2.1 Lexical category

We cannot do justice here to the richness of work on lexical categoryhood undertaken in prior work. For recent overview and literature review, see Baker & Croft (2017). Our goal in this section is instead simply to illustrate the kinds of differences in lexical category of property concept lexemes that has been observed which feed into our study. This literature has shown that property concept lexemes vary cross-linguistically according to their basic lexical category. In many well-studied languages, property concept lexemes typically belong to an adjectival class that has distinct morphosyntactic properties from both nominal and verbal categories.¹ In English for example, adjectival property concept lexemes such as tall cannot function as an independent predicate of a clause, but require the copula be.

Adjectival property concept lexemes also differ from nominal predicates in that they cannot head a noun phrase, but only function as attributive modifiers.²

In other languages such as Hausa (Parsons 1955), Basaá (Bantu; Jenks et al. 2018), and Warlpiri (Pama–Nyungan; Bittner & Hale 1995), property concept lexemes are typically nominal. In contrast to adjectives, nominal property concept lexemes exhibit the morphosyntactic properties of other nouns in a given language. In Hausa for example ((7), repeated from (1)), nominal property concept lexemes such as ƙarfī ‘strength’ must therefore occur in possessive constructions as other nouns do (e.g., zōbe᷅ ‘ring’) when used predicatively.

Finally, there are languages, including Tongan (Austronesian; Churchward 1953), Koasati (Muskogean; Kimball 1994), and Paumari (Arauan; Chapman & Salzer 1998) in which property concept lexemes are typically lexicalized as stative verbs. In such languages, property concept lexemes appear in predicate position without a copula, as exemplified by Tongan below (see also Koontz-Garboden 2007).

While some languages uniformly realize property concept meaning by lexemes of a single category, many other languages exhibit category splits with a subclass of property concepts being realized by lexemes of another category (see Dixon 1982 and Stassen 1997 for cross-linguistic tendencies; also van Lier 2017). In Daakaka (Oceanic, Austronesian) for example, property concept lexemes such as mwelili ‘be small’ are verbal, while others such as bur ‘deaf’ are adjectival, as indicated by the obligatory presence of the copula only in the latter (von Prince 2015).

In addition, languages may exhibit multiple categorial strategies to realize the same property concept without a clear derivational relationship between the forms. Such a case can be observed in Édò (Benue-Congo, Niger-Congo) where property concepts like ‘beautiful’ can be realized by either verbal (10a) or adjectival lexemes (10b), which differ in their tonal pattern and their requirement to combine with the copula ye in predicate contexts (Baker & Stewart 1997).

In summary, property concept lexemes can be lexicalized as various parts of speech — adjectives, nouns, or verbs — both within and across languages.

2.2 Direction of derivation between state and change of state

There is often morphology deriving a word describing a state into a word describing a change into that state (or vice versa; Koontz-Garboden 2005: 94–99). Depending on the kind of state (e.g., property concept state or result state), there are differences in the direction of such derivation (see Beavers et al. 2021 for a typologically robust demonstration). On the one hand, purely stative property concept lexemes like red tend to be unmarked, with the change of state verb derived. On the other hand, result states like cracked, which lexically entail a prior change into that state, tend to bear a different morphological relationship to the verb describing the change into the state, in the case of English with the result state adjective derived from the verb describing the change into the state the adjective describes.

The present study focuses on property concept lexemes, which primarily show three different types of derivational relationship to their change of state forms across languages.³

The first derivational type is equipollent, in which both stative and change of state forms are derived from a common root.⁴ In Ulwa (Misumalpan) for example, property concept roots like sang- take the suffix -ka in their stative form, which is replaced by the suffix -da in their (intransitive) change of state form (see also Doron 2003: 61 on templatic morphology in Hebrew).

In the second derivational type, the change of state form is derived from the simple state, such that the stative form is morphophonologically less marked than the change of state form. This pattern is familiar from property concept adjectives in English where change of state forms are commonly derived by the suffix -en (see Dixon 1982 for a discussion of change of state morphology in English).

Cross-linguistically, the derivation of change of state forms from stative forms is not only found with adjectival property concept lexemes. Nominal property concept lexemes can also be derived into change of state verbs by designated morphology in some languages, as exemplified by Warlpiri. Here, the suffix -jarri verbalizes the nominal property concept lexemes and introduces change of state semantics (cf. Hale & Keyser 1998).

Change of state forms of verbal property concept lexemes may also be morphologically marked. This is the case in Yup’ik (Eskimo-Aleut), where change of state verbs are formed by the suffix -li:

Wáːšiw, as discussed in §4, has verbal property concept lexemes that behave similarly. The derivation of change of state forms from stative property concept lexemes with overt morphology thus appears to be insensitive to the lexical category of the source property concept lexeme.⁵

The final derivational type is labile, in which there is no morphophonological difference between state and change of state forms.⁶ We have already seen this state of affairs in Tongan ((2a), (3) above), in which property concept lexemes like loloa ‘long’ exhibit the same form in both stative and change of state contexts:

This pattern has been noted in the typological and descriptive literature, usually with the differences in meaning observed in the context of different aspectual marking on the polysemous word (e.g., Bybee et al. 1994: 75–76; Tatevosov 2002: 340ff.; Talmy 1985: 92; Smith 1997: 70; Wetzer 1996: 189; and Chung & Timberlake 1985: 217).

Applied to states, closure implies a complete change of state, specifically inception rather than cessation. Thus, languages that have a morphological category (traditionally called perfective) to specify closure for processes often use the same category to signal inception of a state.

(Chung & Timberlake 1985: 217)

State/change of state lability has been observed in several unrelated languages, including Mandarin (Sino-Tibetan; Tham 2013), Paraguayan Guarani (Tupi-Guarani; Califa 2018), and Gã (Kwa; Campbell 2022):

A reviewer notes that while the Tongan state/change of state pairs show identical aspectual marking, the ones in (18)–(20) show differences. Whether the difference is tied to aspectual marking, as suggested for Niuean by Matthewson et al. (2015) or to broader context requiring state change, as suggested for Tongan by Koontz-Garboden (2007) is a matter of current debate. See Smith et al. (2024) for recent discussion. Our goal here is simply to examine lability and category, and the resolution of this debate does not, therefore, impact on our observations below.

In summary, we observe three different types of derivational relationship between state and associated change of state crosslinguistically: equipollent, change of state derived from stative form, and labile.

2.3 A correlation?

The preliminary cross-linguistic picture based on languages discussed immediately above suggests that lability correlates with verbal category of property concept lexemes; adjectival and nominal property concepts instead appear to be less likely to be labile. From these preliminary observations, we arrive at the following hypothesis:⁷

If this hypothesis is supported by further investigation, it requires explanation. To preview, we show that it is indeed supported and argue that it is explained by the lexical semantic underpinnings of verbhood as the only category of word that can lexicalize relations between individuals and dynamic events, and therefore the only category of word that can lexicalize relations between individuals and change of state events. Any non-verbal property concept state (e.g., adjective, noun) therefore has to undergo category change in the derivation of a change of state predicate from it. Because both the introduction of change of state meaning and category change can be (separately) morphologically marked, and because derivation of a change of state requires only the introduction of the change meaning component, but not necessarily category change, state/change of state derivational relations are more likely to be labile when the property concept state is a verb than when it is a non-verb, because in the verbal case there is only one operation to potentially be morphologically marked, whereas in the non-verbal case, there are two, given the necessity of verbal derivation in order to lexicalize change of state meaning.

3.1 Data

In order to test the hypothesis, we draw on the database “Verbal Roots Across Languages” (https://verbal-roots.la.utexas.edu). This database was designed to systematically investigate the morphological derivational relationship between stative lexemes and their semantically related change of state lexemes crosslinguistically (see Beavers et al. 2021 for details). It includes information on 88 languages of the WALS 100 language list (Dryer & Haspelmath 2013), mostly representing the areas and families of the WALS 100, covering 72 root meanings. For the study reported here, we have examined the property concept meanings listed in (24):

For the purposes of the present study, what is important to know about the database is that for each root meaning in (24) in each of the database’s languages, it lists state/change of state pairs, along with any underlying root for languages in which these are derived from a bound root.⁸ The database contains 3365 entries like these with data of some kind (i.e., paradigms where not all cells are empty) based on property concept roots. The database website has a search function which allows for the searching of entries according to the direction of derivation between pairs, so that for example, we can extract all state/change of state pairs, henceforth called paradigms, that are related equipollently, where the latter is derived from the former, or crucially for the purposes of our study, in a labile fashion. From these 3365 paradigms, we extracted all property concepts in all languages in which there was at least one entry where state and intransitive change of state showed a labile direction of derivation to one another.⁹ It is important to highlight that the database treats lability, as do we, as based entirely on superficial morphological form. So, for example, while some might assume for English that there is zero morphology relating e.g., adjectival state dry and verbal change of state dry, these are treated by the database as showing a labile relationship to one another, and are included in our sample. The 41 languages included by this criterion and from which paradigms were therefore extracted are listed in the Appendix. If a language had even a single paradigm in which state and change of state showed a labile relationship, we have included in our study all entries from that language that have non-empty cells. So, to take an example, there is one paradigm in the Verbal Roots database for Paumarí that shows lability: ihamahiki is listed with the stative meaning ‘be angry’ and the change of state meaning ‘get angry’. We therefore looked up in Chapman & Salzer (1998: 347) the lexical category for ihamahiki, which was listed as verb. Furthemore, because there was this one entry from Paumarí showing lability, we included in our analysis all other paradigms in the Verbal Roots database, looking up the lexical category for the property concept word in the remaining thirty two property concept state/inchoative paradigms.

While the database is in most ways well-suited for our objectives, there is a key way in which it is not: it does not include lexical category information. Because such information is crucial to investigate our hypothesis, category information was extracted for each state in each of the paradigms meeting the above criteria from one of the original sources listed in a manner described in the Appendix. Entries for which it was impossible to find a categorization were excluded, as were a small number of other cases described in the appendix (where, for one reason or another, we could not make a conclusion about the category of the state despite there being some information).¹⁰ There were a few entries for which it was clear that the state was one of two categories (see Table 2), but it was not clear which. We have coded these separately, giving each of them two different codings. We integrate these into the analysis with the benefit of a statistical procedure described in §3.2.

With all of these considerations in mind, after the categorization procedure, and collapsing the equipollent and derived patterns into “non-labile”, the exercise yields the distribution of raw data in Table 1 for those entries that had a single categorization, and Table 2 for those few that had two. Leaving the data in Table 2 to the side for the time being, the data in Table 1 do seem to suggest, at least superficially, that verbhood does favor lability, as only 19% of nominal cases and 19% of adjectival cases are labile, while 52% of verbal cases are. In the section that follows, we explore whether this apparent unequal distribution of lability across categories deviates significantly from what could occur by chance.

3.2 Analysis & results

Our analysis was preregistered with the Open Science Framework (https://osf.io/pdnzb/?view_only=bb2fcae6a7d14ea19a676759c50e45c0). We fit a Bayesian mixed effects logistic regression model to the data compiled as above with labile (1) and non-labile (0) as the outcome variable. The category of the property concept state was included as a fixed effect, and language included as a random effect on both the intercept and this slope (allowing us to interpret any observed effect of category as generalisable across language). The model was fitted using STAN via brms (Bürkner 2017). A weakly informative prior was employed for the effects of category (normal distribution with a mean of 0 and a standard deviation of 20). The default LKJ prior with shape parameter 1 was used for the random effects. As in a small number of cases multiple categories were reported, our estimates are made by collapsing together 4000 iterations for each of 20 versions of the dataset (a single category was chosen for each of the ambiguous datapoints in each).

Figure 1 shows the estimated log odds of lability for each category (verb, noun, adjective). The dot for each category represents the best estimate for the log odds of lability. The bars represent the uncertainty around this estimate: they represent the range in which the value falls with 90% probability. Since the best estimate for adjectives and nouns falls outside the interval for verbs, we can say that there is < 0.05 probability that their log odds of lability are as high as that of verbs. We can therefore conclude that the data support the generalization in (25):¹¹

Figure 1: The log odds of lability for each category (verbal, nominal, adjectival) with 90% credible intervals.

4.1 Some assumptions and explanation

We believe the asymmetry in morphological behavior between verbs on the one hand and non-verbs on the other follows from relatively uncontroversial, but non-trivial, assumptions about the lexical semantics of nouns, adjectives, and verbs. We start from the assumption of a model with a domain that includes a variety of different sorts of e-type objects. The foundation is ordinary atomic individuals, assumed as part of any model since Montague (1973). As motivated by the semantics of mass nouns due to Link (1983), we assume there are also mereologically ordered portions (which perhaps includes pluralities, though nothing rests on that here). Finally, we assume there are also Davidsonian objects (Davidson 1967), i.e., states and events, which are standardly assumed to stand in some mereological ordering to one another (Champollion 2017: 27). We assume that states and events differ in that the former are not dynamic, while the latter are. A key consequence of this is that states can have temporal traces that are a single moment of time, while events have temporal traces that are minimally intervals of at least two moments, as a minimum of two moments is what is required for a change to occur, as discussed at length by Dowty (1979: 166ff). It is therefore no surprise that the truth of sentences headed by stative predicates can be evaluated at a single moment of time, while the truth of sentences headed by eventive predicates can only be evaluated at intervals larger than a single moment (ibid.). Many of the celebrated state/event linguistic diagnostics laid out by Dowty (1979: 55–56), such as ability to occur in the progressive construction and truth at a single moment in the English present tense, can be tied to this temporal property. This also presupposes the existence (again standard) of times, and an ordering of them relative to one another. The simple model that these assumptions give rise to is summarized in (26).

With this as background, we now articulate some assumptions about the meanings of words in different categories which we believe are common and widespread, but not generally articulated, and from which we believe our crosslinguistic observations regarding lability follow. Nouns, we assume, denote sets of entities of any sort. There are obviously those that denote sets of ordinary individuals, e.g., dog, cat, computer, house. There are those that denote sets of (ordered) portions, such as mass nouns like e.g., water, strength. Some denote sets of (ordered) events, such as nouns like exam (see §5.1 for references and discussion). Finally, there are nouns with a variety of sorts of entity in their denotation, nouns such as thing, for example. The idea that common nouns denote sets of some sort or another is assumed since at least Montague (1973) and is the foundation of work on generalized quantifier theory (Barwise & Cooper 1981), and much work in formal semantics more generally. There are differing views on what adjectives denote. While some have argued they denote contextually-sensitive sets of individuals (Kamp 1975; Klein 1980) others have argued they denote degree relations (Cresswell 1977; von Stechow 1984; Heim 1985) or measure functions (Kennedy 1997). Alternatively, others have argued for a view incorporating Davidsonian states, so that they denote, for example, sets of stative eventualities (Wellwood 2015; 2019). While these views all differ significantly in their substance and to some degree in predictions they make, distinguishing between them is not important for the claims we want to make. What is important for us is simply that on no theory do adjectives relate individuals to dynamic eventualities (which we will claim is only possible for verbs). Finally, what the denotations of verbs is depends on one’s theoretical starting point and views about the nature of the syntax/semantics interface. Some kind of event semantics seems by now orthodoxy in the syntax/semantics interface literature. But the exact kind impacts on what kind of meaning verbs have. On the standard Davidsonian view (Davidson 1967) verbs denote relations between individuals and events. In the modern syntactic literature dating to Kratzer (1996), the external argument is treated in a neo-Davidsonian fashion, while the internal argument remains a lexical argument of the verb, making some verbs denote sets of events (unergatives) and others denote relations between individuals and events (transitives, unaccusatives). Lurking behind this discussion, and we believe behind much work in formal semantics generally, is the assumption that there are certain kinds of meanings that only words of particular categories can have. While we believe there are many meaning/category pairings, one crucial to us here is (27):

More explicitly, couched in a standard Davidsonian theory, the assumption is that only verbs can denote individual/event pairs (intransitive eventive verbs), individual/individual/ event triples (transitive eventive verbs), or individual/individual/individual/event 4-tuples (ditransitive eventive verbs). Although we use the term lexicalize in (27), this is for expository convenience, and it is important to note that our claim is not tied to any particular theory of the syntax/semantics interface, including a lexicalist one. That is to say, we use this term loosely. On a lexicalist theory, we intend it as standardly assumed, to cover the denotations of words in the lexicon. On a non-lexicalist theory such as Distributed Morphology, the force of (27) holds of categorization of acategorial roots, so the idea is that roots that relate individuals to dynamic eventualities can only have as their first categorizing functional head v, as opposed to a or n.¹²

We assume that constraints on lexical semantics/categorial mapping of the kind in (27), and crucially (27) itself, are unidirectional. Our claim is that there is a particular kind of meaning that can only be lexicalized as a verb, and can never be lexicalized as a word of any other category. The fact that it is unidirectional, however, entails that verbs can have other meanings as well, as indeed they do, as with, for example, stative verbs. To put it slightly differently, our claim is not that verbhood entails dynamicity, but rather that dynamicity entails verbhood. As discussed below, this idea plays an important role in the understanding of the distribution of lability across categories and languages that we have observed above.

4.2 Inchoativization and category change

There are two separate processes that occur when a change of state verb is derived from a noun or an adjective, which are often conceptually lumped together. The first of these is inchoativization—a change in meaning from state to change of state. The second, in the event that the stative predicate is non-verbal, is category change, namely from a noun or an adjective to a verb.

Crucially, we know that these two processes are independent of one another, among other reasons because there exist inchoatives that are overtly derived from (stative) property concept verbs. Such a derivation is found in Wá·šiw (also spelled Washo, Washoe; Hokan/isolate, USA), where all property concept lexemes are verbs (see Hanink & Koontz-Garboden 2021), and where some change of state predicates are derived from them with the suffix –etiʔ (Jacobsen 1964):

In the above examples, no category change is necessary to form the change of state expression, as the simple state is already verbal. What these data point to, then, is that inchoativization can happen independently from category change; the two are not necessarily tied up together.

More acutely, there are cases in Wá·šiw where we see overt verbalization prior to – and thus separately from – inchoativization. This state of affairs is found with bound property concept roots, which must be verbalized by the suffix –iʔ (Jacobsen 1964; Hanink & Koontz-Garboden 2021) prior to inchoativization. Here we find that the inchoativizing suffix –étiʔ appears outside the verbalizing morphology:¹⁴

What these data show is that category change and inchoativization are two separate processes at play in derivation from state to change of state.¹⁵ We argue in the remainder of this section that this is what explains the predominance of labile derivational relations with verbal property concept lexemes, and the lack thereof with nouns and adjectives.

Our argument is as follows. Change of state is dynamic (if not necessarily temporally so; Sweetser 1997; Gawron 2009; Koontz-Garboden 2010; Deo et al. 2013), and can therefore only be expressed verbally. It therefore follows that if a single lexeme is going to have both a state and a change of state sense, it can only be a verb, since a noun or an adjective cannot have a change of state meaning. If a property concept lexeme is a noun or an adjective, therefore, to derive a word with change of state meaning from it, both inchoativization and category change are implicated—two processes that can be marked independently by morphology, as seen above. As already stated, by contrast, category change is not implicated in the derivation of a word with change of state meaning from verbal property concept lexemes; only inchoativization is. Given that category changing derivations and inchoativization can both be morphologically marked, it follows that change of state verbs derived from verbal property concept lexemes are less likely to be morphologically marked, since they only undergo inchoativization, rather than both inchoativization and category change.¹⁶ To put it more simply, with nominal or adjectival property concepts there are two possible reasons for morphological marking in COS derivation, while with verbal property concepts, there is only one, thereby increasing the likelihood of marking/decreasing the likelihood of lability of the former relative to the latter.¹⁷ That we therefore observe in Table 1 (and in the statistical analysis tied to it in Figure 1) more labile relations among verbs than among non-verbs (adjectives and nouns) is consistent with this logic. If, by contrast, we had observed more labile relations with non-verbal predicates than with verbal ones, this would have been inconsistent with our claims.

In summary, because only verbs can relate individuals to dynamic eventualities, category change is a necessity in deriving change of state verbs from underlying nominal and adjectival property concept lexemes, leading to an increased likelihood of morphological marking relative to derivation of a change of state verb from a property concept verb. Before concluding, we turn in the next section to address some potential counterexamples to the claim underpinning this explanation.

5.1 Derived nouns

At first glance, derived nouns can appear to relate individuals to a dynamic event. If this were the case, it would counterexemplify the claim that only verbs are able to do so. There are many different kinds of nominalizations (see Grimshaw 1990; Zucchi 1993; Alexiadou 2001 for overviews); here we focus on three types that have featured prominently in the literature in order to demonstrate they are not true counterexamples.

We begin with the deverbal construction known as acc(usative)-ing nominalizations. The potential problem can be demonstrated by considering the nominalization within the bracketed material in (31). Here, it appears that the nominal itself relates a patient (the wall) as well as an agent (Kim) to a punching event, apparently contrary to our claim that only verbs can relate individuals and dynamic events in this way:

Indeed, this type of nominalization shows robust argument structure. For example, removal of the direct object (the punchee) in (31) results in ungrammaticality (32a), as does removing the agent (32b). This behavior appears to suggest that this type of nominalization is capable of relating individuals to a dynamic eventuality in the same way that verbs can.

However, it is a widely accepted generalization that only nominalizations that are derived from verbs display argument structure in this way (Grimshaw 1990; Zucchi 1993; Alexiadou & Grimshaw 2008; cf. Borer 2013). To dispel the concern that this construction is a counterexample, it is sufficient to show the category of any word that lexicalizes a meaning where dynamic eventualities are related to individuals is verb, as opposed to noun. To state the case more specifically to this empirical backdrop, if we can show that a nominalization, whatever meaning it might have, is derived from a verb with the relevant meaning, then it is the verb’s meaning that matters for (27), not the noun’s, since whatever meaning the nominalization has, it does not have it in virtue of lexicalization, but rather from (morphological or syntactic) derivation.

In fact, there is a widely accepted analysis due to Harley (2009) that has it that precisely this is the case. Building on Kratzer (1996) (and Abney 1987), Harley proposes that nominalization in the acc-ing construction occurs high in the structure: after the verb takes its internal argument(s) and combines with Voice, as schematized in the structure in (33). This accounts for, among other things, accusative case licensing, the obligatoriness of the direct object, and the ability for an agent to appear. Fundamentally then, we are dealing with a verb taking its subcategorized arguments as usual, with the entire structure being nominalized at a higher position.

In (34) we show the denotations of the key projections of the structure in (33). This derivation shows that it is indeed a verbal projection that relates the object to an event (34b) before a subject argument is introduced (by Voice) (34d).

We now turn to another deverbal construction that makes use of the same suffix: ‘of’-ing nominalizations, as in (35).

This type of nominalization shows some argument structure as well, albeit limited to internal arguments. While both constructions involve the nominalizer -ing, the above cited works have argued that the two types of nominalization involve different heights of attachment of -ing, with of-ing attaching at the level of the VP only, thus excluding the potential for an agent-introducing Voice layer to be included in the nominalization. For structural reasons, ‘of’-ing then differs from acc-ing in that the apparent ‘subject’ in the former is not truly an external argument as it is in (33), but a possessor. That this is the case can be diagnosed by available non-argument readings: unlike agent interpretations, the possessive relation is well-known to be contextually conditioned, as Harley (2009: 324) has observed in the context of just this kind of nominalization, and as shown by data like (36):

Harley’s (2009) proposed structure for of-ing nominalizations is schematized in (37), in which nominalization applies at a lower level than in acc-ing nominals, but still at a point after the core verb has selected its internal argument(s).¹⁹

As above, (38) offers a line-by-line derivation of (37), showing that the structure is derived from a verb that relates individuals to events, whatever the meanings at the nodes derived from it:

At the level of nominalization (38d), the meaning is a set of events that can then be related to a possessor, the source of the non-argument possessor in such examples, and the key to capturing the non-argument reading in contexts like (36). Adopting for example Barker’s (1995) theory of nominal possession, a possessor can compose with the nP as follows, making the whole NP denote the set of punching of the wall events that stand in some abstract possessor relation (π) to Kim.²⁰

In summary, in both types of deverbal -ing nominalizations, it is the underlying verb that relates the object to an event, not the nominal itself. For reasons of space we do not address complex event nominalizations (e.g., examination of the patients), but similar arguments apply there, as well (see Grimshaw 1990; Alexiadou 2001; Alexiadou & Grimshaw 2008).

Finally, we turn to a type of nominalization that is typically taken not to be derived from a verbal source: result nouns such as exam in (40).²¹

Result nouns such as this lack the verbal properties that their corresponding verbs (and deverbal nominalizations) do. For example, they do not allow by-phrases as passives (41b) or complex event nominalizations do (41c):

Since they are not deverbal (cf. Harley 2009; Borer 2013; Iordăchioaia 2020), could result nouns then be a true counterexample to the claim that nouns cannot relate individuals to events? After all, Kim in (40) does stand in some relation to exam. Like the ‘of’ –ing nominalizations of the type illustrated in (36), this nominalization type does not license true argument structure. Compare for example the felicity of the internal argument in complex event nominalization in (42a) to the result noun in (42b): internal arguments are not possible in result nominalizations.

As for the would-be external argument in this construction, we find a similar state of affairs to ‘of’-ing nominals in that that they also allow for contextually conditioned possessor readings, as already mentioned, and illustrated in (36). For instance, in (43a), Kim can be the examiner, the examined, someone who wrote the exam, etc., while in (43b) and (43c), Kim can only be the examiner (cf. Grimshaw 1990: 48).

The plausible relation is therefore again, possessive, as illustrated in (44), drawing on the kind of analysis of possessive NPs as above.²²

In summary, result nouns do not pose a challenge to (27); at the level of nominalization, we believe the meaning is a set of events (that can be related to a possessor).

While various types of verb-related noun appear superficially like they could be counterexamples to (27), they are either derived from verbs that have the kind of meaning in question, or, we believe, don’t have the kind of meaning that would challenge (27) in the first place.

5.2 Derived adjectives

Another potential counterexample that we briefly address comes from deverbal adjectives, which are of adjectival category but likewise appear at first glance to relate an individual to a dynamic event (Kratzer 2000). For example, in (45b), the promoted subject door is related to an opening event:

Parallel to the case of -ing nominalizations above, it is widely accepted that deverbal adjectives like (45b) that appear to relate a dynamic event to an individual are derived from a verbal core. Similarly to result nouns on the other hand, there are apparently deverbal adjectives that can be shown not to relate an individual to an event in the same way (Embick 2004; Alexiadou & Anagnostopoulou 2008; Alexiadou et al. 2015; cf. Beavers et al. 2021).

Such a case has been claimed to exist in Greek, in which there are two distinct kinds of (agreeing) adjectival morphology: -menos and -tos (Alexiadou & Anagnostopoulou 2008; but see Anagnostopoulou 2016 and references therein for a more complex picture and discussion). As (46) shows, adjectivalizations formed with the suffix -menos permit an agentive by-phrase, while those formed with -tos do not.

While detailed discussion is beyond the scope of this paper, the two types of adjectivalizations are distinguishable by a variety of diagnostics that lend support to the idea that those showing argument structure (with -menos) do so only because they are deverbal, much like ACC –ing nominalizations above. The suffix -tos as in (46) on the other hand reflects a structure and meaning more like result nouns, and does not involve relation of individuals and dynamic events. See Alexiadou & Anagnostopoulou (2008) for further discussion.

Consequently, like the nominalizations discussed above, neither type of deverbal adjective constitutes a counterexample to our claim about the kind of meaning exclusive to verbs and verbal projections. Again, the claim is that such adjectives, if they do introduce a relation to their argument, do so only at a point in the relevant structure where a verb or verbal projection can do it, and if they do not show a relation (as with –tos), then they do not threaten the claim in the first place.