HOW IT WORKS
9 forensic linguistic analysis layers running in parallel. No black box — every decision is explainable.
THE ANALYSIS PIPELINE
Tense Consistency
Weight: 25%Verb phrases are extracted and tense mapped across every sentence. Unexplained shifts from past to present mid-narrative indicate the speaker is constructing rather than recalling events.
Agent Deletion
Weight: 15%Passive voice constructions remove the actor from sentences. When over 40% of sentences use passive voice, the speaker is likely avoiding ownership of their actions.
Pronoun Consistency
Weight: 10%Genuine personal accounts maintain consistent first-person narration. Dropping 'I' or shifting to 'we' signals psychological distancing from the described events.
Lexical Diversity
Weight: 12%Fabricated stories reuse a limited vocabulary. A Type-Token Ratio below 0.6 on texts of 30+ words suggests rehearsed or constructed language.
Negation Clustering
Weight: 15%Multiple negations in a single sentence reveal defensive thinking. Preemptive denials — denying things never accused — are a particularly strong deception signal.
Narrative Structure
Weight: 15%Truthful accounts have three parts: context before the event, the event itself, and aftermath. Missing prologue or epilogue suggests a fabricated account with no real memory to draw from.
Information Density
Weight: 10%Liars over-detail irrelevant parts while being vague about the central event. VERITY measures word distribution across narrative sections to detect this imbalance.
Cognitive Load
Weight: 12%Phrases like 'honestly' and 'to be frank' signal the opposite. Memory hedges and veracity emphasis words are catalogued and density measured per sentence.
Contradiction Detection
Weight: 15%Subject-verb pairs are extracted and compared across the entire text. Contradictory claims about the same subject are flagged automatically.
HOW SCORES ARE CALCULATED
WHAT VERITY CANNOT DO
VERITY does not verify facts. It analyzes HOW something is said, not WHAT is said.
Rule-based analysis has limits. Subtle deception by practiced liars may score low.
Context matters. Some linguistic patterns are cultural or stylistic, not deceptive.