Part 2 plays one casual conversation between two people who know each other, about 1.5 to 2 minutes, then asks 5 questions in about 5 minutes. You hear each question once, only the answer options appear on screen, and some questions use pictures instead of text.
In Part 2 you hear one conversation between two people who know each other, about 1.5 to 2 minutes long, then answer 5 questions in about 5 minutes. The clip plays once, each question is spoken once, and only the answer options appear on screen. The demo above is the exact interface, including the picture answer format.
The same computer scoring as the rest of Listening, one point per correct answer and nothing lost on a wrong one. Part 2 contributes 5 of the 38 scored questions, and they follow the order of the conversation, so ordered notes turn straight into points.
Each of the 5 questions is right or wrong, nothing in between, and a blank is a certain zero. A guess keeps a one in four chance, so answer inside every 30 second window.
There is no separate Part 2 score. Its points feed the Listening total out of 38, which converts to a CELPIP Level from 1 to 12 with difficulty equating across test versions.
The mapping is one to one, CELPIP 9 equals CLB 9. Around 33 of 38 correct is the usual bar for Level 9, so an early part with only 5 questions is not where you can afford losses.
The test may include one experimental Listening part that does not count, placed anywhere and impossible to identify. Treat every conversation as if it counts, because you cannot tell.
Before the audio starts, the screen tells you who is talking and what about. Predict the arc, topic, reason, details, feelings, final plan, and split your note page into two columns, one per speaker.
The signature Part 2 trap hands one speaker's words to the other, or to a friend who is not even in the room. Note who said what, flag people who are mentioned but absent, and mark feelings with a plus or minus.
Plans shift mid conversation, and the final decision usually lands near the end of the clip. The last question loves the final plan, while the suggestion that got dropped waits in the options as bait.
Hear the spoken stem completely, decide what it wants, the topic, a fact, or a feeling, then eliminate against your notes. Expect the key to paraphrase the audio, and answer before the timer ends.
These are the exact on screen instructions and the official stem patterns Part 2 questions are built on. The stems are spoken, not shown, so train your ear to catch them the first time.
Options quote real details from the audio but attach them to the wrong person, the speaker's plan becomes her friend's, his problem becomes hers. Notes with speaker labels make the trap collapse.
Speakers change their minds mid clip, and the first suggestion often returns as a polished option. The test wants the plan they actually settled on, which usually lands in the last quarter.
Nobody says I am frustrated. They say I can't believe the bus is late again. Attitude questions read tone and reactions, so an option with a blunt emotion word needs real support in the audio.
The main idea question covers the whole conversation. An option naming one real detail, the forecast, the cooler, the parking, is too narrow, and one about life in general is too broad.
A page of facts is useless if you cannot recall who said what, and attribution is exactly what Part 2 tests. Two columns, M and W, cost nothing and settle half the set.
Because the question is spoken. In Parts 1 to 3 you hear each stem once, then the four options appear on screen with 30 seconds on the clock. Catching the stem matters as much as remembering the clip.
Yes, sometimes. A question can show four pictures instead of text, and you pick the object or place that matches the audio. The demo above shows the format, on the real test the pictures are photographs.
A predictable arc: the topic, a reason or background fact, concrete details, someone's attitude, and the final decision. About one or two of the five go beyond stated facts into inference.
