> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/documentation/annotations/text-annotation.md).

# Text Annotation

Text annotation turns language into structured entities, relations, and classifications. Quality depends on token-boundary policy, overlapping spans, relation direction, normalization, and ambiguity handling.

### What this editor is designed to preserve

* The exact source text and selected span boundaries.
* Entity type, relation direction, and document- or item-level classifications.
* Context required to interpret pronouns, repeated names, formatting, and cross-sentence meaning.

### Read the Workbench before annotating

The text workbench supports:

* entities;
* relations between entities;
* pan/navigation;
* comments;
* font-size and line-height controls;
* shortcuts;
* item properties and tags;
* normal workflow completion.

One uploaded text file remains one work item. Large text is delivered to the editor in character windows rather than being split into separate project tasks. Entity offsets and relations remain anchored to the full file. A project setting can allow annotators to edit the source text when the workflow requires correction as well as labeling.

### Entities, relations, and classifications

* **Entity:** a span with domain meaning, such as organization, medication, defect, or clause.
* **Relation:** an explicit connection between entities, such as `treats`, `caused_by`, `belongs_to`, or `answers`.
* **Item property:** a whole-document decision such as language, document type, sentiment, escalation state, or completeness.

In a document-and-audio layout, text entities and their relations can remain visible while annotators review temporal audio events below. Text structure can therefore be interpreted alongside another modality instead of being handled in a disconnected process.

### Text-quality design

Entity guidelines should define inclusion boundaries, punctuation, nested mentions, discontinuous concepts, pronouns, abbreviations, and uncertain spans. Relation guidelines should define direction, allowed source/target classes, and whether the connection must be explicit in the text or may be inferred from context.

### From assigned task to review

{% stepper %}
{% step %}

#### 1. Read the full context

Understand the passage, task instructions, and required ontology before selecting spans.
{% endstep %}

{% step %}

#### 2. Select the exact span

Apply the token, punctuation, whitespace, and nesting policy consistently.
{% endstep %}

{% step %}

#### 3. Assign the entity

Choose the ontology class and complete required properties.
{% endstep %}

{% step %}

#### 4. Create relations

Link the intended source and target with the correct direction and type.
{% endstep %}

{% step %}

#### 5. Apply classifications

Complete text-, document-, or item-level values that do not belong to a span.
{% endstep %}

{% step %}

#### 6. Review for coverage

Check missed mentions, duplicate spans, boundary drift, conflicting relations, and unresolved ambiguity.
{% endstep %}
{% endstepper %}

### Modality-specific quality controls

| Decision           | Production guidance                                                                 |
| ------------------ | ----------------------------------------------------------------------------------- |
| Span boundary      | State whether determiners, punctuation, units, titles, and whitespace are included. |
| Overlaps           | Define whether nested or overlapping entities are allowed.                          |
| Relation direction | Make source, target, and relation semantics explicit.                               |
| Ambiguity          | Use the escalation route instead of choosing a convenient but unsupported label.    |

### Route the result

Submit only when entity coverage, span boundaries, relation direction, classifications, and required properties are consistent with the current Instructions and ontology.