Communication

Abstract

Communication is a form of behavior: the exchange of information between a sender and a receiver, which Shannon reduced to a signal passed through a noisy channel (Shannon, 1948). Human communication, however, is more than a code, because speakers mean more than they say and listeners infer intentions through shared maxims of cooperation (Grice, 1975). Conversation is a joint action in which partners build common ground, grounding each contribution until it is mutually understood (Clark, 1996). Dialogue then proceeds by interactive alignment, as interlocutors converge on shared representations at every level (Pickering & Garrod, 2004), a coordination mirrored in the coupling of their brains (Hasson, Ghazanfar, Galantucci, Garrod, & Keysers, 2012). This article surveys communication as code, as inference, and as coordinated joint action.

Keywords: communication, information theory, common ground

Two people in conversation accomplish, without apparent effort, something no engineered system fully matches: they transfer thoughts from one mind to another through a thin stream of sound, filling in what is unsaid, repairing misunderstandings on the fly, and trading turns with gaps measured in milliseconds. Communication is the behavior that achieves this. It can be described at three levels that do not compete but stack: as the physical transmission of a signal, as the inference of a meaning the signal only points at, and as the coordinated joint action by which two agents bring their mental states into register. Each level answers a different question, and cognitive psychology has a theory for each.

What Communication Is

Communication is the transfer of information from one agent to another by means of signals. The definition is deliberately broad: it covers a warning cry, a printed page, a raised eyebrow, and a spoken sentence, because all of them move information across the gap between two systems. What varies is how much structure the sender and receiver bring to the exchange. At its most austere, communication needs only a shared code and a channel to carry it. At its richest, as in human conversation, it recruits a theory of the other mind, a shared history, and a capacity to mean one thing by saying another.

Three distinctions give the term precision. First, communication is not the same as language. Language is one richly structured code that human communication exploits, but gesture, facial expression, and tone communicate without it. Second, communication is not the same as information in the engineering sense. Shannon’s theory measures how much a signal reduces the receiver’s uncertainty, but a signal can carry many bits and no meaning, or one bit and a life-changing message. Third, communication is cooperative at its root. Tomasello argues that uniquely human communication grew out of shared intentionality — the capacity to form joint goals and attend together to the same thing — so that even a prelinguistic point presupposes that two minds are working toward a common end (Tomasello, 2008). MeSH classifies the construct under behavior, placing it with the other observable, goal-directed acts rather than with the linguistic or perceptual systems it recruits.

Types of Communication

Within the Medical Subject Headings vocabulary, the descriptor Communication is a form of behavior and sits at the head of a large family of narrower descriptors. They partition the field along several unrelated axes — the code used, the medium, the social function, the technology — rather than naming subtypes of a single underlying process, so they are better read as the many facets of communication the biomedical literature indexes than as a branching taxonomy of one mechanism.

The classification is an indexing convenience, not a theory of communication. MeSH is built to organize the biomedical literature, so its branches track how research is catalogued rather than how the process works. The axes are not orthogonal: a single act of communication can be verbal and persuasive and health-related at once, so a descriptor's placement reflects where the literature files it, not a partition of the behavior itself. Links lead to the subtypes that are treated elsewhere on this site; the remaining descriptors are listed for completeness, each a different cross-section of the same broad behavior.

The Channel Model and Information Theory

The foundational formal account of communication comes not from psychology but from engineering. In 1948 Claude Shannon defined communication as the reproduction at one point of a message selected at another, and reduced it to a chain of five elements: an information source that selects a message, a transmitter that encodes it into a signal, a channel that carries the signal, a receiver that decodes it, and a destination (Shannon, 1948). A sixth element, a noise source, corrupts the signal in transit. The model’s radical move was to detach the quantity of information from its meaning: information is measured by how much it reduces the receiver’s uncertainty about which message was sent, in units of bits, where one bit is the uncertainty of a fair coin flip.

Two consequences follow directly. First, a source with more equally likely messages carries more information per symbol, because each selection resolves more uncertainty; the measure of this average uncertainty is the entropy of the source. Second, because noise destroys information, reliable communication over a noisy channel requires redundancy — extra, predictable structure in the signal that lets the receiver detect and correct errors. Natural language is massively redundant, which is why a garbled text message or a sentence heard in a noisy room is usually still readable. What Shannon’s theory deliberately leaves out is as important as what it captures: it treats the message as a selection from a set, with no account of what any message means. That gap is exactly what the psychology of communication must fill.

2 bits corrupted. With no redundancy every hit slot flips a bit and the error reaches the destination uncorrected. Turn redundancy on to see the same noise absorbed.

Shannon’s theory measures information as reduced uncertainty in bits and says nothing about meaning. Its one engineering promise is that redundancy buys error-correction over a noisy channel — which is why natural language is so repetitive and still readable in a noisy room.

Meaning Beyond the Code

If communication were only coding and decoding, understanding a sentence would be a matter of looking up its words and composing their meanings. It is not. A speaker who says It is cold in here may be requesting that a window be shut, and a competent listener recovers that request even though no word of it is uttered. The philosopher Paul Grice explained this with the cooperative principle: participants in a conversation assume that each is trying to contribute appropriately to the shared purpose, and from that assumption, together with a set of maxims of quantity, quality, relation, and manner, a listener can infer meanings the words do not literally encode (Grice, 1975). These inferred meanings are implicatures. When a reference says only that a candidate “has excellent handwriting,” the maxim of quantity — say as much as is required — licenses the inference that there is little else good to report.

Relevance theory recast this machinery around a single principle. Sperber and Wilson argued that an utterance comes with a presumption of its own relevance: the hearer is entitled to assume the speaker has made the utterance worth processing, and so stops at the first interpretation that yields enough cognitive effect for the effort expended (Sperber & Wilson, 1995). On this account the Gricean maxims collapse into one: communication exploits the hearer’s automatic search for relevance. Both accounts share the central claim that human communication is inferential rather than merely a matter of encoding and decoding — the signal is evidence of the speaker’s intention, not a container for a meaning.

Grounding and Common Ground

Inference needs a foundation of shared knowledge to work from, and conversation builds that foundation as it goes. Clark and Brennan called the shared knowledge, beliefs, and assumptions that partners take to be mutual their common ground, and the process of adding to it grounding: the collaborative work of establishing that a contribution has been understood well enough for current purposes before moving on (Clark & Brennan, 1991). Grounding is why conversation is full of mm-hm, right, and clarifying questions — these are not noise but the receipts by which partners confirm mutual understanding and repair it when it fails.

Clark generalized this into a view of language use as joint action: an utterance is not something a speaker does to a listener but something the two perform together, like a handshake or a duet (Clark, 1996). The classic evidence is the referential-communication task, in which a director repeatedly describes an abstract figure for a matcher to pick out. Over successive references the partners converge on a shortened, shared name — a tangram first called “the angel with its arms raised” becomes, by the sixth trial, simply “the angel” — because each has grounded the term and both now treat it as part of their common ground. Reference, on this view, is something speakers and addressees arrive at together, not a label the speaker transmits.

Trial 1 of 6. First mention: the director gives a full, hedged description and the matcher accepts it only after a clarifying exchange.

The referring expression collapses from a full description to a single word not because the speaker decides to be brief, but because both partners have grounded the label and now treat it as common ground. Reference is something they arrive at together.

Dialogue as Joint Action

How do two people coordinate closely enough to ground anything? Pickering and Garrod proposed that dialogue is easy because it runs on interactive alignment: as partners talk, they automatically come to use the same words, the same syntactic structures, and the same conceptual models, and this alignment at each level propagates to the others (Pickering & Garrod, 2004). A speaker who hears the burglar was arrested by the police is more likely to use a passive herself a moment later; having settled on a word for an object, partners reuse it rather than renegotiate. Alignment is largely automatic and resource-cheap, which is why dialogue is fluent while monologue — speaking at length with no partner to align to — is comparatively hard.

They later grounded this in a mechanistic proposal: interlocutors use the forward models of their own production system to predict both what they themselves and what their partner will say, so that comprehension and production become tightly interwoven rather than separate (Pickering & Garrod, 2013). Prediction also solves a timing puzzle. Levinson documented that across languages the gap between conversational turns averages only about 200 milliseconds — far shorter than the half-second or more it takes to plan an utterance — which means listeners must be predicting the end of the current turn and preparing their response while the speaker is still talking (Levinson, 2016). Turn-taking is thus a predictive, anticipatory system, not a reactive one.

Planning overlaps A’s turn by 400 ms. With a 200 ms gap fixed by the data, a responder who needs 600 ms to plan must begin while A is still talking — so B is predicting how A’s turn will end.

The measured gap between turns averages about 200 ms across languages, well under the half-second or more needed to plan a reply. Turn-taking must therefore be predictive: listeners project the end of the current turn and prepare while it is still underway.

The Neural Basis of Communication

If two people align their representations, the alignment should be visible in their brains, and it is. Hasson and colleagues recorded brain activity from a speaker telling an unrehearsed story and from listeners hearing the recording, and found speaker–listener neural coupling: the listener’s cortical responses came to mirror the speaker’s, slightly delayed, and in some higher-order regions actually preceded the speaker’s, reflecting prediction. Crucially, the strength of this coupling predicted how well the listener understood the story (Hasson, Ghazanfar, Galantucci, Garrod, & Keysers, 2012). Communication, on this evidence, is literally a process of bringing two brains into a shared dynamical state; where coupling breaks down, understanding does too.

What the brains are aligning is not merely the signal but a shared conceptual space. Stolk and colleagues argued that mutual understanding rests on conceptual alignment: communicators continually update a shared model of what their signals are meant to refer to, drawing on the same brain regions regardless of whether they are communicating with words, gestures, or novel improvised signals (Stolk, Verhagen, & Toni, 2016). This explains how people establish meaning even for signals that have no pre-agreed code — the hallmark of genuinely open-ended human communication. The neural work thus converges with the behavioral: whether measured in shortened tangram labels, in reused syntax, or in coupled cortical activity, successful communication is the alignment of two minds on a common representation.

Worked Example

Information theory lets us put a number on how much a message tells us. The information carried by a single outcome is its surprisal, I = −log₂(p), measured in bits: the less likely the outcome, the more its occurrence tells us. Take a weather source with four outcomes and these probabilities: sunny 1/2, cloudy 1/4, rainy 1/8, snowy 1/8. The surprisals are

- sunny: −log₂(1/2) = 1 bit - cloudy: −log₂(1/4) = 2 bits - rainy: −log₂(1/8) = 3 bits - snowy: −log₂(1/8) = 3 bits

A forecast of snow carries three times the information of a forecast of sun, because snow is the rarer, more uncertainty-reducing event. The entropy of the source is the average surprisal, each outcome weighted by its probability:

H = −Σ p log₂ p = (1/2)(1) + (1/4)(2) + (1/8)(3) + (1/8)(3) = 0.5 + 0.5 + 0.375 + 0.375 = 1.75 bits per observation.

Compare this with a source whose four outcomes are equally likely, each with probability 1/4. There every surprisal is −log₂(1/4) = 2 bits, and the entropy is the maximum possible for four outcomes, 2 bits. The real weather source carries only 1.75 bits, so its redundancy is 2 − 1.75 = 0.25 bits per observation — the predictable structure that an efficient code could exploit, and that a noisy channel needs in order to survive errors. The gap between 1.75 and 2 is small here; in natural language, where some letters and words are far more probable than others, the redundancy is enormous, which is exactly why we can read a sentence with its vowels removed or hear it across a crowded room.

Discussion

The study of communication has advanced by recognizing that one word names three different problems. The transmission problem asks how a signal can be carried reliably across a noisy channel, and Shannon’s information theory answers it with a precise, meaning-free calculus of bits, entropy, and redundancy. The inference problem asks how a receiver recovers a meaning the signal only gestures at, and pragmatics answers it with the cooperative principle and the presumption of relevance, treating an utterance as evidence of intention rather than a container of content. The coordination problem asks how two agents bring their mental states into enough register to communicate at all, and the grounding, alignment, and neural-coupling accounts answer it by treating communication as a joint action that partners perform together.

These levels are complementary, not competing, and the field’s progress has come from stacking them. Information theory supplies the channel on which the rest rides; pragmatics explains why the thin, redundant signal suffices, because so much of the message is inferred rather than transmitted; the coordination accounts explain how partners build and maintain the shared context that inference presupposes. The deepest open question runs through all three: how a system built for transmitting signals became a system for sharing minds. Shannon’s channel says nothing about meaning, yet meaning is the whole point of talking — and bridging that gap, from bits to shared intentions, is the central project of the cognitive psychology of communication.

Current Directions

A major contemporary theme treats the pressures of real-time communication as a fundamental constraint on language itself. Christiansen and Chater’s Now-or-Never bottleneck argues that because the memory for raw input is fleetingly short, speakers and listeners must compress and commit to interpretations immediately or lose the signal, and that this single constraint shapes how language is processed, acquired, and historically structured (Christiansen & Chater, 2016). On this view the design of language is a solution to the engineering problem of communicating through a lossy, time-pressured channel — a direct descendant of Shannon’s concerns, now turned on the cognitive system rather than the wire.

A second front generalizes the coupling findings into a framework for social interaction at large. Hasson and Frith proposed that coupled dynamics — the mutual, bidirectional influence of two or more brains and bodies over time — is the right level of analysis for understanding communication and other joint behavior, rather than studying each brain in isolation (Hasson & Frith, 2016). The complementary line pushes on the alignment mechanism: conceptual alignment shows that communicators can establish shared meaning even without a shared code, improvising signals and converging on their interpretation through a common neural machinery for mutual understanding (Stolk, Verhagen, & Toni, 2016). The open problems are to join the fast, automatic alignment mechanisms to the slower, deliberate inference of pragmatics, and to extend models validated on controlled laboratory exchanges to the full, multi-party, multimodal richness of everyday conversation.

Common Misconceptions

Communication is just the transmission of a message from sender to receiver.

That is only the signal layer. Human communication is inferential: the signal is evidence of the speaker’s intention, and the receiver recovers far more than the words encode, through cooperative inference and the search for relevance (Grice, 1975; Sperber & Wilson, 1995).

More information, in Shannon’s sense, means more meaning.

Shannon’s theory measures how much a signal reduces uncertainty, in bits, and is deliberately silent about meaning. A string of random noise has high information content and no meaning at all; the quantity and the significance are distinct (Shannon, 1948).

Understanding a sentence is decoding its words.

Decoding is necessary but not sufficient. Reference and meaning are established collaboratively against a shared common ground that partners build and update as they talk, which is why the same words can mean different things to different pairs of people (Clark, 1996).

Listeners wait for a speaker to finish before planning a reply.

They cannot. Turn transitions average about 200 milliseconds, far less than the time needed to plan an utterance, so listeners predict the end of the current turn and prepare their response while it is still underway (Levinson, 2016).

Glossary

Channel.

The medium that carries a signal from transmitter to receiver in Shannon’s model, and the locus at which noise degrades the signal.

Common ground.

The knowledge, beliefs, and assumptions that conversational partners take to be mutually shared and draw on to interpret one another.

Cooperative principle.

Grice’s principle that participants assume each other to be contributing appropriately to the conversation’s shared purpose, enabling inference of unstated meaning.

Entropy.

The average information, in bits, carried by a source; the expected surprisal across its possible messages, and maximal when all messages are equally likely.

Grounding.

The collaborative process by which partners establish that a contribution has been understood well enough for current purposes before proceeding.

Implicature.

A meaning a speaker conveys without stating it, inferred by the listener from the cooperative principle and the conversational maxims.

Information theory.

Shannon’s mathematical account of communication, which quantifies information as the reduction of uncertainty and defines entropy, channel capacity, and redundancy.

Interactive alignment.

The largely automatic convergence of dialogue partners on shared representations — words, structures, and conceptual models — at every linguistic level.

Joint action.

An activity two or more agents perform together by coordinating their individual contributions, the frame in which Clark casts all language use.

Noise.

Any unwanted disturbance that corrupts a signal in the channel, reducing the information the receiver can recover.

Pragmatics.

The study of how context and inference determine the meaning of utterances beyond their literal, encoded content.

Redundancy.

Predictable structure in a signal beyond the minimum needed to carry its information, which allows errors to be detected and corrected.

Shared intentionality.

The capacity to form joint goals and attend together to the same thing, which Tomasello argues underlies uniquely human communication.

Surprisal.

The information carried by a single outcome, −log₂ of its probability; rarer outcomes carry more.

Turn-taking.

The near-universal system by which conversational partners exchange the floor with minimal gap and overlap, relying on prediction of turn ends.

Key Researchers

Herbert H. Clark

(living). American psycholinguist at Stanford University whose account of language use as joint action — grounding, common ground, and the collaborative shaping of reference — reframed conversation as something partners accomplish together rather than messages they take turns transmitting. ORCID - Google Scholar - Faculty page - Wikipedia - Wikidata

Paul Grice

(1913–1988). British philosopher of language whose cooperative principle and conversational maxims founded modern pragmatics, showing how listeners infer a speaker’s intended meaning beyond the literal content of an utterance through implicature. Wikipedia - Wikidata

Uri Hasson

(living). Neuroscientist at Princeton University whose work on speaker–listener neural coupling showed that successful communication is accompanied by the alignment of a listener’s brain activity to the speaker’s, offering a mechanism for sharing a mental state. ORCID - Google Scholar - Faculty page - Wikidata

Stephen C. Levinson

(living). Linguist and cognitive scientist at the Max Planck Institute for Psycholinguistics whose work on pragmatics and the universal turn-taking system of conversation established the tight timing and inferential machinery that coordinate human dialogue. Google Scholar - Faculty page - Wikipedia - Wikidata

Martin J. Pickering

(living). Psycholinguist at the University of Edinburgh who, with Simon Garrod, developed the interactive-alignment model of dialogue, in which comprehension and production become coupled as interlocutors align their linguistic representations. ORCID - Google Scholar - Faculty page - Wikipedia - Wikidata

Claude Shannon

(1916–2001). American mathematician and engineer whose 1948 paper founded information theory, defining the channel model of communication and giving a precise, quantitative measure of information in bits. Wikipedia - Wikidata

Michael Tomasello

(living). Developmental and comparative psychologist at Duke University whose work on shared intentionality traced the cooperative origins of human communication, arguing that even prelinguistic pointing and pantomime rest on a uniquely human capacity for joint goals. ORCID - Google Scholar - Faculty page - Wikipedia - Wikidata

Frequently Asked Questions

What is communication in psychology?

Communication is the exchange of information between agents by means of signals. Psychology studies it at three levels: as the physical transmission of a signal, as the inference of a meaning the signal only points at, and as the coordinated joint action by which two minds come into register. Each level has its own theory, from information theory to pragmatics to the study of common ground (Shannon, 1948; Clark, 1996).

What is Shannon’s channel model?

It is the foundational engineering account of communication: an information source selects a message, a transmitter encodes it into a signal, a channel carries the signal while a noise source corrupts it, and a receiver decodes it for a destination. The model measures information in bits as the reduction of uncertainty, and deliberately says nothing about meaning (Shannon, 1948).

What is the difference between information and meaning?

Information, in Shannon’s sense, is how much a signal reduces the receiver’s uncertainty, measured in bits. Meaning is what a signal is about. The two are independent: random noise carries a great deal of information and no meaning, while a single expected word can carry little information yet convey an important message (Shannon, 1948; Grice, 1975).

What is common ground in a conversation?

Common ground is the body of knowledge, beliefs, and assumptions that conversational partners treat as mutually shared. Partners build it collaboratively through grounding, confirming that each contribution has been understood before moving on, and they rely on it to interpret new utterances. It is why the same words can mean different things to different pairs of people (Clark and Brennan, 1991; Clark, 1996).

What is interactive alignment?

Interactive alignment is the largely automatic process by which dialogue partners come to use the same words, syntactic structures, and conceptual models as they talk. Alignment at one level spreads to the others, making mutual understanding cheap and dialogue fluent. It is thought to rely on the same prediction mechanisms that link comprehension and production (Pickering and Garrod, 2004; Pickering and Garrod, 2013).

How do people take turns in conversation so quickly?

The gap between conversational turns averages only about 200 milliseconds, far shorter than the time needed to plan an utterance. This means listeners do not wait for a speaker to finish; they predict where the current turn will end and prepare their response while it is still underway, making turn-taking an anticipatory rather than a reactive system (Levinson, 2016).

What does brain-to-brain coupling mean?

During successful communication a listener’s brain activity comes to track the speaker’s over time, a phenomenon called neural coupling. The degree of coupling predicts how well the listener understands, suggesting that communication works by bringing two brains into a shared dynamical state. Where coupling breaks down, understanding tends to break down too (Hasson, Ghazanfar, Galantucci, Garrod, and Keysers, 2012).

Is communication unique to humans?

Many species communicate, but human communication is distinctive in resting on shared intentionality, the capacity to form joint goals and attend together to the same thing. On this account even a prelinguistic point presupposes that two minds are cooperating toward a common end, which supports the open-ended, inferential character of human communication (Tomasello, 2008).

References

Christiansen, M. H., & Chater, N. (2016). The Now-or-Never bottleneck: A fundamental constraint on language. Behavioral and Brain Sciences, 39, e62. https://doi.org/10.1017/S0140525X1500031X

Clark, H. H. (1996). Using language. Cambridge University Press. https://doi.org/10.1017/CBO9780511620539

Clark, H. H., & Brennan, S. E. (1991). Grounding in communication. In L. B. Resnick, J. M. Levine, & S. D. Teasley (Eds.), Perspectives on socially shared cognition (pp. 127–149). American Psychological Association. https://doi.org/10.1037/10096-006

Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics, Vol. 3: Speech acts (pp. 41–58). Academic Press.

Hasson, U., Ghazanfar, A. A., Galantucci, B., Garrod, S., & Keysers, C. (2012). Brain-to-brain coupling: A mechanism for creating and sharing a social world. Trends in Cognitive Sciences, 16(2), 114–121. https://doi.org/10.1016/j.tics.2011.12.007

Hasson, U., & Frith, C. D. (2016). Mirroring and beyond: Coupled dynamics as a generalized framework for modelling social interactions. Philosophical Transactions of the Royal Society B, 371(1693), 20150366. https://doi.org/10.1098/rstb.2015.0366

Levinson, S. C. (2016). Turn-taking in human communication — origins and implications for language processing. Trends in Cognitive Sciences, 20(1), 6–14. https://doi.org/10.1016/j.tics.2015.10.010

Pickering, M. J., & Garrod, S. (2004). Toward a mechanistic psychology of dialogue. Behavioral and Brain Sciences, 27(2), 169–190. https://doi.org/10.1017/S0140525X04000056

Pickering, M. J., & Garrod, S. (2013). An integrated theory of language production and comprehension. Behavioral and Brain Sciences, 36(4), 329–347. https://doi.org/10.1017/S0140525X12001495

Shannon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27(3), 379–423. https://doi.org/10.1002/j.1538-7305.1948.tb01338.x

Sperber, D., & Wilson, D. (1995). Relevance: Communication and cognition (2nd ed.). Blackwell.

Stolk, A., Verhagen, L., & Toni, I. (2016). Conceptual alignment: How brains achieve mutual understanding. Trends in Cognitive Sciences, 20(3), 180–191. https://doi.org/10.1016/j.tics.2015.11.007

Tomasello, M. (2008). Origins of human communication. MIT Press. https://doi.org/10.7551/mitpress/7551.001.0001