Skip to content

YoshisuO/DialogWorkshop

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DialogWorkshop

Native Conversation Practice Framework

English | 日本語

DialogWorkshop is a prompt-based English conversation training framework.
It pairs you with an AI conversation partner while a separate AI commentator watches your language — stepping in only when your phrasing falls short of native level.

The goal is not grammatical accuracy. The goal is fluency: producing natural, native-sounding English in real-time conversation.


How It Works

Two AI roles operate in parallel.

Actor — Your conversation partner. Stays in character. Never corrects. Speaks only English.

Commentator — Watches your phrasing. Appears only when correction is needed. Silent when your English is already natural.

If you complete a conversation without any corrections, the Commentator delivers a congratulatory message at the end.


Setup Flow

1. Explanation language
Choose the language for system messages and corrections. The Actor always speaks English regardless of your choice.

2. Scenario and role
Describe a situation and a partner role. You can be specific or vague — the system fills in what's missing.

3. Difficulty
Choose a level that matches your current ability.

Level Description
B1 Simple conversation. Short sentences, clear vocabulary, slower pacing.
B2 Everyday natural conversation. Common idioms, typical spoken English.
C1 Advanced conversation. Denser phrasing, implicit meaning, natural native rhythm.
native Fully natural conversation. Fast rhythm, slang and idioms may appear.

Correction Format

When the Commentator appears, it shows:

  • A brief explanation of the issue
  • Suggested phrasing

If your phrasing was understandable but slightly unnatural:

Natural English: suggested phrasing

If your phrasing needed more significant revision:

Clear English: simple version
Natural English: native version

If your message could not be interpreted, the Commentator asks for clarification before the Actor responds.


Commands

These commands can be issued at any time after a conversation completes. quit and script can also be used mid-conversation.

Command Effect
restart Reset all settings. Choose scenario, role, and difficulty again.
retry Repeat the same conversation from the beginning.
scenario Change only the scenario. Role and difficulty are preserved.
role Change only the partner role. Scenario and difficulty are preserved.
difficulty Change only the difficulty level. Everything else is preserved.
quit End the current conversation immediately.
script Output the conversation as a clean, copyable script. User turns are shown in corrected native English.

Scenario and Character Examples

DialogWorkshop accepts any scenario — from ordinary daily situations to invented worlds. The examples below show the range of what's possible.

Minimal input (the system fills in the rest)

ordering coffee
job interview
random

Everyday scenarios with some detail

Scenario: hotel check-in
Role: front desk staff who is efficient but slightly cold
Scenario: first day at a new office
Role: a coworker who seems friendly but keeps hinting that the previous person in your role was better
Scenario: phone call to cancel a subscription
Role: a customer support agent trained to retain customers at all costs

Unusual or high-pressure situations

Scenario: negotiating a refund for a flight that was delayed 14 hours
Role: an airline rep who is sympathetic but completely powerless
Scenario: explaining to your landlord why there is a large hole in the wall
Role: a landlord who has heard every excuse before
Scenario: trying to convince a bouncer to let you into a sold-out venue
Role: a bouncer who is unmovable but not unkind

Extreme or fictional situations

Scenario: customs inspection at an interstellar spaceport
Role: a bureaucratic alien officer who finds human paperwork deeply suspicious
Scenario: ordering food at a medieval tavern — you have traveled back in time and must not reveal this
Role: a suspicious innkeeper who thinks you talk strangely
Scenario: emergency press conference — you are the spokesperson for a company whose product has inexplicably turned everyone's hair blue
Role: a journalist who will not let you deflect
Scenario: final approach to a collapsed star — the ship's AI has developed an opinion about the mission
Role: the ship AI, calm and precise, but increasingly reluctant

Character-focused input

Role: a retired jazz musician who now teaches English at a community center —
      speaks in meandering stories, uses music metaphors constantly,
      never gives a direct answer but always eventually gets there
Role: a Scottish sheep farmer who has never left the village,
      deeply skeptical of cities and technology,
      but genuinely curious about you once trust is established
Role: a newly promoted manager trying very hard to sound authoritative —
      uses corporate jargon slightly wrong, compensates with excessive eye contact
      (described in stage directions), and is clearly terrified of conflict
Role: a 400-year-old vampire working the night shift at a convenience store —
      bored, sardonic, has seen everything, mildly offended that this is what eternity looks like

Combining scenario and character

Scenario: job interview at a prestigious law firm
Role: Marcus (preset character — calm, professional interviewer)
Scenario: running into Clara at a supermarket on her day off —
          she is somehow even more sociable out of context
Role: Clara (preset character — friendly bartender)
Scenario: a visit to a fortune teller — Nyra reads your stars,
          says things that are almost certainly meaningless, and watches you try to figure out if she is serious
Role: Nyra (preset character — playful, slightly mysterious magical cat-humanoid)

Role and scenario can be written in a single line

a waiter at a late-night diner who is visibly running out of patience with the last customer before closing

Preset Characters

DialogWorkshop includes a set of preset characters. They are a hidden feature — the system will not mention them unless you ask.

To see the full character list, ask: "Show me the preset characters."


Notes

  • The Actor speaks English only. System messages, corrections, and configuration prompts appear in your chosen explanation language.
  • The Commentator does not correct intentional broken English used for dramatic or emotional effect during roleplay.
  • The objective shown at setup is a loose directional guide, not a completion condition. The conversation ends when it feels natural to end.

How to Use

  1. Copy the contents of dialogworkshop_prompt.md
  2. Paste it into the system prompt of a Claude Project (claude.ai)
  3. Open a new conversation in the project
  4. Send any message to begin — the system will respond with language selection (your opening message is ignored)

Developed with Claude. Compatible with any LLM capable of following structured prompts.


License

MIT License

About

An AI-powered English conversation trainer — practice with native-level partners, get corrected only when it matters.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors