Zero-suffixes and their alternatives: A view from French

2.1 Distributed Morphology

Marantz (1997) argues that the Lexicon as a place for derivational morphology should be dispensed with. Building on Chomsky’s (1970) “Remarks on nominalizations”, he argues in favor of the insertion of lexical category-less roots in syntactic structures, with functional projections determining the syntactic behavior of nominalizations ending in -tion and -ing and nominalizations such as growth. While all are dominated by DP, they differ in the presence of a verbalizing functional projection directly dominating the root. The agent-projecting v-1 serves to “verbalize” roots. Marantz argues that it is not only present in sentences containing a verb (1), but also in -ing nominalizations (2). In the representations that he gives, the verbalizer is absent in -tion nominalizations (3) and in growth (4). The absence of the agent-projecting v-1 in growth is based on the crucial aspect of Chomsky’s (1970) analysis that the root √GROW is non-agentive. Marantz assumes that v-1 is absent with -tion nominalizations as well. To account for the difference between John’s destruction of the city and *John’s growth of tomatoes he proposes that with externally caused change of state roots such as √DESTROY, the possessor may be interpreted as an agent, while with internally caused change of state roots such as √GROW this is not possible:

In later work Marantz also introduces the categorizers “little n” and “little a”, besides “little v”, as functional heads that determine the category of a word, see, e. g., Marantz (2001). In the structures of (2), (3) and (4) the categorizer “little n” is merged below D, as in (5)–(6):

In another analysis of nominalizations proposed within the framework of Distributed Morphology, the one put forth by Alexiadou (2001), nouns like destruction are ambiguous between an event reading and a result interpretation (Grimshaw 1990). In the first case Alexiadou assigns them a structure as in (2), whereas in the second case they receive a structural analysis as in (3).

Marantz (2001) states that words can be formed in the domain of the root by attaching a morpheme to the root. This happens in the case of destruction and growth. Words can also be formed outside the domain of the categorizer by merging a morpheme, as in the case of destroying, where the morpheme -ing is merged outside “little v. The morphemes -ing, -tion and -th are inserted into the “n” node. While in these cases there is an overt n, if there is no overt nominalizing suffix, as in the case of the noun glory, Marantz assumes that there is a null (or zero) n:

(7)

In Distributed Morphology, Vocabulary Insertion, a post-syntactic procedure, associates a phonological form with functional nodes. A functional n node can spell out as -ation, but also as Ø, a phonologically null suffix, in which case there is null exponence. The categorial nodes, besides specifying a category, are also associated with some syntactico-semantic information.

2.2 The Exo-Skeletal model

Although some linguists, like Marantz (2001), have argued that categoryless roots are necessarily dominated by a categorizer in cases such as (7) (e. g., Panagiotidis 2011), others have argued against them, more specifically because of the use of zero-suffixes. Objections against the use of zero-suffixes in syntactic approaches to morphology such as Distributed Morphology concern, among other things, a proliferation of the use of zero-suffixes and the issue of directionality of derivation.

Borer (2013) shows that in a model such as Distributed Morphology an adjective such as formal could be associated either to the representation in (8) or to the one in (9):

(8)

(9)

Besides stacking, a second objection against zero-morphemes is that zero-morphemes can have all kinds of categories and can provoke all kinds of syntactico-semantic changes to the root. A third objection concerns directionality of derivation. In theory it should be possible to add a zero-suffix to an overt suffix. But Borer (2013) shows that, in English, this is not possible:

Because of these problems with zero-morphemes, Borer proposes a model that does not make use of zero-morphemes.

In Borer’s (2003) Exo-Skeletal model, which is a syntactic approach to morphology just like Distributed Morphology, zero-affixes are completely banned. In this model, category-less roots can get a category in two ways: either by means of the merger of functional projections such as DP and TP (and thus not by categorizing functional categories, dominating a zero-affix), or by being subcategorized by a derivational morpheme. Besides specifying its own category, the derivational morpheme also determines the category of its complement, namely if it is a root. A derivational morpheme can also be merged above a root that has already been categorized by dominating functional projections. Crucially, in both cases, the derivational morpheme cannot be a zero-suffix.

Borer observes that two counterarguments can be raised to her theory: V → N conversions in English and parallel suffixed A/N derivations. Borer tries to dismiss both counterarguments.

V → N conversions in English are nouns like drive, walk, break and murder. Borer (2003) shows that they cannot be combined with complements. She argues therefore that they are result nominals and not event nominals (which she calls Argument Structure nominals):

Borer argues that the absence of the Argument Structure reading of V → N nominals follows from the assumption that zero-suffixes do not exist in her theory. Only an overt derivational morpheme can nominalize a root, i. e., the L-D, the lexical domain, dominated by verbal projections. This explains why Argument Structure nouns as in (12) are possible, but why those in (11) are not. Borer’s analysis of nominalizations with -ing corresponds to Marantz’ analysis in (2) in the assumption that the nominalization contains a verbal projection. In the same way Borer’s analysis of destruction corresponds to Alexiadou’s (2001) analysis:

In a footnote (fn. 13) Borer (2003) observes that the generalization that V → N nominals can only function as result nouns has some counterexamples. She notes that there are at least some speakers who reject (11-a)–(11-d), but that others find them acceptable. The use of adjectives like constant and frequent without a plural, theta-assignment, aspectual modifiers and by-phrases as arguments are diagnostics used by Grimshaw (1990) to distinguish Argument Structure nominals from result nouns:

Borer (2013) enumerates a longer list of deverbal Argument Structure nouns in English (for an overview see Iordăchioaia (2020)):

Borer notes that these cases are problematic for her theory and that they should possibly involve some kind of null suffix.

Another problematic issue is the availability of parallel A/N suffixes (Borer 2013: §7.5). The problem of their existence for syntactic approaches has been raised by Lowenstamm (2014) and in work on Dutch by De Belder (2011). De Belder analyzes all derivational suffixes as roots. Borer (2013) acknowledges that parallel A/N suffixes also occur in English, as in republican , socialist , accessory and Iraqi . She argues, however, that parallels especially concern A/N, and that in Dutch there are only three cases involving other categories, which suggests that suffixal homonymy apart from A/N is rare. Most suffixes have one category. Furthermore, in order to reduce the number of A/N cases, Borer argues that some of the A/N cases could be reanalyzed as cases of nominal ellipsis.

Although Borer’s Exo-Skeletal model is very attractive because it tries to eliminate zero-suffixes, I will signal some problems with the model in the next sections and I will propose a slight adaptation of the model, still avoiding the use of zero-suffixes. The argumentation will be based on French.

3.1 Nominalized infinitives

Nominalized infinitives occur in languages such as Spanish, Italian, German and Dutch. They are analyzed for instance within the framework of Distributed Morphology by Alexiadou et al. (2011); see also Alexiadou (2020). The authors distinguish two types of nominalized infinitives: verbal nominalized infinitives and nominal infinitives. Alexiadou et al. argue that between these two types of nominalized infinitives, but also between languages, the number of functional projections may differ. Besides nominalized infinitives in Spanish and German, they analyze the English gerund. Whereas Alexiadou’s (2001) analysis of deverbal nouns like destruction is based on Marantz’ (1997) analysis of deverbal nominalizations, which may involve the categorizer v, Alexiadou et al.’s (2011) analysis of nominalized infinitives is rather based on Marantz’ later work, such as Marantz (2001), presented in Section 2. In their analysis, both “little v” and “little n” are used, which besides other functional projections, dominate the root. A verbal nominalized infinitive only contains “little v” in the analysis that they present:

Structure (15), containing TP, is proposed for Spanish, where the verbal nominalized infinitive can contain a subject:

For the nominalized verbal infinitive in German and the English verbal gerund, structure (15) without TP is proposed:

Nominal infinitives and the nominal gerund contain both “little v” and “little n”, in Alexiadou et al.’s analysis:

Structure (18) represents phrases like (19) in English or (20) in German, small details omitted:

Old French also had nominalized infinitives (Buridant 2008). Sleeman (2010) shows that, just like the nominalized infinitive in Spanish and German and the English gerund, they could be verbal, see (21) with a direct object and (22) with an adverb, or nominal, see (23) with a plural:

Mixed verbal and nominal properties are also possible. This means that there can be five types of infinitives in Old French (or in other languages with nominalized infinitives): purely verbal infinitives, verbal nominalized infinitives, mixed nominalized infinitives, nominal infinitives, purely nominal infinitives, i. e., lexicalized nominal infinitives such as le rire ‘the laughter, the smile’. As has been argued in the literature, the syntactic behavior of nominalized infinitives is reflected in the behavior of overtly suffixed deverbal nominalizations: they can also be verbal or nominal to various degrees (see, e. g., Sleeman and Brito 2010 and Sleeman 2021).

The analysis in Sleeman (2010) is based on Alexiadou et al.’s (2011) analysis. In this paper I propose an analysis partly based on Borer’s Exo-Skeletal model, omitting the use of “little v” and “little n”. I propose, however, also a slight adaptation of Borer’s model.

As shown in (12-b), Borer analyzes -ing forms introduced by a determiner as a nominal suffix. Whereas for English -ing it could be argued that it is a nominal suffix because it also creates nouns (a living), infinitival suffixes do not do so. If infinitival suffixes cannot be analyzed as derivational suffixes (C-functors), Borer’s theory has to be adapted, because otherwise the category change from verbal to nominal cannot be accounted for. I propose that the infinitival suffixes are simply verbal, as they are in their normal verbal use. This means that in the case of nominalized infinitives there are no suffixes (C-functors), but only functional projections (S-functors in Borer’s theory). This is represented by Alexiadou et al.’s (2011) analysis of verbal nominalized infinitives as well, because there is no “little n”: there are only S-functors, in Borer’s terms, apart from “little v”. Differently than in Grimshaw’s (1991) Extended Projections model, it should thus be allowed in Borer’s model that “nominal” functional projections dominate “verbal” ones, without intervening categorizers, as in (24). This was, in fact, already the case in Marantz’ (1997) proposal. See also Borsley and Kornfilt (1999). Alexiadou (2020) also argues that mixing of functional layers should be possible, although she still makes use of “little v” and “little n” in her analysis of mixed categories.^[2]

If categorizers are omitted, the structures of verbal and nominal nominalized infinitives only contain functional projections (S-functors), besides the root, but not the categorizers v and n. Borer calls a functional projection (S-functor) that licenses an object ASP_Q. Iordăchioaia (2020), presenting her analysis in the Distributed Morphology framework, calls it a ThemeP. There would also have to be S-functors licensing the complement of a nominal infinitive. The designations of the S-functors are not important. For the illustration they can simply be called F. Verbal nominalized infinitives (24) would be dominated by more S-functors than nominal infinitives (25). Their exact number or nature is not important here, but the S-functors in (24) would be “verbal” and the one in (25) “nominal”:

Such structures may not only be proposed for languages with nominalized infinitives, but also for the English gerund preceded by a determiner, but it means that -ing would not be an overt derivational suffix then, but rather an inflectional suffix, that is, a verbal suffix. This would need an adaptation of Borer’s theory, because in Borer’s model S-functors dominate roots or are extended projections of a C-functor, but mixing of nominal and verbal S-functors, such as DP and the verbal S-functors in (24), is not possible. The proposed adaptation consists thus in allowing categorial mixing of S-functors.

In the next subsection, another type of deverbal nominalization is discussed and analyzed, namely a type without any derivational or inflectional suffix.

3.2 V → N derivations without suffixation

Besides V → N derivations with overt suffixation (-ation, but also -ing in Borer’s analysis, see Section 3.1), Borer (2013) also discusses V → N derivations without overt suffixation, more specifically in English. Borer distinguishes two types. Those in (26) can function both as a verb and a noun and do not take complements:

V → N conversions have been argued by Kiparsky (1982 and further work) to involve directionality. In Kiparsky’s analysis N is derived from V by means of a zero-suffix (see also Don 2004, for Dutch). Borer argues that in (26) the verb and the noun share the same category-less root and that they are categorized by means of S-functors. Since they share the same root, the noun is not derived from the verb or vice versa. This means that the noun is a result noun. If it would be an Argument Structure noun, there would have to be a nominalizing suffix (a C-functor) dominating verbal S-structure. Borer uses the fact that the nouns in (26) are result nouns as an argument against zero-suffixes, i. e., zero C-functors.

There are, however, also deverbal nouns that are more problematic for Borer’s Exo-Skeletal model. These are deverbal nouns without a suffix, but that behave like Argument Structure nouns. Some examples, from Borer (2003, fn. 13), were already provided in (13). Another one is shown in (27). The Argument status of these nouns is suggested by the use of adjectives like frequent, complements and by-phrases (see Grimshaw 1990 or Alexiadou and Borer 2020 for the precise list of diagnostics):

The list given by Borer (2013) in (14) contains verbs such as change, release, resolve, and use. In Borer’s Exo-Skeletal model cases such as these cannot be accounted for. In the case of Argument Structure nominals, there has to be a root dominated by some verbal S-structure, which is categorized subsequently into a noun by means of a nominal C-functor, i. e., by an overt suffix. Borer observes that the list is rather small and that verbs as those in (14) are rather exceptional in English. Borer does not have an explanation for these cases, but suggests in a footnote (2013, fn. 13) that there could be “a highly marked categorizing affix which is phonologically real enough, although perhaps not strictly segmental”. Such an affix would not be available for roots as those in (26).

Iordăchioaia (2020) shows that V → N conversions are much more common in English than thought. On the basis of an extensive corpus research, she argues that many deverbal nouns in English are Argument Structure nouns, because in the corpus they occur with arguments, adjectives like frequent or constant, or with predicates expressing an event. She argues that this possibility is based on the type of root.

In French, deverbal nouns are also abundantly present. In a small dictionary research, I found already more than a hundred cases of deverbal nouns with argument structure (at least one argument). For this small research I searched in the online version of the French dictionary Le Trésor de la Langue française ^[3] for masculine nouns that contained the words “action” or “action” and “résultat” in their description. I checked these words in the French dictionary Le Petit Robert (Rey-Debove and Rey 2010) on their presence, interpretation and on their occurrence with argument structure in the examples that are given in the dictionary. All could be translated by a nominalized infinitive or a gerund.^[4] A more extended research in dictionaries and corpora could provide more cases and more evidence for the Argument Structure nature of these nouns (see Sleeman 2021). For the argumentation of this paper, this small research seems, however, to suffice. Examples of Argument Structure nouns are presented in Table 1. Some correspond to the Argument Structure nouns in English listed in (14).

There are also some more archaic cases such as le ravage d’une région par des pillards ‘the devastation of a region by looters’. Most of the nouns in Table 1 belong to the non-Latinate part of the French vocabulary, first attested in the Old French period. Some of them exist next to a suffixed alternative, with a slightly different meaning:

Some of these deverbal nouns bear a suffix in English, but possibly with a slightly different meaning:

The French deverbal nouns in Table 1 are all masculine, just like the nominalized infinitive in Old French. Masculine gender is the default gender in French.^[5]

Within a lexicalist approach, it has been proposed by, e. g., Zwanenburg (1993) that deverbal nouns in French like those in Table 1 contain a zero-suffix, which forms a paradigm of synonyms with overt suffixes forming deverbal Argument Structure nouns: -ation, -age, -(e)ment, -ure, Ø. In an attempt to get rid of zero-suffixes, as Borer does, I propose, however, a similar analysis as what I have proposed for nominalized infinitives to account for the abundant occurrence of Argument Structure nouns without an overt suffix in French: they are formed by a variable number of functional projections (only S-functors) above the root. They differ from nominalized infinitives in that they do not contain a functional projection for the verbal inflection, which in the case of nominalized infinitives is expressed by infinitival inflection. Since there is neither infinitival inflection, as in the case of nominalized infinitives, nor derivational inflection (a C-functor), as in the case of deverbal Argument Structure nouns like destruction, non-suffixed deverbal nouns simply have the form of roots. Just like nominalized infinitives, as in (24)–(25), non-suffixed deverbal nouns can have a more “verbal” or a more “nouny” interpretation. In their more verbal interpretation, there is verbal S-structure dominated by DP, while in their “nouny” interpretation as a result noun, there is simply a DP dominating the root.

Such an account for deverbal nouns has been discarded by, e. g., Fábregas (2012). Discussing the problem of counterarguments formed by deverbal nominals for Borer’s Exo-Skeletal model, Fábregas (2012: 74), in the spirit of Grimshaw’s Extended Projections theory, argues that “using a functional functor will give rise to ungrammaticality, because the functional category needs to be compatible with the label of the selected structure, and classifiers (being part of the functional structure of nouns) are not directly combinable with aspect (which belong to the verbal domain)”. Fábregas therefore proposes another alternative for deverbal nominalizations that avoids the use of zero-categories, namely spell-out of different heads at the same time as one portmanteau morpheme.

It has, however, also been proposed in the literature that Grimshaw’s Extended Projection theory is too strong, and that mixing of functional projections should be allowed to some extent, e. g., by Borsley and Kornfilt (1999) (cf. Alexiadou 2020, with a comparable view but taken from a slightly different perspective, see Section 3.1). Categorial mixing of functional projections, permitting the avoidance of the use of zero-suffixes, can easily be incorporated into the Exo-Skeletal model. Therefore, I propose this adaptation instead of Fábregas’ solution.

In the next section, we turn to another challenge for the Exo-Skeletal model, namely suffixed A/N parallels.