Future auxiliaries present a challenge to the classical analysis of modal expressions as existential or universal quantifiers over a contextually provided set of possible worlds: these expressions come with a distinct modal flavor, but their interaction with negation and the fact that future judgments come in degrees of confidence is unexpected if the classical analysis were correct. I show how a uniform quantifier analysis of modal expressions can accommodate the distinct empirical phenomena surrounding future auxiliaries. The resulting analysis extends to address a set of related challenges that have been observed for a quantifier analysis of ability modals.
Joint inquiry requires agents to exchange public content about some target domain, which in turn requires them to track which content a linguistic form contributes to a conversation. But, often, the inquiry delivers a necessary truth. For example, if we are inquiring whether a particular bird, Tweety, is a woodpecker, and discover that it is, then our inquiry concluding in this fact would conclude in a necessity, and the form “Tweety is a woodpecker” expresses this necessary truth. Still, whether Tweety is a woodpecker seems a perfectly legitimate object of study, and the answers we accrue can be informative. But the dominant model of inquiry (Stalnaker, 1978, 1984) treats this situation as linguistically deviant, and diagnoses our ignorance and subsequent discovery as metalinguistic: we were ignorant, and ultimately discovered something, about the meaning of our terms. Rather than linguistic deviation, we argue this situation is the norm, and one that calls for an alternative model of inquiry. This paper develops such a model. It shows that to capture how agents can learn something informative about the world—and not merely language—even when inquiry concerns necessary facts, it's key to track how moves in discourse contribute public content onto the conversational record, but also, crucially, how those moves are connected by coherence relations to one another and to real-world situations they are about. This allows us to capture that utterances contribute determinate, public content, while representing the information states of the interlocutors who may have only partial access to the evidence and content of the conversation, without making their ignorance metalinguistic. It lets us give precise explanations why some discourses can be transparently convincing in the conclusions they underwrite. The model thus precisifies the role of public context and shared content in anchoring an inquiry. It allows for imperfect tracking of linguistic contributions that are binding for how inquiry unfolds, and it allows for an inquiry into the status of necessary truths to be both informative, and involve empirical, rather than metalinguistic, ignorance.
We can use “reason,” with its normative sense, as both a count noun (“there is a reason for her to Φ”) and a mass noun (“there is plenty of reason for her to Φ”). How are the count and mass senses of “reason” related? Daniel Fogal argues that the mass sense is fundamental: Just as lights are merely those things that give light and anxieties are merely those things that give anxiety, reasons are merely those things that give reason. In this article, I develop an opposing analysis of the mass noun “reason” that puts reasons first. Just as the detail on the Mona Lisa is composed of particular details (brushstrokes and colors) and the crime in L.A . is composed of particular crimes (pickpocketings and speeding offenses), so the reason for you to go to the dentist is composed of your reasons to go. Reasons stand to reason as parts to a whole. Such a picture makes reasons fundamental once more, but it has a cost of entry. In order to accommodate the behavior of “reason” in comparative constructions, you need to abandon the idea that reasons are facts we can count up. On the contrary: They're not facts, and you can't count them.
I explore constructions which contain an embedded disjunction or , which is interpreted as , where is a possibility modal whose flavor is epistemic, circumstantial, or deontic. I show that no extant theory can account for these interpretations. I argue that the best way to do so is with a theory on which or simply means . In addition to accounting for the novel cases I discuss, this theory explains both wide- and narrow-scope free choice inferences, in a similar way to the theories of Zimmermann (2000), Geurts (2005), and Goldstein (2019) which it builds on. It also accounts for recent observations about the relation between disjunction and possibility from Degano et al. (2025) and Feinmann (2023).
This paper gives a new account of the actuality entailments of ability claims. We observe that, in the environments which give rise to the actuality entailment, ability claims carry a presupposition of trying. We show that, given this presupposition, the actuality entailment is straightforwardly predicted by a conditional theory of ability. We give a theory of why this presupposition arises, showing how to derive it from a presupposition of strong settledness. We then show how to predict contextual variation in the precise content of our presupposition; how to extend our account to apparent deontic and teleological actuality entailments; and how to adapt our account to more sophisticated conditional accounts of ability.
This article challenges conventional boundaries between human and artificial cognition by examining introspective capabilities in large language models (LLMs). Although humans have traditionally been considered unique in their ability to reflect on their own mental states, we argue that LLMs may not only possess genuine introspective abilities but potentially excel at them compared to humans. We discuss five objections to machine introspection: (1) the lack of direct routes to self‐knowledge in training data, (2) the conflict between static knowledge and dynamic mental states, (3) the distorting effects of reinforcement learning on self‐reports, (4) LLMs own denials of inner experience, and (5) arguments that LLMs simply mimic language without understanding. We think all these arguments fail and that there are deep parallels between human and machine introspection. Most provocatively, we propose that LLMs superior processing capabilities and pattern recognition may enable them to develop more sophisticated theories of mind than humans possess, potentially making them more reliable introspectors than their creators. If we are right, this has significant implications for artificial intelligence (AI) alignment, transparency, and our understanding of the nature of AI.
We explore the consequences of a natural and well‐motivated modeling assumption of Bayesian epistemology, according to which the objects of credence are sentences in the agent's language. We show that this assumption is inconsistent with two further natural Bayesian idealizations: those of Logical Perfection (the logical‐deductive consistency and closure of the sentences in which the agent is certain) and of perfect access to (i.e., certainty regarding) the presence or absence of certainty in any particular sentence in the agent's language—an assumption we call “Luminosity.” That is, it turns out that a luminous agent is an irrational agent. After presenting our main result and some related impossibility results concerning variations on Logical Perfection and Luminosity, we briefly examine the range of candidate solutions to the puzzles that these results present us with.
A popular idea in ethics is that subjective normative concepts play an important role in moral deliberation: They are taken to be action‐guiding. It is generally assumed that in order for these concepts to be able to guide an agent's actions, they need to be “informationally accessible” to the agent in a substantive sense. That is, access holds: access: Subjective normative notions are accessible to agents. access has been spelled out in various ways, for example, via knowledge, justified belief, and evidence. We present a novel argument against access on any precisification. Our argument is distinctive for at least two reasons. First, most discussion in the literature on action‐guidance concerns obligation. But we draw attention to subjective permission. We show that on virtually any substantive construal of “accessible,” subjective permissions are not always accessible to agents. Second, existing criticisms of access are motivated by general epistemological considerations, for example, the anti‐luminosity argument. By contrast, our argument against access is motivated by properties that are particular to normative concepts, and is neutral on more general epistemological debates.
When we see a movie or a play, do we see the fictional entities and events depicted? On the one hand, it seems incredibly natural to think we do. For instance, it seems obvious that one thing that differentiates Smith, who watches Star Wars, from Bob, who merely reads the novelization of Star Wars, is that Smith, but not Bob, has seen Darth Vader kill Obi‐Wan Kenobi. Yet, no philosophers working on fiction think this is literally true. And they have good reasons to be skeptical. For, if you have seen Darth Vader kill Obi‐Wan Kenobi, then it seems to follow that Darth Vader must have killed Obi‐Wan Kenobi, in which case, it follows that both were at one point living, flesh‐and‐blood, entities. But if Darth Vader is a flesh and blood being, then he must be spatiotemporally located, in which case, where is he? In this paper, I argue that we do in fact literally see (and hear) fictional entities when we see films. I do so in three stages. First, I argue against various error theories that attempt to account for the intuitions that we do see fictional entities in film. Then, I sketch a metaphysics of fictional entities, which vindicates our genuinely seeing them. Finally, I explore some of the interesting controversies and objections raised to this ontology of the fictional.
In a pure event semantics for natural language, the domain of quantification and predication is limited to events and states. I offer pure event semantic analyses of several phenomena, some of which have not been treated before in formal semantics. In the pure event semantics sketched in the second section, nouns are state predicates, and this provides the starting point for the analyses. The phenomena involve grammatical number, the mass-count distinction, adjectival modification, count adjectives, diminutives, lexical plurals, duals, and mass gender. In the conclusion, there is a brief discussion of potential metaphysical or psychological ramifications of doing semantics this way.
To measure is to err. Serving both numeric and non‐numeric measurement, the language of measurement refers to margins of error, within which measurement reports locate their measurements. Such reports and reasoning from them invoke what is known and what is known to be known about error‐strewn measurement to derive and contrast the implicatures of bare and comparative measures.
Generics about social categories, like “Black people are tall” and “Women are nurturing,” have, often rightfully, been subject to serious criticism. Some theorists have even adopted the prohibitionist stance that all social generics should be avoided. Others hold that this goes too far and that some social generics, such as “Black Americans are economically disadvantaged” and “Women are expected to want children,” can be helpful for describing our social world. There is not, however, enough clarity on what determines whether a social generic is problematic or helpful. I distinguish between essentializing social generics, which are almost always harmful, and positional social generics, which tend to be helpful. I then argue that this distinction cannot be drawn at the level of the generic sentences in isolation, but must appeal to context and specifically question under discussion (QUD). I discuss how this key idea can be implemented on approaches that rely on either pragmatic reasoning or semantic context-sensitivity. Overall, I show that the responsible use of social generics depends on careful attention to what is at issue in the particular conversation.
Quotation marks in natural language that do not function straightforwardly as devices for securing reference to linguistic objects have generally been categorized as instances of either mixed quotation or scare quotation . I argue that certain uses of quotation marks in natural language resist assimilation to either of these two theoretical categories, as well as to the more familiar categories of pure and direct quotation. It follows that we should recognize a further type of quotation in natural language, which I call dummy quotation because it involves the contribution of semantically minimal “dummy meanings” to composition. I develop a semantic theory of dummy quotation and show that it is better able than rival proposals to account for the troublesome examples in question.
