Functional Clause and Deontic Phraseology Annotations for Polish, English and Spanish Mutual Confidentiality Agreements

This dataset contains the data, documentation and reproducibility materials accompanying the study “Confidentiality Agreements in Polish, English and Spanish: A Corpus-Based Comparative Study of Functional Organisation and Deontic Phraseology”. The study is based on a corpus of 30 original-language mutual business-to-business confidentiality agreements, balanced by document count: ten Polish agreements associated with Polish law, ten English agreements associated primarily with England and Wales, and ten Spanish agreements associated with Spanish law. The source-type composition was not matched across the three language subcorpora. The final analytical corpus contains 953 functionally validated contractual clauses and 50,032 words. The deposited materials include: document-level metadata recording provenance, year, source type, jurisdiction, source URL, capture information and available cryptographic hashes; the final clause-level functional annotations; the functional annotation codebook; a multilingual dictionary of Polish, English and Spanish deontic markers; the complete candidate-level deontic annotation dataset; a legal-deontic subset containing 457 expressions classified as obligations, prohibitions, permissions, entitlements or absence of obligation; functional-segmentation and deontic-annotation audit records; scripts documenting text preparation, duplicate control, functional corpus construction, candidate extraction and quantitative analysis; documentation describing the data structure, analytical decisions and reproduction workflow. The extraction procedure identified 620 potential deontic expressions. All candidates were reviewed in their complete functional-clause and document context. The final reference dataset contains 457 expressions classified as legally deontic and 163 retained but analytically excluded definitional, stipulative, non-deontic or template-related uses. The annotations include language, document and clause identifiers, canonical and surface markers, construction type, deontic type, polarity, legal-deontic status, operativity, actor, actor scope, conditionality and annotation provenance. The final analysis script reproduces the four tables and the document-level statistical comparisons reported in the accompanying article from the deposited frozen datasets. The earlier corpus-preparation, segmentation and candidate-extraction scripts document the processing pipeline but require the original third-party source documents and certain intermediate files, which are not redistributed with this record. The public dataset provides derived annotations, document metadata, source information and available cryptographic hashes. Complete copies of third-party confidentiality agreements are not redistributed where reuse rights could not be established. Any third-party contractual text or excerpt retained in the deposited files remains subject to the rights of its respective owner. The licences attached to this record apply only to the original annotations, derived metadata, codebooks, documentation and software produced for the study and do not grant rights in the underlying source documents. Manual validation was conducted by a single researcher. The deposited audit records and codebooks support procedural transparency and inspection of individual decisions, but they do not constitute an independent estimate of inter-annotator agreement.