can be readily queried dissilex_schema.md / dissilex_schema.pdf: schema documentation README.md: full dataset description, statistics, and SQL examples ATTRIBUTION.md: license and attribution notices. LICENSE-DATA: Full CC BY-SA 4.0 license text. Funding, attribution and licence DISSILEX is developed by the Dissident Networks research group (DISSINET) at Masaryk University and has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme, grant agreement No. 101000442, project “Networks of Dissent: Computational Modelling of Dissident and Inquisitorial Cultures in Medieval Europe”, and from the European Regional Development Fund, grant agreement No. CZ.02.01.01/00/22_008/0004595, project “Beyond Security: Role of Conflict in Resilience-Building”. DISSILEX is released under CC BY-SA 4.0. It incorporates data from external resources: LiLa Lemma Bank: CIRCSE, Università Cattolica del Sacro Cuore (Milan). Licensed CC BY-SA 4.0. The database redistributes a subset of LiLa lemma forms (subset-selected and format-converted, not otherwise modified); the ShareAlike clause is honoured by this release's CC BY-SA 4.0 licence. URL: https://lila-erc.eu/ Princeton WordNet 3.0 / 3.1: We distribute offset IDs and glosses. WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved. WordNet License. https://wordnet.princeton.edu/ LatinCy/spaCy: We redistribute output from the LatinCy model. The `spacy_lemma` field contains lemmas generated by the LatinCy spaCy pipeline `la_core_web_lg` (Patrick J. Burns). The model is MIT-licensed (<https://huggingface.co/latincy/la_core_web_lg/blob/main/README.md>); the spaCy library is MIT-licensed (<https://github.com/explosion/spaCy/blob/master/LICENSE>). Full notices, including the Collaborative Inter-Lingual Index (CILI) and Latin WordNet, are in ATTRIBUTION.md.