Pragmatics of Language

Learning Objectives

By completing this book, you will be able to:

1. Explain the distinction between semantics (the meaning of words and sentences) and pragmatics (meaning in context), and why pragmatics is indispensable for understanding real communication.

2. Articulate Grice's Cooperative Principle and its four conversational maxims, and diagnose which maxim is violated when communication fails.

3. Distinguish between what is said, what is implicated, and what is presupposed, and explain why these three layers of meaning frequently diverge.

4. Identify conversational implicatures in real requests and explain how an AI system can capture them or miss them.

5. Recognize implicit presuppositions in formulations of intention and use them as diagnostic tools for unexamined assumptions.

6. Apply pragmatic analysis to human-AI interaction, identifying the systematic sources of loss of meaning between what the user communicates and what the system processes.

1. Semantics vs. Pragmatics: Two Kinds of Meaning

To understand pragmatics, you must first understand what it is not. Semantics is the branch of linguistics that studies the meaning of words and sentences in the abstract, independently of who says them, to whom, where, or when. Semantic meaning is what you can find in a dictionary and a grammar: “the cat is on the mat” means what it means regardless of context.

Pragmatics, by contrast, studies how meaning is constructed, negotiated, and transformed in the real use of language. It is the study of what people mean when they speak, which is frequently more—and sometimes different—from what their words mean literally. Pragmatics recognizes that meaning does not reside in the words alone but in the interaction between the words, the context, the speaker, the hearer, and the situation.

“Semantics: the sentence ‘It's hot in here’ describes a property of the environment (elevated temperature). Its truth value depends on the actual temperature of the place. Pragmatics: the same sentence, said by someone sitting next to a closed window, can mean ‘open the window.’ Said by someone who has just come in from the street, it can mean ‘how pleasant it is in here.’ Said by someone looking at their watch, it can mean ‘let's go somewhere else.’ Semantics sees a sentence; pragmatics sees a communicative action that requires context to be interpreted.”

Charles Morris, the American semiotician, established in 1938 the tripartite distinction that still structures linguistics: syntax studies the relationship between signs (grammar), semantics studies the relationship between signs and what they represent (meaning), and pragmatics studies the relationship between signs and their users (use). Intention Architecture operates fundamentally at the pragmatic level: what matters to us is not only what the user's words mean, but what the user is doing with those words.

2. Grice and the Cooperative Principle

Herbert Paul Grice, the British philosopher, delivered in 1967 the William James Lectures at Harvard that would become one of the founding texts of modern pragmatics, eventually published as “Logic and Conversation” (1975). Grice's question was deceptively simple: how is it possible that humans communicate successfully when what they say literally is so frequently different from what they mean?

His answer was the Cooperative Principle: when people converse, they operate under a tacit agreement to cooperate with one another so that communication works. It is not a moral principle (people can lie); it is a descriptive principle of how conversation works: participants assume that the others are contributing rationally to the exchange, and they use that assumption to infer what is not said explicitly.

Grice formulated this principle as one general directive and four conversational maxims that, together, define what it means to ‘cooperate’ in a conversation.

2.1 The Maxim of Quantity: Say Enough, Not More

Make your contribution as informative as is required for the purposes of the exchange, but not more informative than is required. Neither too little nor too much.

“A tourist asks: ‘Where is the train station?’ An answer that satisfies the maxim: ‘Two streets to the right.’ An answer that violates it by deficiency: ‘In this city.’ (Insufficient: technically true but useless.) An answer that violates it by excess: ‘Two streets to the right, it was built in 1923 by the architect Gustavo Méndez, it has 14 platforms, and the 5:30 train is usually late on Fridays.’ (Too much information: it raises the suspicion that something relevant lies in the excess.)”

The maxim of quantity is especially relevant to interaction with AI. When a system gives an excessively long and detailed answer to a simple question, it violates this maxim. And when it gives a cryptically brief answer to a complex question, it violates it too. The user calibrates the amount of information they expect to receive according to the complexity of their question, and deviating significantly from that expectation generates confusion or distrust.

2.2 The Maxim of Quality: Say What Is True

Do not say what you believe to be false. Do not say that for which you lack adequate evidence. In short: be truthful.

“If someone asks you ‘what is the capital of Australia?’ and you answer ‘Canberra,’ you satisfy the maxim. If you answer ‘Sydney’ believing it to be correct, you make an error but you do not violate the maxim (you believed what you said). If you answer ‘Sydney’ knowing it is Canberra, you violate the maxim: you lie. And here is the subtle part: if an AI system generates an answer that sounds confident but has no sufficient basis—what is called a ‘hallucination’ in the technical language of AI—it violates this maxim not by lying but by asserting without adequate evidence.”

The maxim of quality connects directly with the epistemology of Module II: the demand for evidence, falsifiability, epistemic humility. Every assertion in a cooperative conversation carries an implicit guarantee: ‘what I am saying is true to the best of my knowledge.’ When that guarantee breaks, conversational trust is damaged—it is a stock (Topic 1 of Module III) that erodes with every violation.

2.3 The Maxim of Relation: Be Relevant

Make your contribution relevant to the exchange in progress. Do not change the subject without reason. Do not introduce information that is beside the point.

“If someone asks ‘when does the next bus arrive?’ and you answer ‘Lovely day, isn't it?’, you violate the maxim of relation. But if you think about it pragmatically, the hearer tries to rescue the relevance: perhaps you are suggesting that the day is so lovely they should walk instead of waiting for the bus, or perhaps you are dodging the question because you do not know the answer and do not want to admit it. The hearer works actively to find the relevance—and that is the Cooperative Principle in action.”

Dan Sperber and Deirdre Wilson took this maxim further in their Relevance Theory (Relevance: Communication and Cognition, 1986), arguing that relevance is the central principle of all human communication: the brain automatically assumes that any information it receives is relevant and spends cognitive effort finding that relevance. When it does not find it, it discards the information or distrusts the speaker.

2.4 The Maxim of Manner: Be Clear

Avoid obscurity of expression. Avoid ambiguity. Be brief. Be orderly. In short: communicate clearly and comprehensibly.

“If a doctor says to a patient ‘There exists an advanced-stage malignant neoplasm with multiple metastases,’ the information is technically precise but pragmatically obscure for most patients. If she says ‘You have a cancer that has spread to several parts of the body,’ she better satisfies the maxim of manner. But if she says ‘You have a little problem we're going to keep an eye on,’ she violates the maxim of quantity (insufficient) and that of quality (misleading), even though she satisfies manner (it is clear). The maxims can come into tension with one another.”

3. Implicature: What Is Communicated Without Being Said

Grice's most powerful and influential concept is conversational implicature: the meaning a speaker communicates without stating it explicitly, and which the hearer infers on the assumption that the speaker is cooperating. Implicature is the gap between what is said and what is communicated—and that gap is enormous in everyday human communication.

3.1 How Implicature Is Generated

A conversational implicature is generated when the speaker ostensibly violates a maxim. The hearer, assuming cooperation, reasons: ‘this person appears to violate the maxim, but I assume they are cooperating; therefore, they must mean to communicate something that is not in the literal words.’ It is an inferential process that occurs automatically and with astonishing speed.

“Ana asks Beatriz: ‘What did you think of Carlos's presentation?’ Beatriz answers: ‘Well, the PowerPoint had nice colors.’ Literally, Beatriz has said something true about the colors. But pragmatically, she has communicated that the presentation was bad. How? Beatriz violates the maxim of quantity (she says less than necessary about the quality of the presentation) and that of relation (she talks about the colors when asked about the presentation as a whole). Ana, assuming cooperation, infers: if the best thing Beatriz can say is the colors, the content must have been poor. The criticism was never spoken—it was implicated.”

What is critical is that implicatures are cancellable: if Ana presses, Beatriz can say ‘I was only talking about the colors; the presentation as a whole was fine.’ This cancellability distinguishes implicature from logical implication: a logical implication cannot be cancelled without contradiction, but an implicature can, because it was never stated explicitly.

3.2 Implicature in Interaction with AI

Implicature is the territory where human-AI communication meets its deepest fissures. Humans implicate constantly: we leave things unsaid because we assume our interlocutor will infer them. AI, which predominantly processes literal meaning, can miss implicatures—or worse, can generate unintended implicatures in its own responses.

“A user writes: ‘We have been working on this problem for months.’ Literally: a temporal assertion about the duration of an effort. Possible implicatures: (1) ‘We have not found a solution yet’ (if they had found one, they would not say they had spent months). (2) ‘We are frustrated’ (the duration implies difficulty and fatigue). (3) ‘We need urgent help’ (‘months’ implies they have exhausted their internal resources). (4) ‘Any solution must differ from what we have already tried’ (if the obvious thing worked, they would have solved it already). A system that processes only the locution might answer: ‘Understood, you have been working on this for months.’ A pragmatically competent system would catch the implicatures and respond accordingly.”

4. Presupposition: What Is Taken for Granted

Beyond what is said and what is implicated, there is a third layer of meaning: what is presupposed. A presupposition is information the speaker treats as given, as already shared, as the background against which the utterance is built. Presuppositions are neither asserted nor implicated; they are taken as fact. And precisely because they are not asserted, they are extraordinarily difficult to question.

“The question ‘When did you stop beating your wife?’ presupposes that you were beating your wife. If you answer ‘on Tuesday,’ you accept the presupposition. If you answer ‘I never stopped,’ you accept the presupposition from the other direction. The only answer that rejects the presupposition is: ‘Your question presupposes something false.’ But this answer is socially costly, cognitively demanding (it requires that you detect the presupposition), and frequently does not even occur to us.”

Presuppositions are activated by certain linguistic triggers: factive verbs (‘knows,’ ‘regrets,’ ‘discovered’ presuppose the truth of their complement), definite descriptions (‘the problem’ presupposes that a problem exists), temporal clauses (‘before the project fails’ presupposes that it will fail), change-of-state verbs (‘stopped doing X’ presupposes that X was being done before).

4.1 Presuppositions in the Formulation of Intention

Presuppositions are perhaps the most valuable pragmatic tool for Intention Architecture. Every formulation of intention contains presuppositions—assumptions the user treats as given but which may not be true. Detecting and questioning presuppositions is an act of epistemic rigor that connects directly with the audit of assumptions in Module II.

“‘We need to improve the quality of our leads.’ Presuppositions: (1) There is a lead-quality problem (not a quantity problem, not a conversion problem, not a product problem). (2) ‘Our leads’ are something that can be measured and categorized by quality. (3) Improving leads is a viable means to the end the user is seeking (which is probably more sales). (4) The team has the data to distinguish good leads from bad ones. Each presupposition is an unexamined hypothesis. If presupposition (1) is false—the real problem is the capacity of the sales team, not the quality of the leads—the entire formulation points in the wrong direction.”

Note the connection with Module II: pragmatic presuppositions are the epistemic assumptions of Module II expressed linguistically. Assumptions about the situation (how things are), causal assumptions (what causes what), solution assumptions (what would resolve the problem), and competence assumptions (I know enough) are encoded as presuppositions in the user's language. Detecting them pragmatically is the linguistic version of the epistemic audit.

5. Deixis: The Words That Depend on Context

Deixis is the linguistic phenomenon by which certain words only have meaning in relation to the context of their utterance. ‘I,’ ‘you,’ ‘here,’ ‘now,’ ‘this,’ ‘yesterday’—all of these words change their referent depending on who says them, where, and when. They are the most frequent words in the language and the most dependent on context for their meaning.

“The note ‘Back in 5 minutes’ stuck to a door is semantically clear but pragmatically useless without deictic context: who is ‘I’ (who wrote it)? When is ‘now’ (when was the note put up)? If you read the note at 3 PM and it was written at 10 AM, the ‘5 minutes’ expired hours ago. Temporal deixis (‘back,’ ‘in 5 minutes’) requires knowing when it was anchored to the world in order to be interpreted.”

Karl Bühler, a psychologist and linguist, was the first to analyze deixis systematically in his theory of the deictic field: every linguistic utterance is anchored in an origin point defined by ‘I’ (person), ‘here’ (place), and ‘now’ (time). Without this anchor, the sentence floats without reference.

5.1 Deixis in Interaction with AI

Deixis is a critical challenge for human-AI interaction because the system must reconstruct the deictic context the user takes for granted. When the user says ‘our team,’ ‘this quarter,’ ‘the problem,’ ‘here,’ they are using deictic expressions that point to specific referents only the context can identify.

“A user writes: ‘I need you to analyze this and tell me what we can do about it.’ ‘This’: what thing? An attached document, a data point mentioned earlier, a situation described in the previous conversation? ‘Me’: who is the user, what role do they have, what capacity for action? ‘We’: who are ‘we’? The user and their team, their company, their department? ‘About it’: about which dimension of the problem? Each deictic expression is a hole that requires context to be filled. Without context, the answer will be generic; with context, it can be precise.”

Deixis reveals a fundamental asymmetry in human-AI communication: the human speaks from a rich context (they know who they are, where they are, what they are talking about, with what urgency) and assumes the interlocutor shares that context. The AI receives only the text and must reconstruct the context from the available fragments. Every unresolved deictic expression is information lost.

6. Practical Application: Pragmatic Analysis of Requests

6.1 Case Study: What Was Not Said

Roberto, operations director of a logistics company, writes to an AI system: “Our delivery time has gotten worse. Clients are complaining. We need a solution.”

Complete pragmatic analysis:

What is said (semantics): Three assertions: delivery time has gotten worse, clients are complaining, a solution is needed.

Violations of the maxims (Grice): He violates the maxim of quantity: by how much did it worsen? From what figure to what figure? Since when? On which routes or products? The vagueness may be involuntary (Roberto assumes the system shares his context) or deliberate (Roberto does not know the exact figures and does not want to expose that).

Implicatures: (1) ‘Has gotten worse’ implicates that it was previously acceptable (there is a baseline). (2) ‘Clients are complaining’ implicates that the situation has already reached a critical point (otherwise he would not mention the complaints). (3) ‘We need a solution’ implicates urgency and that previous attempts failed or do not exist.

Presuppositions: (1) The problem is delivery time (not communication with the client, not client expectations, not product quality). (2) A ‘solution’ exists (rather than an inevitable trade-off or a wicked problem). (3) The AI system can provide this solution (a preparatory condition). (4) Roberto has the authority and the resources to implement whatever is recommended.

Unresolved deixis: ‘Our’ (the whole company or his department); ‘has gotten worse’ (relative to what period and what metric); ‘the clients’ (all of them or a specific segment).

“A pragmatically competent system would respond: ‘Roberto, to give you something useful I need to resolve some ambiguities. When you say delivery time got worse: do you have specific data (for example, from 48 hours on average to 72)? Does this affect all routes or specific ones? And the complaints: are they from large clients or from the general segment? I also want to question a presupposition: are you sure the central problem is delivery time, or could it be expectation management (that the client did not know it would take longer)? Sometimes the problem is not speed but communication.’”

6.2 A Pragmatic Analysis Protocol

1. Separate what is said from what is communicated. What does the utterance say literally (semantics) and what does it communicate beyond the words (pragmatics)?

2. Audit Grice's maxims. Does the utterance satisfy the maxims of quantity, quality, relation, and manner? Where are the violations? Are they accidental (the user does not know they are being vague) or deliberate (the user is communicating something through the violation)?

3. Extract implicatures. What is the user communicating without stating it explicitly? What can be inferred from what they say and from what they omit?

4. Excavate presuppositions. What does the user take as given? What assumptions about the situation, causality, solution, and competence are encoded as presuppositions? Could any of them be false?

5. Resolve the deixis. Which deictic expressions require context to be interpreted? Can you resolve them with available information, or do you need to ask?

6. Respond to what is communicated, not only to what is said. The response must address the implicatures, question the suspect presuppositions, and resolve the deixis—without losing sight of the fact that the objective is to help, not to interrogate.

7. Glossary of Key Terms

8. Final Reflection and Self-Assessment

Before advancing to the next topic (Speech Acts, with Austin and Searle), verify that you can apply pragmatic analysis as a tool for decoding intentions.

9. Bibliography and Recommended Readings

Essential Readings

• Grice, H.P. (1975). “Logic and Conversation.” In P. Cole & J. Morgan (eds.), Syntax and Semantics, Vol. 3: Speech Acts. Academic Press.

• Levinson, S. (1983). Pragmatics. Cambridge University Press. [The standard reference in linguistic pragmatics]

• Sperber, D. & Wilson, D. (1986). Relevance: Communication and Cognition. Blackwell. [2nd ed. 1995]

Complementary Readings

• Grice, H.P. (1989). Studies in the Way of Words. Harvard University Press. [The complete collection of Grice's essays]

• Horn, L. & Ward, G. (eds.) (2004). The Handbook of Pragmatics. Blackwell.

• Huang, Y. (2007). Pragmatics. Oxford University Press. [An accessible introduction]

• Morris, C. (1938). Foundations of the Theory of Signs. University of Chicago Press. [The original tripartite distinction: syntax, semantics, pragmatics]

Next Topic

The next book in this series addresses Speech Acts: J.L. Austin's revolution, which demonstrated that to speak is to do, and John Searle's systematization, which classified the actions we perform with language. If this topic taught you that meaning depends on context, the next will teach you that language does not only communicate—it transforms. You will discover that every utterance from the user is simultaneously something said, something done, and something produced, and that the ability to distinguish these three levels is central to Intention Architecture.