2.1 Er-nominalisations

Prototypically, er-nominalisations are interpreted as individuals who have the thematic role assigned to the subject of their parent verb – so a worrier is an individual who worries, an eater is an individual who eats, and so on (Rappaport Hovav & Levin Reference Rappaport Hovav, Levin, Stowell and Wehrli1992). Following much of the literature on nominalisations, I refer to the argument of the parent verb that is picked out by the nominalisation as the R-argument.Footnote ⁷ The range of thematic roles that can be assigned to the R-argument in the nominalisation is just the same as the range of roles which the parent verb can typically assign to its subject. So just as the subject of open can be an agent or an instrument, as in (3), so too can an opener be an agent or an instrument, as in (4). That is, the er-nominalisation ‘inherits’ the argument structure of its parent verb.

1. (3)

   1. (a) Mary opened the can.

   2. (b) The new gadget opened the can.

2. (4)

   1. (a) I am an expert opener.

   2. (b) This here is an excellent opener.

For the purposes of this article, it is important that CoPTErs contain argument and event structure, and this is what I focus on in the remainder of this subsection.

Some theoretical work has proposed that er-nominalisations can have eventive and non-eventive readings (Rappaport Hovav & Levin Reference Rappaport Hovav, Levin, Stowell and Wehrli1992). In eventive er-nominalisations, there is an implication that the nominalised event actually took place, while non-eventive ones lack this implication. This difference is argued to correlate with whether or not a syntactic complement is licensed: eventive er-nominalisations license a syntactic complement, as in (5a); non-eventive ones do not license a syntactic complement, as in (5b) – note that the adjoined noun life in (5b) is not a complement, on which see Borer (Reference Borer, Everaert and Marelj2012).

This distinction follows the contours of the distinction between complex event nominals and referring nominals described by Grimshaw (Reference Grimshaw1990) – complex event nominals contain a representation of an event and inherit the argument structure of their parent verb, thus requiring them to project an internal argument position, if their parent verb does so. In contrast, referring nominals do not contain a representation of an event, and cannot license an internal argument.

However, Alexiadou & Schäfer (Reference Alexiadou and Schäfer2009, Reference Alexiadou, Schäfer, Alexiadou and Rathert2010) argue that the distinction in (5) is not really about the presence vs absence of an event – they argue instead that all er-nominalisations, including those in (5b), contain a representation of an event. Roy & Soare (Reference Roy and Soare2014) investigate this claim in more detail and reach a similar conclusion: (many of) those er-nominalisations classed as ‘non-eventive’ do contain the representation of an event. Their evidence comes from the availability of internal readings of certain classes of adjectives that modify er-nominalisations.Footnote ⁸ Some examples are given in (6)–(7). As these examples show, the internal reading of the adjective – that is, when it is interpreted as modifying the event contained within the nominalisation – is available both in the presence and absence of a syntactic complement. Note that the internal reading of these adjectives should be distinguished from their adverbial reading (‘we were occasionally met by a dolphin-trainer’), and their intersective reading (‘a car-dealer who is a big person’).

On the assumption that the internal (event-related) readings of these adjectives require there to be some representation of an event within the nominalisation, this shows that er-nominalisations both with and without complements contain an event representation.

Given this, Alexiadou & Schäfer (Reference Alexiadou, Schäfer, Alexiadou and Rathert2010) argue that the difference between those er-nominalisations with ‘eventive’ interpretations, which license syntactic complements, and those with ‘non-eventive’ interpretations, which don't license syntactic complements, is actually about what kind of aspectual operator binds the event variable. When an episodic aspectual operator binds the event variable, an ‘eventive’ interpretation obtains; when a dispositional aspectual operator binds the event variable, a ‘non-eventive’ interpretation obtains. Crucially, constructions with dispositional, habitual or generic interpretations generally permit object-drop much more freely. Example (8) exemplifies this using the English habitual:

1. (8) The sewing instructor always cuts Ø in straight lines.

It is the availability of object-drop in dispositional, habitual and generic contexts that leads to the correlation shown in (5), between ‘eventivity’ (in fact, episodicity) and having a syntactic complement. It's not about eventivity.

So, given that prototypical er-nominalisations appear to contain a representation of event structure, which they inherit from their base verb, I adopt the basic analysis of Alexiadou & Schäfer (Reference Alexiadou and Schäfer2009, Reference Alexiadou, Schäfer, Alexiadou and Rathert2010) (similar to the analysis of Roy & Soare Reference Roy, Soare, Iordăchioaia, Roy and Takamine2013, Reference Roy and Soare2014), as shown in (9). Er-nominalisations (at least, those of the variety that we are interested in) contain a full AspP, wherein event and argument structure are reified in the functional projections v and Voice. The nominalising head n serves to nominalise the structure and bind the R-argument, represented as ‘x’.Footnote ⁹

1. (9)

In prototypical er-nominalisations, the R-argument always corresponds to the external argument of the parent verb, so we can analyse the bound argument (‘x’) as being in Spec-VoiceP. The Asp head introduces either an episodic or dispositional operator, which binds the event/eventuality variable introduced in v, thus giving rise to the distinction that Rappaport Hovav & Levin (Reference Rappaport Hovav, Levin, Stowell and Wehrli1992) identified as [+/–event]. The complement to v is then projected just as it would be outside of nominalisation environments, but crucially, in a dispositional context the null object ‘Ø’ may fill the slot instead.Footnote ¹⁰ I assume that the argument and event structure of a CoPTEr is the same as with any other er-nominalisation – the problem comes when attempting to put a CP in the internal argument position projected within the nominalisation. In the next part of this section, I discuss the nature of this incompatibility.

2.2 Nouns with CP complements

Some nouns with CP complements are non-derived, as in (10), and others are derived from verbs, as in (11).Footnote ¹¹

1. (10)

   1. (a) I like your idea [that Barry is a vampire].

   2. (b) I don't believe the rumour [that Barry is a vampire].

2. (11)

   1. (a) I don't understand the belief [that Barry is a vampire].

   2. (b) I heard her suspicion [that Barry is a vampire].

The important point in this section is that nouns cannot take their CP complements as ‘true’ arguments – something noted by various authors (Stowell Reference Stowell1981; Pesetsky & Torrego Reference Pesetsky, Torrego, Guéron and Lecann2004; Moulton Reference Moulton2009, Reference Moulton2015; Elliott Reference Elliott2020). Instead, it appears that CP complements combine with their host noun as a modifier in some sense, rather than as an argument.Footnote ¹²

An important piece of evidence for the non-argumenthood of CP complements to nouns comes from the fact that they are not interpreted, semantically, like arguments of those nouns. Consider the pairs of sentences in (12)–(13). They show that the CP complement to a noun like idea or belief in some sense identifies the content of the noun, allowing the noun and CP to be connected by the copula (Higgins Reference Higgins1973; Stowell Reference Stowell1981).

1. (12)

   1. (a) the idea [that Barry is a vampire]

   2. (b) The idea is [that Barry is a vampire].

2. (13)

   1. (a) the belief [that Barry is a vampire]

   2. (b) The belief was [that Barry is a vampire].

This should be contrasted with what happens with true arguments of nouns. In the previous subsection, we discussed so-called complex event nominalisations (later referred to as Argument Structure Nominals, or ASNs), which ‘inherit’ the full argument structure of their parent verb, and must take the same obligatory arguments that their parent verb takes (Grimshaw Reference Grimshaw1990). Thus the nominal complement of a nominalisation like destruction, as in (14a), can function as a true argument of the nominalisation (once supplied with Case by a preposition like of). Unlike CP complements, true argumental complements (here, an of-DP) cannot be connected to their host nominalisation with the copula:

1. (14)

   1. (a) the destruction [of the city]

   2. (b) *the destruction was [of the city]

Perhaps it is expected that non-derived nominals like idea, as in (12), don't take true CP arguments, since they don't have a parent CP-taking verb from which to inherit argument structure. But it is surprising for deverbal CP-taking nominalisations like belief, as in (13) – we might expect belief to be able to form an ASN, and thus preserve the argument structure of its parent verb believe, at least in some environments. However, Moulton (Reference Moulton2009, Reference Moulton2013, Reference Moulton2015) shows that it is a systematic property of CP-taking nominalisations (belief, suspicion, explanation, etc.) that they just don't form ASNs. By way of evidence, Moulton notes that CP-taking nominalisations are incompatible with Aktionsart-modifying for/in-PPs:Footnote ¹³

1. (15)

   1. (a) They observed [that the butler was likely the killer] for several weeks.

   2. (b) their observation [that the butler was likely the killer] (*for several weeks)

2. (16)

   1. (a) John claimed for years [that the earth was flat].

   2. (b) John's claim (*for years) [that the earth was flat]

3. (17)

   1. (a) I decided [that he was a fraud] in five minutes.

   2. (b) my decision [that he was a fraud] (*in five minutes)

ASNs, by contrast, do allow Aktionsart modifiers, as in (18). In fact, Moulton (Reference Moulton2015) notes that the same nominalisation (e.g. observation) can function as an ASN when it takes an of-NP complement (18b), but must be a non-ASN when it takes a CP complement (15b).

1. (18)

   1. (a) the destruction [of the city] (in three hours)

   2. (b) Their observation [of the butler] (for several weeks) led to a conviction.

Following Moulton (Reference Moulton2015), I assume that CP-taking nouns, both those that are deverbal nominalisations and those that are not, have the reduced structure in (19). Unlike the structure for er-nominalisations in (9), no event or argument structure, in the form of the verbalising head v or the Voice head, is embedded under the nominalising head n. Instead the CP combines with the nP via Predicate Modification (I refer the reader to Moulton's work for a formal semantic analysis).

1. (19)

Further evidence for the non-argumenthood of CP complements to nouns, and thus that the structure in (19) is on the right track, comes from the unavailability of the CP pro-form so.Footnote ¹⁴ Moulton (Reference Moulton2015) points out that while a great many CP-taking verbs can take so – some examples are given in (20) – there is no variation among CP-taking nouns, which uniformly reject so-complements. Some equivalent examples are given in (21).

1. (20)

   1. (a) I believe so.

   2. (b) γ Mother India's Cafe: No gluten-free food even if they claimed so.Footnote ¹⁵

   3. (c) She would not admit so to DYFS because she feared the consequences. (Moulton Reference Moulton2015: 308)

2. (21)

   1. (a) *my belief so

   2. (b) *my claim so

   3. (c) *my admission so

Moulton argues that so, unlike full CPs, does saturate the argument slot of the verbs that select it. So therefore cannot combine with clause-taking nouns, because clause-taking nouns do not have an argument slot that can be saturated.Footnote ¹⁶

As an additional interesting point, it appears that nouns’ inability to take CP arguments is not a solely syntactic restriction, but stems from their interpretation. The examples in (22)–(23) show that idea/belief-type nouns cannot take of-DP arguments, if those DPs are substitutes for propositions (Moulton Reference Moulton2013):

1. (22)

   1. (a) I don't understand the idea [that Barry is a vampire].

   2. (b) *Barry – a vampire? I don't understand the idea of that.

2. (23)

   1. (a) I've never understood the belief [that Barry is a vampire].

   2. (b) *Barry – a vampire? What causes the belief of/in that?

If the ban on CP arguments was a purely syntactic constraint against elements of category C occupying an argument position, then the above examples should be fine – or at least, (23b) should be fine, since belief should be able to form an ASN. But instead, it appears that there is a slightly broader ban, not just on CP arguments within nouns, but on CP-taking nominalisations like belief forming ASNs in the presence of a proposition-denoting complement.Footnote ¹⁷ When we turn to CoPTErs in the next section, we will see that their inability to take CPs as arguments is more straightforwardly syntactic, since they can take propositional of-DPs as arguments (section 3.2).

Let's summarise this section. We saw first that er-nominalisations contain the representation of an event and inherit argument structure from their parent verbs, although this is sometimes obscured in dispositional er-nominalisations, where the dispositional aspectual operator licenses a null object. I adopted Alexiadou & Schäfer's (Reference Alexiadou, Schäfer, Alexiadou and Rathert2010) analysis of er-nominalisations, in which the nominalising head n merges with an AspP containing argument and event structure, introduced by the functional heads v and Voice. We then turned to CP-taking nouns, and saw several arguments that such nouns do not – indeed, cannot – take their CPs as arguments. Instead, CPs can only combine with nouns as modifiers. Either as a cause of this, or a consequence of it, CP-taking verbs don't form ASNs, but instead form nominalisations with the more minimal syntactic structure in (19), wherein the CP is an adjunct at the nP level. Note that I do not offer an explanation for why nouns cannot take CPs as arguments, and I refer the reader instead to Moulton (Reference Moulton2015) and Elliott (Reference Elliott2020) for recent proposals.Footnote ¹⁸

In the next section I discuss the intersection of these two topics: CP-taking er-nominalisations (CoPTErs), exemplified in (1). I argue that several of their properties can be explained as a consequence of a fundamental tension: er-nominalisations are ASNs, but nounhood is incompatible with CP arguments.

3.1 Er-nominalisations reject CP arguments

I present four pieces of evidence that er-nominalisations do not take CPs as arguments, even when used as predicate nominals. The first two pieces of evidence come from two CP pro-forms: so (as discussed in section 2.2) and the null CP pro-form that occupies the internal argument position of the verb in Null Complement Anaphora contexts, which I refer to as ‘Ø_NCA’. I show that both of these CP pro-forms are unavailable within er-nominalisations, and I attribute this to the pro-forms’ status as syntactic CPs. The third piece of evidence comes from the fact that the CP doesn't reconstruct for wh-movement. The final piece of evidence is somewhat weaker than these, and comes from the distribution of null complementisers.

We saw in section 2.2 (example (21)) that CP-taking nouns reject the CP pro-form so. Moulton (Reference Moulton2015) attributes this to the fact that so can only saturate an argument slot, but cannot combine with nouns as a modifier. Er-nominalisations similarly reject so:

1. (35)

   1. (a) *I am a firm believer so.

   2. (b) *She is a consistent claimer/claimant so.

   3. (c) *He is a one-time admitter so.

I follow Moulton in assuming that so is a CP pro-form that can only saturate argument slots. And while er-nominalisations do have an available argument slot for so to saturate (unlike other CP-taking nouns), they simply can't license CPs.

I now apply similar reasoning to Ø_NCA, the pro-form that occurs in the complement of those verbs which license Null Complement Anaphora (NCA). First I argue that Ø_NCA is (at least sometimes) a CP pro-form (following Haynie Reference Haynie2010 and Depiante Reference Depiante, Craenenbroeck and Temmerman2019). Then, I show that Ø_NCA cannot serve as an argument of er-nominalisations, even when the parent verb of the nominalisation lexically selects for NCA. I propose that this is because, as above, er-nominalisations can't take CP arguments.

Let's start with some background. The term Null Complement Anaphora (NCA) was introduced by Hankamer & Sag (Reference Hankamer and Sag1976) to describe the phenomenon in (36) (though Shopen Reference Shopen1972 had earlier described the phenomenon as definite constituent ellipsis). It's when the complement of certain verbs may be omitted, and recovered anaphorically.

1. (36)

   1. (a) I told them to take out the trash, but they refused Ø_NCA.

   2. (b) I didn't tell her that I was going to leave, but she found out Ø_NCA.

Analyses of NCA abound, but some analyses, including Hankamer & Sag (Reference Hankamer and Sag1976), Depiante (Reference Depiante2000) and Haynie (Reference Haynie2010), hold that a null pro-form (here ‘Ø_NCA’) replaces the complement of certain lexical verbs (e.g. refuse, find out). Crucially, Haynie argues that the null pro-form may be of several syntactic categories, including CP, but not DP.

This is very fortunate for our purposes, because this is just the opposite set of categories from those which are permitted in er-nominalisations – recall that er-nominalisations allow of-DP arguments and ban CP arguments. We therefore expect that the ability of a lexical verb to license Ø_NCA should disappear when that verb undergoes er-nominalisation – Ø_NCA is a CP, and CPs aren't licensed in argument positions within the noun. And indeed, this is what we find. In the (b) sentences in (37)–(40), the missing complement can only be interpreted as non-specific (i.e. a forgetter of things, a promiser of things, etc); it can never be interpreted as anaphoric.Footnote ²⁷

1. (37)

   1. (a) Sarcasm doesn't translate well? Yes, I frequently forget Ø_NCA.

   2. (b) Sarcasm doesn't translate well? Yes, I'm a frequent forgetter #(of that).

2. (38)

   1. (a) Will we go see a film tomorrow? Yes, I promise Ø_NCA.

   2. (b) We'll go see a film tomorrow? Yes, I am a chronic promiser #(of that).

3. (39)

   1. (a) We should rejoin the EU? Yes, I agree Ø_NCA.

   2. (b) We should rejoin the EU? Yes, I've been a consistent agreer #(with that statement) for half a decade now.

4. (40)

   1. (a) Mary says she's the one who burned down the old house? Yeah, she confessed Ø_NCA.

   2. (b) Mary says she's the one who burned down the old house? Yeah, she's an occasional confessor #(to that), in her more candid moments.

Let's now turn to the final piece of evidence that CoPTErs don't take their CPs as arguments: the CPs don't reconstruct. Consider first (41a), adapted from Kuno (Reference Kuno, Horn and Ward2004: 335), in which an R-expression (John) is c-commanded by a coindexed pronoun (he), creating a Condition C violation. In (41b), the NP containing the R-expression has been wh-fronted across the pronoun. Crucially, in order for (41b) to be grammatical and obviate a Condition C violation, the NP must not reconstruct into its base position.Footnote ²⁸

1. (41)

   1. (a) *He₁ tried to get [one psychiatrist's view that John₁ was schizophrenic] expunged from the trial records.

   2. (b) [Which psychiatrist's view that John₁ was schizophrenic]_i did he₁ try to get t _i expunged from the trial records?

CP complements to wh-fronted nouns can be contrasted with CP complements to wh-fronted verbs, as in (42) from Moulton (Reference Moulton2009: 63). The relative unacceptability of the coindexation indicated here, compared with that in (41b), indicates that CP complements to wh-fronted verbs do reconstruct, and thus that wh-movement of a clause-taking verb fails to obviate Condition C.Footnote ²⁹

1. (42) *[Whose loudly claiming that Bob₁ is the murderer]_i did he₁ not hear t _i?

Moulton (Reference Moulton2009, Reference Moulton2013) argues that the CP complement to a wh-fronted noun fails to reconstruct because it is merged late – that is, the CP is only merged with its host noun following wh-movement (Lebeaux Reference Lebeaux1988; Fox Reference Fox2002). And what lets it merge late is the fact that it is not an argument of the noun. By contrast, the CP complement to a verb is an argument of it, and so late merge is not an option.

CP complements to er-nominalisations pattern in just the same way as CP complements to other nouns. Example (43) shows that these CP complements do not reconstruct.

1. (43)

   1. (a) [Which firm believer that Mary₁ wasn't telling the truth]_i was she₁ constantly having to contradict t _i?

   2. (b) [Which frequent complainer that the police chief₁ was corrupt]_i did he₁ eventually assassinate t _i?

Thus, by Moulton's reasoning, the CP complement to an er-nominalisation can be merged late, indicating that it is not a true argument of the nominalised predicate. Note that the examples in (43) require placing CoPTErs in argument positions, something that is generally not grammatical (see (25)–(27), (28)–(30)). However, as noted in footnote 25, certain determiners, including which, do improve the acceptability of CoPTErs in argument positions.

A final, weaker argument for the non-argumenthood of the CP comes from the alleged unavailability of the null complementiser. It has been claimed that the null complementiser is only available with the complements to verbs and adjectives, never with the complements to nouns (Stowell Reference Stowell1981; Pesetsky & Torrego Reference Pesetsky, Torrego, Guéron and Lecann2004). This is illustrated by examples like (44), from Stowell (Reference Stowell1981: 398).

1. (44)

   1. (a) I distrust the claim (*that) Bill had left the party.

   2. (b) John's belief (*that) he would win the race was misguided.

To the extent that this generalisation holds, CP complements to er-nominalisations pattern with CP complements to other nouns. In my judgment, the CP complements in (45) require an overt that.

1. (45)

   1. (a) Sorry, I'm a frequent forgetter (*that) sarcasm doesn't translate well.

   2. (b) γ David Brooks, a one-time believer (*that) red and blue America demonstrated “no fundamental conflict”…Footnote ³⁰

Therefore, if the unavailability of the null complementiser can be derived from these CPs' status as non-arguments of their host noun (as Stowell Reference Stowell1981 argues), then we have a further argument that the CP complements of er-nominalisations are, similarly, not true arguments of their host noun.

However, Moulton (Reference Moulton2015: 318) points out that the empirical picture is not so clear-cut and that counterexamples abound, as in (46).

1. (46)

   1. (a) …in the belief he was buying a kilo of skunk cannabis. (Moulton Reference Moulton2015: 318)

   2. (b) γ Tis the season to be jolly (careful) …..With the announcement we ‘should’ be able to open next week on Thursday 3rd December.Footnote ³¹

But to the extent that the generalisation holds, we have yet another way in which CP complements to er-nominalisations differ from true argumental CPs.

To summarise, we've seen several pieces of evidence that CoPTErs do not take their CPs as arguments. Whatever the overt CPs in (1) and (24) are, they aren't arguments of the nominalisation. In the next subsection, I show that DPs, unlike CPs, can serve as the arguments of er-nominalisations. Ultimately, I integrate this finding into my analysis in section 4: the adjoined CP binds a variable whose syntactic category is D.

3.2 Er-nominalisations accept DP arguments

We have seen so far that CP-taking non-ASNs like idea and belief cannot take propositional DPs as complements (see (22)–(23)). And we have also seen that, by contrast, CoPTErs can take propositional DPs as complements (see (37)–(40)). The following pairs of sentences explicitly compare the two – note in particular that the er-nominalisations are in argument positions, and are not being used as predicate nominals:Footnote ³²

1. (47)

   1. (a) *Eleanor – a werewolf? I don't understand the belief of that.

   2. (b) Eleanor – a werewolf? I once met a serious believer of that.

2. (48)

   1. (a) *Eleanor – a werewolf? I don't understand the claim of that.

   2. (b) Eleanor – a werewolf? Yes, I once met a serious claimer of that.Footnote ³³

Non-ASNs like belief and claim can't take a propositional DP as a complement simply because they can't take arguments (cf. section 2.2). CoPTErs like believer and claimer, by contrast, are ASNs (cf. section 2.1), and so can take a propositional DP as an argument. Then, for CoPTErs, the relevant difference between a DP complement (acceptable) and a CP complement (unacceptable) is that a DP can be syntactically licensed in the er-nominalisation's internal argument position (by inserting of or a lexically selected preposition), while a CP cannot be syntactically licensed in the argument position by any means (and so, I suggest in this article, can only be licensed by additional functional structure external to the nominal).

In the next section, I provide a tentative analysis, originally sketched in (9), which captures both the non-argumenthood of CP complements to CoPTErs and why CoPTErs are acceptable only in predicate position.