Configuration File

[DylanBulmer]

Fields:

• type (classification, code-labeling, etc)
• guidelines
• input model
• export model

[vijayantajain]

Fields of config file:

• Input
• Output
• Guidelines

[DylanBulmer]

Mock UI JSON model:

// classification and translation to be implemented first
type TaskType = "classification" | "labeling" | "code-labeling" | "translation";
type InputField = "text" ;
type OutputField = "short-text" | "long-text" | "radio" | "multiple-choice" | "range";

// what the researchers provide to display to users
interface Input<V> {
  type: InputField;
  language?: string;
  value: V;
}

// where the user provides their annotation
interface Output<V> {
  type: OutputField;
  range?: number[]; // [min, max]
}

interface Configuration {
  "type": TaskType;
  "guidelines": string;
  "display" {
    "inputs": Input[];
    "outputs": Output[];
  }
  "model": Record<string, unknown>; // for validating incoming input
}

interface ClassificationConfiguration extends Configuration {
  ...
}

[vijayantajain]

Type of OutputFields:

• Short-Text
• Long-Text
• Range
• Radio
• Multiple Choice (checkboxes)

[vijayantajain]

Columns in the dataset:

(task-independent columns)

1. Id of the sample
2. Status of annotation
3. User Annotation/ Output => (ideal for classification, translation). Not ideal for (NER and other custom tasks)
4. Button to go to the sample