A weaker word, a stronger inference
When a speaker could have chosen a stronger word but uses a weaker one, listeners may infer that the stronger alternative is probably not true. Some of my neighbours signed the petition therefore usually suggests that not all did. Logically, some does not exclude all, but conversation also relies on the expectation that speakers provide enough information for the current purpose.
This inference is not part of the word’s literal meaning; it arises from the choice of word. It can therefore disappear with a change of context or be explicitly withdrawn without contradiction.
Some of my neighbours signed the petition.
Some of my neighbours signed the petition.
The usual inference is that not all the neighbours signed. Some was chosen instead of all.
The coffee is warm.
The coffee is warm.
On the warm — hot scale, the weaker description can suggest “not hot”.
She sometimes cycles to the office.
She sometimes cycles to the office.
The weaker frequency term suggests “not every day” or “not always”.
Scales: stronger and weaker alternatives
On a classical scale, the stronger alternative is more informative: for a nonempty group, all entails some, and and entails inclusive or. Examples include ⟨some, most, all⟩, ⟨possible, certain⟩ and ⟨or, and⟩. Evaluative sequences such as good — excellent and like — love can produce similar inferences, but depend on context and are not always strict logical scales.
Angle brackets ⟨…⟩ represent an ordered set of alternatives. Not any two words form a relevant scale: they must concern the same property, be relevant, and share an understood stronger alternative. Synonymy or opposition alone does not guarantee the inference.
weaker alternative X → “not stronger alternative Y”
It is likely that the board will agree.
It is likely that the board will agree.
The weaker point on ⟨likely, certain⟩: the speaker does not claim certainty.
You may park here.
You may park here.
On ⟨may, must⟩, permission can suggest “but there is no obligation”.
I like her work.
I like her work.
On ⟨like, love⟩, the speaker stops short of expressing strong enthusiasm.
How the listener reasons
The reasoning has three steps. The speaker could have used a stronger expression. That expression would be relevant to the conversation. The speaker probably knows whether it is true. The inference is that the stronger expression probably does not apply. This is one aspect of the general expectation of sufficient information; the broader theory is a separate topic.
The third step determines the inference’s strength. If the speaker is assumed to know the whole picture, a strong “not all” inference arises. If they have seen only part of the evidence, it weakens to “I don’t know whether all”: nothing is asserted about the rest.
Some of the fire doors in this block are blocked.
Some of the fire doors in this block are blocked.
If the inspector checked every door, the listener can infer that not all are blocked.
I have only inspected two flats in the block; at least some flats in the block have damp.
I have only inspected two flats in the block; at least some flats in the block have damp.
The whole building has not been inspected, and at least explicitly leaves open the possibility that other flats have the problem too.
Cancellation: a test for implicature
An implicature can be withdrawn without contradiction: Some, and in fact all, of the files can be recovered. The “not all” inference is cancelled while the literal lower bound, “at least some”, remains.
The literal assertion cannot be withdrawn without contradiction. He sometimes answers emails — in fact, never is contradictory because sometimes asserts at least some occasions. As a practical test, add “in fact, even more”. If the continuation is natural, you are cancelling an implicature; if it creates a contradiction, you have challenged the literal content.
You can cancel “not all” with stronger all, or suspend it: some, perhaps all / some, but I don’t know whether all. In fact, indeed and or even serve different functions and do not always require one scale. If not can introduce a stronger assessment or mean concessive “although not”; read the context.
Some, and in fact all, of the files can be recovered.
Some, and in fact all, of the files can be recovered.
“Not all” is cancelled, while the lower bound “at least some” remains.
The project is possible — indeed, it is already under way.
The project is possible — indeed, it is already under way.
Possible does not prevent continuing with a stronger claim.
When the inference does not arise
In some contexts, the “not stronger” inference disappears. This often happens in conditionals and patterns such as anyone who… or whenever…. The weaker expression can set a minimum condition, so the all case also falls within the rule.
When complete knowledge or relevance of the stronger alternative is not established, the usual inference can weaken or fail to arise. This is not a mechanical ban on implicatures in a particular grammatical construction; the specific context still matters.
If some of the lights fail, the backup switches on.
If some of the lights fail, the backup switches on.
The condition does not exclude all the lights failing. There is no “not all” inference in this reading.
Anyone who has tried some of her recipes will understand.
Anyone who has tried some of her recipes will understand.
The rule also covers people who tried all the recipes, so “not all” is not inferred here.
Negation that corrects the strength of a description
Negation sometimes rejects an overly weak description rather than the literal property: The meeting wasn’t merely useful — it transformed our plans. Merely explicitly signals this strengthening. Without it, context and intonation matter, because changed plans do not by themselves prove that a meeting was useful.
Similarly, not all does not mean none. Not all of the data were lost literally says only that some survived. The usual inference that some were nevertheless lost is cancellable, not compulsory meaning. Inferring none from not all is a common logical error.
The meeting wasn’t merely useful — it transformed our plans for the better.
The meeting wasn’t merely useful — it transformed our plans for the better.
Merely and for the better make the positive strengthening explicit: the restriction “only useful” is rejected.
She doesn't like jazz — she lives for it.
She does not merely like jazz — she lives for it.
The description is corrected upwards on a scale such as ⟨like, love⟩.
Not all of the data were lost.
Not all of the data were lost.
This does not mean none. Literally, some data survived; the usual additional inference is that some were lost.
Choosing words deliberately in your own speech
When a stronger result matters to the question and you know it, the weaker word may mislead. But some is not always inappropriate when all is known: if only the existence of a few cases matters, the weaker answer may be sufficient. Follow the conversation’s purpose.
To leave room for uncertainty, choose the weaker word deliberately or add an explicit qualification: at least some, if not all, I can’t speak for the rest. If listeners may wrongly infer “not all” or “not excellent”, correct that openly, for example in fact, all of them.
At least some, and probably most, of the samples passed.
At least some, and probably most, of the samples passed.
The lower bound is explicit, while the upper bound remains open.
Every one of the tests passed.
Every one of the tests passed.
Every one explicitly states complete coverage, useful when it matters whether all tests passed.
I can't speak for the other branches — some of them may differ.
I can't speak for the other branches — some of them may differ.
The speaker explicitly marks the limits of their knowledge.
Common mistakes
Incorrect:Some of the delegates arrived — in fact, none of them did.
Correct:Some of the delegates arrived — in fact, all of them did.
None contradicts literal some. All is compatible with some and cancels the usual “not all” inference. Perhaps all can instead leave full coverage uncertain.
Incorrect:The proposal is adequate, if not outstandingly.
Correct:The proposal is adequate, if not outstanding.
Both assessments of proposal need adjectives: adequate and outstanding. In this reading, if not introduces a possible stronger assessment; elsewhere it can be concessive.
Incorrect:Not all of the staff were informed, so nobody was informed.
Correct:Not all of the staff were informed, so some may still be unaware.
Not all does not equal none. Denying the upper end of the scale does not imply that the lower end is false; nobody does not follow.
Incorrect:Only some of the committee voted for it, and in fact all of them did.
Correct:Some of the committee voted for it, and in fact all of them did.
Only makes the upper limit part of the assertion, so a continuation cannot cancel it without contradiction.
Incorrect:I’ve checked some of the invoices, so checking every invoice is proved complete.
Correct:I’ve checked some of the invoices; I haven’t yet established whether every invoice has been checked.
Some does not establish that every invoice was checked. Moreover, checking something does not guarantee a positive result; that must be stated separately.