Evaluates every dialogue exchange in a screenplay for subtext quality. Subtext is the gap
between what a character says and what they actually mean. Great dialogue operates on two
levels simultaneously. AI-generated dialogue almost always fails at this.

## Key Points

- Score the exchange, not individual lines
- Context matters: the same line can be subtext-rich or on-the-nose depending on what the
- Action lines between dialogue affect subtext (a pause, a look, a physical action can add a
- Do not reward obscurity for its own sake — if subtext is present but undecodable by the
- Speeches beginning with "As you know...", "Remember when...", "The thing is..."
- One character explaining to another character something both already know
- Information delivered that has no dramatic purpose in the current scene
- Long uninterrupted monologues that exist solely to inform the audience
- "I feel..." followed by named emotion
- "The truth is, I'm scared/angry/sad/confused"
- Characters narrating their own psychology
- Therapist-speak in non-therapy scenes

Dialogue Subtext Analyzer

Evaluates every dialogue exchange in a screenplay for subtext quality. Subtext is the gap between what a character says and what they actually mean. Great dialogue operates on two levels simultaneously. AI-generated dialogue almost always fails at this.

When to Use

Use when the user asks to "check my dialogue", "find on-the-nose dialogue", "improve subtext", "make dialogue less expository", "fix flat dialogue", or when a general screenplay audit flags dialogue quality issues (Module 5 findings).

The Subtext Scale

Score every significant dialogue exchange (2+ lines between characters) on this scale:

Score	Label	Definition	Example
1	On-the-nose	Characters say exactly what they mean and feel. No gap between text and intent.	"I'm angry at you because you lied to me about the money."
2	Lightly veiled	Surface meaning is close to true meaning but slightly softened or redirected.	"You know, trust is a funny thing. Once it's gone..."
3	Functional subtext	What's said differs from what's meant. Reader can decode with context.	"Did you ever fix that leak in the basement?" (meaning: are you handling your problems?)
4	Rich subtext	Multiple layers. Characters talk about one thing while the scene is about something else entirely.	A couple argues about how to load a dishwasher. The scene is about their failing marriage.
5	Deep subtext	What's said and what's meant are completely different. Requires the full scene context and character knowledge to decode. Silence and non-answers carry meaning.	Character cheerfully discusses weekend plans immediately after learning devastating news.

Scoring Guidelines

• Score the exchange, not individual lines
• Context matters: the same line can be subtext-rich or on-the-nose depending on what the audience knows
• Action lines between dialogue affect subtext (a pause, a look, a physical action can add a whole layer)
• Do not reward obscurity for its own sake — if subtext is present but undecodable by the audience, it fails

AI Dialogue Anti-Patterns

1. Exposition Dumps

Characters deliver backstory or plot information directly to the audience through dialogue.

Detection signals:

• Speeches beginning with "As you know...", "Remember when...", "The thing is..."
• One character explaining to another character something both already know
• Information delivered that has no dramatic purpose in the current scene
• Long uninterrupted monologues that exist solely to inform the audience

Example: A character delivers her entire backstory in one speech ("As you know, my father was a cop for thirty years before he was killed..."). The fix: show her touching her father's old watch while deflecting a question. The audience learns more from what she avoids saying.

2. "As You Know, Bob" Speeches

Characters tell each other things they both already know, purely for the audience. Detect via "You remember that...", recapping witnessed events, or explaining relationships both characters are in. Fix by converting to conflict — if known information must be discussed, give the characters a reason to disagree about it.

3. Feeling Monologues

Characters who announce their emotional state directly.

Detection signals:

• "I feel..." followed by named emotion
• "The truth is, I'm scared/angry/sad/confused"
• Characters narrating their own psychology
• Therapist-speak in non-therapy scenes

Flag and suggest: Replace with behavior. Show the emotion through action, deflection, or dialogue about something else entirely.

4. Identical Voice

All characters use the same vocabulary, sentence structure, and emotional register.

Detection signals:

• Swap character names and the dialogue still reads identically
• All characters use complete sentences and proper grammar
• No one interrupts, stammers, deflects, or refuses to answer
• Everyone is equally articulate about their feelings

Flag and suggest: Audit each character's voice profile: education level, regional speech patterns, verbal tics, what topics they avoid, how they handle conflict (fight/flight/freeze manifested as verbal behavior).

5. Ping-Pong Dialogue

Characters take turns delivering equal-length, fully formed thoughts with no overlap, interruption, or asymmetry.

Detection signals:

• Every speech is 2-3 lines
• Perfect alternation with no interruption
• Each speech is a complete thought
• No character dominates or withdraws

Analysis Procedure

Step 1: Build the Exchange Map

Read the full script and identify every dialogue exchange (scene-by-scene). An exchange is an unbroken conversation between two or more characters, including any action lines within it.

Step 2: Score Each Exchange

For each exchange, assign a subtext score (1-5) and note:

• What the characters are literally saying (text)
• What they actually mean (subtext, if present)
• What the audience learns that the characters don't say (dramatic irony, if present)
• Whether the exchange advances plot, reveals character, or both

Step 3: Identify Patterns

• Calculate the script's average subtext score
• Identify the weakest exchanges (all 1s and low 2s)
• Flag the anti-patterns listed above with specific line citations
• Note which characters consistently have the flattest dialogue

Step 4: Generate Rewrite Suggestions

For every exchange scoring 1 or 2, provide a rewrite suggestion that:

• Preserves the same information transfer (audience still learns what they need to)
• Adds at least one subtext layer
• Maintains character voice
• Does not make the scene longer (ideally shorter)
• Uses action lines, pauses, and non-answers as tools

Output Format

Report the average subtext score, per-exchange scores, and a breakdown of anti-pattern occurrences by type and scene. For each exchange scoring 1 or 2, include: the scene reference, what is said, what subtext is missing, which anti-pattern applies, and a suggested rewrite. End with a character voice distinctiveness table and anti-pattern summary counts.

Calibration Notes

• A well-written feature script averages 3.0-3.5 on the subtext scale
• Comedy allows more on-the-nose dialogue for joke delivery (adjust expectations)
• Kids animation (ANIM-K) operates at lower subtext levels by design — age-appropriate directness is not a flaw
• Documentary narration is expository by nature — score interview dialogue, not narration
• A script with 0% score-1 exchanges is unrealistic; some information must be direct
• The target is not score-5 everywhere — that would be exhausting and opaque

Anti-Patterns for This Skill

• Rewarding obscurity. Subtext that the audience cannot decode is not good subtext.
• Ignoring genre conventions. A courtroom drama has more direct dialogue than a marriage drama. Adjust expectations by genre.
• Penalizing comedic directness. Comedy often works by saying the quiet part loud. Score comedic intent separately from dramatic subtext.
• Over-rewriting. Suggestions should be minimal and targeted. Do not rewrite the entire scene — show the principle on the weakest lines.