Guide to: Running a Text Relationships Job

Overview 

The Text Relationships tool (cml:text_relationships) allows users to create a job that annotates relationships between spans of text with a custom ontology, please contact your Appen representative to gain access.

Glossary 

• Span - a string of text with an assigned class label; the output of a model or contributor judgment.

• Relationship - consists of two spans (from a span and to a span) and the relation between them.

• Relation - the name/type of the relation between two spans as defined in the job's ontology.

• From span - the starting span in a relationship

• To span - the ending span in a relationship

Upload Data

The source data of a Text Relationships job can come from two different sources:

• The output of a previous text annotation job on the Appen platform

  • The output of a previously ran text annotation job on the Appen platform can be uploaded to a text relationships job directly without any modification.

• Data created externally

  • A source file containing data created externally can be uploaded to a text relationships job, but an identical JSON format (hosted in a URL and a CORS configured bucket) as the output of a text annotation job on the Appen platform will be required.

Note: There is an example file attached below on how to format source data.

Build a Job

Parameters

Below are the parameters available for the text relationships tool. Some are required in the element, some can be left out.

• source-data (required)

  • The name of the source data column containing the data to be annotated

• name (required)

  • The result header where the result links will be stored

• context-column(optional)

  • The name of the source data column containing the context of each data row

• review-data (optional)

  • The name of the source data column containing pre-annotated text relationships

• task-type (optional)

  • Please set task-type=”qa” when designing a review or QA job. This parameter needs to be used in conjunction with review-data . See this article for more details.

• direction (optional):

  • Renders text in a specific direction

  • Accepts rtl and ltr for right-to-left and left-to-right scripts respectively

  • Defaults to left-to-right if not set

Ontology

• The Ontology Manager allows job owners to create and edit the ontology within a Text Relationships job. Text Relationships Jobs require an ontology to launch.

• When the CML for a text relationships job is saved, the Ontology Manager link will appear at the top right of the Design page.

• The ontology of Text Relationships jobs allows you to create relationship restrictions to each span class.

• The ontology of a Text Relationships job can be copied from a Text Annotation job or another Text Relationships job, via download and upload.

Ontology Manager Best Practices 

• The limit of ontology is 1,000 classes, however, as best practice, we recommend not exceeding 16 classes in a job to ensure contributors can understand and process the different classes. 

• Choose from 16 colors pre-selected or upload custom colors as hex code via the CSV ontology upload.

Important Note: If there is no relationship restriction defined to a class, the class will not able to relate to any other classes in the job.

Nested Spans

Text Relationships tool (cml:text_relationships) supports nested spans. However, there is one restriction for creating relationships for nested spans.

Consider this example:

Here we have one root span

and two sub spans.

We can create Relationship between sub spans

But between the root span and the sub span we cannot create a relationship.

 Results

• Results are links to a JSON file that contains a list of relationships.

• The links are found in the Full or Aggregated report under the column header that was specified as the value for the name attribute.

• Result links will expire 15 days after generation; to access results after the links have expired, you will need to re-generate the result report.

• Each relationship instance is an array of five attributes:

  • id: the unique ID of each relationship instance

  • name: the class name of the relation

  • from_span: contains the details of from_span

  • to_span: contains the details of to_span

  • annotated_by: indicates whether a relationship instance is pre-loaded into the job or manually added by an annotator