What Is a Signal?
A signal is a lexical cue associated with fear. We organize
these into six families:
Audio: scream, silence, music. Sound
cues that create atmosphere or shock through auditory elements.
Visual: dark, blood, shadow. Visual
elements that set mood or show visceral imagery.
Pace: sudden, rapid, silence. Terms
that indicate tempo, rhythm, or abrupt changes in pacing.
Threat: knife, danger, violence.
Direct dangers, weapons, or explicit threats to characters.
Setting: night, isolated, abandoned.
Environmental cues that establish location and atmosphere.
Psyche: dread, trapped, panic.
Psychological states, emotions, and mental distress.
We measure not just frequency, but effectiveness: do these words
actually spike the fear score?
Throughout the visualizations below, we use these signal
families to reveal patterns in horror storytelling: which
families generate the most fear, how signals build tension
across a film's runtime, which signals are most effective, and
how different films combine families to create their unique
horror recipes.
Methodology
Parsing: Scripts were chunked by scene
headings. AI Analysis: GPT-4o extracted
structured JSON: character stats, lexicon hits, and emotional
scoring (Fear, Tension, Sentiment).
AI Output Structure
{
"heading": "INT. ASYLUM - NIGHT",
"signals": { "audio": "scream", "setting": "dark" },
"scores": { "fear": 0.85, "tension": 0.70 }
}
Scoring Metrics: Fear, tension, and sentiment
are normalized to scales of 0 to 1, where 0 represents calm and
1 represents maximum intensity. Signals are detected when horror
lexicon terms appear in the scene text, and we analyze how their
presence correlates with spikes in fear and tension scores to
measure effectiveness. Throughout the visualizations below,
higher scores indicate stronger emotional impact.
Fear vs. Tension: Fear captures sudden shocks;
tension measures slow burning suspense. GPT-4o generates these
normalized scores by analyzing each scene's screenplay content,
horror signals, and emotional cues.
