Zero morphology and change-of-state verbs

1 Introduction

This article proposes an analysis of adjective-verb conversion. The phenomenon is illustrated in (1).

Conversion refers to the fact that the adjective narrow in (1a) has the exact same form as the verb narrow in (1b-c). The verb can have either an inchoative (change-of-state) reading (1b), or a causative one (1c).

Alongside the conversion cases in (1), deadjectival verbs in English sometimes require an overt suffix, as in (2).

Interestingly, the verbs show the same ambiguity between an inchoative and a causative reading as the unsuffixed verbs.

The standard approach to deadjectival verbs attributes to the examples (1) and (2) the exact same bimorphemic structure, (3a,b), where the verb is derived from the adjective. The two classes differ only in whether the verbalising suffix is overt or not.

The idea that we pursue in this paper is that in a case like narrow, the apparent zero marking of the verb arises as a consequence of phrasal spellout, where the relevant phonology is associated to a complex syntactic object, as indicated by the circle in (4). In developing the analysis, we also show how it accounts for the inchoative-causative ambiguity of the derived verbs, which we can observe in both types, the unsuffixed and the suffixed ones.

The paper is organised as follows. In Section 2, we present some of the data that motivate the approach. Section 3 introduces the prerequisites for our analysis. The main proposal is described in Section 4. Section 5 discusses suffixal marking, and ends with some predictions regarding the typology of morphological marking in the triplet adjective-inchoative-causative. Section 6 examines how these predicted patterns are realised in Czech deadjectival verbs. Section 7 briefly touches upon the topic of anticausatives.

2 The data

We start by presenting a representative selection of English deadjectival verbs in Table 1. In the first column, the table contains verbs that are zero-derived from adjectives. The other columns contain suffixed deadjectival verbs and one group of prefixed deadjectival verbs, as indicated in the column headings.

Our main point is that it is to a large extent an arbitrary property of the adjective whether it forms a zero-derived verb, or whether it requires a suffix, and if it requires an affix, which affix.

Let us start by noting that there are a couple of phonological restrictions to be observed. For instance, the suffix -en normally attaches to monosyllabic adjectives ending in a plosive, fricative or affricate (Plag 2003: 117–118, Carstairs-McCarthy 2002: 55–56). However, as Carstairs-McCarthy (2002: 87) noted, there are adjectives that have the right phonology and nevertheless do not take -en. The adjective wet is an example, because *wetten does not exist, but the zero-derived wet does. In sum, while the phonological rule can explain why *greenen will never be formed (green does not end in a plosive or fricative), additional factors are at play for adjectives that do satisfy the condition.

Something similar applies to the suffix -ify: this suffix attaches only to bases that are either monosyllabic, stressed on the final syllable or end in unstressed /ɪ/. However, this does not mean that -ify can attach to all adjectives that obey these phonological conditions. A case in point is the adjective empty, which allows for a zero-derived verb, contrasting with pretty, which does not.^[1]

In sum, even though there are some phonological conditions that restrict the formation of deadjectival verbs in English, it is impossible to predict which adjective allows for zero marking and which one requires an affix. As a result, we conclude that which class an adjective belongs to is to some extent arbitrary, and this information therefore needs to be learned and stored for each adjective in the lexicon.

Similarly, it is not predictable which suffix will appear on a suffixed deadjectival verb. For instance, it is not clear why -ify can derive falsify, while -en cannot derive *falsen, even though the phonological conditions would allow for either of these suffixes (compare coarsen).^[2]

The next point we want to bring out is that all the affixes shown in the table can show the inchoative–causative ambiguity. For reasons of space, we only illustrate this for the suffix -en, (5):

In this paper, we present a Nanosyntactic account of how the difference between zero derivation and suffixation is encoded in the lexicon. Our main aim is to show that the Nanosyntactic framework, where zero-derivation is modeled by means of phrasal spellout, has the right properties to model this distinction in a way that allows us to provide an explanatory account not only for English, but also for other languages. Moreover, the very same type of account will allow us to explain the causative-inchoative ambiguity, which is found in English, but largely absent in Czech, as we discuss in Section 6.^[3]

3 Prerequisites

In this section, we lay out the prerequisites for our analysis. Specifically, we adopt here the structure of deadjectival verbs as proposed in Ramchand (2008). In Ramchand’s work, verbs in general are decomposed into a series of heads which she refers to as Initiaton, Process, and Result. They are organised hierarchically as in (6):

(6)

The Init head introduces the causation event. The Proc head specifies the nature of the change or process, whereas Res gives the result of the event. Each of these heads may license a specifier (not shown in (6)): the specifier of the causation event is the initiator, the specifier of the process is the undergoer, and the specifier of the result is the resultee (Ramchand 2008: 40). For example, in the sentence Mary gave the book to Bill, Mary is the initiator who brings the event about, the book is the undergoer of the Process (it changes possession) and ‘Bill having the book’ is the Result. The Init and Res heads are optional in that some verbs lack them. For instance, the verb get in Bill got the book lacks the Init component.

Against this background, Ramchand (2008: 90,108) analyses deadjectival verbs as in (7). The trees depict the structure of the causative, the inchoative and the adjective, respectively. Recall that each of Proc and Init brings along an argument position, which accounts for the argument-structure alternations.

Building on Hay et al. (1999), Ramchand (2008: 90) proposes that verbs like dry are “a special kind of process verb where the degree of verbal change is mapped onto a property scale of some sort (derived from a basic adjectival meaning). Thus, in their intransitive use, they are classic proc verbs, with the single argument being an undergoer.” This is shown in (7b), which is just a special instance of the generally adopted structure (3b). Note that like Ramchand, we understand the AP in the complement of Proc as representing not directly the positive-degree adjective, but rather a scale on which both the inchoative and the positive adjective are based.^[4]

In Ramchand’s system, the causative in (7a) has an additional Init head, which introduces the Initiator argument. Verbs such as narrow in their causative use thus contain minimally two components of meaning that apparently receive no overt marking (i. e., Init and Proc).

In sum, the syntactic structures in (7) are based on the two assumptions given in (8):

The evidence in support of these assumptions is both morphological and semantic.

We start with morphological evidence in support of (8a). We have already discussed some of the evidence in Table 1, which shows how verbs are derived from adjectives by various affixes. The structural representation of these cases in (9) clearly shows how the verb is more complex than the adjective.

This type of evidence can be replicated for many languages, where verbs related to adjectives are typically morphologically more complex than the adjectives.

Semantically as well, the meaning of the verb contains that of the adjective. For example, the verb ‘to open’ means ‘to become open’, or ‘to cause to become open.’ This type of paraphrase generally works quite well for the verbs in Table 1. In some cases, a paraphrase containing a comparative of the adjective is more appropriate. For instance, atelic sentences such The soup cooled for hours do not entail that the soup ultimately became cool, but they nevertheless entail a change along the relevant scale. As we have made it clear above, this is compatible with our proposal, since the AP at the bottom of the tree does not necessarily correspond to the positive degree, but rather to the scale along which the change proceeds.

Let us now turn to the second assumption in (8b). This is also supported by semantic evidence. Intuitively, to open in its causative sense means ‘to cause to open’, with the second occurrence of open being the inchoative one (but see Harley 2012 for a different view). Lundquist et al. (2016: 2) put this more formally in what they call the Causational Entailment:^[5]

This entailment is what accounts for the (semantic) deviance of (11).

As far as the morphology is concerned, the picture is more complex. There are languages where the morphology supports the claim that the causative contains the inchoative. We reproduce some empirical evidence from Haspelmath (1993: 91) in (12). The table shows cases where the causative is morphologically more complex than the inchoative. We refer to this as the transparant pattern.

As the discussion unfolds, we will also get to see an inverse pattern, which is traditionally referred to as the anticausative, where the inchoative is morphologically more complex than the causative. We come back to anticausatives in Section 7. For now, we focus on the causative-inchoative entailment and the transparent morphological pattern as support for the structures in (7).

4 Zero-marking as phrasal spellout

As highlighted in the introduction, our main goal is to explore the idea that there are no zero morphemes used in the derivation of deadjectival verbs. Our claim is that zero morphemes are an illusion, arising when grammatical meanings expressed by the purported zeroes are actually realized (in a portmanteau fashion) by the root and/or other morphemes.

The idea that roots can (in addition to the conceptual meaning) also express adjacent grammatical categories is perhaps best seen in cases of root suppletion (bad–worse; man–men). These cases can be understood as instances of a scenario where the suppletive root realizes both the lexical category (A or N) and the relevant grammatical category (cmpr or plural). We show this in (13) and (14).

An alternative analysis could (of course) conclude that the relevant grammatical categories in these forms have a “zero allomorph.” While we are aware of the fact that there is no direct link between suppletion and the absence of regular morphology (e.g., bett-er), we think that the frequent absence of regular morphology with irregular roots such as worse or men is not accidental. Therefore, the intuition we shall follow is that whenever regular markers are absent with particular roots (such as in (13) and (14)), this is because the root already realises the relevant meanings, thereby blocking the appearance of regular morphology.^[6]

In the remainder of this paper, we extend the type of analysis in (13) and (14) to deadjectival verbs. In the technical implementation of this idea, we follow the Nanosyntax approach (Starke 2009, 2018). Nanosyntax is a Late-Insertion theory of morphology, where the syntactic structure is assembled first, and then it is spelled out using lexical entries. In this conception, lexical entries are understood as stored links between syntactic representations on the one hand, and a phonology and/or a concept on the other hand. Crucially, the syntax part of the lexical entry consists of a full syntactic tree (rather than a terminal), similarly to what we saw in (13) and (14).

With this background in place, consider now the lexical entry for narrow in (15), which links the phonology /ˈnæ erU / (represented by means of plain spelling) with a particular syntactic tree:^[7]

When syntax builds the structure of a causative verb, as in (16), the lexical tree in (15) is identical to it, and spellout is successful. We represent this by placing a circle around the syntactic structure in (16). Phrasal spellout thus obviates the need for zero morphemes: the init and proc heads do not dominate zero morphemes, but the lexical entry of narrow is such that it may cumulatively realise several heads, including init and proc.

The next question is how the lexical entry of narrow can also realise the structure of an inchoative verb and an adjective. The phrasal-spellout model treats this as an instance of syncretism, i. e., different meanings expressed by the same form. Technically, this is achieved by proposing that lexical entries may lexicalise various different syntactic trees – under the condition that the syntactic tree is contained in the lexical entry. This condition is known as the Superset Principle:

The result of adopting the Superset Principle is that the entry for narrow in (15) can also spell out the inchoative structure (18a) and the adjective structure (18b), because these trees are contained in (15).

In sum, the mechanism of phrasal spellout gives us an elegant way of representing two related phenomena: on the one hand, the zero marking of certain types of deadjectival verbs, and on the other hand the syncretism between the inchoative and the causative verb.

5 Suffixal marking

As we saw in Table 1, only a subset of adjectives allows for zero marking of the corresponding verb. A lexical item like wide contrasts with narrow in that wide is only an adjective and not a verb. The contrast is illustrated in (19).

Recall that whether an adjective behaves like narrow or wide is a matter of lexical idiosyncrasy, i. e., it has to be learned (and stored) for each relevant lexical item. In the phrasal-spellout model, this is achieved simply by associating an adjective like wide to an AP only:

(20)

Since the lexical tree (20) does not contain the syntactic tree of a verb, such a lexical item will not be able to function as a verb: the root wide will need help from an additional morpheme to realise all meaning components, as informally shown in (21).

What is the lexical entry for -en? Recall that in Nanosyntax, lexical entries are understood as memorized links between syntactic structures (constituents) and phonology. Therefore, by proposing that the suffix -en spells out the heads proc and init, as shown informally in (21), we are led to assign to it the lexical entry (22), which corresponds to a constituent containing these two features.

The geometry of the lexical entry reflects the structure (21): specifically, the entry (22) is identical to (21) minus the part of the structure that is spelled out as wide. In effect, the structure (21) literally divides into two parts, where one part is spelled out by wide and the other by -en.

But now a problem arises. Specifically, the lexical tree in (22) does not contain the syntactic tree (21), nor any subpart of it, as a subtree. Therefore, the Superset Principle is not met, and the heads Proc and Init cannot be spelled out by (22) as long as the structure is as given in (21). This problem is resolved by moving the AP node to the specifier of the InitP, yielding (23). This movement is driven by the need to lexicalise the features Init and Proc.

(23)

The initP in (23) is now an exact match for the lexical entry of the suffix, and spellout is successful. Moreover, the affix -en now appears where we want it, namely as a suffix following the root.^[8]

The inchoative version of widen works analogously, except that the syntactic derivation lacks the init head, (24a). Movement of the AP to the left yields (24b). Since the remnant ProcP is a subtree of (22), -en can spell out the ProcP in (24b).

It is relevant to note that if there was a suffix in the English lexicon that would only be specified for the circled ProcP, this suffix would also qualify as a match for (24b), and it would, in fact, take precedence over the ambiguous causative/inchoative -en, on the grounds that it is more specific. There is no such suffix in English, but we will be led to propose such a suffix in Czech in the next section.

To close off this section, we consider some implications of our theory for the typology of marking in the triplet adjective – inchoative – causative. So far, we have discussed two cases, instantiated by the English adjectives narrow and wide, respectively. However, the scenarios allowed for by the system just developed are wider. We show some of the predicted scenarios in Table 2.

Table 2

Five scenarios.

This table should be read as follows. The top row represents the features found in various forms. The scenario where the root is a syntactic adjective is defined by the feature A. This scenario is represented on the lines marked by (a) (i. e. 1a, 2a, etc.). The grey cells stand for heads which are not present in the derivation, namely proc and init. To derive an inchoative verb, we add proc to the adjectival structure, as shown on lines (b). The causative verbs add init (lines (c)).

The horizontal sections in the table (numbered 1–5) represent various root and suffix types predicted by our analysis, ordered according to root size, from the smallest (AP-size) roots in the top section to the largest (InitP-size) roots in the bottom one. A lexically small root like wide is shown in section 1. It can realise no more than the adjective feature A. When it functions as a verb, it needs a suffix, as shown on lines 1b and 1c, which correspond to the derivations we have just seen. The suffix is the same for the causative and for the inchoative verb, giving rise to syncretism between the causative and the inchoative.

We show an InitP-size root, like narrow, in section 5 of the table. As explained above, the Superset Principle ensures that such roots are three-way syncretic: they can realise each member of the triplet adjective-inchoative-causative, as shown on the lines 5a, 5b, and 5c, respectively.

In the final column, the table contains the names given in Haspelmath (1993) to different types of morphological marking of the causative-inchoative alternation. The two English patterns (given in section 1 and 5 of the table) are called ‘labile,’ showing syncretism between the causative and the inchoative. The reason why we have two different labile patterns is that Haspelmath does not consider the derivational relation to the adjective. As a result, his classification slightly deviates from ours, but we include Haspelmath’s terminology here to show that we are dealing with well-established typological patterns.

At least three additional patterns of marking are predicted by our system, shown in sections 2, 3, and 4 of the table. The first of these (section 2) is one where we have different suffixes for the inchoative and the causative, a possibility already pointed out in the discussion surrounding (24). This is a scenario where one suffix (suffix1) lexicalises just proc, whereas a different suffix (suffix2) lexicalises both init and proc, such that suffix1 (marking the inchoative) is replaced by suffix2 in the causative. This pattern is labelled the equipollent pattern in Haspelmath (1993), since it involves a common stem, to which different markers are added in the inchoative and the causative.

The second predicted pattern is in section 3 of the table. In this scenario, the causative suffix lexicalises only init and not proc. Therefore, a causative verb is derived by stacking a causative suffix on top of an inchoative one, as shown on line 3c of the table.

The third additional pattern (section 4) is one where the root is of size procP, i. e. larger than a mere adjective, but smaller than a causative verb. Such roots are predicted to have a zero-derived inchoative, but a suffixed causative. Both patterns 3 and 4 are called causative in Haspelmath (1993), since they involve a causative that is derived from the inchoative.^[9]

In the remainder of this paper, we provide evidence for some of the additional patterns by discussing deadjectival verbs in Czech. Specifically, we argue that Czech instantiates Patterns 1, 2, and 4 of Table 2. Before we turn to this detailed discussion, we want to briefly consider evidence for the existence of all three of the predicted additional patterns in Turkish. The relevant patterns are shown in Table 3.^[10]

The numbering of the patterns links them to the relevant sections in Table 2. The relevant suffixes (which we discuss below) are in bold; the bracketed suffix is the infinitival ending.

Starting with Pattern-2 roots, we can see that they have different suffixes for the inchoative and the causative, with the causative -t replacing the inchoative -n.^[11] Pattern 3 roots have a causative suffix -tir that stacks on top of the inchoative -leş. Pattern 4 roots have a zero-marked inchoative and the causative suffix -t. The particular interest of Turkish resides in the evidence it provides for Pattern 3, which is one that Czech (to be discussed in detail in the next section) does not have. Pattern 3 is also interesting in that we consider it to be one that faithfully reflects the underlying meaning composition, with the inchoative derived from the adjective, and the causative further derived from the inchoative.

6 Three inch/caus scenarios in Czech

We start by describing causative verbs in Czech, because the picture here is rather simple: there is a single suffix -i, which derives causative verbs. We turn to inchoatives in Section 6.2.

6.2 The inchoative

Having discussed the formation of causatives, we turn to their inchoative counterparts. Here, the situation is more complex, in that at least four different classes may be distinguished. In this section, we focus on three classes, which instantiate the predicted patterns in Table 2. We repeat the table here as Table 4, but with the three Czech verb classes added.

Table 4

Three verb classes in Czech.

The Czech Class I is characterised by the fact that both the causative and the inchoative are marked by the same suffix (namely -i-), which is analogous to the English wid-en class. An example is the verb derived from the adjective levn-ý ‘cheap’. In (29a), we give a causative sentence based on this verb, and in (29b) we present the inchoative counterpart. The verb is the same in both sentences.

(29)

a.

Škoda	Auto	z-	levn	-i	-la	své	dva	hlavní	modely.
Škoda	auto	pfx-	cheap	-caus	-pst	its	two	main	models
‘Škoda Auto has made its two main models cheaper.’

b.

Vodka	z-	levn	-i	-la.
vodka	pfx-	cheap	-inch	-pst
‘Vodka got cheaper.’

Table 5 lists the roots in our database showing this inchoative-causative syncretism. This is a relatively minor class and we could therefore only illustrate it using a prefixed verb in (29).

Table 5

Class I verbs.

gloss	adj	inch/caus
expensive	drah-ý	z-draž-i-l
cheap	levn-ý	z-levn-i-l
firm	pevn-ý	z-pevn-i-l
slow	pomal-ý	z-pomal-i-l
fast	rychl-ý	z-rychl-i-l
calm	mírn-ý	z-mírn-i-l
intensive	intenzivn-í	z-intenzívn-i-l

The analysis of this class is analogous to English wide. The lexical entry of the root is of the smallest size (AP), see (30a). The suffix -i is then specified for the heads init and proc as indicated in (30b). The (partial) derivation of the causative verb is shown in (31).

Class II differs from Class I in that it has a dedicated suffix -ě for the inchoative. A case in point is the adjective hutn-ý ‘dense,’ illustrated in (32). Specifically, (32a) contains the causative verb with -i. Interestingly, the inchoative in (32b) no longer has -i, but -ě:

(32)

a.

Půdu	jsem	hutn	-i	-l	dostatečně.
soil.acc	aux.1.sg	dense	-make	-pst	sufficiently
‘I made the soil sufficiently dense.’

b.

Zvuk	hutn	-ě	-l	a	koncert	ostře	gradoval.
sound	dense	-become	-pst	and	concert	sharply	grew.in.intensity
‘The sound was getting denser and the concert was growing in intensity.’

Table 6 lists all the roots that show the -ě suffix in the inchoative.

We analyse these roots as being of the same size as the Class I roots, i. e. they can realise an AP, see (33). As a result, they need the suffix -i to spell out the verbal heads proc and init, see (34).

Different from the Class I verbs, however, these verbs make use of a specialised inchoative marker spelling out just proc. We show the lexical entry of this marker in (35).

In the inchoative structure (see (36)), the marker -ě wins in competition against -i. The competition is regulated by the Elsewhere Principle, which favours the more specific marker over the more general one. Since -ě only applies in inchoative environments, and -i both in inchoative and causative ones, ě is more specific and wins the competition. The representation for the adjective-inchoative-causative triplet for the Class II roots is therefore as shown in section 2 of Table 2.

Table 6

Class II verbs.

gloss	adj	inch	caus
coarse	drsn-ý	z-drsn-ě-l	z-drsn-i-l
clear	jasn-ý	z-jasn-ě-l	z-jasn-i-l
smooth	jemn-ý	z-jemn-ě-l	z-jemn-i-l
compact	hutn-ý	z-hutn-ě-l	z-hutn-i-l
liquid	kapaln-ý	z-kapaln-ě-l	z-kapaln-i-l
calm	klidn-ý	z-klidn-ě-l	z-klidn-i-l
beautiful	krásn-ý	z-krásn-ě-l	z-krásn-i-l
cultural	kulturn-í	z-kulturn-ě-l	z-kulturn-i-l
mighty	mohutn-ý	z-mohutn-ě-l	z-mohutn-i-l
dead	mrtv-ý	z-mrtv-ě-l	z-mrtv-i-l
national	národn-í	z-národn-ě-l	z-národn-i-l
nervous	nervózn-í	z-nervózn-ě-l	z-nervózn-i-l
affectionate	něžn-ý	z-něžn-ě-l	z-něžn-i-l
ugly	oškliv-ý	z-oškliv-ě-l	z-oškliv-i-l
transparent	průhledn-ý	z-průhledn-ě-l	z-průhledn-i-l
exact	přesn-ý	z-přesn-ě-l	z-přesn-i-l
strict	přísn-ý	z-přísn-ě-l	z-přísn-i-l
dark	temn-ý	z-temn-ě-l	z-temn-i-l
immobile	nehybn-ý	z-nehybn-ě-l	z-nehybn-i-l
unsure	nejist-ý	z-nejist-ě-l	z-nejist-i-l
uncalm	neklidn-ý	z-neklidn-ě-l	z-neklidn-i-l

Finally, the Czech Class III is illustrated in (37) by verbs based on the adjective tich-ý ‘silent.’ This class is characterised by the fact that the past participle has no marker for the inchoative at all, see (37b).

(37)

a.

Ššš,	tiš	-i	-l	mě	otec.
psst	silent	cause	-pst	me.acc	father.nom
‘Psst, my father was making me silent.’

b.

Posměšný	křik	skřetů	tich	-Ø	-l.
mocking	shouting	goblins.gen	silent	-become	-pst
‘The goblins’ mocking shouting was getting silent.’

Table 7 lists the roots that have an unmarked inchoative.^[14]

Table 7

Class III verbs.

gloss	adj	inch	caus
deaf	hluch-ý	o-hluch-l	o-hluš-i-l
brown	hněd-ý	za-hněd-l	za-hněd-i-l
weak	chab-ý	o-chab-l	o-chab-i-l
lame	chrom-ý	z-chrom-l	z-chrom-i-l
blind	slep-ý	o-slep-l	o-slep-i-l
silent	tich-ý	z-tich-l	z-tiš-i-l
dark	tmav-ý	z-tmav-l	z-tmav-i-l
bitter	trpk-ý	z-trpk-l	z-trpč-i-l
tough	tuh-ý	z-tuh-l	z-tuž-i-l
alive	živ-ý	o-živ-l	o-živ-i-l

Given the zero marking in the inchoative, we analyze the roots in this class as capable of spelling out ProcP. The table (38) shows that roots of this size can realise both an adjective and an inchoative structure without any additional marker, see lines (38a) and (38b). The init projection of the causative needs an additional morpheme, see line (38c′).

(38)

(39)

However, recall that the causative marker -i can potentially realise both proc and init features. This offers another possibility of lexicalising the causative structure, namely as (38c″).

We shall now argue that only the analysis in (38c″) is compatible with insertion based on the Superset Principle. Our starting point is precisely the observation that the causative suffix in Class III verbs (namely -i) is identical to the one in Classes I and II. This means that the same lexical item is involved, namely the one given in (30b) above, and repeated in (39). Such an entry cannot spell out a remnant initP that only contains Init but not Proc. Such a constituent is in (40), but it cannot be spelled out by -i, because the InitP is not a subtree in the lexical entry (39). This makes the analysis in (38c′) impossible.

Since the Init feature fails to be lexicalised in a derivation like (40), the causative must be derived in such a manner that the root does not reach its full lexicalisation potential, stopping at AP, just as it does in the simple adjective. This is shown in (41). The suffix -i then lexicalises the same constituent initP as with the Class I and Class II roots (see (31) above).

Technically, this derivation involves backtracking, the details of which we do not address here for lack of space. We refer the reader interested in the details of backtracking to Vanden Wyngaerd et al. (2020). What we do want to point out, however, is that there is some interesting empirical evidence for the analysis in (41). Note first that if (41) is correct, the root spells out the same features when used in the causative (namely A) as when used as an adjective. This is different from the inchoative, where it spells out a bigger constituent, A+proc.

This has consequences for root suppletion. Following Caha et al. (2019), we assume that suppletive roots are differentiated by the number of features they realize. If that is so, we expect suppletive adjectival roots to differentiate between the inchoative environment on the one hand (the root spells out AP+proc) and, on the other hand, the adjective and the causative environment (the root spells out AP). Interestingly, there is a set of roots in Czech with exactly this type of root distribution, see Table 8.^[15]

Table 8

Adjectives with mildly suppletive forms in the inchoative.

gloss	adj	inch	caus
dense	hust-ý	z-houst-l	z-hust-i-l
dry	such-ý	u-sch-l	u-suš-i-l
young	mlad-ý	o-mlád-l	o-mlad-i-l
weak	slab-ý	ze-sláb-l	ze-slab-i-l
golden	zlat-ý	zlát-l	zlat-i-l
yellow	žlut-ý	za-žlout-l	za-žlut-i-l

Summarising, Czech provides two reasons for analysing zero marking as an instance of phrasal spellout. The first reason is that Czech fills two gaps in the predicted typology of marking. Most notably, it features a set of roots, which are of ProcP size, with an overt causative, but a zero inchoative (Class III).

The second reason has to do with the pattern of root suppletion exemplified in Table 8. The pattern shows that the suppletive root is found where the regular marking is absent, much as in the bad-worse case. Specifically, what we see is that when proc and init are spelled out by suffixes, we get no suppletion (The final column in Table 8). Suppletion only arises when the root spells out proc. Such roots, then, clearly reveal that zero marking does not come for free, and the root must do a part of the job.

7 The inverse pattern: anticausatives

As we have already noted, there is also a pattern where the inchoative is formed on the basis of the causative through the addition of a marker, often a reflexive.

Haspelmath calls this the anticausative pattern. This type of pattern is also found in Czech and it instantiates the fourth class of verbs, which was left undiscussed in Section 6.2. In (43), we give an example of a causative derived from the adjective tup-ý ‘blunt.’ In order to express the inchoative reading (‘get blunt’), Czech must add a reflexive marker se to the causative, see (43b). There is no non-reflexive form that would be able to express this meaning.

This pattern is apparently problematic in that a structure with fewer features (the inchoative) requires more markers than the semantically richer structure (the causative). In this section, we highlight two possible solutions. The choice between them is left for future research.

The first possible line of attack would capitalize on the fact that this type of anticausative marking involves a marker of reflexivity. Koontz-Garboden (2009: 80) observes that also cross-linguistically, reflexivisation and anticausativisation ‘seem almost always to be marked in a morphologically identical fashion’. Following Chierchia (2004), Koontz-Garboden (2009), Beavers and Koontz-Garboden (2011), one could then analyse the reflexive-marked verb, like prodloužil se ‘lengthened’ in (43b), as a reflexivized version of the causative verb. We represent this schematically as in (44).

This is different from a true inchoative verb, which we take to have a representation as in (45):

Under this analysis, anticausatives like the one in (43b) are structurally different from English-type inchoatives like widen, in the sense that they are based on the causative structure. The reflexive marker se is, then, not a derivational affix on the verb, but a particular way in which the external or internal argument is realised. Koontz-Garboden (2009) supports this by pointing to potential meaning differences between anticausatives and true inchoatives. Specifically, the reflexivisation operation yields a predicate that is true if the single argument is both the Effector and the Theme. That is, the single argument is not simply undergoing some change (as a Theme), but it is at the same time also responsible for its own undergoing of the change. A consequence of this is that for anticausatives, the Causational Entailment (10) is predicted to not always be valid, as schematically shown in (46).

The example (47) (Koontz-Garboden 2009: 117) may clarify this. Spanish has both inchoative verbs like empeorar ‘worsen’, and anticausative ones like romperse ‘to break’ (with the reflexive se). With the inchoative, the Causational Entailment is valid, as shown by the deviance of (47a). But with the anticausative, speakers accept sentences like (47b), suggesting that the Causational Entailment is not always valid.

However, there has been a debate as to whether this kind of approach is correct (see Schäfer and Vivanco 2015).^[16] Even if it were, the question remains how it can be extended to languages where the anticausative marker differs from the reflexive marker. This is notably the case in such languages as Turkish (Key 2013), Hungarian (Márkus 2015), and Korean (Jeong 2018). Let us therefore sketch here an alternative approach to the anticausative conundrum, explored in work by Márkus (2015).

The main idea of the alternative approach is that causative and inchoative structures do not live in a vacuum, so to speak. In syntax, argument structure projections are always embedded under aspectual and temporal projections, as shown in a simplified form in (48).

Suppose now that a language has a lexical entry like the one in (49). The lexically stored tree of this hypothetical entry does not only contain the argument structure projections, but also the aspectual projection Asp. This lexical entry can spell out all the projections in (48), see (50). However, the entry (49) cannot be used to spell out all the projections of the inchoative structure, as shown in (51). The reason is that the AspP given in (51) is not contained (as a constituent) inside the lexical entry (49). The lexical entry does contain the procP, so that ProcP can successfully lexicalise, as indicated by the circle in (51). But this leaves the Asp head without lexicalisation, which is marked by the grey circle in (51).

As a result, an additional morpheme must be used to spell out Asp, and the inchoative ends up needing more morphemes than the causative. The reason for the extra morpheme is that the missing init head prevents constituent matching between the lexical tree of (49) and the syntactic structure.

The consequence of this proposal is that the reflexive is not associated to an argument-structure projection, but to an aspectual layer higher up in the structure. A piece of data suggesting that this may be correct is that in nominalisations (which presumably have less structure than verbs), the inchoative form loses the reflexive marker seen in (43b). We show this in (52), where the nominalisation in (52a) is derived from a causative verb (as the translation suggests), while (52b) is inchoative. What is remarkable is that including the reflexive marker in (52b) leads to ungrammaticality.^[17]

It is worth pointing out that the corpus does contain the string tupení se ‘blunting refl,’ but this is attested only in a reflexive/reciprocal meaning, not in the anticausative one. While we find this evidence suggestive, we shall not elaborate on any detail here, merely noting that the anticausative pattern of marking may receive an explanation even under the view (argued for here) that the structure of the causative always contains the structure of the inchoative.

8 Conclusion

The main idea of this paper is that derivational ‘zero affixes’ do not exist, and that conversion arises as an effect of cumulative exponence, where the root realizes the relevant meaning ingredients. We have implemented this idea in Nanosyntax, a framework with phrasal lexicalisation with matching governed by the Superset Principle. In addition, we adopted a relatively widespread idea that causatives contain inchoatives, and that inchoatives contain adjectives. Combining the Superset Principle and the two independent ideas allowed us to account for the following set of facts. (i) Not all adjectival roots function as verbs, and which roots may do so is largely unpredictable. We achieved this by relying on arbitrary lexical storage: only roots that spell out Proc and/or Init function as verbs. (ii) Some zero-derived verbs are ambiguous between the causative and the inchoative reading. We encoded this by saying that these verbs spell out both Proc and Init (as in English). (iii) Czech zero-derived verbs are only inchoative (but not causative). This is because they only spell out Proc, but not Init. (iv) Zero-derived verbs in Czech may be (mildly) suppletive, while non-zero-derived causatives are not. This is because the adjectival root in zero-derived inchoatives pronounces the Proc component, which may be reflected via suppletion. (v) Finally, our assumptions allowed us to capture the different patterns of the inchoative/causative alternation that are known from the typological literature, including the superficially problematic anticausative pattern.