Role-Play Practice: One Scenario at a Time, Errors at the End

Today's Goals

1. Run a scenario role-play with AI under the practice-partner brief, one scenario at a time.
2. Ask AI to summarise at most three errors by category at the end of the exchange.
3. Tell real errors apart from regional variants and style preferences.

On Tuesday 27 October, Lin practises scenario C2: arriving at the hotel in Valencia on Monday 16 November and finding the booking is for the wrong dates. Role-play is where an AI partner shines. It can play the receptionist as many times as Lin likes, adjust its speed and vocabulary, and change details so the scene never becomes a script. It is also where the brief from Day 1 earns its keep. Without it, the assistant interrupts to correct, drifts into English, and fixes things that were not broken. Today Lin runs the scene once, collects the feedback at the end, sorts real errors from regional differences, and runs it again.

Plain-Language Walkthrough

Set up one scenario at a time

A role-play works best when it has one scenario, one clear task, and a way to know whether the task succeeded. "Let's have a conversation in Spanish" gives the assistant nothing to aim for, and conversations like that drift into whatever the model finds easy to talk about. Lin sets it up precisely, after pasting the brief, the goals, and yesterday's log line:

Role-play. You are the receptionist at a hotel in Valencia. I am checking in on Monday 16 November. In your system my booking is for a single room from 15 to 17 November. Do not tell me about the mistake; read the booking back and let me notice. Stay in the role until I write "fin."

The success criterion for C2 is concrete: by the end, the booking says 16 to 18 November, and Lin knows whether anything else changed, such as the price or the room. Writing the criterion down before starting makes it obvious afterwards whether the task was completed, which is the first item on the rubric.

One scenario per session keeps the practice focused. A conversation that wanders from the hotel to restaurants to the weather practises nothing in depth, and its errors are too scattered to show a pattern.

Sign in and link WeChat to read

Sign in, then link your WeChat mini program account on the account page. Linked accounts share one free article per day. Unlock additional articles with rewarded videos in the mini program.

Scan in WeChat to open Dayward AI. Open the same course and chapter, choose the rewarded video, then refresh this page after completion.

Unlocked — refresh to readSign in / scan with WeChat →