Skip to content

data

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

data

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
Sample
Sample
 
 
260331-Arguinas-Test-v2.json
260331-Arguinas-Test-v2.json
 
 
260331-Arguinas-Train-v2.json
260331-Arguinas-Train-v2.json
 
 
README.md
README.md
 
 

README.md

Arguinas Dataset

Dataset Splits

Split File Count
Train 260331-Arguinas-Train-v2.json 2,934
Test 260331-Arguinas-Test-v2.json 241
Total 3,175

Format

Each file is a JSON array of objects. Every object represents a single debate argument and its argument reconstruction. The argument reconstructions are generated by Claude Sonnet 4.5 based on our Generalized Automatic Argument Reconstruction (GAAR) framework.

Top-Level Columns

Column Type Description
messages array Chat-format conversation with two entries: a user message containing the debate topic, background, argument text, and instruction, and an assistant message containing the argument reconstruction (premises + conclusion).
title string The debate topic / motion title.
background string Background context for the debate topic. "None" when no background is provided.
argument string The raw argument text to be reconstructed.
fallacy_info object | null Fallacy annotation when the argument contains a fallacy; null otherwise. See sub-columns below.
sections object Detailed reconstruction pipeline outputs broken into four sub-sections (see below).
explicit_premises array[string] List of explicit premises in reconstruction
implicit_premises array[string] List of implicit premises in reconstruction

messages Structure

Each entry in messages is an object with:

Field Description
role Either "user" (input prompt) or "assistant" (model response).
content The full text content for that role.

fallacy_info Sub-Columns (when non-null)

Field Type Description
type string Category of the fallacy — either "formal fallacy" or "informal fallacy".
rationale string Explanation of why the argument contains the identified fallacy.

sections Sub-Sections

sections.reconstruction

Field Type Description
premises string Natural-language premises (P1, P2, …).
intermediate_conclusions string Intermediate conclusions derived from subsets of premises (IC1, IC2, …).
conclusion string The final natural-language conclusion of the argument.
definition string Symbol definitions mapping natural language to first-order formal logic.
formalized_premises string Premises expressed in formal logic notation.
formalized_intermediate_conclusions string Intermediate conclusions in formal logic notation.
formalized_conclusion string Final conclusion in formal logic notation.

sections.check_validity

Field Type Description
necessary_formalized_premises string Python dictionary mapping premise labels to their Z3-compatible formal expressions.
final_formalized_conclusion string The conclusion expressed for Z3 validity checking.
z3_program string Full Z3 Python program that checks deductive validity and finds minimal premise sets.
validity string Result of the validity check — "valid" or "invalid".
valid_formalized_premises array[string] List of formalized premises that are necessary for the valid argument.

sections.streamlined

Field Type Description
valid_premises array[string] Final curated list of natural-language premises in the valid reconstruction.
valid_conclusion string Final natural-language conclusion of the valid reconstruction.

sections.check_faithfulness

Field Type Description
faithfulness boolean Whether the reconstruction faithfully represents the original argument.
feedback_faithfulness string Detailed evaluation explaining the faithfulness judgment.