Modification and context

3.1 Function and form

In their introduction to the analysis of the Noun Phrase in the F(D)G tradition, García Velasco and Rijkhoff (2008, 15) comment on the problem of categorizing in linguistic description:

The basic problem with names for linguistic categories seems to be that they tend to be based either on formal or on functional properties and that there is usually no direct relationship between them. The same constituent may occur in different functions, and the same function may apply to different forms or constructions.

This general problem seems to be particularly evident in the case of modification. Thus, in his recent introduction to modification, Morzycki (2016, 8) notes a difficulty in the characterization of the notion, which in his view stems from the fact that any linguistic unit may be characterized internally (i.e. by their intrinsic properties) or externally, by the role they play in linguistic expressions. There is, however, no novelty in this proposal other than the use of different terminology, for Morzycki’s observation again seems to revolve around the function/form opposition in linguistic categorization. However, the problem with modification is exacerbated in that both the functional and formal sides of the notion seem to correspond to a particularly unclear communicative or grammatical function on the one hand, and to a plethora of constructions with different morphosyntactic properties on the other. Thus, the grammatical function of modification is difficult to characterize and the formal properties of modifiers correspond to a number of disparate units which seem to defy consistent categorization. Morzycki illustrates this very clearly:

The internal sense of ‘modifier’, then, to a very crude approximation, may amount to just this; you’re a modifier if you’re an adjective or an adverb (…). The external sense of ‘modifier’ has to do with crosscategorial parallels in the role an expression plays. You’re a modifier if you are adjoined to something you’re not an argument to. (Morzycki 2016, 9)

The vagueness of this characterization is revealing. On the one hand, modifiers are identified internally as specific lexical classes such as adjectives or adverbs; however, this immediately raises the question of how to characterize those lexical classes. If one is to define the class “adjective” as a modifier of a nominal head (see below), the definition of modification becomes circular, and if adjectives are to be identified on the basis of their morphosyntactic properties, there will surely be languages in which modification thus understood would play no role if they lack adjectives altogether. Similarly, the external sense of modifier is based upon the argument/adjunct status of linguistic units, but this is not always clear for some participant types (directionals, beneficiaries, instrumentals, etc.).

The functional or external role of modification can also be characterized in communicative or actional terms as a process (i.e. a function) which is performed by speakers, on a par with other linguistic “actions” such as predication and reference. However, distinguishing modification from reference and predication does not seem to be an easy matter. Referring can be generally defined as the identification of an entity with linguistic means. In order to do so, speakers characterize an entity by ascribing properties, a process which should help addressees identify or construe the intended referent in joint cooperative communication (Dik 1997, 132). For that purpose, nouns, adjectives, prepositional phrases, relative clauses, and additional units may be appropriate, which proves that predication and modification are similar processes in actional terms. In fact, given that in FDG all lexemes are treated as property-denoting items which are inserted in property frames defined by variable “f” (note also that in old Functional Grammar all lexemes were treated as predicates), there is in practice no semantic or interpersonal distinction between nouns and adjectives, as they are both used to predicate or ascribe a property or relation and restrict the denotational value of an entity variable.

Several authors, however, have tried to distinguish modification from predication and reference in actional terms. Thus, Croft defends the existence of three basic propositional acts with communicative value: predication, reference, and modification:

The act of REFERENCE identifies a referent and establishes a cognitive file for that referent, thereby allowing for future referring expressions coreferential with the first referring expression. The act of PREDICATION ascribes something to the referent. Predication does not establish a cognitive file for the state of affairs that is predicated, but instead prototypically reports relatively transitory states of affairs, often in a narrative sequence. The act of MODIFICATION (of referents) functions to enrich a referent’s identity by an additional feature of the referent, denoted by the modifier. (Croft 2001, 66)

In the FDG tradition, Rijkhoff (2014) has elaborated on Croft’s proposal and argues extensively for the inclusion of the Subact of Modification along with the already established Subacts of Ascription and Reference.^[3] One of the reasons in favour of this proposal is that languages make use of a dedicated lexical category, that of adjectives, for the basic Subact of Modification (but see Section 3.2). Thus, in the typological work of Hengeveld (1992, 2013), for example, which is also behind the FDG approach to lexical categories, the adjective is functionally characterized as the prototypical modifier of a nominal head. Rijkhoff (2014, 133) defines the Subact of Modification as a basic communicative function whose role is “to modify (‘enrich’, ‘supplement’) core linguistic material used in the act of predicating and referring.” Similarly, Dixon (2004, 10) claims that adjectives serve two main functions in the grammar of languages: (i) assigning a property to an individual (as in copula constructions), and (ii) attributive modification, which is characterized as “a specification that helps focus on the referent of the head noun in an NP that relates to a predicate argument.”

There seems to be an inherent problem with Rijkhoff’s proposal, though. While predication/ascription and reference can be properly defined in communicative terms as independent and basic “actions” with communicative import, modification presupposes the existence of a referential or ascriptive act, which can be further enriched through the specification of additional features by the modifiers. In other words, modification involves a prior act of ascription in the identification of a referent (see Croft’s quotation above). This leads naturally to the view of modification as a secondary function of an optional nature and casts doubt on its analysis as a basic communicative action on a par with predication and reference. In fact, given that both nouns and adjectives have a predicative function, they can only be distinguished by the priority they obtain in the construction of referents: nouns come first and modifiers come afterwards in the dynamic construction of noun phrases, as was already noted in classical FG, in which nouns were considered “first” restrictors. If we view reference construction as the stacking of “clues” (in the form of predicates) leading to the identification of a referent, the actional role of modifiers and nominal lexemes is the same, and there seems to be no obvious reason to add a new propositional (Sub)act of Modification.^[4] In other words, the alleged communicative difference between heads and modifiers is, I would argue, not of an actional nature, but a natural consequence of the inherent properties of the semantic process of entity description. There is thus no interpersonal difference between predication and modification, and it is only the difference in semantic priority between first and secondary restrictors that is reflected in morphosyntax through the obligatoriness of heads and the optionality of modifiers. Therefore, units at the RL will be regarded as those that compositionally contribute to the description of a given denotational unit, irrespectively of the way that contribution is instantiated. Modification can indeed be characterized as the enrichment of the descriptive or denotational content of a referent, but this involves no difference in actional or interpersonal terms. Compositionality thus emerges as the only requirement to construct the designation of a denoting unit. As shown in Section 4, this solves some problems discussed in previous literature (Giomi 2020, Keizer 2019) with respect to the restrictive/non-restrictive, intersective/non-intersective nature of attributive adjectives and modifying units.

3.2 The miscellaneous formal nature of modifiers

A second problem in the characterization of modification derives from the miscellaneous nature of the units that function as modifiers. This problem is undoubtedly linked to the previous discussion. As the number of entity-denoting lexemes in a language is obviously limited, the construction of a referential unit begins with the selection of that which best approximates the described entity for the purpose of referent identification. If this is felt insufficient for the addressee to grasp the intended referent, more specifications will follow in the form of modifiers. However, the ways in which entities may be further specified are initially unlimited, and it is virtually impossible for languages to lexicalize all potential properties of all entities in the world of discourse, both concrete and abstract. Dixon (1977; 2004) notes that languages with a small adjective class lexicalize four main semantic types: “dimension” (big, small, etc.), “age” (old, new, etc.), “value” (good, bad, perfect, etc.), and “colour” (black, white, red, etc.). In languages with a larger adjective class, additional semantic types include “physical property” (hard, soft, heavy, etc.), “human propensity” (jealous, happy, clever, etc.), and “speed” (fast, slow, etc.). He also notes that in languages with a small adjective class, tendencies can be observed, so that a particular semantic type will typically be expressed by another part of speech. For example, if the language lacks adjectives for physical property terms, this category will usually be expressed by verbs or nouns (e.g. structurally complex units such as PPs, relative clauses, etc.). Thus, the existence of significant differences in the size of adjective classes and the need for other constructions to assume the modifying function, together with the different morphosyntactic properties of adjectives across languages, casts doubt on the very existence of the adjective as a universal part of speech, to the extent that some typologists prefer the term “property word” (Van Lier 2017). However, although there is some controversy as to whether the Noun–Verb distinction is present in all languages, it is at least acknowledged to be a strong universal tendency deriving from the basic actional distinctions of predication and reference (see e.g. Lyons 1977, 429).

In principle, then, all those units which contribute to the description of a referent other than the head may be considered a modifier, regardless of their final formal expression. Thus, in Rijkhoff’s typological work on the structure of the Noun Phrase (2002, 2014), “modifier” is used as a cover term for “the various functional modifier categories discussed here, which include both grammatical and lexical modifiers and which cover all the ‘dependents’ that are not arguments or complements, both in the NP and the clause” (Rijkhoff 2014, 134). As noted by Van de Velde (2007, 204fn), the notion of modifier in FDG may be used to refer to both grammatical and lexical dependents, or to lexical modifiers only (as was the case of the notion “satellite” in classical FG), while the grammatical expression of additional specifications of predicates and referents at different levels are treated as operators.

The diverse formal realizations of modifiers are also evident in the analysis of levels and layers in FDG. In the theory, all levels of linguistic analysis are formalized on the basis of the following general template, which provides positions for modifiers and operators in each layer:

1. (π v₁: [h (v₁)_Φ]: [σ (v₁)_Φ])

As mentioned earlier, implicit in this schema is the recognition of a primary Subact of Ascription ([h (v₁)_Φ]), which may be elaborated by additional and optional modifiers ([σ (v₁)_Φ]). In other words, modification requires the previous inclusion of a head unit as the first restrictor of the entity variable (v). Given that this schema is applied to all levels and layers of linguistic description, the diversity of modifiers and their relation with modified units seems difficult to characterize in a universally valid manner.^[5] As a consequence, a definition of modification which covers all the different uses and functions of modifiers in linguistic expressions can only be provided in very general and somewhat loose terms. In FDG, then, modification can be defined as the optional specification or description of a grammatical head. In the same way as “headedness” is a grammatical relation, modification is seen as a grammatical relation too, potentially relevant at all levels and layers of linguistic description. Incidentally, this supports a non-actional interpretation of modification, as predication and reference are interpersonal actions, but irrelevant to the processes of morphosyntactic and phonological encoding.

To sum up, in spite of the different attempts at providing a definition of modification on interpersonal, functional, or formal grounds, the notion still seems extremely elusive and defies a precise characterization. The identification of modifiers needs to be made with reference to the notion of grammatical head, and the exact nature of the modifying function very much depends upon the particular level of linguistic analysis. The following section is devoted to particular aspects of modification at one such level in FDG with special reference to the interaction between semantic and contextual knowledge.

4.1 The local context: adjectives

In predicate logic, an expression such as an old rich man, which denotes the intersection of three predicates, receives the following analysis:^[6]

1. ∃x MANx & RICHx & OLDx

The FDG analysis at the RL resembles this notation and conjoins predicates as lexical properties which restrict the variable for individuals “x” (Hengeveld and Mackenzie 2008, 241):

• (3) an old rich man

•   (1x_i: [(f_i: man_N (f_i)) (x_i)]: [(f_j: rich_A (f_j)) (x_i)]: [(f_k: old_A (f_k)) (x_i)])

An initial difference between the representations in (2) and (3) is that in FDG the order in which predicates are stacked onto the individual variable is relevant. Thus, the first restrictor is the head of the construction, which is usually a noun in the noun phrase. The notation does not make any initial distinction between heads and modifiers, other than their position in the structure.

A basic objective in truth-conditional semantics is to show how linguistic meaning follows from linguistic expressions independently of the context of use. Thus, the analyses in (2) and (3) warrant valid entailments in all contexts of use. Each one of the predicates denotes a property which can be independently predicated of an individual, given the fully compositional and intersective nature of their semantic contribution:

• (4) Geoffrey is an old rich man

•   Therefore: ‘Geoffrey is old/rich/a man’

Additionally, the analysis of those lexemes as predicates in representation (3) is in accordance with the potential to use them in predicative expressions such as he is rich or he is old.

However, it has long been noted that not all adjectival predicates are intersective, which poses a problem for logical analysis. Following work by Partee (1995), introductory semantic books such as Morzycki (2016) and Kroeger (2019) distinguish subsective and non-subsective adjectives, whose main properties will be discussed in the following sections.

4.2 Vague modification, degrees, and standards

The interaction between the denotational properties of lexemes and context is not limited to non-subsective adjectives. Many apparently intersective adjectives seem to behave subsectively too, as they denote a subset of the extension of the head noun. Standard gradable adjectives such as tall or large need to be modulated in context: a tall boy is tall for a boy and a large fly is large for a fly. The degree of height and size these expressions entail depends on the noun they modify (what is large for a fly may not be for a boy) and need to be worked out in context. Partee (1995, 330) illustrates the problem with the following pattern of inference:

• (10) Win is a tall 14-year old

•   Win is a basketball player

•   ??Therefore: Win is a tall basketball player

The invalid inference in (10) suggests that the adjective tall is non-intersective. According to Partee, however, it seems inappropriate to treat adjectives like tall as subsective. Instead, they are said to introduce a “range of indeterminacy,” which needs to be resolved in context relative to the semantics of the noun it modifies. For this reason, gradable adjectives are treated as vague and context-sensitive expressions as it is necessary to adjust “the interpretation of the modifier in the light of the local context created by the head noun” (Partee 1995, 333).^[9]

A similar situation is observed in the analysis of degree words which I provided in García Velasco (2013). In that work, I noted that the function of degree words is to further specify a property with reference to a standard of comparison along a given dimension scale (see also Giomi [2022], who additionally treats degrees as quantities). In fact, an analysis along those lines had already been suggested by Hengeveld and Mackenzie (2008, 455), and slightly modified in Hengeveld and Mackenzie (2021). Consider the following example:

• (11) John is more intelligent than his brother

•   [(f^c ₁: (f₁: Adj (f₁): (f₂: more_Adv (f₂) (f₁)) (x₁)_Standard)) (f^c ₁)) (x₂)_U)]

The representation in (11) assumes the existence of a standard of comparison, which is realized by the individual denoted by his brother in this particular example. As I argued in García Velasco (2013), degree expressions may be classified according to whether they introduce an explicit standard of comparison in the form of a than-phrase as in (11) or a result clause (e.g. “so intelligent that she got an A+”), or whether the standard of comparison is provided by the context. In that article, I further argued that the standard of comparison that degree words introduce replaces the one provided by gradable adjectives, which are inherently comparative. Thus, a boy is tall if he is taller than a height which we consider “unremarkable” for some group (Murphy 2010, 228). The particular properties of the standard of comparison are provided by the context or general knowledge. Of course, privative and other non-subsective adjectives are not inherently comparative, but one could also argue that their coercive effects on the denotational properties of the noun they modify are also a function of the context of use (either linguistic or extra-linguistic), which needs to be captured in the RL, as they clearly have compositional effects.

Interestingly, the grammar of languages provides examples of cases where access to context is necessary to construct the truth-conditional interpretation of a linguistic expression. One obvious case of a construction whose truth conditions cannot be obtained on the basis of conventional literal meaning only is that of English N + N compounds. At least a subset of them, typically those whose interpretation is not constrained by thematic relations, can receive multiple readings which can only be resolved in context. Thus, as is well-known in the literature, a compound such as marble museum (Plag 2003, 150) may have several readings, including “museum built with marble” or “museum in which marble statues are displayed.” More interpretations are possible, but all of them will refer to a museum of some kind, since it is a basic fact of English word-formation that the rightmost element functions as the head of the entire construction. As Partee (1995) notes, however, given the absence of semantic constraints, the burden of the interpretation of nominal compounds is put onto the addressee’s general knowledge, which should help arrive at the most plausible interpretation in the relevant context (see also Kamp and Partee 1995, 144).

Two important aspects about N + N compounds are worth discussing. First, they illustrate a productive pattern in the grammar of English. This is not a case of an occasion-specific unit (e.g. Clark and Clark’s (1979) contextuals) created on the spur of the moment to satisfy specific communicative needs. It shows that the grammar of languages may contain productive and regular patterns of grammatical constructions and rely on the addressee’s ability to infer a significant part of the denotational meaning of the expression in actual usage. Second, N + N compounds are not incompatible with the principle of compositionality as long as one assumes a weak interpretation of the principle. Weiskopf (2007) discusses several approaches to the semantics of nominal compounds including semantic underspecification, free enrichment, or multiple ambiguity, and proposes to treat them as context-sensitive structures which contain an indexical element that needs to be satisfied in context. In spite of the technical differences, all the proposals that Weiskopf considers preserve compositionality and are compatible with the context-sensitivity of many natural language expressions. Although the interaction between context and compositional meaning as represented in the RL may be a problem for FDG under standard assumptions, the following section will hopefully show that compositionality is fully compatible with context-sensitive semantics.

6.1 Non-subsective modification in FDG

The problem posed by subsective adjectives was addressed in FDG a long time ago (Hengeveld 2008, Escribano 2008) and then taken up again more recently in Giomi (2020) and Keizer (2019); see also Keizer (2022) and Kemp and Hengeveld (2022). Hengeveld and Mackenzie (2008, 230) note that some adjectives modify the “f” variable and create a complex property which restricts the denotation of the individual variable “x” as a whole. This is the analysis proposed in (7b) for the subsective interpretation of the expression gay activist. However, the authors extend this representation to privative adjectives like former in the expression former neighbour, and argue that the same analysis should be used for classifying adjectives such as medical in a medical student (Hengeveld and Mackenzie 2008, 230, 242):

• (12) a former neighbour

•   (1x_i: [(f_i: neighbour_N (f_i): [(f_j: former_A (f_j)) (f_i)])] (x_i))

However, adjectives like former and medical have different properties. Whereas former indicates that the property denoted by the noun was but is no longer applicable, a medical student still is a type of student. The qualifying adjective medical is thus subsective as it denotes a subset of the extension of the head noun. The adjective former, however, changes the denotation of the noun and it is thus said to belong to the group of so-called “privative” adjectives. Recent developments in FDG suggest a potential analysis for adjectives of the former type. As mentioned above, the set includes others such as present, erstwhile, previous, or old (Morzycki 2016, 45), which denote a point or period in time and are called “temporal-ordering” modifiers in that work. Given that the option of being located in time is a property of States of Affairs, one could reanalyse former as an event modifier. The expression He is my former neighbour warrants the inference He was my neighbour. If we thus analyse the property “neighbour” as a State of Affairs “e” restricting an individual “x” variable, the resulting analysis would be as follows:^[10]

• (13) (1x_i: [Ant e_i: [(f_i: neighbour_N (f_i) (e_i))]: [(f_j: former_A (f_j)) (e_i)] (x_i))

In (13), the property “neighbour” restricts the SoA variable “e”. This is intended to represent the meaning “be a neighbour”, which is temporally set as anterior to the designation of the property denoted by the temporal-ordering adjective “former.” The gist of this representation is thus the reanalysis of a typically individual-restricting property as an event-restricting property, due to the requirements of the temporal modifier “former.” In other words, this analysis can be seen as a case of type-coercion, which in FDG is defined as “a process whereby the class of a lexeme is adapted to the requirements of its syntactic position. Coercion applies to lexemes that are used in a non-default position at the Representational Level and need to be prepared for their function at the Morphosyntactic Level” (Keizer 2015, 300). Additional evidence for this analysis derives from the fact that coercion in FDG typically brings about the addition of a derivational affix at the ML. In fact, Hengeveld and Mackenzie (2008, 404) state that “much of derivational morphology can be interpreted as the formal reflection of coercion effects.” The use of derivational prefix ex- in expressions such as ex-boyfriend can thus be seen as morphologically equivalent to syntactic modification by the adjective former.

In fact, a coercion analysis for privative adjectives has been suggested in the formal semantics tradition but, crucially, with a slightly different interpretation of coercion processes. Indeed, it seems difficult to see how the FDG analysis in (13) above could be extended to other non-temporal privative adjectives. Predicates like fake or counterfeit, which also cancel the denotation of the main predicate allow predicative uses and cannot be said to modify an event under any reasonable interpretation:

• (14) a. This painting is a counterfeit Vermeer

•    Therefore: This painting is not a Vermeer

•   b. A fake gun

•    Therefore: It is not a gun

Partee (2007, 153; see also Morzycki 2016, 27) notes that adjectives such as fake or counterfeit coerce a loose interpretation on the noun they modify: Hence, even though the entailment “this is not a gun” follows from the noun phrase a fake gun, the question “is this gun real or fake?” makes perfect sense, which might suggest that fake behaves subsectively after all. This can be seen in situations in which it would seem unnatural to refer to a gun as a fake gun, as in speaking to a child, for whom the toy is real in their imaginary world. Partee (2007) then explains that adjectives need to be interpreted relative to the context created by the head noun in a dynamic way. This is formulated in the so-called The Head Primacy Principle (HPP):

In a modifier-head structure, the head is interpreted relative to the context of the whole constituent, and the modifier is interpreted relative to the local context created from the former context by the interpretation of the head.

However, privative adjectives seem to be particularly sensitive to another general principle, The Non-Vacuity Principle (NVP): “In any given context, try to interpret any predicate so that both its positive and negative extension are non-empty.” The two principles are not absolute and usually cooperate in interpretation, but in the case of privative adjectives, the NVP seems to outrank the HPP, which coerces the meaning of the head noun in a fake gun to include real guns too. As a consequence, Partee argues, privative adjectives can be reanalysed as subsective after all (see also Kamp and Partee 1995).

Partee then assumes a dynamic construction of noun phrases in which the conventional meaning of predicates is calibrated as the process proceeds. In fact, this dynamic construction of noun phrases is very much reminiscent of the stacking of predicates in term structure (see example (3)), which is found in Dik (1997, 149). In Dik’s system (but see Giomi 2020 for an opposing view), the order of predicates in the construction of terms is crucial for defining scope relations and arriving at the appropriate interpretation of the complex term. Assuming this dynamic process of term construction and in line with previous discussion on sense modulation before compositionality, the coercive effect of privative adjectives can be easily incorporated in FDG; we only need to allow the system to modify the semantic interpretation of the predicates as term construction proceeds.

A similar strategy can be followed in the analysis of vague adjectives such as tall (see Szabó 2001 for a detailed analysis along these lines). Recall that these adjectives are interpreted relative to a standard of comparison, which is explicitly introduced in the FDG analysis of comparative constructions. Similarly, one could propose that vague adjectives can participate in frames such as this:^[11]

• (15) (f₁: tall_A (f₁)) (x₁)_Standard

Example (15) would be paraphrased as “tall for x,” where the value of “x” has to be fixed in context. The analysis of “a tall boy” is then given in (16):

• (16) A tall boy

•   (1x_i: [(f_i: boy_N (f_i)) (x_i)]: [(f_j: tall_A (f_j)) (x_i) (x_j)_Standard])

This analysis solves the problem noted by Partee (i.e. a tall 14-year old who plays basketball is not necessarily a tall basketball player), as the entailment that would follow from an expression such as John is a tall 14-year old boy is “tall for a 14-year old” and not simply “tall”. As Morzycki (2016, 10) notes, the standard will be represented by the class of elements designated by the head noun itself. In FDG, this could be analysed by spelling out the standard of comparison with the property “boy” and the universal quantifier (∀):

• (17) (∀x_j: (f_j: boy_N (f_j)) (x_j) _Standard)

The standard of reference in this case is then fixed by all entities to which the property “boy” applies. The exact value (i.e. what most people would consider an “unremarkable” height for boys) of the standard of reference needs to be determined against contextual or general knowledge.

Note that if this analysis is extended to evaluative adjectives, the problem noted by Keizer (2019) that some representational modifiers are not restrictive can be solved. Keizer (2019, 383) provides the following corpus examples:

• (18) a. There were seven of us, my three kids, wife, my father-in-law, my old mother and me

•   b. Our friendly staff is here to make sure that you have an outstanding experience.

•   c. The prolific Toni Morrison returned this year with her first novel set in the current time.

Keizer notes that the adjectives old, friendly, and prolific in these examples are not restrictive, as the referent of the noun phrase in which they participate would be the same if the adjectives were not present, and they just serve to add a property to an already defined set of entities. She therefore proposes to represent them with an additional propositional content within the RL.

I agree with Keizer that the non-restrictive modifiers in (18) should be analysed as representational units, given their descriptive nature and the fact that they contribute compositionally to the semantics of the unit in which they participate. However, another possible analysis for those non-restrictive adjectives which merely add a descriptive specification would be to assume that, unlike in ordinary uses, they do not introduce a standard of comparison since they just denote what is considered an intrinsic property of the head noun. This is evident in another example discussed by Keizer: “white snow”. The adjective “white”, although gradable (e.g. “snow is whiter than cotton”), need not be interpreted relative to an unremarkable standard of “whiteness for snow” in that expression. The difference between the restrictive and non-restrictive uses of, for example, adjective old (e.g. “old mother” in (18a) vs “old computer”) may be interpreted again in terms of the Dikkian approach to term construction. Dik (1997, 130) introduces an opposition between constructive and identifying reference. In constructive reference, the speaker assumes that a referent is not available to their addressee and therefore provides clues for its construction in the form of restrictive predicates. In identifying reference, the referent is presumably available to the addressee, and the function of restrictive units is to guide the addressee to the activation of the intended referent. One could thus assume that a standard of comparison is only necessary in constructive reference, and not in identifying reference. This accords well with Keizer’s observation that the identity of the referent of my old mother in (18a) is guaranteed both by the sense of the noun “mother” and by the presence of the definite possessive determiner “my”. The definite (known) status of the referent is even more obvious in the case of prolific Tony Morrison in (18c), given the interpersonal and context-unique nature of proper nouns.

The analysis of (restrictive) gradable adjectives can thus be accommodated in FDG under the assumption that the modified noun establishes the class of entities against which the adjectival property needs to be modulated. Subsective adjectives can be similarly analysed, with the only difference that the standard of reference now ranges over properties and not individuals. Thus, an expression such as a skilful surgeon, which is paraphrased as “skilful as a surgeon”, would receive the following analysis:

• (19) A skilful surgeon

•   (1x_i: [(f_i: surgeon_N (f_i): [(f_j: skilful_A (f_j)) (f_k) (f₂)_Standard])] (x_i))

The relevant difference between subsective and intersective gradable adjectives thus derives from the fact that the class of entities that set the standard of reference is of the property type in the former case and of the individual type in the latter.

Finally, consider the case of non-subsective adjectives of the alleged type. I argued in Section 4 that the analysis of alleged as an interpersonal modifier is inadequate under the view of the RL adopted in FDG. Alleged in an alleged murderer opens the possibility for the description of an individual as a murderer not to be the case. They thus alter the denotation of the head noun and have truth-conditional and compositional effects. One possible analysis would simply assume that the adjective modifies the nominal property, as in (20):

• (20) an alleged murderer

•   (1x_i: [(f_i: murderer_N (f_i): [(f_j: alleged_A (f_j)) (f_i)])] (x_i))

In (20), it is then the lexical semantics of the adjective that indicates that the property (f_i) it modifies may or may not be the case. However, this analysis is parallel to subsective representations (cf. 7b) and is therefore inadequate, for an alleged murderer does not belong in the class of murderers. Moreover, the expression John is an alleged murderer cannot be paraphrased as “John is alleged as a murderer”, and it would also seem odd to claim that the property “alleged murderer” is being predicated of John.

As mentioned earlier, in FDG adjectives of the alleged type are treated as modifiers of a Subact of Ascription and are therefore considered interpersonal units. This follows Van de Velde’s (2007, 218) analysis, who notes that these adjectives, and others like reputed, ostensible, false, or possible, put “the felicity of the ascriptive subact at issue.” In my view, however, the function of alleged is in fact to reject commitment to the evaluation of an individual with the property designated by the noun. In other words, it is the veracity (truth) of the description which is at issue (not the Subact of Ascription itself), and therefore alleged should be treated as a propositional modifier. Evidence in favour of this analysis comes from the interaction of alleged with factive verbs, whose complement clauses are presupposed to be true, as opposed to non-factives, with which the speaker does not commit him/herself to the veracity of the propositional content. Thus, (21b) is unnatural, because the use of the non-factive verb believe in the main clause signals the speaker’s lack of commitment with the truth of an embedded proposition containing the adjective alleged, which in turn indicates non-commitment with the description of an individual as a murderer. This results in a pragmatic anomaly, which contrasts with the behaviour of factive know in (21a):^[12]

• (21) a. They know that John is an alleged murderer

•   b. # They believe that John is an alleged murderer

Note, additionally, that other ascriptive operators (sort-of) and modifiers (so to speak) sound natural in the same context (see Hengeveld and Mackenzie 2008, 111–12):

• (22) a. They believe that John is a sort-of murderer

•   b. They believe that John is a murderer so to speak

Further evidence comes from the FDG treatment of predicate negation, which in a way has a semantic effect parallel to that of privative adjectives. Expressions such as non-issue or non-happy/not happy are analysed by Hengeveld and Mackenzie (2018, 30) as instances of property negation:

• (23) (neg f_i: issue_N (f_i))

Just like alleged, property negation changes the denotation of the property introduced by the head noun. The denotation of non-issue is vacuous in effect, but the communicative value of the expression derives from the interpretation of the empty set it denotes with reference to the property introduced by the noun issue. The noun sets a frame or context with reference to which the entity described as a non-issue is to be interpreted. The interpretation of the expression does not derive from the infelicity or lack of appropriateness of the description, but is in fact a description in negative terms: it is not the case that something is an issue. The speaker is thus sure of the felicity of not ascribing that property to a given entity. The function of alleged is somewhat similar. In the expression an alleged murderer, the noun murderer sets the frame against which an individual is to be described. The speaker is sure that somebody’s condition as a murderer is what is at stake. This is confirmed by the fact that an alleged murderer is typically to be found in conversational contexts in which there is evidence, although not necessarily conclusive, that someone is a murderer. Consequently, the proposed representation is as follows (compare (13)):

• (24) (1x_i: [ p_i: [(f_i: murderer_N (f_i)) (p_i))]: [(f_j: alleged_A (f_j)) (p_i)] (x_i))

The gist of the analysis is then the insertion of the noun murderer in a propositional slot, thus restricting a proposition “p”, whose truth of falsehood is described as “alleged”, that is, potentially true or false due to lack of evidence.

6.2 Sense modulation in FDG: the Contextualizer

The role of context in the construction of linguistic interpretation is acknowledged in FDG with the introduction of an ancillary Contextual Component which informs the grammar of those aspects of the text and context that have grammatical impact in the linguistic expression. Hengeveld and Mackenzie (2014, 2021) argue that the Contextual Component includes two types of information: discoursal and situational. Situational information is restricted to a “language-specific selection of those details of the speech situation that have relevance for Formulation” whereas discoursal information is recorded in the Contextual Component, as it may be needed for anaphoric reference to units in previous utterances. What both dimensions have in common is that they are dynamic, in the sense that they are constantly being updated and adapted to the needs of communication as the verbal interaction evolves. In García Velasco (2014, 2018), I argued that there is an inherent tension between FDG’s static approach to grammar and the necessary dynamism of language use and verbal interaction (see also Cornish 2013, 2022). In fact, several authors have argued that some linguistic processes cannot be fully understood under the static conception of the grammar proposed in FDG (see e.g. Butler 2013, Connolly 2007, 2014, Cornish 2009, 2013, 2022, Keizer 2014). In different ways, all these proposals defend the idea that the role of context in grammatical description needs to be taken into account even if it leads to probabilistic choices in the grammar only. The present contribution, however, goes one step further, as the examples of meaning modulation discussed show that the rigid separation of semantics and pragmatics is inappropriate both on architectural and methodological grounds.

The relation between the Contextual Component and the grammar is mediated by a process known as Contextualization. This operation contains a complex contextualizer which connects strata in the context with corresponding levels of representation in the grammar. In the representational stratum, for example, information about perceived entities is transmitted onto the grammar so that referential units can be linked with observable referents in the speech situation.

Sense modulation then entails that the contextualizer should be allowed to dynamically interact with the RL for the purposes of compositional meaning construction. This respects the static nature of conventional meanings and the context dependence of compositional meanings, but raises the question whether what is being modelled in this proposal is a representation of psychological processes in speech production. Hengeveld and Mackenzie (2008) have made it clear that FDG should not be seen as a production model, and that its dynamic implementation should be seen as a reflection of the sequence of steps taken by the analyst in order to “clarify the logic of the relations among the layers, levels and components” (Mackenzie 2014, 251: see also Hengeveld and Smit 2009). This, in Hengeveld and Mackenzie’s view, should be enough to meet the standard of psychological or cognitive adequacy which was proposed by Dik (1997).

I believe the modulation of meaning in the RL is fully compatible with the dynamic implementation of FDG as proposed in Hengeveld and Smit (2009):

the dynamic construction of the representational level occurs in close interaction with the dynamic construction of the interpersonal and morphosyntactic levels, respectively. Note that in proposing a dynamic implementation of the grammar we do not imply that FDG should be conceived as a procedural speaker model. We merely refer to the dynamic and stepwise procedure in which the grammar generates underlying representations. (Hengeveld and Smit 2009, 1118; emphasis mine)

The dynamic implementation is thus restricted to the grammar component and the generation of underlying representations. It is seen as an internal process, but there is no contradiction for the grammar to consult the Contextual Component in the process. As mentioned above, Hengeveld and Smit (2009, 1124) observe that in the dynamic construction of the RL, the head places constraints on the way it may be restricted by non-primary restrictors. They illustrate the issue with selectional restrictions between operators and heads, but nothing would prevent the dynamic implementation of the grammar to consult the ancillary components in the dynamic process of linguistic generation.