1. Introduction

How do we effortlessly understand utterances such as “I porched three newspapers today”?¹ These sentences, which defy conventional meaning yet somehow make sense, reveal a fundamental truth about human communication: understanding is not merely decoding preexisting rules or inferring pure contextual cues. It is an act of joint scenario-building, where speakers and hearers collaboratively construct plausible interpretations by blending fragments of convention with creative elaboration.

Traditional theories of linguistic understanding fall into two broad camps. Rule-based approaches claim that meaning derives from fixed lexical entries and compositional semantics. Yet these struggle to explain novel combinations like “to porch”, since there is no rule for verbing “porch”. Pure-context theories hold that meaning is entirely situational, emerging ad hoc from shared circumstances. But this cannot explain why “Mark is a lion” reliably conveys courage rather than, say, a mane or a roar; some conventional anchors must constrain interpretation. This paper offers a third path. On the scenario-building view I am here proposing, understanding is a dynamic process in which (1) speakers provide cues-words, gestures, shared knowledge-to seed a plausible scenario; (2) hearers integrate these cues with conventions and contextual creativity to construct interpretations; and (3) alignment occurs when the hearer’s scenario sufficiently overlaps with the speaker’s intended one.

By integrating insights from the epistemology of understanding into the philosophy of language, this framework reframes understanding not as decoding or guessing, but as co-creation: language conventions provide the scaffolding for imaginative collaboration. To argue for, and develop, this position, I begin by examining the limits of rule-based (section 2) and pure-context (section 3) theories, two categories that build on François ^{Recanati’s (2004}; ²⁰⁰⁵⁾ distinction between literalist and contextualist positions. I will argue that these theories are both unable to explain how agents are able to understand utterances that include deviant uses of terms. Then, in section 4, I develop the scenario-building framework. The final section shows how this framework can be used to model degrees of understanding, based on the overlap of the constructed scenarios.

2. Rule-Based Theories

In “Literalism and Contextualism: Some Varieties” (2005) and Literal Meaning (2004), François Recanati introduces, contrasts, and discusses the distinction between two broad categories of positions concerning language understanding, which he calls “Literalism” and “Contextualism”. Within these categories falls an array of positions that range from those that most adhere to the principles of Literalism to those that most adhere to Contextualism. At each limit of the spectrum we find Strong Literalism and Meaning Eliminativism. The tension between these positions is at its heart a disagreement about the nature of the elements (syntactic/semantic/pragmatic) that are involved in the understanding and the interpretation of an utterance, especially in situations of ordinary communication.²

According to Literalism, to understand an utterance is to grasp the proposition, or propositional content, it conveys. This grasping is a matter of semantic interpretation. This is the process whereby the interpreter uses her knowledge of the language and its rules to assign the values to indexical expressions and to determine the meaning of the whole expression on the basis of the meaning of its parts. This is the process of deductively establishing the truth conditions of the sentence, i.e., its propositional content. Rules play a crucial role in the literalist framework. Indeed, in its strongest form, Literalism was espoused by philosophers of the first half of the twentieth century (such as Frege, Russell, and Carnap, to name just a few) who maintained that language in general should only be studied via formal languages of the type Russell and Whitehead presented in the Principia Mathematica. Since their syntactic structure is transparent, these languages allow for a direct access to their compositional rules, contrary to natural languages that can be opaque, and which can only express a determinate content in context. Rules also played an important role in the later development of formal semantics by philosophers such as ^{Montague (1970}; ¹⁹⁷³⁾ and ^{Davidson (1967}). Applying the methods of their predecessors, they conceived of natural languages as systems of rules for determining the meaning of expressions (their truth conditions) based on the meanings of their parts (the conventional assignment of an entity to that symbol).

Even in its weaker forms, rules play an important role in the literalist framework. With the advent of ordinary language philosophy through the works of Grice and Austin, among others, philosophers began to focus on the pragmatic aspects of natural languages -since this is the main feature that distinguishes them from formal languages-and aimed to show that these are important features of these languages rather than defects because such utterances are used to do things in addition to the fact that they mean something. Thus, the understanding involved in ordinary communication would require, in addition to semantic interpretation, a process of pragmatic interpretation, that is, a process whereby communicational intentions are attributed to the agent in order to make sense of the communicated content. While, as we will see, some philosophers completely embraced the shift by moving away from the rule-based analysis and into Contextualism (although with variations in strength), others such as Herman Cappelen, Ernie Lepore, and Emma Borg, argued for Minimalism. This is the view that “there are minimal propositions (or minimal truth conditions, or minimal-whatever-else-a-semantictheory-might-run-on) which are the result of little or no pragmatic processing and which are available to provide the literal meaning of sentences” (^{Borg 2007, p. 340}). In other words, according to Minimalism, “‘what is said’ […] departs from the conventional meaning of the sentence (and incorporates contextual elements) only when this is necessary to ‘complete’ the meaning of the sentence and make it propositional” (^{Recanati 2004, p. 7}).

According to ^{Cappelen and Lepore (2005)}, most sentences are “semantically insensitive”, i.e., their truth conditions do not vary with subtle changes in context, beyond the cases where context supplies values for formally encoded parameters such as in the case of indexicals like “I”, “here”, “now”, or demonstratives. Once those parameters are set, the sentence’s truth conditions are fixed by the linguistic rules, regardless of what the speaker intends or what the hearer might pragmatically infer. Thus, according to Cappelen and Lepore, contextual contributions to the semantic content are syntactically triggered, e.g., by the presence of an indexical. Indeed, following ^{Kaplan (1989)}, it has become generally accepted that the assignment of semantic values to indexical expressions is a function from the context of utterance to the referent. Since indexicals such as “I” (but also “here”, “now”, etc.) refer to whomever the speaker is, and vary across contexts (when speakers are different), determining the meaning of an indexical expression requires fixing its referent depending on the context. The search for the referent is, therefore, triggered by the presence of this indexical expression. For Cappelen and Lepore, all contextual contributions operate in this way. They are bottom-up pragmatic processes, i.e., processes by which one completes the meaning of the sentence through the assignment of contextually determined values and are said to be “bottom-up” because they are triggered by elements present in the sentence. While Borg agrees that this is a tenet of Minimalism, she also argues that contextual contributions to semantic content are formally tractable, that is, she restricts the idea that there could be “top-down pragmatic processes”, or “pragmatic modulation” (as ^{Recanati (2004}, and in many of his works) calls them), which affect the truth conditions of the utterance in a way that exhibits a context sensitivity that is not under linguistic control because they are not triggered by elements present in the sentence. These include, most notably, free enrichment, i.e., the process by which the meaning of the sentence is fleshed out through the addition of caveats and other contextual information which, for example, delineate the boundaries of a quantifier. These processes serve to eliminate potential ambiguities that result from semantic underdetermination. For example, if I say “I went to the bank”, there is an ambiguity concerning the type of bank in question. So, through free enrichment, one makes the interpretation of the utterance more specific than what the sentence literally says. To take another example, suppose John says to the barista “More coffee, please”. We, as well as the barista, directly understand him as asking for more brewed coffee of the drinkable kind and not, for example, coffee beans or a coffee tree. This precision is not triggered by any element in the sentence. Unlike the case of an indexical expression, it involves the importation of prior external information that serves to determine the content being communicated, at least in part, by determining the speaker’s intended communicated content. Determining the speaker’s intention is in the realm of pragmatics because, given a certain utterance u, pragmatic interpretation is concerned with the question of why a/this speaker would rationally utter u on this occasion. To answer such a question would require the import of information that goes beyond the literal content of u, i.e., information that is semantically intractable. Hence, ^{Borg’s (2007)} tractability thesis. We can understand moderate literalists such as Borg, Cappelen and Lepore as saying that while (at least some) contextual elements must be taken into consideration in an analysis of the understanding achieved in communication, these must ultimately be anchored in semantic and syntactic aspects. These are rule-based approaches since the literal content of an utterance is determined by the linguistic rules governing word meanings, syntactic combination, and the assignment of values to formally encoded context-sensitive expressions like indexicals, demonstratives, or tense. These rules are taken to yield a complete, truth-evaluable proposition before any pragmatic interpretation takes place-with pragmatic interpretation only taking place if the semantic interpretation alone is not sufficient to determine the propositional content; and they do so independently of the speaker’s broader communicative intentions. The role of rules here is foundational: they constitute the publicly accessible code that speakers and hearers can both rely on, ensure a compositional mapping from syntax to truth conditions, and sharply delimit the boundary between semantics and pragmatics. On this view, pragmatics operates only after the rules have fixed the literal truth conditions, enriching or reinterpreting them to recover the speaker’s meaning, but never feeding back into the determination of what is literally said. By contrast, contextualism treats the rules of language not as fully determining truth conditions, but as supplying only a partial and often skeletal starting point for interpretation. On this view, advocated in different forms by ^{François Recanati (2004}; ²⁰¹⁰⁾, Charles ^{Travis (1985}; ²⁰⁰⁸⁾, and ^{Dan Sperber and Deirdre Wilson (1995)}, the content that a sentence expresses on a given occasion cannot be fixed without appeal to pragmatic processes that are sensitive to the speaker’s communicative intentions and the specifics of the conversational setting. The linguistic rules provide cues-lexical meanings, syntactic structure, and perhaps a few formal context parameters-but these cues require substantial top-down enrichment from context to yield a complete, truth-evaluable proposition. In this way, pragmatics is not merely post-semantic but constitutive of truth-conditional content, and the boundary between semantics and pragmatics is correspondingly porous.

Before moving to Contextualism, a central problem for Minimalism must be addressed. Hearers seem to be able to understand utterances that not only employ terms which deviate from their conventional use-such as when the term “play” began to be used to mean “deceive” in sentences such as “I played him”-but that also do not follow any existing rule-such as verbing the term “porch” in Gibbs’ example of “George managed to porch the newspaper yesterday” (^{Gibbs 1984, p. 10}). Minimalism encounters difficulty with non-standard uses of figure terms because it presupposes a stable, context-independent semantics, which cannot straightforwardly accommodate novel or metaphorical uses since these uses lack established lexical entries or grammatical rules. The root of this difficulty is Minimalism’s foundational commitment to a context-invariant semantics, which is at odds with the context-driven nature of these interpretations.

Emma ^{Borg (2004)} argues that such novel uses arise through pragmatic enrichment processes that go beyond minimal semantic content. She emphasizes that hearers rely on pragmatic inference to construct intended meanings when literal lexical meanings are insufficient or violated. This pragmatic enrichment is guided by conversational principles and contextual cues, enabling flexible interpretation of novel or metaphorical usages. Thus, Minimalism’s fixed semantic core turns out to be insufficient, and pragmatic mechanisms are essential. Similarly, Cappelen and Lepore, in their critique of contextualist theories, defend a minimalist semantics but acknowledge the need to separate semantic content from pragmatic processes. They argue that Semantic Minimalism can still hold by restricting semantics to context-invariant content while relegating all flexibility in interpretation to pragmatics. However, their view requires a sharp distinction between semantics and pragmatics, where pragmatic inference handles all non-conventional uses without altering the semantic content. This creates tension because pragmatic inference must be highly sophisticated to handle novel verbing or shifted meanings reliably. Furthermore, in these cases, but especially in the case of “to porch”, it seems that pragmatic interpretation plays a central role rather than a secondary one in the interpretation of the utterance. Indeed, in their study on nonce verbs,^{3 Clark and Gerrig (1983)} note:

Listeners are sophisticated about handling word senses. They consider the information the addressees share with the speaker, evaluate the salience and coherence of potential interpretations against this common ground, and select or create interpretations as specific as this common ground allows-all with deceptive ease. They mix sense creation with sense selection even in the absence of sentential cues. For a theory of language comprehension to be correct, it must characterize how they manage this. (p. 607, my emphasis)

Clark and Gerrig’s findings point to the idea that hearers do not survey the lexicon when they seek to understand an utterance containing a novel term, but rather create it on the spot by accessing information from the common ground-a pragmatic notion and aspect of communication.⁴ If this is correct, then pragmatic interpretation would not be post-semantic since it would run even without sentential cues. Hence, this highlights Minimalism’s struggle: the semantic system alone cannot explain how hearers interpret utterances with novel or shifted figure terms. Instead, they point to the necessity of pragmatic context-sensitivity and inferential processes to fill the gap, motivating more context-driven frameworks like Contextualism.

3. Contextualism

In very broad terms, contextualism is the view that modulation processes are essentially involved in the interpretation of utterances. Without contextual modulation no proposition could be expressed. Contextualists maintain the view that utterances cannot have literal meaning, because it does not make sense to speak of the meaning of an utterance in general. Utterances, therefore, do not have two levels of meaning-one that is determined through semantic interpretation and another that is determined by pragmatic interpretation. Pragmatic interpretation is always involved in the process of utterance production and understanding. Now, where they disagree is concerning the level at which the utterance is affected by these processes. The two positions I will be focusing on are the Pragmatic Composition view (hereafter PC), a version of which I will be defending through the development of the scenario-building framework, and Meaning Eliminativism (ME), whose rejection will shed light on those problems of contextualism that my framework is able to avoid. What should become clear is that while they all maintain that the utterance as a whole is affected by modulation, they disagree concerning whether the understanding and use of individual words also involves modulation.

According to PC, while words can have literal meanings-their sense can be nothing more than the expressed sense-modulation is always involved in the production and understanding of an utterance because the process of semantic composition requires that the appropriate adjustments take place in order to make the parts fit together. In other words, while words such as “cut”, “the”, and “grass” may have literal meanings that one could, say, look up in the dictionary, when we consider the utterance “Cut the grass” as a whole we cannot understand it without first making sense of what it means for one to cut the grass, as opposed to cutting a cake, or cutting hair. While the word “cut” is not ambiguous, nor would it be used in a homonymous or polysemous way in “cut the grass” and “cut the cake”, it makes in each instance a different contribution to the truth conditions of the sentence. According to PC, this is because background assumptions-what Searle calls “the Background”⁵ -play a role in fixing the meaning of the term in the context of the sentence, or what philosophers of language call its “satisfaction conditions”, i.e., what counts as cutting in the context of the utterance. In order to argue for the necessity of background assumptions in the process of understanding an utterance as a whole, Searle proposes his now famous example of “Cut the sun”, a combination of words that is so far-fetched that we have no background assumptions to import. According to Searle, there is no way to understand this utterance precisely because we do not have background assumptions that would allow us to make sense of what it means to cut the sun. In other words, we cannot discriminate between possible worlds in which the satisfaction conditions of the utterance have been met and those in which they haven’t.⁶ Therefore, without these assumptions it would be impossible to retrieve from memory a possible sense of “cut” such that it would allow us to construct a scenario that makes sense of the utterance “Cut the sun”. Unlike with our other examples involving the word “cut”, here we cannot go beyond the literal meaning of this word in order to adjust it to the object clause quite simply because the literal meaning of the verb clause and the object clause, when taken together, don’t make sense.

It should be noted that PC’s claim that word meanings are adjusted contextually depending on the other words with which they cooccur finds empirical support in the previously discussed 1983 study on nonce verbs by Clark and Gerrig. Their findings showed that participants interpreted novel verbs, e.g., ‘to pilk’, by pragmatically modulating their meaning based on direct objects (here the newspaper). If this empirical evidence does in fact support PC, then it would show that this approach, unlike literalism, is able to explain-without needing to resort to ad hoc caveats about pragmatic processes-how hearers can understand utterances which employ deviant terms, such as in the example of “George managed to porch the newspaper yesterday”.

Now, while PC maintains that words have literal meanings that need to be adjusted to fit co-occurring terms (a position that is at the heart of the scenario-building framework I will develop in the next section), the most extreme form of contextualism, ME, argues that word meanings emerge solely from situated use. According to ME, communication involves no retrieval of stored lexical meanings, but rather generates meaning entirely through contextually-sensitive processes operating on prior usage episodes. On this view, what appear to be conventional meanings are merely consistent patterns of use, with no independent linguistic status. This radical form of Contextualism has, most notably, been defended by ^{Travis (2008)}.

Luca ^{Gasparri (2013)} presents an interesting opposition to this view. He rejects it on the basis that it is too cognitively demanding to reflect how we actually understand language. He presents two arguments against ME: the first is from the psychological indispensability of lexical concepts and the second is from the linguistic indispensability of conventional lexical schemes. Concerning the former, Gasparri first remarks that we have the intuition that “our lexical system can be functionally represented as a network of entries under which semantic information is stored in the form of lexical concepts” (^{Gasparri 2013, p. 1019}). This is an intuition that ME directly undermines because it rejects the idea that we could have lexical concepts generated through an abstraction process which takes as its input the past uses of the corresponding term. According to ME, interpretation is a one-step abstraction/modulation process which takes past uses as its input and which has, as its output, a particular use of the term. Gasparri remarks that if we accept such an account then there is in principle “no reason why speakers’ procedures of lexical access should comply with any constraint of regularity” (^{Gasparri 2013, p. 1021}). The set of possible interpreted values we can assign to a word is in principle infinite if it is not constrained by its lexical meaning, and, according to Gasparri, this means that there is nothing preventing a speaker from using, for example, the word “compass” instead of “blue” to describe a blue object. He, then, imagines that proponents of ME would construct a reply that attempts to show that knowledge of past uses would constrain the possible uses of a term. His counterargument is two-fold. Firstly, he points to the incoherence of such a view with empirical evidence that tends to show that the recollection of past lexical uses of a term is a matter of episodic memory,⁷ and that deficits in episodic memory do not covary with deficiencies in lexical access. The second issue Gasparri raises is that the step of comparing past situations in which a term was employed with the target situation-in order to judge whether these two are sufficiently similar when it comes to aspects that are salient to the application of that word-makes the model too laborious from a computational standpoint and, therefore, incompatible with the existence of fast and seemingly unreflective semantic phenomena such as semantic priming and word-association.

Gasparri’s second argument-the one from the linguistic indispensability of conventional lexical schemes-raises the issue that there seems to be more stability in the assignment of values to expressions by members of a same language community than the ME model suggests. He rejects from the outset the strategy that would consist in appealing to past uses to argue that ME can deal with this phenomenon because the idea that there can be deviant uses of a term entails that the term has possible conditions of application, and this would mean that there is a type of semantic information accessible to users that goes beyond recollection of past uses. We can understand this as saying that the use of a term not being part of the set of past uses of that term is not a sufficient condition for the new use to be qualified as deviant. However, Gasparri doesn’t seem to take this line of argumentation. He rather appeals to competent speakers’ “strong intuition” (^{Gasparri 2013, p. 1022}) that they can identify deviant uses of a term not because of a recollection of past uses but because they know the meaning of the term. Furthermore, he rejects the idea that we could account for the stability of the lexical use of a term by appealing to the correction that the language community exercises because this would not allow us, for example, to explain which figurative uses of a term are correct. Gasparri privileges the following model:

The ability to evaluate that a genuinely metaphorical and non-ironical use of the expression “Mark is a lion” is based on a felicitous association between a property (say, Mark’s outstanding moral strength) and a class of objects (in this case, lions) seems to depend directly on the knowledge of the lexical/encyclopedic dependability of the expression “lions are strong animals”, not on the ability to build an online abstraction of the property “being strong” from the inspection of all the referents associated in past uses to occurrences of lion. (^{Gasparri 2013, p. 1022})

He maintains that this model can best account for the speed, effortlessness, and confidence with which speakers produce these kinds of judgments. It also aligns with both empirical evidence concerning how young children acquire the ability to process and employ figurative uses and empirical evidence indicating that “proficiency in metaphor acquisition and comprehension is highly covariant with lexical skills rather than with abilities related to intention ascription, episodic processing, and theory of mind” (^{Gasparri 2013, p. 1022}).

Gasparri’s description seems to capture something correct about how we process figurative uses of an expression. When processing such expressions, the success and failure of understanding seems to be determined by whether we can find similarities or links between the properties we assign to the subject and the predicate terms. This process of finding parallels between properties can be illustrated with a graded series of examples using the term “disarm”:

The term “disarm” is, first and foremost, associated with the idea of taking physical objects away from a person or country, such as in (1). When the term “arm” is itself taken in the figurative sense of “resources used for offensive or defensive measure”, then we can get something like (2). We typically associate the idea of arms with that of things that can be used to cause harm or distress, so the use of “disarm” in association with the term “opponent” makes intuitive sense. Both arms and opponents share the property of potentially causing harm or distress so their association feels natural. Now, when we move on to (3) the association of “disarming” and “slides” requires a more imaginative exercise to be grasped, unless one already knows that it here means “diffuse”. However, when we consider the spatial properties of slides, that they are horizontally extended, and when we can incorporate the idea that these are inflatable slides, and the idea that we are on a plane, we can imagine the scenario of an inflatable slide shooting out. So we can associate slides with being disarmed through this property. However, we do need some contextual information to make this association. With regards to (4) we need to know that there is a gun on the plate of fricassee in order to understand the utterance. There is no sense in which a fricassee can be armed. Fricassees do not have any properties that can, by any stretch of the imagination, be associated with arms unless either the fricassee is itself an arm, in which case it would not actually be a fricassee (this is a sort of Trojan Horse scenario in which someone has disguised, for example, a bomb to look like a fricassee), or, more cartoonishly, a scenario in which the fricassee is a sentient being holding a weapon.¹⁰ It seems that the less a term is used in combination with other terms with which we typically associate it, the more understanding requires an imaginative exercise consisting of finding parallels between the properties of the ideas associated with the terms. This imaginative exercise consists in constructing scenarios in which the association makes sense.

The ME framework doesn’t seem to provide the resources necessary to account for the construction of these scenarios. According to this framework, there needs to be a sufficient resemblance with regards to the salient properties of the predicate P between past situations in which P was applied and the target situation. The set of source situations is the semantic potential of the predicate, and it is the target situation that must possess certain features making it sufficiently similar to the source-situations. The problem with the one-step abstraction/modulation process is precisely that it requires similarities between past situations and the target situation. By accepting this token-to-token relation, ME actually proposes a framework that bears more resemblance to Literalism than meets the eye. Indeed, let’s consider the example from Gibbs: “George managed to porch the newspaper yesterday” (^{1984, p. 10}). In his argument against the notion of a sentence’s literal meaning, Gibbs remarks that according to this view of interpretation one would have to look up the term “porch” in the lexicon, see that it is there only classified as a noun, and be unable to continue interpreting the sentence. Now, dictionaries don’t just catalog conventional meanings, they also catalog conventional uses. But if we were to look up these past uses of the term “porch”, we would not find an example of an utterance in which the noun is used as a verb. Thus, Gibbs’ utterance is itself a target situation which seems insufficiently similar to past situations with regards to the application of the noun “porch” precisely because it is there not employed as a noun.

Furthermore, in “I sat on the porch” and in “George managed to porch the newspaper yesterday”, it is clear that we do not have a case of homonymy, but nor do we have a case of polysemy. What we have is a deviant use of the term “porch” that is, nonetheless, made felicitous by an exercise of the imagination. However, this imagination is not free floating, it is restricted by a type of semantic information accessible to users that goes beyond recollection of past uses and beyond literal or conventional meaning. And unless it comes to be conventionalized, the expression “to porch” makes no sense outside of the context of the utterance as a whole. Indeed, an utterance such as “I porched” would be impossible to understand without the addition of an object clause because there would be insufficient information for modulation to take place. This is clear by how unenlightening all of these exchanges would be:

The point here is not to say that contextual information is insufficient for understanding, but rather to point out that this contextual information has to be the right kind of information, that is, information that is enlightening as to the application of a term. By “enlightening” I mean information that is relevant to the construction of a scenario that makes sense of the use of a term in context.

By “scenario” I do not mean a mental picture or a simulationist reconstruction of a concrete situation, but a structured representational situation: a constrained subset of possible situations compatible with the interlocutors’ shared assumptions, linguistic conventions, and communicative aims. Scenario-building, therefore, should not be understood as the speaker and hearer’s joint attempt to imaginatively simulate an event, but rather as their cooperative effort to delimit a space of possible interpretations that satisfies the utterance under mutually recognizable conditions. The distinction between enlightening and unenlightening contextual information, i.e., between contextual information that supports scenario-building and context that doesn’t, becomes clear when we consider examples 5 to 9: the contextual information provided would clearly be unenlightening to any person who had never encountered Gibbs’ sentence. But this is not a point for ME because an understanding of Gibbs’ sentence is possible regardless of the fact that one may never have encountered a similar use of certain terms-as is remarked by ^{Clark and Gerrig (1983)} in the passage cited in the previous section. It is the structured, cooperative process of scenario-building depending on the right kind of contextual information, not a mere recollection of past uses, that allows for the understanding of novel utterances.

Now, the question we need to address is that of how we determine what the right context is.

4. Understanding a Communicated Thought

When we consider the debate within the philosophy of language, it’s easy to get lost and forget that it is firstly about the nature of the content that is transmitted and received in communicational exchanges. According to strong Literalism, in ordinary communication the content of an utterance is a contextless proposition; it is produced by semantic interpretation, i.e., the process of determining the truth conditions of a proposition expressed by any sentence of a language the knowledge of which is like the knowledge of a system of rules from which the truth conditions are deductively established. If this contextless proposition is true, then it would be true regardless of the time or place in which the utterance takes place, of the speaker who is making the utterance, and so on. According to weak Literalism, i.e., Minimalism, all utterances have a minimal propositional content-a content that is not affected by any pragmatic processes-and pragmatic processes only operate in a post-semantic, and semantically restricted, manner. Finally, according to Contextualism, there is no such semantics/pragmatics distinction because there is only one kind of content, the communicated content, that is essentially affected by modulation processes. Now, the question of our relation to contents is, at least partially, an epistemological question: to determine what and how much information is required to understand the content conveyed by an utterance is to ask what the hearer must know, believe, or Φ in order to achieve this understanding. So, let’s look at what epistemologists working on the relation we call “understanding” have to say about the matter.

There are not many works within the epistemology of understanding that center on the topic of language, perhaps because epistemologists tend to focus on cognitive relations to contents in the abstract propositional sense. These are sometimes said to be of special interest for epistemologists.¹¹ However, the article “Understanding a Communicated Thought” stands out in this regard. ^{Carter et al. (2021)} argue that linguistic understanding and propositional understanding are distinct yet both involved in understanding a communicated thought, i.e., in understanding a communicated content, as philosophers of language would call it. They define linguistic understanding as the result of a capacity to “recognize what is being said or what is the meaning of a given utterance of a sentence in a language one knows” (^{Carter et al. 2021, p. 12147}). Thus, supposing that Alex did not possess the relevant concepts, or that she was unable to grasp the relations between the constituents of the proposition “The cytoplasm is enclosed by the cell membrane”, she will fail to achieve propositional understanding, but could nonetheless linguistically understand an utterance of “The cytoplasm is enclosed by the cell membrane” as long as she is sufficiently proficient in the English language to succeed in identifying what has been said. In other words, if she is able to identify the proposition being communicated, regardless of the fact that she is unable to propositionally understand it. And supposing that Alex is a biologist who only speaks Polish, she would be able to propositionally understand that the cytoplasm is enclosed by the cell membrane although she would be unable to linguistically understand an utterance of this sentence. Now, while these two types of understanding are distinct, Carter et al. maintain that they can co-occur, and their co-occurrence is required for understanding a communicated thought. So if I said “The cytoplasm is enclosed by the cell membrane”, Alex would understand the thought I am communicating if (1) she linguistically understands it, and (2) if she has some degree of propositional understanding. Thus, understanding a communicated thought requires both identifying what is said, i.e., representing the content of the utterance, and grasping the content.

Now, we might ask just what is a communicated thought, or more precisely “a thought as communicated through a given utterance” (^{Carter et al. 2021, p. 12148}). Firstly, we know that it is not identical with what has been said, since the ability to recognize what has been said, what proposition was expressed through the utterance, or to recognize the meaning of the utterance, is not sufficient for the ability to understand a communicated thought. One may be able to recognize p as being what has been said, yet fail to understand the communicated thought because one is unable to achieve a propositional understanding of p. But why is this insufficient? Carter et al.’s position is that linguistic understanding is insufficient for understanding a communicated thought because of the lack of propositional understanding. Yet, why is this also required? Why do I need to have a grasp of the expressed proposition in order to understand the thought being communicated? Consider an utterance of the sentence “Your cell phones are spying on you”, and suppose that on one side you have a speaker S1 who is ignorant concerning most technology related subjects-S1 doesn’t know how computers work, has no knowledge of cameras, and is more generally the type of person that could be afraid of 5G. On the other side you have S2 who is a specialist in cybersecurity. Could you say that S1 and S2 have communicated the same content by uttering this sentence? I don’t think so. It seems that while they have said the same thing they have communicated vastly different contents. The content communicated by S1 is something like “[insert name here] is looking at you through your camera lens”, while the content communicated by S2 would be something like “Through the collection of your searches, response time, activity, and data more generally, [insert name of any search engine LLC] collects incredible amounts of information about you that it gives or sells to third party private companies, states, governmental institutions, etc.”. A hearer who knows the English language can acquire linguistic understanding whether the utterance was made by S1 or S2. However, only a hearer who has the required knowledge about how search engines work can understand S2’s utterance. This is Carter et al.’s claim. Now, there is another point to take into consideration here. Among the information required to understand the thought being communicated, we will find some information that is not found within the proposition itself. In our example, it would be information concerning the speaker. To see why, imagine you were given the information about S1 and S2 that I provided above. But this time suppose that you do not know, and have no way of knowing, which speaker is S1 and which one is S2. You would be unable to understand the thought being communicated because you would be unable to identify which thought is being communicated. S1 and S2 don’t mean the same thing by “spying”, although the senses they pick out are related. S1 means looking at you, watching you in the visual sense, while S2 means something more indirect, like non anonymous data collection. These two senses are clearly part of a family of senses associated with the term “spying”, and it seems that to achieve the second requisite for understanding a communicated thought-propositional understanding-the hearer would have to know which sense the speaker has in mind. Yet, the hearer would be unable to do so if she did not import non-linguistic information concerning the speaker.

This dual structure of linguistic and propositional understanding provides the groundwork for explaining how interlocutors can, in ^{Longworth’s (2025)} terms, think together. Recent accounts of thought-sharing and communicative alignment (^{Pickering and Garrod 2004}; ^{Zavaleta 2023}; ^{Zaefferer 2019}) similarly view understanding as a process of coordinating representational perspectives rather than transferring fixed content. Yet unlike most accounts of thoughtsharing, which remain largely descriptive, the model here proposed aims to specify the structural conditions under which such coordination succeeds or fails thereby grounding the gradability of understanding in the measurable overlap of interpretive scenarios. The next section discusses the means by which this overlap can be measured.

5. Modeling Degrees and Conditions of Understanding

There are two senses in which one could fail to have propositional understanding. In the first sense, the objective sense, the proposition expressed by the utterance imposes on the speaker its conditions of understanding (or, perhaps we could call these, interpretive success conditions). In this case S1 would have a lower degree of propositional understanding than S2. In the second sense, the more relative sense, it is what the speaker means that determines the conditions of understanding: S1’s utterance is satisfied in all possible worlds in which someone is in fact watching you through your phone’s camera, while S2’s utterance is satisfied in all worlds in which your data is being collected in a non-anonymous manner. If we take seriously the idea that propositional understanding is gradable, then we will have to maintain that it is so in both cases. In the second one, the gradability will depend on how close the hearer has gotten to the representation that the speaker has in mind while making the utterance (or that leads her to make this utterance more generally). But, whatever picture we endorse, it seems that to understand the thought being communicated, one would have to understand the thought that the speaker is intending to communicate. For suppose that S2 hears S1’s utterance of “Your phone is spying on you” and takes it to mean that his data is being collected. Has he understood the thought being communicated? I don’t think so. Whether we maintain that this is because he has a too high degree of objective propositional understanding or the wrong picture all together, it is clear that there is a fair amount of pragmatic modulation to make sense of which propositional content the speaker is actually conveying. This modulation would have to take place depending on the Background, that is, a certain set of assumptions that the speaker has. This is reflected in the sense she takes each term to have and the relations she takes to hold between them.

You may object that this is too high a standard for ordinary communication, and to this I would respond in the following manner: the understanding we achieve in ordinary communication is usually approximate. We rarely if ever understand exactly what other people are saying unless (a) we know them very well, (b) the conversation has been set up in an extensive manner, (c) we are in the kind of situation that precludes the need to extensively set up the conversation,¹² or (d) we are describing a state of affairs to which we are (or have been) jointly attending. Communication is often messy and requires a certain amount of correcting, modifying, and specifying that can only take place with more communication (whether verbal or otherwise). This being said, we have to ask whether S1 and S2 do not understand each other at all, i.e., whether there is absolutely no overlap between S1’s representation and S2’s representation. This question is pertinent whether one endorses the first view described above, or the second one. If you endorse the view that propositional understanding fails because S1 and S2 have different degrees of objective propositional understanding, then the question is whether those degrees are ranges that may, nonetheless, overlap at some points. And if you endorse the view that it is what the speaker means by her utterance that determines the conditions of understanding, then the question is whether the hearer’s representation may, nonetheless, overlap with that of the speaker regardless of the fact that it does not capture it completely.

To answer this question, let us begin with the second case. S1 and S2 are communicating different contents when they say “Your phone is spying on you”. To clarify the conceptual structure of this graded understanding, we can formally model it using the tools of possible worlds semantics. S1’s utterance is satisfied in all possible worlds in which someone is watching you through your phone’s camera, while S2’s utterance is satisfied in all worlds in which your data is being collected in a non-anonymous manner. If we maintain that S1 and S2 do not understand each other at all, then we would have to say that there is no overlap between the two sets of possible worlds. This would mean that although S1 and S2 are using the same words, they nonetheless divide this modal space in a way that yields two completely distinct sets of possible worlds. Let’s call these PWs₁ and PWs₂ to refer respectively to the sets of possible worlds in which S1’s utterance is satisfied and those in which S2’s utterance is satisfied. This would, again, mean that when S1 uses the word “spying”-the problematic term in this example-its use would not be satisfied, so to speak, in any of the possible worlds making up PWs₂. In that case, whether S1 uses the word “spying” or a nonsensical word such as “blanking” there should be no difference. But there is clearly a difference. The word “spying” is one that the hearer knows or, at the very least, it is one that she can look up in a dictionary. Although the cybersecurity expert uses the word in a more technical manner, this use is nonetheless related to the standard definition of the word. Let’s take this standard definition to be the following minimal definition of the verb “to spy”: to secretly collect and report information about activities.¹³ Given this definition, there are clearly worlds at which what S1 means by “spying” and what S2 means overlap. These are the worlds at which the use of the term is satisfied by the secret collection of visual information, or images.¹⁴ Therefore, PWs₁ and PWs₂ do overlap, which, if the analysis proposed so far is correct, means that S1 and S2 do understand each other’s utterance to a certain degree. Furthermore, it seems that the degree of understanding and the area of the modal space at which PWs₁ and PWs₂ overlap will covary. When PWs₁ and PWs₂ perfectly overlap, there is the highest degree of understanding, and when they partially overlap there is a lower degree of understanding and, conversely, a certain degree of misunderstanding.

This conception aligns with, yet diverges from, recent accounts of partial or imperfect understanding in the philosophy of language. For example, ^{Picazo’s (2022)} analysis of “impoverished contents” in gradual communication explains communicative success under conditions of semantic incompleteness: hearers grasp an attenuated but truthapt version of what the speaker means. By contrast, the scenariobuilding model relocates the source of gradience from propositional impoverishment to the degree of scenario overlap between interlocutors. Understanding fails not because the content communicated is underdetermined, but because the interpretive scenarios constructed by speaker and hearer fail to sufficiently converge.

Now, it should also be noted that while two parties may communicate the same content when saying “Your phone is spying on you”, and, thus, have perfectly overlapping sets of possible worlds in which their utterance would be satisfied, they will only count as understanding each other’s utterance-in this case to the highest degree-if they in fact attribute this representation to each other. In other words, for understanding to be achieved it does not suffice that S2 and S3-another cybersecurity expert-mean the same thing, i.e., communicate the same content, they also have to attribute that content to one another. If S2 says “Your phone is spying on you” to communicate that your data is being collected in a non-anonymous manner, the fact that S3 would have used that sentence to communicate the same content will not suffice for (perfect) understanding to be achieved. She will also have to identify that content as being the content that S2 has communicated, or intends to communicate.

Otherwise, we would get the same outcome as when PWs₁ and PWs₂ only partially overlap. To see why this is a necessary condition, let us get back to the case in which there is only partial overlapping. S1 makes the utterance to communicate the content that someone is looking at you through your phone camera. S2, being a cybersecurity expert, would mean something quite different if she were to make this same utterance. Now, suppose she knows that S1 is not an expert in this field, better yet that S1 has very little, if any, knowledge of these matters and that S1 has covered her device’s camera. Furthermore, she knows, whether explicitly or implicitly, the minimal definition of “spying” I proposed above. Granted S1 is speaking in a language that S2 understands, S2 is able to identify the thought being communicated by S1 when she says “Your phone is spying on you” although this is not the thought she herself would have communicated through this utterance. This means that, using the information she has acquired, S2 is able to construct the scenario in which S1’s utterance would be satisfied. But if S2 had falsely believed that this information applies to S3, and that S1 is the other expert in the room, then she would have attributed this thought to S3 and the thought about data collection to S1, and we could plausibly claim that she has misunderstood them. Therefore, the identification of the thought being communicated cannot be sufficient for understanding an utterance because attributing that thought to the speaker is another necessary condition for understanding to be achieved. This entails that two speakers need not use an utterance to communicate the same thought in order to understand each other, i.e., the possible worlds in which their utterances are satisfied need not perfectly overlap, as long as they can identify the thought that the other is trying to communicate as the thought that she is intending to communicate, i.e., when they are able to identify that (or those) possible world(s) in which it would be satisfied.

6. Concluding Remarks

This article has developed and defended a scenario-building framework for linguistic understanding in which meaning emerges through the interplay of conventional scaffolding and pragmatic co-creation. Drawing on insights from philosophy of language, epistemology, and psycholinguistics, the framework aims to preserve the strengths, and avoid the weaknesses, of both rule-based semantics and pure contextualism, particularly in addressing their shared difficulty with deviant uses of terms in communication.

Scenario-building, though flexible, operates under two constraints: (1) speakers must supply sufficient conventional scaffolding, and (2) hearers must share enough conventions to anchor interpretive creativity. When these conditions fail, misalignment results, which explains, for example, the incomprehensibility of Searle’s “Cut the sun”. This framework builds on the PC view, and thus finds its intuitive roots in Contextualism. It nevertheless rejects both the top-down enrichment favored by strong contextualism and the strictly bottomup inferences of literalist pragmatics. Instead, it treats conventions and context as co-constitutive: terms such as spying are not merely enriched but reimagined within jointly constructed scenarios guided by shared knowledge and speaker intention.

This approach departs from standard contextualism by making mutual construction central, rather than treating hearer inference as the primary driver of interpretation. It narrows the semantics- pragmatics gap by applying the same scenario-building process to both routine and deviant cases, avoiding the minimalist bifurcation between literal content and intended meaning. Furthermore, the scenario-building framework is not intended as a loose metaphor for contextual enrichment but a structured process with identifiable stages, success conditions, and failure modes. Beyond explaining deviant utterances, I believe that it also offers tools for addressing pressing questions in the philosophy of language, such as how polysemy is resolved in practice and whether communication involves genuine thought-sharing.

Finally, the scenario-building framework can be seen as contributing to the contemporary debate on thought-sharing and communicative alignment (^{Pickering and Garrod 2004}; ^{Zavaleta 2023}; ^{Zaefferer 2019}). These discussions converge on the idea that communication involves more than the transfer of fixed contents: it requires coordination between interlocutors’ cognitive perspectives. The scenario-building approach provides a concrete model for this coordination: speakers and hearers do not merely exchange propositions but collaboratively delimit interpretive scenarios whose overlap constitutes shared understanding. By modeling communicative success as graded scenario alignment rather than as the transmission of identical thoughts, this framework extends existing accounts of thought-sharing (^{Picazo 2022}) and situates linguistic understanding within a broader picture of cooperative cognition.