Skip to content

Goldilocks Range / Dual-Thumb Slider

The range slider schema presents a dual-thumb slider where annotators select a minimum and maximum value on a continuous scale, defining a range rather than a single point. This is sometimes called the "Goldilocks" task: annotators mark the acceptable range of a scalar property (not too low, not too high — but somewhere in between).

Overview

Standard sliders elicit a single point estimate. Many human judgments are more accurately described as ranges:

  • A sentence is "formal enough" somewhere between 60–80 on a formality scale
  • A paraphrase is acceptable if its meaning similarity is between 70–90%
  • A response length is appropriate if it falls between 2–4 sentences

The dual-thumb slider captures this interval structure directly. Both thumbs are independently draggable; the selected range is highlighted between them.

Research Basis

  • Pavlick, E., & Kwiatkowski, T. (2019). "Inherent Disagreements in Human Textual Inferences." Transactions of the Association for Computational Linguistics (TACL) 7. Shows that annotator disagreement on inference tasks reflects genuine semantic indeterminacy that is better modeled as a range than as a point.
  • Jurgens, D. (2013). "Embracing Ambiguity: A Comparison of Annotation Methodologies for Crowdsourcing Word Sense Labels." NAACL 2013. Demonstrates that allowing annotators to express uncertainty via ranges or soft labels improves dataset quality compared to forcing single-point choices.

Configuration

Options

Option Default Description
annotation_type — Must be range_slider
name — Schema identifier (required)
description — Task instruction
min_value 0 Minimum end of the scale
max_value 100 Maximum end of the scale
step 1 Increment between valid positions
left_label "" Label displayed at the left (min) end of the track
right_label "" Label displayed at the right (max) end of the track
show_values true Display the current numeric range values above the thumbs
label_requirement.required false Require range selection before proceeding

The widget opens with no range: the thumbs sit at the two ends in grey, no band is drawn between them, and the readout says "No range set". It stays that way until the annotator moves a thumb. There is no starting position to configure. A pre-drawn range is not stored, so an annotator who agrees with what is on screen leaves the item unanswered and is not told.

YAML Example — Formality Range

annotation_schemes:
  - annotation_type: range_slider
    name: formality_range
    description: "Select the range of formality scores that would be acceptable for this text in a professional email context."
    min_value: 0
    max_value: 100
    step: 1
    left_label: "Very informal"
    right_label: "Very formal"
    show_values: true
    label_requirement:
      required: true

YAML Example — Acceptable Confidence Interval

annotation_schemes:
  - annotation_type: range_slider
    name: confidence_interval
    description: "Select the range of values you believe the correct answer falls within."
    min_value: 0
    max_value: 100
    step: 5
    left_label: "0%"
    right_label: "100%"
    show_values: true
    default_low: 25
    default_high: 75

YAML Example — Sentence Length Acceptability

annotation_schemes:
  - annotation_type: range_slider
    name: length_range
    description: "How many words is acceptable for a summary of this article? Drag both thumbs."
    min_value: 10
    max_value: 200
    step: 5
    left_label: "Too short"
    right_label: "Too long"
    show_values: true

Output Format

The selected range is stored as two separate key-value pairs:

{
  "formality_range": {
    "range_low": "25",
    "range_high": "75"
  }
}

Both values are string-encoded numbers. range_low is always less than or equal to range_high.

Use Cases

  • Vague scalar adjective annotation — measure acceptable range of "tall", "expensive", "fast"
  • Acceptable paraphrase similarity — what similarity range qualifies as a valid paraphrase
  • Quality threshold elicitation — what range of scores would a human accept as "good enough"
  • Toxicity boundary marking — where does mildly offensive become clearly offensive
  • Summary length norms — what length range is appropriate for a given source document
  • Temporal expression annotation — mark the range of time a vague expression like "recently" refers to

Troubleshooting

Both thumbs overlap at the same position: This is a valid degenerate case (zero-width range = point estimate). If your task requires a non-zero range, add a validation check in the instruction text but note that the schema does not enforce a minimum range width by default.

An item comes back with no range: The widget records nothing until a thumb moves, and the unset state says so on screen. Set label_requirement.required if every item needs one.