Skip to content
AL | Apatheia Labs
A

Argumentation

Toulmin Structure Analysis

Deconstructs claims into their logical components using the Toulmin model. Identifies unsupported assertions, missing warrants, and logical gaps.

Claim ExtractionWarrant AnalysisGap DetectionRebuttal MappingStrength Scoring

The Problem

Claims are presented as fact with no visible evidence chain. "The father has anger management issues" appears in a professional report — but where is the evidence? What connects observation to conclusion? The Toulmin model exposes the skeleton of every argument: claim, grounds, warrant, backing, qualifier, and rebuttal. When components are missing, the argument is unsupported.

How It Works

  1. 1Identify assertions requiring support
  2. 2Extract grounds (evidence) for each claim
  3. 3Evaluate warrant (logical connection) validity
  4. 4Find claims with missing grounds or implicit warrants
  5. 5Rate argument quality from evidence to claim

Inputs

  • • Document corpus
  • • Claim extraction
  • • Evidence mapping

Outputs

  • • Argument maps
  • • Strength scores
  • • Gap identification report

What You Get

CLAIM: "The father has anger management issues"
  Source: Doc 23, p.7, para 3 | Author: Social Worker B

GROUNDS (evidence): [NONE FOUND in document corpus]
WARRANT (reasoning): [MISSING]
BACKING (warrant support): [MISSING]
QUALIFIER (certainty level): Stated as fact (no hedging language)
REBUTTAL (counter-evidence): [NONE — no alternative explanations considered]

STRENGTH SCORE: 0.08 / 1.0

Finding: Evaluative claim stated as established fact with no identified evidential basis, no reasoning chain, and no consideration of alternatives.

Works With

KContradiction

Identifies claims that conflict across documents — the Argumentation Engine then assesses which version has stronger evidential support.

BBias Detection

Uses argument strength data to determine whether weak arguments systematically favour one direction.

LAccountability

Consumes argument maps to identify where professionals made claims that violate evidence-based practice standards.

Use Cases

Expert report analysis

Deconstructing a psychological assessment or expert witness report to identify which conclusions are evidence-based and which are unsupported professional opinion presented as fact.

Court judgment deconstruction

Mapping the logical structure of judicial reasoning to identify where findings of fact rest on contested evidence, hearsay, or assumptions not tested at trial.

Policy document evaluation

Assessing whether institutional policies and recommendations are grounded in cited research and evidence, or whether they rest on unstated assumptions.

Technical Approach

  • Claim extraction using fine-tuned classification models that distinguish propositional claims from factual observations, procedural descriptions, and hedged statements
  • Toulmin decomposition maps each claim to the six Toulmin components, identifying which are present, absent, or implicit
  • Gap detection prioritises findings by severity — a conclusion stated as fact with no grounds scores lower than a hedged opinion with partial evidence
  • Strength scoring weights completeness, evidence quality, warrant soundness, and acknowledgment of limitations into a single 0.0–1.0 metric