Inference Questions: What Was Meant, Not Said
Speakers rarely announce their feelings. They complain about a bus, hesitate before agreeing, or answer a different question from the one asked, and you are expected to draw the conclusion.
Recognising an inference question
Three question types run through this section: general meaning, which is about the clip as a whole; specific information, which is about one stated fact; and inference, which asks for a conclusion that was never stated directly. Every question you meet is one of the three, and knowing which one you are looking at changes how you search.
Inference questions announce themselves in the wording. They carry probability language, such as probably, most likely, or would suggest. They ask about feelings and attitude: the man sounds something about the new schedule. They ask what someone would agree with, or what they will do next.
That is a useful tell because it tells you immediately not to search your notes for a matching fact. There is no line in the audio that says what you are being asked for. Searching for one wastes the seconds you need and tends to produce a specific-information distractor, which is a true statement that answers a different kind of question.
Instead, an inference question sends you to a different question: what did the speaker say, and why would someone say it that way?
Feelings arrive indirectly
This is the central design fact behind inference items. People in the clips do not say 'I am frustrated'. If they did, the question would stop being an inference item and become a specific-information one, so the test cannot allow it.
What they say instead is 'I cannot believe the bus is late again', and the frustration lives in the word again, in the choice of cannot believe, and in the intonation. What they say is 'I suppose we could try that', and the reluctance lives in suppose and in the hesitation before it.
So the evidence for an inference question is always present in the audio; it is simply carried by word choice, emphasis and rhythm rather than by content. Candidates who listen only for propositional content hear the sentence and miss the attitude entirely.
There is a short list of carriers worth listening for. Repetition and the word again. Hedges such as I suppose, I guess, maybe, if you think so. Understatement, as in that is not ideal. Delay, where a speaker answers a question after a noticeable pause. Topic change, where someone responds to a suggestion by talking about something else, which is almost always a refusal.
The evidence test
The rule that keeps inference answers honest is simple to state and requires discipline to apply: you must be able to point to the thing in the audio that supports your answer.
Not a general impression of the speaker. Not what a person in that situation would plausibly feel. A specific moment, a particular phrase, a hesitation you actually noticed. If you cannot name it, you are not inferring, you are inventing, and the option you are drawn to was probably written for exactly that mistake.
This test is what separates the two failure modes. Under-reading means answering only from stated content and choosing the flat, literal option that misses the attitude. Over-reading means constructing a psychology for the speaker that the clip does not support.
Over-reading is the more common problem among strong candidates, because good listeners naturally build a rich model of a conversation, and a rich model generates conclusions the audio never licensed. The evidence test is the check on that.
Before choosing an inference answer, name the exact phrase, hesitation or word choice that supports it. If you cannot point to one, you are constructing a psychology the clip did not give you, and that is the mistake the distractor was written for.
Prediction questions
A distinct subtype asks what a speaker will probably do next, or what will most likely happen. These are answerable and they are answerable from a narrower range of evidence than candidates assume.
The evidence is almost always the last thing decided. Conversations in Parts 1, 2, 3 and 5 end by resolving something, and the prediction question is usually asking you to carry that resolution one step forward. If they agreed she would phone the office in the morning, the prediction is about phoning the office.
This is why the closing seconds of a clip matter so much, and why attention dropping at the end is expensive. The prediction item and the general-meaning item both depend disproportionately on material that arrives when a candidate has started to relax.
The distractors here are usually options that were genuinely discussed but not chosen, and options that extend the decision further than the speakers did. If they agreed to phone, the answer is not that she will visit in person, even if visiting was mentioned as a possibility earlier.
On a would-most-likely-agree question, write the speaker's position in one short sentence from your notes before reading the options. Reading the options first lets them shape your summary, which is how the overstated distractor gets in.
Would most likely agree
Part 5 and Part 6 add a harder variant: which statement would this speaker most likely agree with. You are not being asked what they said, you are being asked to apply their position to something new.
Work it in two steps rather than one. First, from your notes, state the speaker's position in your own words as one short sentence. Do this before reading the options, because doing it after reading them lets the options shape your summary. Then test each option against that sentence.
The options in this type are usually a plausible spread: one that matches the position, one that matches a different speaker's position, one that overstates the position into something more absolute, and one that is true of the discussion generally but not attributable to this speaker.
The overstatement is the one to watch. A speaker who said a proposal was worth trying does not agree that it should be adopted immediately, and a speaker who raised a cost concern does not necessarily oppose the whole plan. Strength of commitment is preserved in the correct option and inflated in the distractor.
Training the skill
Inference is trainable, but not by doing more multiple-choice questions, which mostly tells you whether you got it right rather than why.
The useful drill is the attitude pass. Take a conversation you have already listened to for content and play it again with a single question in mind: how does each speaker feel about the thing being discussed, and what exactly tells you? Write the feeling and the evidence side by side. If the evidence column is empty, that is the finding.
A second drill targets the carriers directly. Play a clip and write down only the hedges, the repetitions and the pauses. Ignore content entirely. This feels strange and it does something valuable, which is to make the attitude signals audible as a layer separate from the words.
Over weeks, what changes is not that you reason better about the clips but that you notice more while they play. Inference on a single-pass test is largely a perception skill wearing a reasoning skill's clothing.
Listen for topic changes in response to suggestions. When someone answers a proposal by talking about something else, that is a refusal, and it is one of the most consistent attitude signals in conversational audio.
Practise it
Five questions on reading attitude and drawing conclusions the audio supports.