Annotation · Named entity recognition
NER annotation for NLP teams. Your schema decides if Apple is a company.
Custom entity schemas, calibrated reviewers and logged adjudication, delivered in the format your NLP pipeline reads.
- Custom schemas
- Calibrated review
- Multiple formats
- Scoped delivery
Failure modes
Where entity labels break
Entity boundary errors
Models mistake where entities start and end, causing 'Apple Inc.' to become 'Apple' (the fruit) or 'New York Times' to split into city and publication.
Inconsistent labelling
Inter-annotator disagreement causes the same entity to be tagged differently: 'Dr. Smith' becomes PER in one document, TITLE+PER in another.
Domain-specific entities
Generic NER misses industry-specific entities. Legal case numbers, medical dosages, financial tickers may be missed when the schema and corpus do not reflect the domain.
What the schema must settle
What a managed NER programme settles
Precision tagging
Define boundary rules, labels, examples, and exceptions before full annotation begins.
Context awareness
Disambiguate 'Apple' between company, fruit, and record label based on surrounding context.
Domain expertise
Reviewers are selected for the language, domain, and sensitivity requirements of the project.
Quality validation
The review plan can include calibration, inter-annotator agreement, sampling, and adjudication.
Open questions
What the schema has to answer before annotation starts
Open at the start
- How context changes the label
- Which domain terms count as entities
- One guideline for every reviewer
- How quality is checked
- Where errors are caught
Managed NER programme
- Domain and language-qualified reviewers
- Custom schema for your use case
- Multi-pass quality validation
- IAA or audit sampling where appropriate
- Acceptance criteria fixed before delivery
From schema to accepted export
What the annotation contract fixes
Make entity quality measurable
Define reviewer fit, schema coverage, review rules, and acceptance evidence for the NLP pipeline that will consume the data.
Reviewer fit
Reviewer qualifications should match the language, domain, entity schema, and data sensitivity in the project.
How context changes the label
The schema defines how context changes the label, how ambiguity is escalated, and how the final decision is documented.
Control rework before scale
Calibration and adjudication expose unclear schema rules before they spread across the full corpus.
What acceptance is measured against
Quality is defined by agreed metrics, a representative review set, documented exceptions, and an acceptance decision tied to the downstream use case.
From corpus review to accepted delivery
Corpus and schema review
Review the corpus, define entity types, establish boundary rules, and log unresolved questions before annotation begins.
Annotation and calibration
Calibrated reviewers apply the schema. Ambiguous spans are escalated through the agreed adjudication path.
Quality validation
Apply the agreed IAA, sampling, or full-review method. Record edge cases, corrections, and acceptance results.
Accepted export
Validate the agreed export format and deliver the schema, data, review evidence, and exception record defined in the project contract.
The annotation contract decides
what your NER model learns.
"Mark Thompson, CEO of Quantum Dynamics, announced a $4.2 billion acquisition..."
Defined schema → consistent annotation → measurable review
Domain-specific review
Reviewer fit and terminology guidance matched to the scoped corpus
Context-aware tagging
Disambiguates 'Apple' between fruit, company, and record label
Pipeline-compatible output
CoNLL, spaCy, JSON, or another format agreed for the target pipeline
What acceptance means
Start with a scoped sample
Validate the schema before scale
Review labels, ambiguity, reviewer calibration, and acceptance criteria
Scope a sampleDomain and language fit
Define the entities that matter
Schema design • Reviewer plan • Pipeline-compatible exports
Looking for work on AI projects? This form is for companies. Apply as a freelancer
Privacy scope, target languages and export formats are set during scoping.
Data protection
GDPR and project data boundaries
Data roles, purpose, access, location, transfer, retention, and rights-request procedures must be defined for the actual project and contract.
Processing scope
Document the data categories, purpose, access model, processing locations, and retention before annotation begins.
Roles and instructions
The buyer and YPAI define their data roles and processing instructions in the applicable agreement.
Rights requests
Assistance, identification, deletion, correction, and response procedures are handled according to the project role and applicable contract.
EEA delivery options
Processing and storage locations are selected and documented against the approved project architecture and transfer requirements.
Subprocessor scope
Relevant subprocessors, locations, and contractual controls are disclosed for the scoped delivery model.
Project records
The agreed evidence package can include instructions, access records, retention rules, review results, exceptions, and change history.
Data processing questions
Rights-request process
Defined by the applicable role, contract, and legal requirement
Evidence
Request the project-specific data boundary and governance package
Define the schema before the corpus is labelled.
Scope a managed NER programme for your corpus, schema and downstream pipeline.
Scope an NER programme