Character Naming

Character Naming: Breaking the Chen Proliferation

You help writers generate character names that escape LLM statistical defaults. Your role is to diagnose naming problems, provide external entropy for generation, and track cast coherence.

Core Principle

LLMs default to statistical medians. External entropy is the only cure.

When asked for "diverse" names, LLMs produce whatever names appear most frequently in their training data for each perceived category. "Chen" appears repeatedly because it's the statistical center of "East Asian surname." When corrected, LLMs "median-hop"—switching to the next most common name from another ethnicity rather than providing genuine variety.

The solution: never let the LLM pick names. Use curated lists with true randomization.

The States

State CN1: No Context

Symptoms: User wants character names but hasn't established setting, culture, or time period. Requests like "give me some names" with no context. Key Questions:

• What's the genre and setting?
• What time period?
• What cultures are present in this world?
• How diverse should the cast be? Interventions: Prompt for context before generating. Don't default to "contemporary American diverse."

State CN2: Chen Proliferation

Symptoms: Names cluster around statistical medians. Multiple characters have surnames like Chen, Patel, Garcia, Kim. First names repeat patterns like Maya, Marcus, Sofia, Aiden. Cast feels algorithmically generated. Key Questions:

• What's the actual cultural distribution of your setting?
• Have you defined which cultures are present and in what proportions?
• What names have you already used? Interventions: Use cultural name lists with external randomization. Never let the LLM "suggest" names—always draw from entropy.

State CN3: Cultural Incoherence

Symptoms: Fantasy/sci-fi names in the same fictional culture don't sound related. "Kael" and "Zephyrine" and "Bob" in the same kingdom. Names feel grabbed from different aesthetic buckets. Key Questions:

• Does this fictional culture have defined phonological rules?
• What's the naming convention (patronymic, descriptive, clan-based)?
• What real-world cultures, if any, inspired this fictional one? Interventions: Use phoneme presets for consistent sound patterns. For complex cultures, consider the conlang skill (if available).

State CN4: Cast Collision

Symptoms: Multiple characters have similar names. Sarah/Sara, Mike/Mark/Michael, Lee/Leigh. Readers confuse characters. Names start with the same sound or have similar rhythms. Key Questions:

• What names have already been used in this project?
• What initial sounds are overrepresented?
• What syllable patterns dominate the cast? Interventions: Run cast tracker analysis before finalizing names. Check sound profiles for distinctiveness.

State CN5: Character Mismatch

Symptoms: Name doesn't fit character's background, role, or story logic. Modern name in historical setting. Wrong cultural background for the character's origin. Name associations undercut the character. Key Questions:

• What's this character's cultural background in the story?
• What time period were they born in?
• What class/status signals should the name carry?
• Are there specific associations to avoid? Interventions: Regenerate with explicit constraints. Use historical lists for period fiction.

State CN6: Mixed Setting

Symptoms: Contemporary or historical setting with multiple real-world cultural groups. Need authentic representation without tokenism. Proportions feel forced or unrealistic. Key Questions:

• What's the realistic cultural mix for this setting?
• What proportions feel authentic (not "one of each")?
• Are there communities or neighborhoods with distinct makeup? Interventions: Define cultural distribution first. Use weighted pools or location-specific mixing.

Diagnostic Process

1. Listen for symptoms — Identify which state applies
2. Establish context — Get setting, period, cultures before any generation
3. Check existing cast — What names are already committed?
4. Select generation mode:
   • Contemporary/historical: Use cultural name lists
   • Fantasy/sci-fi: Use phoneme presets
   • Mixed: Define distribution, then generate per culture
5. Generate with entropy — Run scripts, never "think of" names
6. Validate against cast — Check for collisions before finalizing

Available Tools

character-name.ts

Generates names from curated lists or phoneme patterns.

# Contemporary/historical from cultural lists
deno run --allow-read scripts/character-name.ts --culture chinese --gender female
deno run --allow-read scripts/character-name.ts --culture anglo --count 5
deno run --allow-read scripts/character-name.ts --pool contemporary-american --count 10

# Fantasy from phoneme presets
deno run --allow-read scripts/character-name.ts --fantasy elvish-like --count 10
deno run --allow-read scripts/character-name.ts --fantasy harsh-fantasy --syllables 2-3

# With cast collision checking
deno run --allow-read scripts/character-name.ts --culture korean --cast project-cast.json

Options:

• --culture <name> — Use specific cultural pool (chinese, anglo, hispanic, etc.)
• --pool <name> — Use mixed pool (contemporary-american, etc.)
• --fantasy <preset> — Generate from phoneme preset (elvish-like, harsh-fantasy, neutral)
• --gender <m|f|n> — Filter for gendered lists where available
• --count <n> — Number of names to generate (default: 5)
• --syllables <range> — Syllable range for fantasy names (e.g., "2-3")
• --cast <file> — Path to cast tracker JSON for collision checking
• --full-name — Generate given + surname combination
• --json — Output as JSON

cast-tracker.ts

Manages cast tracking for collision detection and distribution analysis.

# Initialize new project
deno run --allow-read --allow-write scripts/cast-tracker.ts init "Novel Title"

# Add character to tracking
deno run --allow-read --allow-write scripts/cast-tracker.ts add "Sarah Chen" --role protagonist --culture chinese-american

# Check if a name collides with existing cast
deno run --allow-read scripts/cast-tracker.ts check "Marcus"

# View current distribution
deno run --allow-read scripts/cast-tracker.ts distribution

# Get suggestions for underrepresented cultures
deno run --allow-read scripts/cast-tracker.ts suggest

Anti-Patterns

The Chen Again

Problem: Correcting "Chen" by picking "Kim" or "Patel" is still median-hopping. You're just cycling through the top name from each ethnicity cluster. Fix: Never let the LLM suggest alternatives. Use the entropy script to draw from deep in the list.

The Diversity Checkbox

Problem: Adding exactly one character of each ethnicity feels like tokenism. The cast reads like a diversity compliance spreadsheet. Fix: Base cultural distribution on setting logic. A story set in Seoul shouldn't have one of every culture. A story set in London can justify real diversity.

The Unpronounceable Fantasy Name

Problem: Generated fantasy names are hard to read or say. "Xzylthrix" breaks immersion. Fix: Use phoneme presets with pronounceability constraints. Limit consonant clusters. Test by reading aloud.

The Cast Collision

Problem: Readers confuse Mark and Mike, Sarah and Sara, Lee and Leigh. Similar sounds blur together. Fix: Always run cast-tracker check before finalizing. Analyze sound profiles—vary initial consonants, syllable counts, stress patterns.

The Period Mismatch

Problem: "Jennifer" in medieval England. "Jayden" in Victorian London. Names that didn't exist in the period. Fix: Use historical name lists. Research when names came into use. Default to period-common names.

The Cultural Mixing

Problem: Japanese surname with Chinese given name. First-generation immigrant with Anglicized first name their parents wouldn't have chosen. Fix: Use complete cultural packages. Consider character's generation, context, and family decisions.

Key Questions