Institutional Language Detection — Open Data

unmask.tools maintains a continuously growing dataset of phrases used to obscure, downplay, or reframe harmful events in political, corporate, and military communication. Each phrase is crowd-sourced and expert-curated, with confirmation counts, geographic spread, industry sector mapping, and references to applicable international law.

The dataset contains no personal data of any kind — no email addresses, no identifiers, no individual behavioural data. It is a structured vocabulary of institutional language, suitable for NLP training, media monitoring, compliance tooling, and academic research.

Unique phrases
Confirmed detections
Detected today
EN · DE · FR
Languages (growing)

How phrases enter the dataset

1
Candidate identification

Phrases are nominated from three sources: the LDM Core Library (expert-curated), community suggestions submitted through the extension, and AI-assisted pattern analysis of news and institutional texts.

2
Crowd confirmation

Extension users who have opted in to global sharing can confirm or dismiss detections in real-world context. A phrase graduates to the global_index once it reaches a confidence threshold based on confirm/dismiss ratio, domain spread, and country spread.

3
Severity classification

Each phrase is assigned a severity: critical (direct masking of violence, displacement, or deprivation of rights), high (significant obfuscation of labour or economic harm), or medium (euphemistic but context-dependent). Classification is informed by references to international law — Geneva Conventions, Rome Statute, UDHR, ILO conventions.

4
Monthly snapshot

On the 1st of each month, the live dataset is automatically snapshotted. Licensed parties receive access to monthly exports and optionally a live query endpoint, giving them both historical depth and current coverage.


Dataset fields

Every record in the dataset contains the following fields. All fields present in the full licensed dataset; the free sample includes all columns below.

Field Type Description
phrase string The exact phrase as it appears in text
normalized_phrase string Lowercase, punctuation-stripped form for matching
severity enum critical | high | medium — based on degree of obfuscation and applicable law
lang string ISO 639-1 language code (en, de, fr, …)
confidence_score decimal 0.0 – 1.0 confidence score based on crowd confirmation ratio, domain spread, and country spread
confirm_count integer Number of times this phrase was confirmed as intentional obfuscation in real context
dismiss_count integer Number of times dismissed as a figure of speech or innocent use
domain_count integer Number of distinct registered domains on which the phrase was confirmed
country_count integer Number of distinct countries from which confirmations were received
isic_sections string[] ISIC Rev. 4 section codes for industry sectors where the phrase is commonly observed (e.g. ["O","K"] for public administration + finance)
legal_ref string Primary international law reference (Geneva Conventions, Rome Statute, UDHR, ILO, Paris Agreement, etc.)
source_document string Originating source (LDM Core Library for expert-curated entries, or community submission reference)
first_seen timestamp UTC timestamp when the phrase was first promoted to the global index
detection_count integer Number of distinct detection events recorded by the extension across all users
total_occurrences integer Total occurrence count across all detection events — a single event may detect the phrase multiple times in one document

Top 15 phrases by confidence score

Free sample — full dataset of 339 phrases available under licence (see below). Data shown is identical in structure to the complete export.

unmask-dataset-sample.json 15 records · JSON · schema_version 1
↓ Download JSON
Phrase Severity Confidence Confirms Domains Countries Legal reference
enhanced interrogation techniques CRITICAL 0.970 631811
shock and awe CRITICAL 0.960 581510
population transfer CRITICAL 0.950 44119
collateral damage CRITICAL 0.940 47128
austerity measures CRITICAL 0.930 511913
workforce reduction HIGH 0.920 542214
surgical strike CRITICAL 0.910 3897
rightsizing HIGH 0.900 411610
kinetic action CRITICAL 0.890 3176
administrative removal HIGH 0.880 33108
redundancies HIGH 0.880 36149
humanitarian intervention CRITICAL 0.880 2986
fiscal consolidation HIGH 0.870 28107
regime change CRITICAL 0.860 2565
carbon offsetting HIGH 0.850 2486

Data access tiers

The 15-phrase sample above requires no key. A free research key gives you the full dataset via API. Commercial and OEM licences add monthly snapshots, CSV export, version history, and redistribution rights.

Research
Free non-commercial, attribution required
  • Full live dataset via API
  • JSON format
  • Instant self-service key below
  • Non-commercial use only
Get API key →
OEM / Embedded
On request per product
  • Embed in your SaaS or data product
  • White-label available
  • Webhook-based update delivery
  • CSV + JSON + snapshot history
  • Custom field extensions

Commercial or OEM licence?

Send us a brief note about your use case and we'll get back within two business days.

hello@unmask.tools Request access →

No personal data — ever

The dataset contains no personal data of any kind. It is derived exclusively from aggregated, anonymised crowd-confirmation signals. No email addresses, user identifiers, device identifiers, or behavioural profiles are included or derivable from this dataset.

Phrases contributed by users through the opt-in global sharing feature are incorporated into the aggregate counts only. Individual contribution records are never exported. The legal basis for commercial licensing of the anonymised aggregate dataset is our legitimate interest (Article 6(1)(f) GDPR), as detailed in our Privacy Policy §5a.

Commercial use of this dataset must not be used to identify, target, profile, or surveil any individual or group of individuals. See Terms of Service §7a for full licensing terms.