Datasets:

Languages:
English
Size:
n<1K
ArXiv:
License:
achieve-the-core / README.md
lucy4's picture
Update README.md (#1)
a3bc393 verified
metadata
license: odc-by
language:
  - en
tags:
  - education
  - math
size_categories:
  - n<1K

Dataset Card for Achieve the Core

This repository includes Common Core math standards, their descriptions, and metadata obtained from Achieve the Core.

Example of a math standard:

{
  "id": "K.CC.B.4",
  "description": "Understand the relationship between numbers and quantities; connect counting to cardinality.",
  "source": "Achieve the Core",
  "level": "Standard",
  "cluster_type": "major cluster",
  "aspects": [],
  "parent": "K.CC.B",
  "children": ["K.CC.B.4c", "K.CC.B.4b", "K.CC.B.4a"],
  "connections": {"progress to": ["1.OA.C.5", "K.CC.B.5"], "progress from": [], "related": ["K.CC.A.2", "K.CC.C.6", "K.CC.A.1"]},
  "modeling": false
}

See MathFish for more details on uses of this data.

This data can be used to evaluate language models' abilities to assess whether math problems enable students to learn specific skills/concepts. Code to support this can be found in this Github repository.

Dataset Details

Dataset Description

  • Curated by: Lucy Li, Tal August, Rose E Wang, Luca Soldaini, Courtney Allison, Kyle Lo
  • Funded by: The Gates Foundation
  • Language(s) (NLP): English
  • License: ODC-By 1.0

Dataset Sources

Dataset Structure

This repository includes two key files: domain_groups.json and standards.jsonl.

We created domain_groups.json because the "domains" we evaluate with for our tagging task do not have a one-to-one mapping to K-8 domains and high school (HS) categories in Common Core State Standards (CCSS). Some HS categories are equivalent or similar to a domain in K-8, and some differences in K-8 domains are difficult to explain a brief description at the domain-level. Thus, a "domain" in our paper sometimes groups multiple actual CCSS domains/categories. We mostly retain the original CCSS K-8 domains and HS categories, but make exceptions for the following: we group OA (Operations & Algebraic Thinking), EE (Expressions & Equations), and A (HS Algebra) into Operations & Algebra, S (HS Statistics & Probability) and SP (K-8 Statistics & Probability) to \textit{Statistics & Probability}, and finally NS (K-8 The Number System) and N (HS Number and Quantity) to Number Systems and Quantity. Since CCSS and Achieve the Core do not provide brief descriptions of domains, we worked with a curriculum specialist to write domains' descriptions.

Within standards.jsonl, each line is a standard, sub-standard, cluster, domain, or grade level:

{
  id: '', # e.g. 'K.OA.A.1'
  description: 'description of standard from achieve the core',
  source: 'Achieve the Core',
  level: '', # one of Grade, HS Category, Domain, Cluster, Standard, Sub-standard
  cluster_type: '', # e.g. major cluster, additional cluster, minor cluster
  aspects: [], # a list containing items such as "Application", "conceptual understanding", "Procedural Skill and Fluency"
  parent: '',
  children: [],
  connections: {''progress to': [], 'progress from': [], 'related': []} # standard-level Achieve the Core connections
  modeling: # True or False depending on whether the standard is a "modeling" standard
}

After downloading each file, you can load them:

import json
with open('domain_groups.json', 'r') as infile:
    domain_groups = json.load(infile)
    print(domain_groups.keys()) # should print the keys of this dictionary

with open('standards.jsonl', 'r') as infile:
    for line in infile:
        this_standard = json.loads(line)
        print(this_standard['id']) # should print the ID of the row in this file

Citation

@misc{lucy2024evaluatinglanguagemodelmath,
      title={Evaluating Language Model Math Reasoning via Grounding in Educational Curricula}, 
      author={Li Lucy and Tal August and Rose E. Wang and Luca Soldaini and Courtney Allison and Kyle Lo},
      year={2024},
      eprint={2408.04226},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2408.04226}, 
}

Dataset Card Contact

[email protected]