Content, modals and attitude predicates

1.1 The problem of content

In the classical possible worlds approach to content, the object of John’s belief in (1-a) is indistinguishable from the object of his belief in (1-b) because the complement clauses, both being logically necessary given a particular mathematical language, are true in exactly the same worlds.

However, it seems intuitively clear that John could possibly believe one of those propositions without believing the other. The problem arises because identifying the content of a proposition with the set of worlds in which is it true is too loose, and generous and does not do justice to the narrow specifics of what each proposition is about. Exact verification is a way of getting at the specific content of a proposition without getting over specific and requiring identity of syntax or word choice (Moltmann discusses the structured propositions proposal of Cresswell 1985 as an example of this overly fine way of representing propositional content).

Of course, this problem has been well known in the philosophical literature for a long time, and has never been satisfactorily solved. Why then have we semanticists, as a field, been ignoring it? I think the reason for this is that descriptively speaking, there are very few instances of actual get-your-hands-dirty semantic analysis that require one to consider such cases. The view seems to be that a well-chosen set of examples can upset the toolbox, but this is of interest only to philosophers (or semanticists with their philosophy hats on), it need not disturb the ordinary day to day business of writing semantic descriptions. This pragmatism, coupled with the belief that someone will eventually figure out a solution and the semantic descriptions in the old system will simply be translatable into the new one, seems to justify persisting with the possible world semantics framework, especially since it seems to be able to do many things quite easily and elegantly.

But, I would argue, the failure of the toolbox on examples like (1) is fatal. The whole business of semantics as a field of endeavor must surely be to give us a theory of why things mean what they do, i.e., a theory of ‘content’. We could imagine many computational systems that perform pretty close to human judgments about content, and overlap with them in the vast majority of cases, but which are not in fact exact representations of the human semantic system. The failure of the possible worlds toolbox on (1) shows us that truth in a world, or even in a set of all possible worlds, is not equivalent to a human speaker’s understanding of the content of a proposition, even though in many cases it looks like a good proxy for it.

So if we are interested in modeling ‘the human speaker’s understanding of the content of a proposition’, then we have to build a more realistic system (see also Pietroski 2005, 2018).

In fact, it is simply not true that the possible worlds toolbox works beautifully and elegantly in all other cases — it is continually throwing up puzzles and paradoxes for natural language semantics which semanticists then solve in increasingly baroque ways (see Ramchand 2018, 2019 on the imperfective paradox as a discussion of one such case).

The reason that Fine’s truthmaker semantics is a better starting point for a theory of content, is that it employs ‘exact verification’. This means that for a situation that verifies a particular sentence, all its parts are relevant, and there is nothing extraneous: “In truthmaker semantics, logically equivalent sentences will have different semantic values whenever they are about different things” (Moltmann, p. 168).

This feature of truthmaker semantics also enables a direct and intuitive account of the notion of partial content (Fine 2017b, 2017c; Yablo 2015). The problems for the standard toolbox arise from the fact that possible world semantics is not really tracking content, but is just a very good proxy for it in many cases. Problems that the standard toolbox have with ‘partial content’ in multiclause utterances, as summarized in Moltmann’s article, highlight a failure to meet a basic desideratum of compositionality — the bare and incontrovertible fact that human speakers of language systematically build more complex meanings from simpler units, in a way that is creative, open ended and reliable across users of the same code. If we cannot track how the partial meanings are contained in the larger ones they are built from, we are missing a basic intuition underlying compositionality.^[1]

Of course, going the exact verification route requires rethinking many of the core ways of thinking about possibility and modal notions in general, which is why the programme is potentially disruptive and revolutionary. After all, the treatment of modality and the intensional complements of attitude verbs has been thought of as one of the success stories of modern semantic theory. Do we really want to throw the baby out with the bath water? As we will see in Section 2, Moltmann proposes a theory that is designed to preserve the advantages of the classical account while still taking advantage of the superior ontological set up of truthmaker semantics.

2.1 On nominal polysemy and content

Moltmann’s analysis of CP complementation in natural language discussed in the previous subsection is part of a larger project on natural language ontology (Moltmann 2017, 2020). She argues that natural language ontology is an important domain within ‘descriptive metaphysics’ (using the term from Strawson 1959), which is distinct from the kind of foundational metaphysics that the philosophical tradition tends to engage itself in. This project analyses the ontological commitments implicit in natural language(s) itself, one of the most pervasive and important reflections of our human engagement with the world, and is a project to which Moltmann has contributed consistently in various contributions over the years (e.g., Moltmann 1997 on part whole structures, Moltmann 2009 on tropes as the particularized property denotations of adjectives, Moltmann 2013 on abstract objects). As Fine (2017a) argues, there is a case to be made that progress in foundational metaphysics relies on a close and nuanced understanding of the descriptive metaphysics involved in natural language ontologies. But even if that were not the case, as a linguist, it seems to me that the project of natural language ontology is crucial if we are to understand the compositional products of meaning and meaning building in language and the mechanisms by which it is embedded in our cognition and cognitive processing more generally. The spare and elegant axiomatization of semantic descriptions anchored just in truth and reference to particulars simply does not do justice to content and partial and incremental contents that we see in language. Exploring natural language ontology in its own right, taking the internal evidence as primary is a prerequisite to getting this kind of deeper understanding. Having said that, exploring these specifically linguistic patterns also raises deep and difficult questions about the relationship between the ontologies required for language specifically and those that are part of a more general cognitive ontology. As a linguist, I am also interested in the question of how the syntactic representation of content relates to these ontological commitments, and in particular to the distinction between lexical and functional content when it comes to the syntax-semantics interface.

The important semantic distinction we have seen so far in the understanding of CP complementation, and the one which has motivated the existence of attitudinal objects in the ontology, is a distinction that we see also in the nominal domain.

Consider the sentences in (8) below. The noun ‘book’ in English and in every other language I know can refer to the book qua object, or the book qua literary content. In (8-c) you can see that the ‘same’-ness of the book can refer either to the identity of physical copy, or to the identity of literary content.

This is not a polysemy unique to names of concrete objects. Even abstract nominals like ‘the argument’ (in the sense of a sequential representation of reasoning) have an equivalent vagueness.

The written manifestation of the argument can be erased, but only the logic behind the argument can be understood. The type of reading that the nominal gets is determined not just by context, but by the selectional pressures of the verb that the nominal is an argument of. Moreover, we know this is systematic vagueness because of the classic zeugma and ellipsis tests targeting the different readings of the same NPs, (10). Unsurprisingly, we do not know of languages which systematically mark the use of a noun as content-denoting versus physical object-denoting with overt morphology.

I think this ubiquitous phenomenon is instructive because it shows us that what is happening in the interpretation of sentential complements is neither uncommon, nor specific to CPs. Moreover, the properties of the alternation are eerily similar: the interpretation of a DP as either object referring or content referring is not signaled overtly, and it is closely tied to the selectional properties of the argument taking predicate. But imagine if we treated the DP case analogously to the CP one. In (11-a), the verb put takes the physical object denoted by the book as its argument. In (11-b), however, the object of understanding could on this view be a special ontological object, let us label it a Thought/Idea for concreteness, a special kind of ‘cognized object’ (on analogy with attitudinal object) whose content is somehow associated with the physical object ‘the book’, because the content of those ideas inheres in the book in question. So in fact, the nominal ‘the book’ is a predicate which describes the content of the thing understood, rather than being directly the object of the verb. The toy neo-Davidsonian representations are given below each sentence.

Following the Moltmann line of attack, these are exactly the semantic representations we should be trafficking in the nominal case as well. Now, maybe (11-b) is actually the right way to go for sentences of this type with special cognition verbs, but there would be more resistance in the linguistic community to positing these kinds of representations here than in the CP domain, and it is worth asking why. The problem is that we as linguistic semanticists often think about rich conceptual contents for DPs and polysemy as part of the connection to truth conditions that we do not reify as part of the semantic representation per se. Conceptual content of roots is part of a domain of meaning that, if it has structure and connections to other cognitive domains, this is part of a black box that the syn/sem representation itself does not engage in. Under this view, the different meanings of ‘book’ in terms of its different contributions to the truth conditions, involve polysemy and the denotation of a DP with the lexical content given is simply vague with respect to a systematically available network of possible meanings. In other words, do we need to invoke an attitudinal object in the representation when we can explicate its effects in the particular choice of precisification of meaning for the DP itself?

There is some reason to be dissatisfied with this kind of ‘black box’-ism, because I think that formal semantics has not really provided satisfying mechanisms for dealing with the phenomenon of polysemy as distinct from pure homonymy for completely distinct meanings (ambiguity), and that what we are seeing here is a kind of systematic polysemy that cuts across category differences, languages, and constructions.

Nevertheless, the ubiquity of the phenomenon of slippage between contents versus denotata that ‘have’ those contents suggests a cognitive generality to the distinction. Once again, I do not know of any language that does not do this effortlessly, and without piling on morphological help to signal it. And while it appears that CP complementation involves a number of syntactic and semantic properties that distinguish it from DP complementation, DP complementation itself is the ‘normal’ and yet seems to rest on the same sort of implicit ontology. So if language never makes this implicit difference in type of meaning explicit, is the distinction really something that is built by the linguistic system at all, or is it rather part of a more general and pervasive cognitive ontology?

There are a host of interesting questions that arise here in the nominal domain around how lexical contents systematically yield a predictable network of metonymically related denotata in cases like (10), even possibly extending to denoting events canonically associated with those denotata (e.g., Rachel began the book). Which aspects of these predictable networks are ‘seen’ by the functional structure of the syn/sem systems of natural language, and which aspects of them reside in more general systems of cognitive uptake?

I don’t know what the answer is here, but I do think that Moltmann in her work has uncovered something deep about the semantics of these items, and that the intuition shared by both Moltmann and the Moulton/Kratzer line of research in emphasizing the difference between propositions and their content is important. I am still unpersuaded about whether these facts really do motivate the existence of this range of different attitudinal objects in the linguistic ontology, as opposed to telling us something deep about cognitive ontology. I am, however, persuaded that there is a binary distinction between interpreting the embedded CP as a the direct propositional theme argument of a verb of communication, and as something that denotes rather the content of the attitude in question.

This brings us back to the notion of content, and the urgent task of any formal semantic theory should be to have a fine enough grained theory of content that it will be able to do justice to the meaning of embedded complements of attitude predicates in the first place. Taking the project of exploring of natural language ontology seriously is the first step towards doing that, and creating a theory that can inform both mentalistic and foundational metaphysical projects.

3.1 Modal auxiliaries and the classical toolbox

Let us first consider the formal semantics of sentences like (12-b), the canonical modal auxiliary sentence type. Kratzer’s early work on the semantics of modals was essentially a semantic account with plenty of room for contextual and discourse factors (Kratzer 1977, 1991). The idea in brief was that the semantics of a modalized sentence could be analyzed with just three factors within a framework of possible world semantics: (i) the quantificational force of the modal in question (ii) the modal base that the modal quantifies over and (iii) the ordering source which imposes a contextual ordering of worlds within the modal base. While the first factor was contributed by the modal itself, the second two came from presuppositional properties of the modal combined with contextual and discourse factors of use. However, it is well known from typological descriptive work and from syntactic cartography that there is a powerful and apparently exceptionless generalization which states that epistemic readings reside higher in the clausal hierarchy, at the same height or possibly outside tense, while circumstantial and dynamic modals are found inside tense. Further, dynamic modality is hierarchically lower than circumstantial modality (Cinque 1999; Nauze 2008).

Recognising that the standard account had no obvious way of accounting for these robust crosslinguistic generalizations, Hacquard (2006) (in her dissertation and in subsequent work) made an important and influential proposal extending the Kratzerian system to account for these patterns. She proposes a system that ties particular types of interpretation to height in the structure. Her idea is to replace the base world from which the modal base is calculated within a standard possible world semantics for modals, with an event instead, and relate the semantic differences to differences in how that event is anchored/bound. When the modal is speaker-oriented, it is keyed to the speech time and receives an epistemic interpretation; when the modal is subject-oriented, it is keyed to the time provided by tense and receives a root interpretation.

Hacquard (2006) keeps otherwise intact the central structure of the Kratzerian solution: modals are functions from sets of possible worlds to sets of possible worlds; a restriction via contextually defined modal bases, ordering sources, existential versus universal quantification. However, while this is an important step forward, it comes up against a rather deep problem due to the very nature of the theoretical framework and its assumptions. Basically, since modals are functions from sets of possible worlds to sets of possible worlds, they need something of the propositional type to combine with. In order to reconcile the syntactic position of the circumstantial modal (i.e., below tense and aspect) with the need for it to combine with the propositional type, Hacquard’s new theory is forced to base generate aspect as an argument of the verb and then move it for type reasons, leaving something of the propositional type. This is technically possible of course, but if we are allowed to do such things why is it that we get the ordering we do in the first place? The ordering Asp > Circumstantial Modal is necessary in this style of implementation if the event relativity of meaning is to be captured by the mechanism of binding, but this is actually at odds with the phrase structure ordering usually found across languages. So syntactic hierarchy requirements are pulling in two different directions here. But since the phrase structure base ordering has robust confirmation in typological facts, it must be the decision to implement the required semantic differences via event binding that is the culprit.

A further indication that Hacquard’s solution simply does not do the job is that the system loosens up only just enough to relativize the base event/world to syntactic height, and still leaves unexplained the correlation of height to type of modal base. For example, it is still possible in Hacquard’s system for ‘high’ modals to take the speech event as the base event, while still choosing a circumstantial modal base. In fact the correlations are exceptionless: the base world is the speech event if and only if the modal base is epistemic; the base world is the VP event if and only if the modal base is circumstantial. Hacquard notices this of course, and offers a stipulation to handle one direction of the correlation, but admits to having no explanation of the other. To date there has been no proposal offered in the literature to fix the Kratzerian system so that it actually delivers the generalizations found in natural language with respect to height of modal and modal base.

3.2 Modal height and syntactic size of prejacent

The obvious desideratum is the following situation, a modal such as must in English, which is ambiguous between a circumstantial (deontic) interpretation and an epistemic one has one basic denotation, let us call it Must just to have a mnemonic label. We want to be able to say that Must can be merged at different heights in the syntactic representation to adjoin to either the phrase B(ig)P or the phrase S(mall)P, where the functional structure expressed by the verbal extended projection SP is properly contained in the functional structure expressed in BP.

(13)

(14)

Further, we want the compositional semantics to be sensitive to the nature of what Must is merging with, such that the two different interpretations arise from the different denotations of BP and SP respectively.

I note here that even though modals in English uniformly seem to end up in the highest inflectional node of the tree (in a position where they can happily host negation and invert with a Subject in questions), the typological evidence of stacked modals uniformly suggests a height difference in terms of first merged position. Moreover, there is distributional evidence that epistemic and deontic readings of a modal like must really do combine with different natural classes of prejacent. As noted in Ramchand (2014), while prejacents like progressive, perfect and stative be allow epistemic readings with must (15), epistemic readings are systematically disallowed with episodic readings of main verbs, and dynamic passives (16). ^[2]

The facts concerning the syntax-semantics interface are telling us something very specific here about the relationship between hierarchical structure of the extended verbal projection and semantic interpretation: prejacents in combination with circumstantial readings contain less structure than prejacents in combination with epistemic readings.

The most minimal way this could be the case is literally that Must takes B(ig)P as its semantic complement in one case and takes S(mall)P as its semantic complement in the other case, where we want to be able to see the content of the SP as part of the content of the BP. Note this is not what the standard toolbox allows you to do because anything that a modal combines with has to be of the propositional type. We put a further natural and desirable constraint on the meaning building here and say that we want the meaning of the whole modalized clause also to have the meanings of SP or BP as a part of it.

So the puzzle now becomes, what kind of meaning must Must have and what must the denotations of BP and SP be to satisfy this desideratum. Under truthmaker semantics with exact verification by situations, we have a fighting chance of setting up a semantics that directly tracks the ontological units that language makes use of to build up to propositions of different types. Intuitively, we want the set of situations that verify (and falsify) (17-a) above to be in a certain natural relationship to the set of situations which are the verifiers (and falsifiers) of BP. Moreover, we want the set of situations that verify (and falsify) the situation denoted by SP not to be the same as for BP, but to also bear a systematic relationship to it, to be part of it in some way.

The solution proposed in Ramchand (2018) starts off by noting that BP is some unit at least the size of TP, while SP is only an AspP.

(18)

(19)

The claim is further that this does make a difference to the denotation in a way that matters for the semantics of the modal. I reiterate, this is never the case with the standard toolbox because modals have to combine with propositions, whose denotations are particular mappings from possible worlds to truth values, regardless of their ‘flavour’.

3.3 Using situations and exact verification to understand content

The set up in truthmaker semantics (Fine 2012, 2014, 2017b, 2017c) is fundamentally different with regard to content, because “Exact truthmaking holds between a situation and a sentence just in case the situation is wholly relevant for the truth of the sentence” (Moltmann, p. 160). Let us remind ourselves of the basic features of the theory that Fine proposes:

1. Declarative sentences are made true by situations that are their exact truthmakers.

2. Situations are mereological parts of worlds and the standard conditions on conjunction, disjunction, existential quantification hold, together with a standard mereological notion of fusion.

3. Truthmaker semantics assigns to each sentence not just its set of verifiers, but also its set of exact falsifiers (or violators) which is used in defining negation.

4. The domain of situations includes impossible situations in addition to actual and possible ones.

Although Fine does not discuss the application of the theory to sentence internal composition, the strength of the exact verification approach is that in principle it makes it possible for us to track the precisification of content within the clause from the lowest lexical elements through to the various functional projections and adjuncts that create the description of the situation finally constructed. Consider the clause in (20) below, bracketed according to syntactic constituency, with standard labeling. The core situational description built involves the lexical verb and its internal and external arguments (which I am here calling the vP) and is the ‘smallest’ coherent situational description built up by the sentence. I am further assuming that the PP ‘for his root canal’ is adjoined to the thematic domain of the clause, and that ‘reluctantly’ is adjoined to the inflectional domain, and that ‘yesterday’ is adjoined after the addition of tense anchoring information (here, past).

The interesting thing here is that ‘Hector visit the dentist’ does in fact already describe a situation, albeit a rather open ended and underspecified one. All declarative sentences underspecify some aspects of the situations they describe, so this it not qualitatively different from what we expect from situations in the first place. Let us call the sub-sentence expressed by the inner vP1 S1, and allow that it already stands potentially in a truth making relation to the world — in other words, its denotation will be the pair which consists of sets of its situational verifiers pos(S1) and falsifiers neg(S1).

Because S1 expresses descriptive conditions for a situation without providing any tense information, there is a question about whether it can have any exact verifiers on the world side at all. This is because temporal information about the situation will always be in some sense extraneous, and one might argue that any such situation would not be ‘wholly’ relevant to the description in question. However, this I think would be not be the correct conclusion. We already have to tolerate some underspecification in the sense that a simple sentence like Hector visited the dentist would indeed be ‘verified’ by a situation in which Hector visited the dentist last week, but also by situations where Hector visited the dentist a year ago. Further, it cannot be that we disallow situations on the side of the world from having detail not mentioned in the sentence, such as what Hector might be wearing, for example. So if Hector is wearing a yellow hat in that dentist visit, the situation in which he does so is still considered an exact verifier for the sentence, as are situations in which he wears a red hat, or a dress, or is whistling a tune, etc. So some degree of ‘detail gap’ is acceptable in exact verification, provided that the details in question are detail regarding the already presupposed necessary features for the situation to be instantiated in the world (situations exist in time and space, bodies have clothing, minds bear attitudes).

I would argue that existing in space and time are part of those necessary features for situations, and that the underspecification of the vP situation in (21) above is no different in kind from the fact that we do not mention whether Hector is wearing a hat or not. Therefore, in (21-b), we have a set of positive verifiers that includes situations from all over the temporal map. Now let us take a look at the slightly larger vP, where the PP has been added to the syntactic representation. We call this sub-sentence S2.

Here, we still are not specific about the temporal instantiation, but the syntactic representation contains a PP which precisifies that Hector’s visit to the dentist was specifically for getting a root canal done. This means that the exact verifiers must not contradict this information, and the falsifiers will include all the events (of whatever temporal anchoring) in which he visits the dentist on a social visit, or just to clean his teeth. Since we have added information in going from S1 to S2, the exact verifiers for S2 will be a subset of the set of verifiers for S1. We can continue the process of adding information, and the AspP adds the information that Hector must be reluctant, and merging in the T node will add the information that the verifying situations must be located in the past with respect to the speech time. Finally, with the addition of the adverb yesterday we narrow down the verifying situations to precisely those that happened yesterday.

The natural conjecture when looking at the hierarchical representation built by a particular language’s syntax, is that information increases monotonically as structure builds. We might even propose a generalization or constraint on the syntax-semantics mapping for compositional meaning, which one might call ‘The Monotonicity of Content’, expressed informally below in (23).

There are two points to be clarified here: firstly, that something like monotonicity of content can probably be expressed in systems that employ loose verification as well, and secondly, that the actual order of the elements that successively precisify situational descriptions does not fall out of just this notion. But all of this is moot, if Monotonicity of Content does not actually hold in the first place. There are obvious problems that immediately arise when it comes to the folding in of speaker oriented adverbial elements (like allegedly), or indeed modal auxiliaries, as we will discuss in Subsection 3.4. The point I would like to make here is that while it is natural to define monotonicity of content over subset relations among verifying situations, it is also possible to extend the notion in constrained ways to include certain other kinds of relations that allow complex meaning building. The phenomenon of embedded clauses shows us that the content of one proposition can be part of another (however we choose to implement this), and the bridging relationships that allow this come from the restricted class of verbs taking propositional ‘complements’. However, language is much pickier when it comes to the kind of content monotonicity it requires while still allowing a single monoclausal representation. In general, I think the problem of sentence internal partial contents is important enough that we should start with the (probably too) strict statement above and see how far it will take us. Note that the statement above is careful to claim subset-ism only in the case of extended projections. In addition, it may be that speaker oriented and non at-issue modifiers will have to be syntactically integrated in different ways from the regular spine (as in Potts) in order to keep to the strictest interpretation of such a principle. It is a much more restricted version of compositionality than the one allowed by function-argument composition, if higher order types are admitted, but one that is more intuitively tied to a notion of content.

3.4 The ontology underpinning modal ‘flavour’

How does this help us to solve the problem of modal height correlating with flavour of modality? Don’t modals immediately cause a problem for such a draconian principle of Content Monotonicity in the first place (at least if it is thought of in terms of subsets and supersets of verifiers)? Space prevents a completely detailed and formal exposition of the solution, but I just sketch here how the challenge of modal sensitivity to height might be addressed using situational verifiers in a way that just doesn’t seem possible with the traditional possible worlds toolbox. This in turn will lead us to propose a constrained and motivated addition to the allowed ways in which content monotonicity can be satisfied.

In Ramchand (2018), I argue that the modal auxiliary introduces a new situation (the base situation, or perspectival situation in the sense of Condoravdi 2002), and that a modal denotes a predicate that asserts that there is a Choice of some kind, against the background of a set of live alternatives or potentials. Under this view, the set of alternatives will be determined by the denotation of the prejacent. Or we could, on analogy with the interpretations advanced for embedded attitudes, say that the modal introduces a special kind of modal object, the ‘live-alternative set’ which is described by the prejacent in some systematic way.

Since the prejacent is a constituent that has a set of situational verifiers, and since those verifiers in the case of ‘small’ prejacents consist of situations with many different time traces, we can say that S2 helps describe the live alternative set by intersecting the verifiers of S2 with the set of undecided situations. Then what the modal asserts is a Choice situation S0 involving ‘Hector’ in this case,^[3] where the alternatives are suitably constrained by the prejacent. In Ramchand (2018), I argue that so called ‘universal’ modals are those in which the choice is exclusive (the only choice), whereas existential modals are those in which the choice is possibly one of many.

The Choice situation is then verified by situations held by Hector at the perspective time, in which Hector’s only viable choice is contained in the set of Live-Alternatives, given some set of background constraints on admissible situations for Hector. (Note that we are not removing the dependence of circumstantial modal interpretation on pragmatic and contextual grounds which provide a shrinking of the domain of situations, we are developing a system where a systematic contribution to the meaning is given by the denotation of the prejacent.)

The reason this works when scaled up to larger prejacents is that the larger prejacent has acquired more content, and by hypothesis, is time anchored, and therefore the set of verifiers of S2 in (25) does not include any situations that are not anchored to the present tense (the same time as the perspectival situation introduced by the modal). The external argument of the Choice predicate here is also not a participant in the prejacent situation (whose participant list has been existentially closed by the time the Choice predicate is merged), but the speaker.

The Choice pivot is the speaker herself, and the Live-Alternatives are the assertoric alternatives open to her. Once again, this alternative set of assertions is ‘described’ by the denotation of S2 which acts as a predicate constraining those assertoric alternatives. The meaning of Must can remain the same: the speaker is in a Choice situation (i.e., is operating against a background of assertoric uncertainty) where the speaker’s only viable choice is contained in that particular Live-Alternative set. The Live-Alternatives here are the different propositions that the speaker could utter, her epistemic alternatives based on all direct and indirect evidence, intersected with the verifiers of S2.

The following table schematically lays out how the Choice predicate generalizes across size domains in a way that is sensitive to (i) the content built up by that point and (ii) the argument available as Choice pivot. (I include dynamic modality here for completeness although there is no space here to discuss this case in detail).

The above is therefore a sketch of a solution to the understanding the dependency of modal interpretation on the size of the syntactic representation of the prejacent. It is only possible with a theory that tracks content in an exact way. Even though Monotonicity of Content can be expressed using loose verification as well, what we lose is the specificity of content. Once we say that the denotation of a chunk of structure corresponds to a mapping from possible worlds to truth values, we are describing the output truth conditions in a way that makes the detailed content opaque and unrecoverable. I do not see a way to use that meaning to provide content to the live alternatives that are built when that very same structure then combines with a modal.

3.5 Return to attitude reports

Working out the semantics of modals while paying attention to the monotonicity of content has revealed a role for the notion of Choice as a pervasive, perhaps universal cognitive category, and one which human appraisers want to label, track, and take notice of. For me, the importance of the truthmaker semantics lies in its ability to track content in a more fine grained way, although taking a leaf out of Moltmann’s book, one might want to make a case for ‘choice’ as a special kind of ontological eventuality, distinct from other situations or actions. It is not merely something like ‘thinking’ or ‘believing’ which come with their own lexical items and can be used without any other verbal support, the Choice notion is parasitic on, and sits on top of the normal situational description. It also, importantly, can be added to ‘normal’ situational descriptions without introducing a whole new clause (even though there is good evidence that there are two, albeit related, situations being expressed). This makes it different from the situation of attitude predicates, which by definition embed a whole clause as their complement with its own independent verbal extended projection. For this reason, the nexus of a situation of Choice and its Alternatives seems to be a way in which Content Monotonicity can be satisfied, in addition to subset relations. At least the evidence from monoclausal utterances crosslinguistically shows us that it is a privileged kind of situation to situation relationship. As we have seen from Moltmann’s own work, the range of attitude predicates is very rich and the range of attitudinal objects as new elements of the ontology is, although smaller, also quite rich. Given that these predicates introduce an attitude whose content can be that of a proposition, it stands to reason that modal notions could also show up with full embedding. This, I submit, is the reason that one can find matrix attitude predicates that seem to mirror or rather share the content of a modal auxiliary. But the interesting and telling fact is that the other direction does not hold — modal notions make good auxiliary verbs while not every notion that can be expressed as an attitude predicate makes a good auxiliary. This means that a verb selecting for something whose content can be described by a propositional predicate is an important generalization in its own right and isolates a certain class of predicates, but it is not enough to guarantee that that ‘verb’ can be integrated with a propositional predicate in the context of a single extended projection, a monoclausal construction. It seems to me that this more restricted subset of verbs that combine with situational descriptions is an additional and intriguing pattern that tells us something about monotonicity and unity of situational contents. So the fact that modals can be auxiliaries as well, showing harmony with some propositional attitude predicates, is predictable because of the common denominator of selecting for propositional contents, but the phenomenon of modal auxiliation specifically needs to also be distinguished and the reasons for its specialness better understood. I have suggested that this specialness arises from the way in which the notion of choice and undecidedness is built into the semantics of natural language statements as a natural extension of how to describe the world.