Shared Semantics


By Jessica Lake


Communication is often described as the transmission of information.

I suspect this is backwards.

Suppose you and I wish to communicate.

We begin with a problem.

Your mind is not mine.

We possess different knowledge, assumptions, histories, goals, observations, vocabulary, and associations.

If communication required me to transmit my internal state to you, we would be in trouble.

Language is pathetically slow.

Fortunately, I don't think that's what we do.

THE HANDSHAKE

Consider two computers attempting to communicate.

Before anything useful happens, they establish whether communication is possible.

A handshake.

Human beings appear to do something similar.

  • Who are you?

  • What do you know?

  • What do we already share?

  • What vocabulary is available?

  • What assumptions are safe?

  • What are we talking about?

  • Where are we?

  • What just happened?

Much of this happens without being stated.

A glance establishes an object.

A gesture establishes direction.

The room establishes location.

The previous sentence establishes context.

Our history establishes almost everything else.

The purpose of the handshake is not to communicate the message.

It is to determine how much message will be necessary.

SYNCHRONIZATION

Once compatibility is established, we begin reducing the disparity between our internal states.

We establish enough common context to make the intended communication possible.

Location.

Time.

Participants.

Goals.

Assumptions.

Terminology.

Shared experience.

I do not need your mind to become identical to mine.

That would be impossible.

I need only synchronize the portion relevant to what comes next.

This changes the problem considerably.

The objective is not to exchange conclusions.

The objective is to establish enough common state that conclusions can be reconstructed.

COMPRESSION

Now something interesting happens.

As shared state increases, transmission decreases.

Suppose you and I are standing beside a table.

I point.

“Red one.”

Two words.

Yet an enormous amount of information has been communicated.

I did not transmit:

There is a table.

There are objects on it.

One of those objects is red.

I am directing your attention toward those objects.

I wish you to identify the red one.

Almost everything required to interpret the message was already available.

“Red one” is merely the delta.

With sufficient synchronization:

  • A gesture may suffice.

  • A glance may suffice.

  • A single word may suffice.

The receiver reconstructs the remainder from shared state.

This explains why old friends can communicate with fragments.

  • Why pointing works.

  • Why face-to-face communication is so powerful.

  • Why jargon is efficient among experts and impenetrable to outsiders.

  • Why culture, friendship, religion, tradition, profession, and shared experience can make communication remarkably economical.

They preload context.

ADDRESSES

This also changes the role of words.

A word need not contain everything it means.

It can function as an address.

Dog.

I do not transmit dogness.

If we possess sufficiently compatible semantic knowledge, the word gives you an address into your own semantic memory.

You supply most of the meaning.

The same is true of far more complicated concepts.

  • Democracy.

  • Evolution.

  • Mother.

  • War.

  • Constraint.

The word is small.

What it invokes may be enormous.

This is why synchronization matters.

If the structures at our respective addresses differ substantially, we may use precisely the same word while communicating different things.

The protocol succeeds.

The communication fails.

COMMUNICATION

A useful model therefore becomes:

1. Handshake

Establish compatibility.

Estimate state disparity.

2. Synchronization

Establish shared context.

Reduce relevant state disparity.

3. Communication

Exchange deltas.

Transmit addresses.

Signal anomalies.

That is strikingly different from imagining communication as one mind packaging meaning and shipping it to another.

Very little meaning may actually cross the channel.

Most of it already exists at both ends.

What crosses the channel tells the receiving system what to activate, what has changed, and where its state differs from the sender's.

Communication therefore becomes increasingly efficient as state disparity decreases.

LANGUAGE

And this produces one final inversion.

Language is not communication.

Language is one mechanism for communication.

Nor is language necessarily the first synchronization mechanism.

A herd of animals can synchronize without a word.

  • One freezes.

  • Others freeze.

  • One runs.

  • Others run.

  • Alarm state propagates.

  • Gesture synchronizes.

  • Expression synchronizes.

  • Posture synchronizes.

  • Proximity synchronizes.

  • Shared attention synchronizes.

Language arrives with an extraordinary advantage: it allows increasingly abstract semantic states to be synchronized across a narrow serial channel.

But the underlying problem is older.

Two autonomous systems possess different internal states.

For coordinated behavior, some relevant portion of those states must become sufficiently compatible.

Only then does the tiny message work.

Perhaps we have communication backwards.

Its great achievement is not how much information we transmit.

It is how little we need to.