Challenge

Where the Wild Features Are

Predict kindergarteners' end-of-year reading scores from literacy screening audio, and find which speech features carry signal beyond correctness scores

Tasks
Reading assessment
Data types
Audio
Tabular
$50,000 in prizes
8 weeks left model submissions
53 joined
Adapted from a photo by Guy Basabose on Unsplash

Overview

The ability to read, write, and comprehend language is fundamental to a person's development, underpinning success in school, opportunities in the workforce, and engagement in community. Early literacy skills are strongly related to later reading success, which makes intervening early a promising way to help children who are at risk for reading difficulty. To find those children, teachers across the U.S. give literacy screeners: short oral tasks such as blending sounds into a word, deleting part of a word, or repeating a sentence. Each response is scored right or wrong by the adult giving the test. A correctness score captures whether a child got the answer, but not how they said it. Speech may carry additional signal about a child's reading development that correctness scores miss.

In this challenge, your goal is to advance the science of early literacy screening by finding which acoustic features of children's speech predict end-of-year reading outcomes. The data come from the Reach Every Reader literacy screener, collected by MIT and Florida State University.

The challenge has two phases. First, develop a model that uses kindergarteners' beginning-of-year literacy screening scores and audio recordings to predict their end-of-year reading scores. Then, in Phase 2, explain which acoustic features provide additional predictive information beyond the screening scores. The leaderboard measures predictive accuracy, but final prizes are awarded based on judged write-ups. The strongest solutions will show, with rigorous analysis, what children's speech reveals about later reading outcomes that the screening scores alone do not. Tools that can better analyze speech could help teachers identify children at risk for reading difficulty and intervene before reading failure occurs.

%%{
  init: {
    'theme': 'base',
    'gantt': { 'tickInterval': '1month', 'topPadding': 66, 'gridLineStartPadding': 50 },
    'themeVariables': {
      'background': '#f9f8f5',
      'mainBkg': '#f9f8f5',
      'titleColor': '#20083f',
      'fontFamily': 'Inter, Helvetica, Arial, sans-serif',
      'fontSize': '14px',
      'gridColor': 'transparent',
      'taskBorderColor': '#4d3965',
      'taskBkgColor': '#daceec',
      'taskTextColor': '#20083f',
      'taskTextLightColor': '#20083f',
      'taskTextOutsideColor': '#20083f',
      'taskTextClickableColor': '#20083f',
      'activeTaskBorderColor': '#1c6b8e',
      'activeTaskBkgColor': '#d3eaf5',
      'doneTaskBkgColor': '#eae6dc',
      'doneTaskBorderColor': '#bcb1a3',
      'critBorderColor': '#1c6b8e',
      'critBkgColor': '#d3eaf5',
      'todayLineColor': 'transparent',
      'sectionBkgColor': '#f9f7fb',
      'altSectionBkgColor': '#f4f9fb',
      'sectionBkgColor2': '#f9f7fb'
    },
    'themeCSS': 'text { font-style: normal !important; } .tick text { display: none; } .sectionTitle { font-weight: 700 !important; } .grid .tick line { display: none; } path.domain { display: none; } .vert { rx: 0 !important; ry: 0 !important; width: 6px !important; fill: none !important; stroke: #4d3965 !important; stroke-width: 1.5px; stroke-dasharray: 4 3; clip-path: inset(2px 0 24px 50%); transform: translateX(-6px); } .vertText { transform: translateY(-28px); font-weight: 700 !important; fill: #20083f !important; font-size: 12px !important; } rect.vert[class*=crit] { display: none; } .vertText[class*=critText] { transform: translateY(-12px); font-weight: 400 !important; fill: #4d3965 !important; }'
  }
}%%
gantt
    title Competition Timeline
    dateFormat  YYYY-MM-DD
    axisFormat  %B

    section Phase 1
    Model submission window                         :phase1, 2026-11-19, 2027-01-19

    section Phase 2
    Write-up window                                 :active, phase2, 2027-01-20, 2027-02-08

    Nov 19                                          :vert, v1, 2026-11-19, 0d
    Jan 19                                          :vert, v2, 2027-01-19, 0d
    Feb 8                                           :vert, v3, 2027-02-08, 0d
    Competition opens                               :crit, vert, e1, 2026-11-19, 0d
    Submissions close                               :crit, vert, e2, 2027-01-19, 0d
    Write-ups due                                   :crit, vert, e3, 2027-02-08, 0d

Prizes

Competition End Date:

Feb. 8, 2027, 11:59 p.m. UTC

Place Prize Amount
1st $20,000
2nd $15,000
3rd $10,000
4th $5,000
Total Prize Pool $50,000

Prizes for this competition will not be awarded based on leaderboard ranking alone. The competition runs in two phases:

  • Phase 1: Modeling. Train models and submit predictions for the test set. Your rank on the private leaderboard determines whether you are invited to Phase 2.
  • Phase 2: Write-ups. The top 15 teams on the private leaderboard when Phase 1 closes will be invited to submit a solution write-up explaining which acoustic features carry signal beyond the screening scores. A panel of subject matter experts will select the 1st–4th place winners based on the write-ups.

The top 4 teams after judging must submit code that reproduces their leaderboard submission to receive a prize. Prize winners must also release their solution under the MIT License.


How to compete

  1. Click the "Compete!" button in the sidebar to enroll in the competition.
  2. Get familiar with the problem through the overview and problem description. You might also want to reference additional resources available on the about page.
  3. Download the data from the data tab.
  4. Create and train your own model. The benchmark blog post is a good place to start, and will be linked here when it is available.
  5. Use your model to generate predictions that match the submission format.
  6. Click "Submit" in the sidebar, and then "Make new submission". You're in! You can make up to 3 submissions per day.
  7. If you are in the top 15 on the private leaderboard when Phase 1 closes, you will be invited to submit a write-up. Summarize your findings as described in the solution write-up requirements, then upload your PDF on the "Write-up submissions" tab. Read the requirements early, since they should shape how you build and validate your model.

The challenge rules are in place to promote fair competition and useful solutions. If you are ever unsure whether your solution meets the competition rules, ask the challenge organizers in the competition forum or send an email to k12-ai-infra@drivendata.org. A few key rules are highlighted below. For more details, see the competition rules.

Teams and registration

Teams can have up to 5 members. Registration, team formation, and team mergers close when Phase 1 closes, on January 19, 2027 at 23:59:59 UTC.

Use of test data

Your predictions must be generated entirely by code, with no manual input. Hand-labeling, manual transcription, or listening to test clips to inform predictions is not allowed. Test data may not be used to train models, including through pseudo-labeling. The top 4 teams' code will be checked against their leaderboard submissions.

External data and models

External data and pretrained models are allowed if they are freely and publicly available to all participants under a permissive open license that does not prohibit commercial use, such as CC0 or CC-BY. This includes pretrained speech models such as wav2vec2, Whisper, and the children's speech recognition models from DrivenData's On Top of Pasketti challenge.

You may not upload competition data to any third-party services or APIs that retain the data. For example, you cannot submit data to ChatGPT, Claude, or Gemini. You can use external models by loading open-source model weights into an environment that you can wipe the data from afterwards, such as your local machine or a cloud compute environment. See the competition rules for details.


This challenge is sponsored by the MIT Gabrieli Lab, the MIT Senseable Intelligence Group, and the Florida Center for Reading Research at FSU.

Sponsor Logos


Image Credit: Adapted from a photo by Guy Basabose on Unsplash