Week 3: Payload and Envelope

“We should get coffee sometime.”

Is that a request to schedule coffee? If you’ve replied “sure, Thursday at 10?” and watched the other person’s face do something complicated, you’ve encountered the problem. The same line can be an invitation or a way to say you’d like to stay on friendly terms.

The specification the protocol claims to follow#

The closest thing the protocol has to published documentation was written by the philosopher Paul Grice. He proposed a Cooperative Principle: make your contribution “such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange”. Under it sit four categories of maxim (Grice, 1975):

Read that as a spec, and the majority’s protocol fails it constantly. “We should get coffee sometime” gives no time, no place and no commitment. It is less informative than it would need to be if it were a plan.

Grice saw this coming. In the same passage, he notes that there are other maxims, “such as ‘Be polite’”, that also generate meaning beyond what is said. And he says that stating his maxims as though talk existed to exchange information is “too narrow”.

So the protocol has a published spec, and a known exception the author flagged himself: politeness is allowed to override the other four. Nobody told you which one wins.

Payload and envelope#

Two terms for the rest of the semester.

• The payload is what the speaker needs you to understand or do.
• The envelope is how it’s wrapped: the words chosen instead of the payload, and the maxim those words bend to get there.

Grice ties his maxims to what he calls conversational implicature: meaning that gets across without being said. The course’s working version is this. When a sentence visibly bends a maxim, assume the speaker is still being cooperative, and ask why a cooperative person would bend that one. The answer is usually the payload.

Why anyone sends an envelope#

If the payload is the point, why not send it bare?

Pinker, Nowak and Lee give three reasons (Pinker et al., 2008). First, an indirect request offers plausible deniability: a cooperative listener can accept it, and an unwilling one can decline without either of you having to acknowledge a refusal. Second, talk does two jobs at once. It carries information, and it negotiates what kind of relationship two people have. A bare request can renegotiate the relationship when all you wanted was the salt. Third, an explicit statement becomes common knowledge between you, in a way a hint never does, even when both of you knew exactly what was meant.

That last reason is why “we both knew” and “you said it” are different events under this protocol.

When does an envelope appear?#

“People are indirect to be polite” is true and nearly useless. It doesn’t tell you when to expect one.

Lee and Pinker built a model that does. Their strategic speaker goes indirect when they are uncertain of whether the hearer is cooperative or antagonistic (Lee & Pinker, 2010). Their worked case is deliberately unsocial, which is why it’s easy to see: bribing a policeman who might be corrupt or honest. A veiled offer can be taken up by the first and ignored by the second. The briber survives either outcome. A direct offer is only safe if you already know which officer you’re talking to.

Ordinary life has the same shape with emotional penalties instead of legal ones, whenever what you want is quietly ruled out by the kind of relationship you’re in.

Ask what is this person not sure of? The next section adds a second question: what are they keeping off the record?

In each example, the sender risks an answer they can’t predict. The envelope gives them a way to retreat.

The part that still works when you both know#

Here’s the clause that makes the model more than politeness in a lab coat. Even when the hearer’s willingness is known, Lee and Pinker argue, indirectness still buys something: a higher-order deniability that preempts certainty, gossip, and common knowledge of the request.

Even when both of you understand it, the request itself remains unsaid. The sender can dispute what the hint meant.

That is why “we both knew” and “you said it” are different events under this protocol, and it is worth knowing for the moments when someone insists that nothing was ever asked of you. Under this protocol, that can be literally true and completely false at the same time.

In the room — find what stays unsaid (8 min)#

Four minutes, in pairs. Use the four envelopes in the table above: coffee, “no worries”, the weekend and an idea for the group. For each, suggest what the sender might not know, or what either person might prefer to leave off the record. Do not assume uncertainty is the only reason to hint.

Two minutes. Compare with the pair next to you. Where you disagree, write down both possible payloads.

Two minutes, the room. One line, argued out loud.

There isn’t a key at the back. If there were, the protocol would be documented, and we would not be here.

The decoder#

Worked examples, each checked against the principle above. The entries are the course’s, not a research table; test them in the lab and in your journal.

Decoding from both ends#

If you decode by reading the words, the table is a lookup until the pattern becomes automatic, and it may never need to. When a sentence might be an envelope and the stakes matter, ask for the payload. Asking isn’t an accusation if you take the blame for the ambiguity:

“Do you mean coffee as a plan, or as a nice thing to say? Either’s fine. I just want to know whether to put it in my calendar.”

If you decode without noticing, the table is you, slowed down. Your exercise this week runs the other way: send one payload bare, to someone who would rather have it that way, and notice what it costs you to do it.

Sending a payload bare is always valid. Under Pinker et al.’s account, it also changes the relationship a little, because it creates common knowledge where a hint wouldn’t. That is a cost, not a mistake. Sometimes it’s worth paying. You can lower it by saying you’re choosing directness: “I’m going to be literal, because I’d rather not get this wrong.”

In the room — three readings, three replies (10 min)#

These cases are fictional. The table is a prompt bank, not a decoder key.

Two minutes, alone: mark what is said and what you infer. Which later turn, if any, checks the inference? Three minutes, in pairs: write one reply to C that makes room for either an available Tuesday or a hesitant one. Three minutes: swap the reply with another pair; they name the options it leaves open. Two minutes, the room: compare two replies and name what is still unknown.

A’s follow-up confirms a request to close the window. B has a stated reason for asking about temperature. C remains unresolved: “Tuesday is bad” is one possible reading, not a fact in the log. “Is Tuesday a yes, or should we find another day? Either works for me” checks it without first declaring what the sender meant.

Who is actually wrong when a hint misses#

Keysar and Henly had people say deliberately ambiguous sentences while trying to convey one specific intention, then asked them to estimate whether the addressee had got it. They overestimated, consistently: they expected to be understood more often than they were (Keysar & Henly, 2002).

Overhearers who had been told the speaker’s intention did not make the same error. So the overestimation isn’t a general fact about watching a conversation. It comes from having been the one who spoke.

The authors’ explanation: a speaker monitoring their own utterance does not act as an unbiased observer. They underestimate how ambiguous their sentence was, and overestimate how much their own extra cues made the intention clear. The abstract calls this a systematic source of miscommunication in ordinary conversation.

Ask the listener what they understood. The sender’s confidence cannot tell you that.

If you send envelopes: when it matters, check what the listener understood. Send a second one that is plainer if needed, and don’t treat having to as a failure on their part.

If you miss them: the sender may have overestimated how clear they were. You can question the hint and still want the decoder. Wanting to get better at something is not an admission that the whole problem was yours.

But is the decoder missing?#

A fair question to ask at this point, if you’ve been told all your life that you don’t get hints: is this the part that’s absent?

Chevallier, Wilson, Happé and Noveck tested one well-studied inference. From “John or Mary will come”, most listeners infer that not both will — although the words permit both. They predicted, on the basis of the existing literature and its reported pragmatic deficits, that autistic participants would make fewer of these inferences than matched controls.

They report the result as contrary to expectations: performance did not differ between the groups (Chevallier et al., 2010).

One experiment tested one inference type in two matched groups. It does not settle wider questions about pragmatics.

In the room — send it twice (7 min)#

Three minutes, alone. Write one request as an envelope. Use a real one you are willing to share, or the fictional request “Please close the window” from Thursday’s bank. Include the setting.

Two minutes. Underneath, write the same request bare. No softening, no “no worries if not”, no exit ramp.

Two minutes, in pairs. Swap and read only the envelope. Your partner writes what they think the payload is. Then you show them the bare version.

If the bare version feels unsendable, write down why. Return to that cost in week 6.

Over-decoding#

The failure the table doesn’t show is hearing an envelope where there’s nothing inside. “Is it cold in here?” is sometimes just a question about the temperature. Treating every sentence as a hint is as nonconforming as treating none as one, and more tiring.

The course’s rule of thumb: decode when a maxim is visibly bent, and the listener is plainly expected to do something about it. Otherwise, take the words at face value, and let the speaker send a second, clearer envelope if they need to.

In the room — the over-decoding audit (6 min)#

Three minutes, alone. Write down two times this week you suspected a sentence was an envelope. For each, answer: was a maxim visibly bent, and were you plainly expected to do something?

You can use the fictional temperature and Tuesday cases above instead. Label them as practice; they are not observations for your journal.

Three minutes, in pairs. Read one out. Your partner’s only job is to say whether the face-value reading also works.

If the face-value reading works and nothing was expected of you, you may have over-decoded. Note how much time you spent on it, and whether the other person had any reason to notice.

What this week added to the vocabulary#

Payload, envelope, plausible deniability and over-decoding — all four are in the glossary and all four come back. An envelope can manage uncertainty or leave a request off the record. Senders can overestimate how clear they were.

Next week: the gaps between turns, and what each length of silence gets read as.

If time remains#

When does leaving a request unsaid give both people room to choose, and when does it leave one person doing all the decoding?

Optional discussion. Finish by 15:55.