Understanding Manner Modification from a Cross-Dependency Perspective

Abstract

This article aims to increase our understanding of the syntax of manner modification by examining it from the perspective of the syntax of anaphoric dependencies. It is proposed that the two grammatical dependencies share certain abstract formal properties and are governed by the same type of principle governing the computational system of human language. Building on the so-called IDI constraint (Inability to Distinguish Indistinguishables), it is proposed that the Computational system of Human Language (C_HL) is unable to distinguish two predicates—for example, a verbal predicate and an adjectival one—if they are in a local domain. Specifically, an adjectival predicate (e.g., quick) cannot merge directly with a verbal predicate (e.g., walk). The C_HL can only deal with two predicates if their linguistic environment allows them to be distinguished as different occurrences. This formal distinctness can be achieved by means of various licensing strategies, including so-called protection, which is a formal strategy that turns the modifier (e.g., quick) into a more complex syntactic object (e.g., quick+-ly). It is shown that the various morpho-syntactic implementations of the protection strategy are quite similar for Rreflexivization and for Manner Modification: (grammaticalized) body nouns, doubling pronouns, and adpositional material can be used for making the dependent element (i.e., the anaphor; the manner modifier) more complex. It is further proposed that superficially bare manner modifiers (e.g., fast) are actually complex syntactic objects, where the complexity comes from the (hidden) presence of a silent element.

1. Introduction

A fundamental notion of the theory of grammar is the dependency relation. Most grammatical relations are dependencies of some sort between an antecedent α and a dependent element δ. In a movement dependency like (1a), for example, α (John) is the moved phrase, and δ (John) is the trace/copy. In an agreement dependency like (1b), α (John) is the agreement controller, and δ (speaks) is the agreeing target. Finally, in an anaphoric dependency like (1c), α (John) is a nominal antecedent, and δ (himself) is a reflexive pronoun.

(1)	a.	John was invited John by Sue.
	b.	John3P.SG speaks3P.SG three languages.
	c.	John saw himself in the mirror.

An important step in our thinking about these superficially disparate dependencies was the idea that we might gain a deeper understanding of their nature by searching for their commonalities rather than focusing on their differences. This cross-dependency approach led to the identification of various shared abstract properties (Chomsky, 1973, 1976; Koster, 1987). It was noted, for example, that so-called A-movement dependencies, which are attested in passive constructions like (1a) and so-called subject-to-subject raising constructions (e.g., John seems [John to invite Sue]), share certain abstract properties with anaphoric dependencies between an antecedent and a reflexive pronoun. Specifically, and as exemplified in (2)–(4), just as a noun phrase can be moved only locally, as shown in (2a) and (3a), a reflexive pronoun can be interpretively connected only to a local antecedent, as in (2b) and (3b). If the antecedent and the dependent are in a non-local relationship, we get an ill-formed sentence; see (4).

(2)	a.	Mary was recognized Mary.
	b.	Mary recognized herself.
(3)	a.	Mary was believed [Mary to be the winner].
	b.	Mary believed [herself to be the winner].
(4)	a.	*Mary was believed that [I recognized Mary].
	b.	*Mary believed that [I recognized herself].

This quest for hidden abstract properties was applied not only across different types of grammatical dependencies—e.g., A-movement dependencies and anaphoric dependencies—but also across dependencies that belonged to a single class, such as wh-movement (i.e., A-bar movement) dependencies. As shown in Chomsky’s (1977) On Wh-movement, the wh-question construction and the relative clause construction, which are superficially quite different, share a set of abstract properties that constitute the signature of a deeper process that underlies the two constructions, namely wh-movement of a phrase to the left edge of the clause (i.e., [Spec,CP]).

Our increased insight into the abstract properties of well-known grammatical dependencies can help us gain a better understanding of dependencies that have had a less prominent position on our research agenda so far. Without wanting to disparage any previous research, I believe we may say that the modifier dependency—that is, the dependency between a modifier and a modifiee—belongs to the class of grammatical dependencies that deserve more in-depth investigation.1 In this article, I hope to contribute to our understanding of this grammatical dependency by studying it from a cross-dependency perspective. Specifically, I reflect on the grammatical nature of the modification dependency by examining it through an “anaphoric lens” (see (1c)). Furthermore, I restrict my investigation to one class of adverbial expressions, namely manner-adverbials. An example of manner-modification is given in (5):

(5)	John [modifiee cried] [modifier softly].

Clearly, the modification dependency in (5) and the anaphoric dependency in (1c) are disparate at the surface. The former involves a dependency between two predicates, namely a verbal predicate (cried) and an adjectival one (soft(ly)); the latter, on the other hand, involves a dependency between two arguments, namely a subject-argument (John) and an object-argument (himself). There is one property, though, which they share. It concerns the form of the dependent element: it can’t be morphologically bare but must be morphologically “dressed” in English; that is, we need additional marking on the pronoun (i.e., him+self) for getting a reflexivity reading and on the adjective (i.e., soft+ly) for getting a manner modificational reading.2 In other words, the two dependencies require a complex form of the dependent element.

(6)	a.	John recognized him*(self).
	b.	John cried soft*(ly).

This extra marking on the manner-adverbial expression raises the question as to why the (English) grammar (i.e., I-language; Chomsky, 1986b) requires it (modulo certain bare adverbs such as fast, far, and well). Other questions that concern this adverbial marking are the following: What is the inner structure of a complex adverbial form like softly? What other adverbial marking strategies do we find cross-linguistically? Why do certain languages lack adverbial marking on the adjective? The Dutch equivalent of (6b), for example, does not show any morphological marking on the manner-adverbial: Marie huilde zacht (Mary cried soft, ‘Mary cried softly’). And, finally, to what extent does the manner adverbial form (inner syntax) matter for the distributional behavior of the manner-adverbial expression (external syntax)? Notice, for example, that while quickly can occur on either side of the verb, as in (7b), fast cannot, as exemplified in (7b). The latter must occur post-verbally.

(7)	a.	John will <quickly> leave the room <quickly>.
	b.	John will <*fast> leave the room <fast>.

My aim is twofold: first of all, I hope to show that one can get to grips with a relatively unknown grammatical dependency, namely manner modification, by studying it from a cross-dependency perspective, where the basis of comparison is a more familiar dependency, namely the anaphoric dependency. Secondly, and also at a more explanatory level, I aim to show that, even though the manner-adverbial dependencies look quite complex prima facie, both as regards their inner form and as regards their outer behavior, their properties can be accounted for by the interplay between general constraints on the computational system of human language—for example, the constraint that a predicate P1 cannot merge directly with a predicate P2—and morphosyntactic strategies that languages use in order to avoid a violation of these general constraints. Let me finally add that many of the arguments brought in favor of my analysis of manner-adverbial dependencies are based on Dutch and English. When appropriate, other languages are included as well.

This article is organized as follows. Section 2 presents a brief discussion of four core properties of anaphoric dependencies. Section 3 argues that these abstract properties are also present in Manner-Modification (i.e., MM-) dependencies. Section 4 continues with a comparative perspective on anaphoric dependencies and MM-dependencies by focusing on the formal manifestation of the dependent element (for example, the English composite forms him-self and quick-ly). Section 5 addresses the question as to why languages prefer complex anaphors (e.g., English himself) over simplex ones (e.g., Frisian him). It presents Reuland’s (2011, 2018) approach to this question in terms of his so-called IDI constraint (Inability to Distinguish Indistinguishables), which excludes identical linguistic elements in a local domain. This section also discusses the licensing strategies that languages use to make identical elements formally distinct. It further presents a brief historical overview of constraints that are similar in spirit to Reuland’s IDI constraint with the aim of showing that avoidance of identical elements in a local domain is a property that holds for different types of grammatical dependencies (e.g., anaphoric dependencies, movement dependencies, complementation). In Section 6, it is proposed that (manner-)modification dependencies are governed by a distinctness requirement as well. A syntactic configuration in which a predicative phrase XP is directly combined (i.e., Merged) with another predicative phrase YP is illegitimate. In Section 7, various licensing (more specifically, protection-) strategies that languages use for making the dependent (i.e., the modifier) formally distinct from the modifiee are discussed, specifically pronominal protection (Section 7.1), protection by doubling (Section 7.2), and adpositional protection (Section 7.3). The protecting element (e.g., English -ly) is analyzed as an adjunct that is combined with the adjectival predicate, this way showing resemblance with self, which has been analyzed as an adjunct attached to him in the English reflexive pronoun himself. Section 8 discusses a Dutch manner-adverbial pattern that encodes a possessive relationship: op z’n Piets (at his Piet’s, ‘in a Piet-like way’). For this manner-adverbial expression, a parallel is drawn with the reflexive pronoun yourself/myself, which also encodes a possessive relationship. Section 9 discusses bare adverbial APs in Dutch and English (e.g., fast). It is proposed that these superficially bare forms have a hidden complex structure and consequently do not violate IDI. In Section 10, I discuss two other types of manner-adverbial licensing strategies that are attested in natural languages, namely Separation and Bundling. Also, for these strategies, a parallel can be drawn between anaphoric dependencies and MM-dependencies. Section 11 concludes this article.

2. Anaphoric Dependencies and Their Properties

Anaphoric dependencies like the one in (8) from Dutch constitute a core grammatical relationship in human language:

(8)	Jan	zag	zichzelf	in	de	spiegel.
	Jan	saw	himself	in	the	mirror.
	‘Jan saw himself in the mirror.’

The reflexive pronoun zichzelf lacks the capacity for independent reference. Therefore, it depends on another expression, namely the antecedent Jan, for its interpretation.

As studies on anaphoric dependencies have shown, these dependencies display several core grammatical properties (Chomsky, 1981; Everaert, 1986; Koster, 1987; Reinhart & Reuland, 1993; Reuland, 2011; Safir, 2004). The first property is the obligatoriness of the antecedent; that is, the dependent element zichzelf must have an antecedent. If such an antecedent is absent, the sentence is ill-formed:

(9)	*Ik	zag	zichzelf	in	de	spiegel.
	I	saw	himself	in	the	mirror
	*’I saw himself in the mirror.’

The second property is the uniqueness of the antecedent; that is, we can only have a single antecedent for an anaphor. Thus, split antecedents are impossible (Giorgi, 1984; Dimitriadis, 2000; Anagnostopoulou & Everaert, 2013). This is exemplified in (10). The reflexive pronoun zich cannot be associated simultaneously with Jan, the subject of the main clause, and Marie, the subject of the infinitival clause. The sentence is well-formed when the plural personal pronoun hen is used.3

(10)	Jan	zag	Marie	de	bal	tussen	*zich/OKhen	inleggen.
	Jan	saw	Marie	the	ball	between	REFL/them	put
	‘Jan saw Marie put the ball between them.’

This property of uniqueness applies only to antecedents and not to anaphors. Thus, a nominal antecedent can be connected to more than one anaphor:

(11)	Jan	sprak	tegen	zichzelf	over	zichzelf.
	Jan	spoke	to	himself	about	himself
	‘Jan spoke to himself about himself.’

The third property that characterizes anaphoric dependencies is the syntactic prominence of the antecedent with respect to the reflexive pronoun. The antecedent must be in a hierarchically more prominent position than the anaphor. In more technical terms, the antecedent must c-command the anaphor (Reinhart, 1976, 1983).4 In (12a), the subject noun phrase Jan can act as an antecedent for the object zichzelf since it c-commands the latter. In (12b), in contrast, Jan cannot act as an antecedent for zichzelf. Being embedded within the subject noun phrase, it does not c-command the direct object zichzelf. The only noun phrase that can function as antecedent for zichzelf is the subject noun phrase de buurman van Jan.

(12)	a.	Jan	haat	zichzelf.
		Jan	hates	himself
		‘Jan hates himself.’
	b.	*[De	buurman	van	Jan]	haat	zichzelf
		the	neighbor	of	Jan	hates	himself
		Impossible reading: ‘Jan’s neighbor hates him (i.e., Jan).’
		Possible reading: ‘Jan’s neighbor hates himself (i.e., Jan’s neighbor).’

The fourth and final property that characterizes anaphoric dependencies is locality. In generative grammar, the standard form of locality for anaphors is given by principle A of the binding theory (Chomsky, 1981): an anaphor must be bound in its governing category. Traditionally, this locality domain is defined as the minimal X^max containing the governor of an anaphor and a subject (subject or Agr) accessible to the anaphor. Consider now the dependencies in (13):

(13)	a.	Marie	mopperde	op	zichzelf.
		Marie	grumbled	at	herself
		‘Marie grumbled at herself.’
	b.	*Jan	hoorde	[dat	Marie	op	zichzelf	mopperde].
		Jan	heard	that	Marie	at	himself	grumbled
		*’Jan heard that Marie grumbled at himself.’
	c.	*Jan	hoorde	[Marie	op	zichzelf	mopperen].
		Jan	heard	Marie	at	himself	grumble
		*’Jan heard Marie grumble at himself.’

In (13a), the main clause functions as the governing category for zichzelf. It is the domain that contains a governor (the adposition op) and a subject (Marie). The latter functions as the antecedent of zichzelf. In (13b) and (13c), the embedded finite clause and the embedded infinitival clause function, respectively, as governing categories, with op as governor and Marie as subject. Since Jan is located outside of the embedded clause, it cannot function as an antecedent of zichzelf.

Importantly, the four properties discussed above hold for anaphoric dependencies involving an anaphor, which belongs to the class of reflexives (or reciprocals). They do not hold for personal pronouns such as hem ‘him’ or its weak form ’m.

(14)	a.	Ik	zag	hem/’m	in	de	spiegel.
		I	saw	him	in	the	mirror
		‘I saw him in the mirror.’							(compare (9))
	b.	Jan	zag	Marie	de	bal	tussen	hen	inleggen.
		Jan	saw	Marie	the	ball	between	them	put
		‘Jan saw M. put the ball between them.’							(compare (10))
	c.	[De	buurman	van	Jan]	haat	hem/’m.
		the	neighbor	of	Jan	hates	him
		‘Jan’s neighbor hates him.’							(compare (12b))
	d.	Jan	hoorde	[Marie	op	hem/’m	mopperen].
		Jan	heard	Marie	at	him	grumble
		‘Jan heard Marie grumble at him.’							(compare (13c))

The well-formedness of (14a) shows that the personal pronoun hem/’m does not require the presence of an antecedent within the sentence. (14b) illustrates that the pronoun hen can have a split antecedent (Jan, Marie). In (14c), it is shown that hem/’m can take a non-c-commanding noun phrase as its antecedent. In (14d), finally, we see that hem/’m cannot have a c-commanding antecedent within its governing category, that is, the embedded clause. However, the main clause subject Jan, which is located outside of this governing category, can function as its antecedent.

Having summarized four core properties of anaphoric dependencies, I explore in Section 3 to what extent this set of properties is also found in Manner-Modification dependencies (MM-dependencies). By taking this cross-dependency perspective, I hope to gain a better understanding of the abstract nature of this type of syntactic dependency.

3. MM-Dependencies and Their Properties

Let’s start our investigation into the nature of MM-dependencies with the Dutch example in (15):5

(15)	Jan	zag	dat	de	kinderen	[(erg)	aandachtig]	luisterden.
	Jan	saw	that	the	children	very	attentively	listened
	‘Jan saw that the children listened (very) attentively.’

The adjectival phrase (erg) aandachtig acts as an MM of the verb luisterden. It indicates that the children listened in a (very) attentive way. MM’s phrasal status is clear from the fact that MM can have a complex structure that consists of a degree word (erg) and an adjective (aandachtig). If we take modifiers to be phrases that are adjoined to the modified phrase (see Chomsky, 1986a; Haider, 2000; Ernst, 2002), the sequence (erg) aandachtig luisterden can be assigned the representation in (16); the label XAP stands for Extended Adjectival Projection (Corver, 1997a, 1997b).6

(16)

dat

de

kinderen

[VP

[XAP

(erg)

aandachtig]

[VP luisterden]].

The MM (erg) aandachtig functions as a dependent element. It can function as a modifier only in combination with the VP (luisterden), which—stretching the notion a little and drawing a parallel with the anaphoric dependencies in Section 2—I qualify as ‘antecedent’, that is, the phrase on which the modifier is dependent for its modifying function. MM’s dependent status is firstly suggested by its optionality. If you leave it out, the sentence is still well-formed: …dat de kinderen luisterden. Notice that if aandachtig had the status of antecedent in (16), one would expect its absence to yield an ill-formed sentence since the dependent element (say, VP) in the clause would be “looking for” it.

Another reason for treating MMs like aandachtig as a dependent element comes from the observation that it cannot combine with all types of verbs. Specifically, while they can combine with verbs denoting events, they typically do not combine with verbs denoting states, which, according to Dik (1975), represent a non-dynamic and non-controlled state of affairs.7

(17)	a.	…	dat	de	kinderen	(*aandachtig)	rekenles	hadden.
		…	that	the	children	attentively	math-lessons	had
		‘… that the children (*attentively) had math-lessons.’
	b.	…	dat	de	kinderen	(*aandachtig)	40	kilo	wogen.
		…	that	the	children	attentively	40	kilo	weighed
		‘… that the children (*attentively) weighed 40 kilos.’

Let us turn now to the second property: the uniqueness of the antecedent; that is, we can have only one VP-antecedent for a MM. In other words, a single MM cannot be associated simultaneously with two VPs (i.e., modifiees).8 Before showing this, consider first the examples in (18), which represent embedded finite clauses in Dutch:

(18)	a.	…	dat	ik	goed	zag	dat	de	kinderen	luisterden.
		…	that	I	well	saw	that	the	children	listened
		‘… that I saw well that the children listened.’
	b.	…	dat	ik	zag	dat	de	kinderen	goed	luisterden.
		…	that	I	saw	that	the	children	well	listened
		‘… that I saw that the children listened well.’

In (18a), MM goed modifies the matrix verb (zag), while in (18b), it modifies the embedded verb (luisterden). With the knowledge that goed is an appropriate modifier for both zien ‘to see’ and luisteren ‘to listen’, consider next the Exceptional Case Marking constructions in (19), where the matrix clause and the embedded clause are “entangled” as a result of the (restructuring) process of Verb Raising (Evers, 1975; Wurmbrand, 2001).

(19)	a.	?…	dat	ik	de	kinderen	goed1	goed2		zag1	luisteren2.
		…	that	I	the	children	well	well		saw	listen
		‘… that I saw well that the children were listening well.’
	b.	*…	dat	ik	de	kinderen	goed	zag	luisteren.
		…	that	I	the	children	well	saw	listen
		Intended reading: ‘… that I saw well that the children were listening well.’

As (19a) shows, it is possible to have two instances of the modifier goed. For getting the right reading, it helps to accentuate goed₁ and add a brief pause right after it, as in: … dat ik de kinderen goed <pause> goed zag luisteren. As indicated, goed₁ acts as the modifier of zag, and goed₂ as the modifier of luisteren. Importantly, and as shown in (19b), it is impossible for a single MM goed to function as a modifier of both the finite verb zag and the infinitival verb luisteren. When goed zag luisteren in (19b) is pronounced as a single prosodic unit, I prefer the reading in which goed is interpreted as a modifier of luisteren; thus, ‘to listen well’. When pronounced as goed <pause> zag luisteren, I prefer the reading in which goed acts as a modifier of zag; thus, ‘to see well’.

Recall from the discussion of anaphoric dependencies in Section 2 that the property of uniqueness applies only to antecedents and not to anaphors. Thus, a single antecedent may be connected to more than one anaphor (see (11)). The same situation holds for MMs: a single VP (the antecedent) can be connected to more than one MM (the dependent).9

(20)		Onze	nieuwe	sorteermachine	kan	de	brieven	[snel	[nauwkeurig	[selecteren]]].
		our	new	sorting machine	can	the	letters	quickly	accurately	select
		‘Our new sorting machine can quickly select the letters accurately.’

In this example, the MM nauwkeurig acts as a modifier of selecteren, and the MM snel as a modifier of nauwkeurig selecteren. In other words, the MM snel has scope over the MM nauwkeurig. This difference in scope of the two MMs is clear, especially in contexts in which the first MM is used contrastively, such as in (21); the small capitals indicate a contrastive accent.

(21)

Ik

heb

liever

een

machine

die

snel

nauwkeurig

sorteert

I

have

preferably

a

machine

that

quickly

accurately

sorts

dan

een

machine

die

[langzaam

[nauwkeurig

[sorteert]]].

than

a

machine

that

slowly

accurately

sorts

‘I would rather have a machine that sorts accurately fast than one which does so (=sort accurately) slowly.’

Note in passing that (21) can be reformulated by using a doet dat ‘does so’ paraphrase:10

(22)	Ik	heb	liever	een	machine	die	snel	nauwkeurig	sorteert
	I	have	preferably	a	machine	that	quickly	accurately	sorts
	dan	een	machine	die	dat	langzaam	doet.
	than	a	machine	that	that	slowly	does
	I would rather have a machine that sorts accurately fast than one which does so (=sort accurately) slowly.’

The fact that doet dat can be interpreted as a replacement of nauwkeurig sorteert (to the exclusion of langzaam) shows that the MM langzaam (and also snel) occupies a hierarchically more prominent position. Schematically, [_VP langzaam/snel [_VP nauwkeurig [_VP sorteert]]].

The third property concerns the syntactic prominence of the antecedent with respect to the dependent. As we saw in Section 2, example (12), the antecedent (Jan) must be in a hierarchically prominent position with respect to the anaphor (zichzelf). In more technical terms, the antecedent must c-command the anaphor. Does this c-command relationship between antecedent and dependent extend to MM-dependencies? Under a definition of c-command in terms of sisterhood—that is, a node c-commands its sister and all the daughters (and grand-daughters and great-granddaughters) of its sister—the lower VP-segment of the adjunction structure [_VP MM [_VP V..]] c-commands MM, and conversely. As noted in Chomsky (1986a, p. 9), this mutual c-command relation can be avoided by resorting to the concept exclude.11

According to the fourth property, locality, the dependency between MM (the dependent) and the modified VP (the antecedent) must be local. As shown in (23), MM-dependencies clearly display this property: when MM occurs in the main clause, it can act as a modifier of the main clause VP but not as a modifier of the VP that is part of the embedded clause (see (23a,b)), and vice versa (see (24a,b)):

(23)	a.	Er	werd	uitvoerig	uitgelegd	[waarom	de	patiënt	onderzocht	werd].
		there	was	thoroughly	explained	why	the	patient	examined	was
		‘It was explained thoroughly why the patient had been examined.’
	b.	*Er	werd	uitvoerig	uitgelegd	[waarom	de	patiënt	onderzocht	werd].
		there	was	thoroughly	explained	why	the	patient	examined	was
		Intended meaning: ‘It was explained why the patient had been examined thoroughly.’
(24)	a.	Er	werd	uitgelegd	[waarom	de	patiënt	uitvoerig	onderzocht	werd].
		there	was	explained	why	the	patient	thoroughly	examined	was
		‘It was explained why the patient had been examined thoroughly.’
	b.	*Er	werd	uitgelegd	[waarom	de	patiënt	uitvoerig	onderzocht	werd].
		there	was	explained	why	the	patient	thoroughly	examined	was
		Intended meaning: ‘It was explained thoroughly why the patient had been examined.’

This locality effect is attested not only in bi-clausal structures but also in mono-clausal structures. Consider, for example, the sentence in (25), where the MM goed is a modifier that is embedded within an adjective phrase, namely goed verstaanbaar (well understandably, ‘clearly’). This larger phrase is, in turn, a modifier of the VP headed by uitgesproken. When the sequence goed verstaanbaar uitgesproken is pronounced as a single prosodic unit, the MM goed can be associated only with verstaanbaar, yielding the meaning ‘clearly’, and not with uitgesproken, which would yield the interpretation ‘pronounced correctly’. In short, modification is restricted to a local domain.

(25)	De	leraar	heeft	de	woorden	[goed	verstaanbaar]	uitgesproken.
	the	teacher	has	the	words	well	understandably	pronounced
	‘The teacher pronounced the words clearly.’

In summary, so far, I have studied MM-dependencies from a cross-dependency perspective. It was shown that anaphoric dependencies (Section 2) and MM-dependencies (Section 3) share certain abstract grammatical properties. Although the elements that are involved in these two types of dependencies are different—namely, two argumental noun phrases in anaphoric dependencies and two predicative phrases in modification dependencies—the abstract form (i.e., configurational structure) of the two dependencies is essentially the same. This finding is important for two reasons: first of all, it confirms the idea that linguistic properties of dependency relations are not construction-specific but rather have a more abstract nature (Chomsky, 1981, 1995). Secondly, this structural parallelism invites us to search for a shared explanation (i.e., unification) for the abstract similarities between both types of grammatical dependencies.

In Section 4, I continue my search for (abstract) similarities between the two types of dependencies by having a closer look at the formal manifestation of the dependent element. In Section 5, Section 6, Section 7, Section 8, Section 9 and Section 10, I explore the path towards a unified approach to anaphoric dependencies and manner-modification dependencies.

4. On the Form of the Dependent Element

This section takes a comparative perspective on anaphoric dependencies and MM dependencies by focusing on the formal manifestation of the dependent element. From a purely descriptive perspective, we can make a distinction between simplex (i.e., non-composite) dependents and complex (i.e., composite) dependents.

Two examples of a simplex anaphor are given in (26):12

(26)	a.	Ringo	waste	zich.	(Dutch)
		Ringo	washed	refl
		‘Ringo washed himself.’
	b.	Ringo	waske	him	(Frisian; Reuland, 2018).
		Ringo	washed	him
		‘Ringo washed himself.’

In the Dutch example (26a), we have the simplex anaphor zich, while in the Frisian example (26b), we have the simplex pronoun him. Importantly, Frisian him in (26b) can receive a reflexive interpretation, as indicated by the English translation. It should be noted that it can also be interpreted as a personal pronoun; in that case, him refers to some individual (say, Peter) in the discourse or situational context.

As noted in Reuland (2001, 2011), these simplex forms are cross-linguistically not very common for the expression of reflexivity. Languages prefer complex anaphors, that is, anaphoric expressions that consist of a simplex anaphor or a pronominal together with some additional element, as for example, in English himself (him+self) and Norwegian ham selv and seg selv ‘himself’. Dutch and Frisian also have complex anaphors. In (26a), for example, the complex anaphor zichzelf can be used instead of zich, and in (26b), the complex form himsels is possible as well.

Besides complex anaphors featuring self as an additional element, there are also complex anaphors that have a body part (e.g., head, spirit, soul, bone) as an extension; see Reuland (2011); König et al. (2013); Schladt (2000). An example of such a body-part reflexive is the Basque complex anaphor bere burua ‘his head’ in (27) and the Urhobo complex anaphor oma-re-oyen ‘his body’ in (27b); see Hualde and Ortiz de Urbina (2003) and Volkova (2014), respectively.

(27)	a.	Jonek		[bere	burua]	gorroto	du.	(Basque, isolate)
		John.erg		his	head	hate	aux.tr
		‘John hates himself.’
	b.	Ijoni	mre	[oma-re-oyen].				(Urhobo, Niger-Congo)
		John	see.pst	body-AssocM-him
		‘John saw himself.’

A third way of forming a complex anaphor is by doubling the reflexive-pronominal element. Examples of such reduplicated pronominal patterns are given in (28). Example (28a) illustrates the doubling phenomenon for Tsakhur (North Caucasian; Toldova, 1999) and (28b) for Avar (North Caucasian; see Rudnev, 2017).

(28)	a.	Rasuly-ē	[wudž-ē	wudž]	yaramališ-aɁ-u.	(Tsakhur, North Caucasian)
		Rasul- erg	refl.1-erg	refl.1.abs	wound-1.do-pf
		‘Rasul wounded himself.’
	b.	ʢalic:a	[žinc:ago	žiwgo]	č’wana.	(Avar, North Caucasian)
		Ali.erg	self. erg:emph	self. abs: emph	kill.aor
		‘Ali killed himself.’

Keeping these simple and complex anaphors in mind, let us now turn to the formal patterns that correspond to what are traditionally called ‘manner adverbs.’ Also, for this category, we find cross-linguistically both simplex forms, which are forms without an adverbial marker, and complex forms, which are forms with an additional element that functions as an adverbial marker.13

Example (29) illustrates the use of simplex forms:14

(29)	a.	…dat	Jan	de	zin	vreemd	uitsprak.	(Dutch)
		that	Jan	the	sentence	strangely	pronounced
		‘…that Jan pronounced the sentence strangely.’
	b.	uqa	mahuc		nu-i-a.	(Amele, Papuan, Madang; Roberts, 1987, p. 158)
		3sg	quick/quickly		go-3-tod.past
		‘He went quickly.’
	c.	ŋaia	ja:ri	da:rŭi.	(Gumbainggir (Australian, Pama-Nyungan; Smythe, 1948, p. 59))
		I	go	good
		‘I am doing well.’

A strategy for making manner adverbs complex consists of the addition of a body part noun such as mind/soul and body. The mind/soul strategy is familiar in Romance languages such as French and Italian, in which the element -ment/-mente is added to the adjective. This adverbial marking historically relates to (the ablative form of) the Latin noun mens ‘mind’.

(30)	a.	Il	a	couru	rapidement.		(French)
		he	has	run	quickly
		‘He ran quickly.’
	b.	Il	professore	spiega	tutto	chiaramente.	(Italian; Brinker, 1997, p. 111)
		the	teacher	explain.3sg.pres	everything	clearly
		‘The teacher explains everything clearly.’

The use of body as an adverbial marker is arguably at the basis of adverb formation in English. The adverbial marker -ly, as in quickly, has its origin in lic(e), which means ‘body/corpse’; see, among others, Jespersen (1933/1964) and Nevalainen (1997).

(31)	John left the room quickly.

Besides complex adverb formation by means of a body-part noun, there are also languages that use reduplication as a strategy for adverb formation:

(32)	[yavaş	yavaş]	yürüyorduk.	(Turkish, Turkic, Lewis, 1967)
	slow	slow	we.were.walking
	‘We were walking slowly.’

As is clear from the above discussion, complex anaphor formation and complex MM formation display similarities in their formation strategies. As for the inner structure of complex anaphors, it has been proposed that the additional element—that is, self, the body-part noun or the doubling element of the reduplicative pattern—constitutes the head of a nominal expression or an adjunct that is adjoined to the pronoun. As noted in Faltz (1985), the distinction between these two types of complex reflexives is not always straightforward. Furthermore, the status of reflexives may even vary within a single language. For example, English himself/herself, which features a personal pronoun, arguably instantiates the ‘self = adjunct’ pattern, as exemplified in (33a), while the anaphoric form myself/yourself, which features a possessive pronoun, plausibly instantiates the ‘self = head-pattern (see (33b)).15 Schematically:

(33)	a.	[dp	[dp him]	[np self]]
	b.	[dp my	[d’ D	[np self]]]

Reduplicative patterns like wudž-ē wudž in (28a) could be assigned the structure in (33b), with wudž-ē in [Spec,DP] and wudž as the nominal head of NP. Thus, [_dp wudž-ē_ERG [_d’ D [_NP wudž_ABS]]].16 Since the second pronoun carries the Case that is compatible with the base position of the direct object noun phrase, it seems plausible to analyze this (absolutive) pronoun as the head of the complex pronoun and the ergative pronoun as the doubling element.

Taking a cross-dependency perspective on complex anaphors and complex MMs, we could ask whether the distinction between adjunct-reflexives (33a) and head-reflexives (33b) can be transposed to the manner-adverbial domain. In other words, are there manner-adverbials featuring an adjunction configuration, as in the schematic representation in (34a), and are there manner-adverbials that display more of a Spec-head configuration, as in the schematic structure in (34b)?

(34)	a.	[xp	[xp quick]	[yp -ly]]	(-ly = adjunct)
	b.	[xp quick	[x’ X	[yp -ly]]]	(-ly = head)

In Section 6, I address this question about the inner structure of manner-adverbial expressions. In Section 5, however, I first explore why languages prefer complex anaphors (e.g., English himself) over simplex ones (e.g., Frisian him) and what we can learn from this for the analysis of MM-dependencies. In short, we once more take a cross-dependency perspective.

5. On Protecting the Pronoun in Anaphoric Dependencies

Having shown that complex anaphors and complex MMs use similar formation strategies, let us next address the following question: why do many languages employ special means (i.e., additional material) to express reflexivity (him-self) and manner modification (quick-ly)? Or, to phrase it differently, why do reflexivity and manner modification need to be licensed by adding an extra marker (i.e., self; -ly)? Focusing on (Standard) English, why does English exclude simple pronouns (35a) and bare adjectives (35b) for the expression of reflexivity and modification, respectively?17

(35)	a.	John	saw	him*(self).
	b.	John	walked	to	the	store	quick*(-ly).

In this section, I address this question for the anaphoric dependency in (35a), highlighting insights from Reuland (2001, 2011); see also Everaert and Reuland (2025). In Section 6, I make use of those insights to answer the question of why bare quick in (35b) is ruled out in (35b) but “dressed” quickly is not.

Reflexivization, as exemplified by John saw himself, can be said to occur if, in a structure, an argument of a predicate (e.g., saw) is bound by another argument of that same predicate. In Reinhart (2006, p. 171), this notion of A(rgument)-binding is defined as in (36).

(36)	A-binding: α binds β iff α is the sister of a λ-predicate whose operator binds β.

When we apply this definition to the antecedent-pronoun dependency in (37a), we get the LF representation in (37b). Rendering his as x (rather than some other alphabetic variant) expresses the bound variable construal of his:

(37)	a.	John saw [his sister] in the garden.
	b.	John (λx (x saw [x’s sister]))

One could imagine that the most direct way to express reflexivization would be binding the personal pronoun, as in John saw him; see (35a). Reuland (2011) refers to this pattern as brute force reflexivization. It turns out, however, that languages generally try to avoid this reflexivization pattern and use various morphosyntactic strategies to license (i.e., mark) reflexivity. One may wonder, of course, why languages need this extra marking for the expression of reflexivity.

According to Reuland (2011), the answer to this question must be sought in a property of the computational system of human language (C_HL). This property consists in the fact that C_HL is unable to distinguish identical elements (e.g., identical variables) from each other if they are in a local linguistic domain. It can, therefore, only deal with two identical elements if their linguistic environment allows them to be distinguished as different occurrences. Reuland expresses this property of C_HL in terms of the IDI principle:

(38)	IDI: Inability to distinguish indistinguishables

The impossibility of applying brute force reflexivization to John saw him follows from IDI. Binding of him by John yields an IDI configuration. At the CI Interface, the representation of binding contains two tokens of the variable x, which instantiate one linguistic object. These tokens can’t be distinguished from each other because they are located in a local domain.

(39)	*John (λx (seeθ1, θ2 (x, x)))

As Reuland notes, the IDI configuration can be avoided by keeping the two arguments formally distinct, namely by adding complexity (i.e., an additional morpheme) to one argument (generally, the object argument). This additional morpheme (morph) provides a form of protection for the bound pronoun, i.e., the variable x in [morph x].

(40)	a.	John saw [him [self]].
	b.	dp (λx (seeθ1, θ2 (x, [morph x]))) (e.g., morph = self)

The pronoun can be protected through embedding in a complex NP structure. This extra structure allows for distinguishing the two tokens of a variable at the CI interface. Cross-linguistically, this licensing of x through protection can happen through a self-element (e.g., [him [self]]), a body part noun (e.g., [bere [burua]] in (27a)), or a doubled pronoun (e.g., [wudž-ē [wudž]] in (28a); see Section 4.

Reuland (2018, 2011) further argues that the morph element must be interpreted so as to yield a function of x, mapping it onto an object that can be a proxy for x. The extra element should make some minimal contribution to interpretation; that is, an interpretation that is sufficiently close to x for a reflexive interpretation. As he notes, self and body part nouns have minimal lexical content and are relational nouns that are (near-)reflexive.18 A self is always some individual’s self, and a body is always some individual’s body. That a doubling pronoun, as in (28), makes a minimal contribution to meaning speaks for itself.

Before addressing in Section 6 the question as to why manner modifiers need to be licensed by adding an extra marker, as in quick-ly, I briefly dwell on the idea that indistinguishable elements should not be present in a local domain. Importantly, this property is not restricted to anaphoric dependencies but is also attested in other types of dependency relations, not only in syntax but also in phonology (see Reuland, 2011). For example, Leben’s (1973) Obligatory Contour Principle (OCP) states that two consecutive identical phonological features (e.g., two high tones) are banned in a local underlying sound representation (e.g., the sound representation of a single morpheme). In the domain of syntax, Kayne (1982) formulates a constraint on head-complement dependencies that states that a non-maximal projection of N must not govern a maximal projection of N, this way ruling out patterns such as the [destruction_N [_NP the city]]. In the same spirit, Hoekstra’s (1984, p. 85) Unlike Category Constraint states that no head (e.g., N) can govern a phrase of the same category (e.g., NP). According to Richards’ (2002) Distinctness Constraint, a constraint that applies at the Syntax-PF-interface, two nodes of type α cannot be linearized at the Syntax-PF interface if they are contained in the same Spell-Out domain. The pattern the destruction the city, for example, is ruled out because the D-head of the complement the city and the D-head of the matrix-DP are in the same Spell-Out domain. Another illustration of the impossibility of having two identical elements in a local domain comes from Abels’ (2003) Anti-locality condition on movement dependencies: the moved element and its trace (i.e., copy) can’t co-occur in a local domain. It is impossible, for example, to move a constituent from the complement position of a head X to the specifier position of that same head. As a final illustration, we could mention Chomsky’s (2013, pp. 43–44) constraint on the labeling of syntactic objects (SO) of the type {XP,YP}. As he notes, the label of this SO can’t be determined since minimal search is ambiguous: both the head X of XP and the head Y of YP could, in principle, function as the element that projects its categorial label to SO. In a way, they are indistinguishable for minimal search. According to Chomsky (2013, p. 43), there are two ways in which SO can be labeled: firstly, by modifying SO (e.g., by means of displacement of YP) so that there is only one visible head; secondly, by identifying a grammatical feature shared by X and Y, which can be taken as the label of SO.

It should be clear from this brief historical overview that the spirit of the IDI property “haunts” several domains of syntax: anaphora, complementation, movement, and projection (labeling).19 In the next section, I hypothesize that this property also plays a role in the expression of manner modification.

6. On Protecting the Manner-Adverbial in MM-Dependencies

Why does manner modification in English need to be licensed by adding an adverbial marker on the adjective; see (41a)?20 Or, to put it differently, while borrowing a term from Reuland (2011), why doesn’t English permit “brute force” manner modification, as in (41b)?

(41)	a.	The children listened attentively.
	b.	*The children listened attentive.

In the spirit of Reuland’s IDI constraint and the other organizational principles that block local dependencies between identical elements (see Section 5), I propose that modification dependencies—that is, dependencies between a (manner-)modifier and a modifiee—are governed by a distinctness requirement as well. Specifically, a syntactic configuration in which a predicative phrase XP is directly combined (i.e., Merged) with a predicative phrase YP is illegitimate. This principle governing adverbial design is given in (42), where I take this SO (Syntactic Object) to result from Merge:21

(42)	*[[XPPRED] [YPPRED]].

In the spirit of the protection strategies discussed in Section 4, I propose that the adverbial IDI configuration in (42) can be avoided by keeping the two predicates formally distinct by adding complexity to the dependent predicate (i.e., the modifier). Schematically:

(43)	[[XPPRED] [ZP [YPPRED] morph]].

The morph-element can be a body part noun like French -ment or English -ly. In line with Reuland’s interpretation of the grammatical role of the morph-element (e.g., self) in complex anaphors, I assume that the adverbial morph-element is a noun with minimal lexical content. In Emonds’s (1985) terms, -ment and -ly are grammatical nouns; that is, nouns that lack purely semantic features (i.e., conceptual meaning) and consist only of what he calls cognitive syntactic features, that is, ‘formal features’ in the sense of Chomsky (1995, pp. 348–379). Following the idea that the lexical category “noun” has a complex structure of the type n+Root (see Marantz, 1997; Borer, 2005), I assume that the grammatical nouns -ment and -ly have the following structure, respectively: [_nP n [√ment]]; [_nP n [√-ly]]. With -ly being phrasal (namely, nP) and attentive also being phrasal (namely, aP), the question arises as to which of these two phrases determines the categorial status of the complex SO attentive+-ly. One analysis would be to say that n(P) determines the categorial nature of this complex SO and that aP functions as an attributive modifier within this nominal expression. Taking an attributive adjective phrase to be adjoined to its nominal host, we obtain the structure in (44):22

(44)	[nP [aP attentive] [nP -ly]].

In the literature on manner adverbs, several empirical arguments have been given in support of (a variant of) this structure. Firstly, if aP is an attributive modifier, the concord phenomenon in Italian and French manner adverbs (e.g., chiara-mente in (30b)) follows immediately: the attributive adjective (chiara) displays feminine agreement with the modified noun (mente). Secondly, as noted in Baker (2003, pp. 234–235), if -ly is a nominal expression that is modified by an attributive adjective, we can easily explain that, like attributive adjectives but unlike predicative ones, (manner) adverbs typically do not take any complement (see Jackendoff, 1977, p. 78). Consider for this the following examples:

(45)	a.	The mountaineers were proud (of themselves).
	b.	The mountaineers showed their skills proudly (*of themselves).
	c.	John and Mary are [proud mountaineers (*of themselves)].
		Intended reading: ‘J&M are mountaineers who are proud of themselves.’

(45a) shows that the predicative adjective proud can take a PP-complement, while (45b) shows that this PP-complement can’t occur in combination with the adverbial form proudly. If proudly has the structure in (44), the ill-formedness of proudly of themselves follows straightforwardly: the PP-complement that is selected by the adjective proud is separated from the adjective by an intervening noun, namely -ly. As shown in (45c), such discontinuous patterns are impossible: the noun mountaineers cannot occur in a position interspersed between the attributive adjective and the PP that is selected by the adjective.

Although the structural analysis in (44) clearly has analytical appeal, it also faces a serious empirical problem. This problem is related to degree modification, and more particularly to modification by enough. As noted in Bresnan (1973), enough has the peculiar property that, as opposed to other degree words (e.g., very, as in very legible), it must follow adjectives; see (46).

(46)	a.	Is this copy of the letter [<*enough> legible <enough>]?
	b.	I need [<*enough> legible <enough>] copies.

Consider now the following examples:

(47)	a.	The students were asked to write [<*enough> legibly <enough>].
	b.	He drove [<*enough> carefully <enough>] to pass the driving exam.

If legibly and carefully had the structure in (44), one would predict that patterns such as [[legible enough] -ly] and [[careful enough] -ly] are possible in English. But they are not. Furthermore, the well-formed patterns legibly enough and carefully enough, in which enough functions as a degree modifier of the adjectives legible and careful, are incorrectly predicted to be ill-formed. In a structure like (44), the degree modifier enough would not be part of the attributive adjectival modifier but be a constituent within the nominal expression headed by -ly. As shown in (48a), this postnominal placement of the degree word enough is impossible when enough is interpreted as a modifier of legible.23 This modification reading is only possible if enough directly follows the attributive adjective and precedes the noun, as in (48b).

(48)	a.	*I need [legible copies enough].
		Intended reading: ‘I need sufficiently legible copies.’
	b.	I need [[legible enough] copies].

Besides this empirical problem, the analysis in (44) raises the question as to what the attributive AP is modifying. According to Déchaine and Tremblay (1996), as well as Baker (2003), the nominal element -ly carries the meaning ‘manner’, just like the nouns manner/way in the adpositional MM-pattern in a legible manner/way. Alexeyenko (2012) rightly points out, however, that such an interpretation of -ly is unlikely given its presence on adverbial elements that belong to different semantic classes: e.g., extreme-ly (degree), local-ly (location), et cetera. Given this wider distribution of -ly, he suggests that -ly is a dummy nominal inserted merely for syntactic reasons, namely the requirement that adjectives need a noun to modify. As he notes, the noun -ly is comparable to the dummy noun one used in contexts like an illegible document and a legible one. Although drawing this parallel has a certain appeal, it should be noted that legibly and (a) legible one display different behavior as regards the placement of the degree modifier enough. As we saw in (47), enough must follow legibly, as in legibly enough, and cannot occur interspersed in between legible and -ly, as in *legible enough-ly. With the dummy element one, we see the opposite distribution: enough can’t follow one but must occur interspersed between the attributive adjective and the modified noun:

(49)	This is not a very good copy of the letter, but [a legible <enough> one <*enough>].

In view of the problem posed by the distribution of the degree word enough, I explore an alternative analysis for English adverbials like attentively and also for its French equivalent attentivement. Specifically, I propose the structure in (50), which is reminiscent of the adjunction structure of the anaphor himself in (33a):

(50)	[aP [aP attentive] [nP -ly]]

According to this structure, attentive determines the categorial nature of the complex Syntactic Object and -ly acts as an adjunct. I assume that -ly, just like self in the anaphoric expression himself, is a semantically weak nominal expression (i.e., a grammatical noun in Emonds’s (1985) sense). I assume it is not completely devoid of semantic meaning, though. At this point, the question seems pertinent as to why the body part nouns -ly and -ment, and not some other noun, developed into adverbial markers. An important feature of body part nouns is their relational nature; they express relations over pairs of entities. For example, a body is always some individual’s body, and a mind is always some individual’s mind.24 Suppose now that this relational property is preserved by the grammatical nouns -ly and -ment.25 Having lost their conceptual meaning, these nouns no longer express a relation over pairs of entities (e.g., John and body, as in John’s body). They are rather used to express relations between predicates. Specifically, in a sentence like The children listened attentively, its grammatical role is to relate the property denoted by the adjectival predicate attentive to the event denoted by the verbal predicate listen. Notice that being a pure “relator”, the syntactic object attentive-ly can be a proxy for attentive; they essentially mean the same thing.26

The previously observed features of manner modifiers can be explained on the basis of the structure in (50). Firstly, the distribution of the degree word enough, as in legibly enough (see (47a)), directly follows from the structure in (50). Since the SO legibly is an adjectival SO, the placement of enough in the final position within the XAP is expected.27 Secondly, the fact that, in languages such as French and Italian, the adjective can reflect the gender of the adjoined noun follows straightforwardly from a theory of concord that analyzes the spread of nominal features in terms of feature copying (Norris, 2014; Polinsky, 2016; Bayırlı, 2017). In (51), the feminine gender feature associated with mente is copied onto its sister, the lower aP-segment. This yields the form rapida.28

(51)	[aP [aP rapida[gend:fem]] [nP -mente[gend:fem]]]

Let us, finally, turn to the intriguing property that adverbial forms like proudly typically do not take an of-complement: proudly *(of themselves); see (45b). I tentatively propose that -ly, being a nominal element, “absorbs” the (genitival) case feature associated with the adjective.29 As a result of this Case absorption, the adjective is no longer able to (Case-)license the internal argument John.

7. Protection Strategies

So far, I have argued that grammatical nouns such as English -ly and French -ment can function as minimal protectors of adjectival predicates that function as modifiers. By adjoining -ly to the adjectival predicate (XAP), the adjectival predicate and the verbal predicate are no longer in a direct and local [_PRED XP]-[_PRED YP] configuration. In this section, I examine a few other protection mechanisms, starting in Section 7.1 with a discussion of what I take to be pronominal protectors, namely the Dutch adverbial marker -jes and the Swedish adverbial marker -t.

7.1. Pronominal Protectors

As shown in (52), Dutch allows manner-adverbial expressions in which the adjective (zacht, ‘soft’) is followed by the element -jes, which can be decomposed into two parts: the diminutive morpheme -je and the element -s. In traditional Dutch grammar, the latter element is referred to by the name of “adverbial -s”; see Royen (1947–1954).

(52)	Jan	duwde	zacht(jes)	tegen	de	deur.	(Dutch)
	Jan	pushed	soft(dim-s)	against	the	door
	‘Jan pushed softly against the door.’

In Corver (2022), the pattern zachtjes is taken to be a nominal expression, as in (53), where wijze ‘way’ is a silent root that moves to the categorial head n, which spells out as -s. The bound morpheme -je is the diminutive morpheme and typically shows up in nominal environments (e.g., huis ‘house’; huis-je, house-dim, ‘little house’). The adjective zacht is analyzed as an attributive AP that functions as a modifier of the (head-movement-derived) small nominal expression wijze-je-s, where wijze is a silent noun; see Kayne (2003).30

(53)	[XP zacht [XP -je [nP [n √WIJZE+n (= -s)] [√wijze]]]]

Notice that this structure is quite similar to the structure in (44) for the English adverbial expression attentively. According to that analysis, attentive is an attributive AP that modifies the nominal element -ly.

The structure in (53) turns out to be problematic when we consider patterns featuring the degree word genoeg ‘enough’; recall at this point our discussion of the English pattern legibly enough (see (47a)). As opposed to other degree words (see (54a)), Dutch genoeg must occur in post-adjectival position (see (54b)):

(54)	a.	Deze	deken	is	[erg	zacht].		(Dutch)
		this	blanket	is	very	soft
		‘This blanket is very soft.’
	b.	Deze	deken	is	[zacht	genoeg]
		this	blanket	is	soft	enough
		‘This blanket is soft enough.’

Consider now the adverbial patterns in (55):

(55)

a.

Jan

duwde

[erg

zachtjes]

tegen

de

deur.

Jan

pushed

very

soft-dim-s

against

the

door

‘Jan pushed very softly against the door.’

b.

Jan

duwde

[zachtjes

genoeg]

tegen

de

deur.

Jan

pushed

soft-dim-s

enough

against

the

door

‘Jan pushed softly enough against the door.’

The placement of the degree word erg in (55a) can be easily accommodated in the structure in (53). Together with the attributive adjective zacht it forms a complex adjectival phrase that modifies the nominal expression -jes. The distribution of the degree word genoeg, however, is more problematic. If genoeg were a modifier of zacht, one would expect to find it in a position in between the adjective and the nominal expression -jes; that is, zacht genoeg-jes. Schematically:

(56)	[XP [zacht genoeg] [XP -je [nP [n √wijze+n (= -s)] [√wijze]]]]

The linear order zacht genoegjes, however, is not possible in Dutch.31 Notice furthermore that the well-formed pattern zachtjes genoeg in (55b) is completely unexpected if one adopts the structure in (56). Deriving this word order would imply that genoeg has shifted rightward from a position within the attributive AP to a position within the noun phrase. Such an operation is impossible, as is shown also by the ill-formedness of example (57b):

(57)	a.	[een	[zacht	genoeg]	duwtje]
		a	soft	enough	push- dim
		‘a soft enough (little) push’
	b.	*[een	[zacht	genoeg]	duwtje	genoeg]

Another adjectival phenomenon that can help us in evaluating the correctness of the structure in (53) is given in (58):

(58)	Jan	duwde	[zo	zacht	mogelijk]	tegen	de	deur.
	Jan	pushed	so	soft	possible	against	the	door
	‘Jan pushed as softly as possible against the door.’

This example features the discontinuous sequence zo…mogelijk, which acts as a modifier of the attributive adjective zacht (see 1997a, Corver, 1997b). The sequence zo + zacht + mogelijk can also occur in an attributive position, as exemplified in (59a). Notice that mogelijk must immediately follow the adjective zacht; it cannot occur in the postnominal position; see (59b):

(59)	a.	[een	[zo	zacht	mogelijk]	duwtje]
		a	so	soft	possible	push-dim
		‘a push which is as soft as possible’
	b.	*[een	[zo	zacht	mogelijk]	duwtje	mogelijk]

Consider now the following patterns featuring the manner-adverbial expression zachtjes.

(60)	a.	*Jan	duwde	[zo	zacht	mogelijkjes]		tegen	de	deur.
		Jan	pushed	so	soft	possible-dim-s		against	the	door
	b.	Jan	duwde	[zo	zachtjes		mogelijk]	tegen	de	deur.
		Jan	pushed	so	soft- dim-s		possible	against	the	door
		‘Jan pushed as softly as possible against the door.’

The analysis in (56) incorrectly predicts that the sequence zo zacht mogelijkjes, as in (60a), is well-formed, and the sequence zo zachtjes mogelijk, as in (60b), is ill-formed.

Based on the above-mentioned distributional properties of genoeg (55b) and zo…mogelijk (60b), I conclude that the structure in (53) is not the correct one for the adverbial expression zachtjes. Instead of taking zachtjes to be a nominal expression featuring an attributive AP, I propose that zachtjes has the structure in (61), which is reminiscent of the structural analysis of attentively in (50).32

(61)	[aP [aP zacht] [nP -jes]]

According to this structural analysis, zachtjes is an adjectival syntactic object that consists of two parts: an adjectival phrase aP and a nominal phrase nP. In line with my analysis of attentively in (50), I take the nP -jes to be a phrase adjoined to aP. Following Corver (2022), I assume that -jes is a composite nominal expression that consists of the diminutive morpheme -je and the element -s, which I take to be a realization of the categorial node n.

One may wonder why the diminutive -je is used as a component of the adverbial marking -jes. Possibly, this is related to the fact that -je is associated with the neuter gender. As exemplified in (62), a common (i.e., non-neuter) noun becomes a neuter noun when the diminutive morpheme is attached to it. This gender change is clear from the selection of the definite article; that is, het instead of de.

(62)	a.	de[-neuter]	fiets[-neuter]
		the	bike
		‘the bike’
	b.	het[+neuter]	fietsje[+neuter]
		the	bike-dim
		‘the small bike’

Given this [+neuter]-property of the diminutive morpheme, we may qualify -jes as a [+neuter] nominal expression, which is adjoined to the adjectival predicate (i.e., zacht in (61)).33 Thus, [_aP [_aP zacht] [_nP -jes_[+neuter]]]. This [+neuter] status of -jes may be less arbitrary than we imagine at first sight. It turns out that in Dutch (but also in other Germanic languages), predicatively used XAPs can be substituted for by a [+neuter] pronoun.34

(63)

a.

Jan

is

niet

slim

en

hij

zal

het

ook

nooit

worden.

Jan

is

not

smart

and

he

will

it

also

never

become

‘Jan is not smart and he will never be.’

b.

Jan

is

slim,

wat

zijn

moeder

vroeger

ook

was.

Jan

is

smart,

what

his

mother

formerly

also

was

‘Jan is smart, just like his mother used to be.’

In (63a), the neuter personal pronoun het ‘it’ replaces the predicative adjective slim ‘smart’, and in (63b), the neuter relative pronoun wat ‘what’ takes the predicative adjective slim as its antecedent. Suppose now that the neuter form -jes in (61) functions as a kind of pronominal doubler of the adjective to which it is adjoined. Clearly, under such an interpretation of -jes, the complex form zachtjes is sufficiently close (i.e., proxy) in meaning to the simplex adverbial form zacht.35

Following up on the idea that -jes is a sort of pronominal doubler, we may characterize its grammatical nature more carefully by making use of Déchaine and Wiltschko’s (2002) typology of pronominals. This typology distinguishes the following three types: pro-DPs, pro-φP, and pro-NP. Each of these pronominals has its own grammatical properties. The pronoun type ‘pro-DP’ encodes the grammatical feature ‘definiteness’ and can occur only in argument positions, on the assumption that DPs can only be arguments (Stowell, 1989; Longobardi, 1994). The pronoun type ‘pro-φP’ encodes φ-features such as gender and number. They are pure spell-outs and lack inherent semantics. Furthermore, there is no inherent restriction on their distribution; they can function either as predicates or as arguments. Finally, the pronoun type ‘pro-NP’ has the same syntax as lexical nouns, and as NPs/nPs, they can occur in a predicate position.36 In view of the fact that diminutive -je represents the neuter gender, I propose that the pro-form -jes instantiates a φP, with je as the φ-head, -s as the realization of the categorial node n, and a silent root, which is represented here as pro. Under this analysis of -jes, (61) can be more precisely represented as in (64):

(64)	[aP [aP zacht] [φP -je [nP -s [pro]]]]

Having argued that the adverbial form zachtjes contains a neuter pronominal element instantiated by -jes, it may not come as a surprise that there are more languages in which the adverbial marking has a neuter form. In Scandinavian languages (see (65a)), but also in Slavonic languages (see (65b)), manner-adverbs are formed through the attachment of a neuter singular agreement suffix to the adjective.37

(65)	a.	Hon		sjunger	vacker-t.	(Swedish, Germanic)
		she		sing.pres	beautiful-neut
		‘She sings beautifully.’
	b.	tja	pée		xubav-o.	(Bulgarian, Indo-European, Slavonic)
		3sg.fem	sing.pres.sg		nice-neut
		‘She sings nicely.’ (Scatton, 1984, p. 345)

In the following, I briefly consider the formation of manner adverbs in Swedish. As shown in (66), the neuter-suffix -t appears also on attributive adjectives that modify a neuter noun (i.e., so-called concord) and on predicative adjectives in, for example, copular constructions:

(66)	a.	ett		vackert		hus
		a.neut.sg		beautiful.neut.sg		house
		‘a beautiful house’
	b.	Det	här	huset	är	vackert.
		this	here	house-the	is	beautiful.neut.sg
		‘This house is beautiful.’

In (66a), the attributive adjective vackert covaries with the neuter-feature of the noun hus and the indefinite feature associated with the indefinite article ett ‘a’, which itself also covaries with the neuter-feature of hus. In (66b), the predicative AP vackert covaries with the phi-features (number and gender) of the subject-noun phrase det här huset. Importantly, in these agreement dependencies, the AP functions as a dependent element (i.e., the target of agreement), and the noun phrase functions as the antecedent, that is, the element that is the agreement controller.

It is tempting, of course, to analyze the adverbial form vackert in (65a) as an agreeing element as well. But which element does it agree with? The agreement controller is typically a nominal expression, as in (66). In (65a), there is a nominal expression present, namely hon ‘she’, but this pronoun, which carries the gender feature [+feminine], clearly does not co-vary with the neuter suffix.

Rather than interpreting adverbial -t as an agreement suffix that co-varies with a nominal antecedent, I propose that -t in the adverbial form vackert is a pro-nominal element that is adjoined to aP vacker, as represented in (67), where I have analyzed -t preliminarily as a pro-nP:

(67)	[aP [aP vacker] [nP -t]]

Evidence in support of this pronominal interpretation of -t comes from Swedish pronouns. As shown in (68), the common versus neuter gender distinction is visible in certain pronouns:

(68)	a.	denCOMMON ‘it’	a.’	detNEUTER ‘it’
	b.	ingenCOMMON ‘noone’	b.’	ingetNEUTER ‘nothing’
	c.	någonCOMMON ‘someone’	c.’	någotNEUTER ‘something’

Observe that the common forms typically end in -n and the neuter forms in -t. In view of this formal difference, it does not seem implausible to analyze these definite pronouns (68a-a’) and indefinite quantificational pronouns (68b-b’, 68c-c’) as composite forms. Following Déchaine and Wiltschko’s (2002) decompositional analysis of pronominals, I assume that the common forms in (68a-c) and the neuter forms in (68a’-c’) have the following phrasal syntax:

(69)	a.	[DP de [φP -n [NP pro]]]	a.’	[DP de [φP -t [NP pro]]]
	b.	[QP inge [φP -n [NP pro]]]	b.’	[QP inge [φP -t [NP pro]]]
	c.	[QP någo [φP -n [NP pro]]]	c.’	[QP någo [φP -t [NP pro]]]

According to this analysis, pronouns are syntactic objects with an inner structure. Each functional layer encodes a particular pronominal property, such as ‘definiteness’ (D), ‘quantity’ (Q), and φ-feature information (e.g., common versus neuter gender). As shown in (70), Swedish also has the neuter indefinite pro-form ett ‘one’ in addition to the common form en ‘one’. I assume these indefinite pro-forms instantiate φP.

(70)	Lars	har	ett	stort	hus	och	Sven	har	ett	också
	Lars	has	a	big	house	and	Sven	has	one	too
	‘Lars has a big house and Sven has one too.’

Following Déchaine and Wiltschko’s typology of pronominals, I propose that the adverbial neuter marking -t on the adjective vacker in (65a) and (67) is actually an adjoined pro-φP. Thus, we have the following structure, which is quite similar to Dutch zacht-jes in (64):

(71)	[aP [aP vacker] [φP -t [nP no [pro]]]]

Since φP lacks inherent semantics and can function as a (nominal) predicate, the meaning of the adverbial form vackert is a proxy to that of the adjectival form vacker.

In summary, in this section, I argued that the morph-element in the structural template in (43)—that is, [[XP_PRED] [_ZP [YP_PRED] morph]]—can take the form of a pro-form, more specifically a φP, which, according to the pronominal typology proposed by Déchaine and Wiltschko, is a type of pronoun that can function either as a predicate or as an argument. I assume that in the adverbial expressions zachtjes (Dutch) and vackert (Swedish), the pro-φPs -jes and -t function as (semantically weak) pro-predicates.38

7.2. Protection by Doubling

This section briefly discusses a second form of protection, namely protection by means of doubling. Recall from Section 4 and Section 5 that an IDI configuration of the type ‘DP V PRON’ can be avoided by doubling the pronoun. This was shown by example (28a), which is repeated here as (72):

(72)	Rasuly-ē	[wudž-ē	wudž]	yaramališ-aɁ-u.	(Tsakhur, North Caucasian; Toldova (1999)
	Rasul-erg	refl.1-erg	refl.1.abs	wound-1.do-pf
	‘Rasul wounded himself.’

The reduplicative pattern wudž-ē wudž consists of two pronouns, of which one bears ergative case (wudž-ē) and the other absolutive case (wudž). Since the latter pronoun carries the Case that is compatible with the base position of the direct object noun phrase, it is plausible to analyze this (absolutive) pronoun as the head of the complex pronoun and the ergative pronoun as the doubling element (i.e., the morph-element). Under the assumption that the doubling element is adjoined to the doubled element, we obtain the following structure: [_YP [_XP wudž-ē] [_YP wudž]]. As it is embedded in a complex pronominal structure, the pronominal head wudž is no longer part of an IDI configuration; that is, the two tokens of a given variable are no longer in the same local domain.39

As shown in (73), the reduplication strategy is also used in the formation of manner-adverbials. In languages such as Turkish and Indonesian, this adverbial strategy is quite common.

(73)	a.	Yavaş	yavaş	yürüyorduk.		(Turkish, Turkic, Lewis, 1967)
		slow	slow	we.were.walking
		‘We were walking slowly.’
	b.	Anak	itu	berteriak	keras-keras.	(Indonesian, Austronesian, Sneddon, 1996)
		child	that	screamed	loudly
		‘The child screamed loudly.’

Notice that, in these examples, the two adjectives that jointly form the adverbial expression are identical in form.

Also, in languages in which the reduplication strategy is less common, manner-adverbial patterns can be found that instantiate this strategy:

(74)	a.	Bits of paper hung higgledy-piggledy on the walls.
	b.	Je	n’aime	pas	que	tu
		I	neg-like	not	that	you
		ranges	tes	jouets	pêle-mêle.
		put.away	your	toys	disorderly
		‘I don’t like it that you put away your toys in a disorderly fashion.’
	c.	Maria	si	avvicinò	pian-piano
		Maria	REFL	approached	slowly
		‘Maria approached slowly.’
	d.	Jan	kon	de	boete
		Jan	could	the	fee
		alleen	handje	contantje	betalen.
		only	hand-dim	cash-dim	pay
		‘Jan could only pay the fee with cash money.’

The English form higgledy-piggledy (meaning: ‘in a disorderly fashion’) probably has the word pig as its base.40 Reference to this animal suggests mess and disorder. The French form pêle-mêle has its origin in the Old French form pesle mesle, which probably has the word mesle, the imperative of mesler ‘to mix’, as its base form.41 The Italian form pian-piano is a reduplicative form of the word piano ‘slow(ly)’, and the Dutch expression handje contantje betalen (hand-dim cash- dim pay, ‘to pay cash’) is a reduplicative variant of the simpler expression contant betalen (cash pay, ‘to pay cash’).

In accordance with the pronominal doubling pattern in (72), I assume that the linearly second element of the reduplication patterns in (73) and (74) forms the head (the doublee), and the first element the adjoining element (the doubler), i.e., the morph-element that turns, for example, Turkish yavaş ‘slow’ and French mêle into a complex adverbial expression. Thus, Turkish yavaş yavaş ‘slowly’ and French pêle-mêle ‘disorderly’ have the following structures, respectively: [_YP [_XP yavaş] [_YP yavaş]]; [_YP [_XP pêle] [_YP mêle]]. Regarding the Dutch reduplicative pattern handje contantje, it should be noted that it features the diminutive morpheme -je on both elements. The attachment of -je to hand is expected since the latter is a noun; the attachment of -je to contant is less so since contant is an adjective, as is clear from the fact that it can occur as a prenominal attributive modifier, as in deze contante betaling (this cash-infl payment, ‘this cash payment’). Normally, adjectives don’t function as hosts for the diminutive morpheme. Possibly, the diminutive morpheme of hand-je is copied onto its host, which could be interpreted as a sort of concord effect. Schematically: [_AP [_NP hand-je] [_AP contant]+je].

7.3. Adpositional Protectors

In Section 4, it was shown that an IDI configuration like John saw him, with a reflexive interpretation of the John-him-dependency, can be avoided by adding complexity to the object argument. The pronoun can be protected through embedding in a complex nominal expression featuring a body part noun, as in [him [self]], or a doubling pronoun, as in [wudž-ē [wudž]]. Furthermore, it was shown that similar protection strategies can be found for manner-adverbial expressions. French rapidement uses the ‘body part noun’ strategy, and Swedish vackert uses the pronominal strategy. From the logic of the argument, one would expect there to be other formal strategies to avoid the IDI effect. For anaphoric dependencies, any representation in which the antecedent (say, John) and the pronoun (say, him) are not strictly co-arguments serves this purpose. Similarly, for manner-modification dependencies, any representation in which the verbal predicate (say, walk) and the bare manner predicate (say, quick) are not merged directly can escape from the IDI effect.

One such alternative protection strategy is the adpositional strategy. As noted in Reuland (2018, p. 99), the language Zande (Niger–Congo) uses this strategy: a reflexive interpretation of the dependency ‘antecedent … pronoun’ is possible when the latter is part of a PP:

(75)	a.	Mì-ímí	tí-rε’.	(Zande, Niger–Congo; Tucker & Bryan, 1966)
		I-kill	on-me
		‘I kill myself.’
	b.	Mì …. [PP tí rε’]

Turning now to manner-adverbial dependencies, we observe that the adpositional strategy is also used in this domain of grammar. Some illustrations are given in (76); see also Loeb-Diehl (2005). Note in passing that I take oblique case markers (e.g., elative (source)) to fall under the adpositional strategy.

(76)	a.	leigh	sé	[go	Cúramach]	i	(Irish, Indo-European, Celtic; Ó Siadhail, 1989, p. 208)
		read	he	to	careful	it
		‘He read it carefully.’
	b.	t’ejt’er-es		mor-i		[mazi-ste].		(Mordvin, Uralic, Finno-Ugric; Loeb-Diehl, 2005, p. 31)
		girl-def		sing-pres.3sg		beautiful-ela
		‘The girl sings beautifully.’

As Loeb-Diehl (2005, p. 33) observes, the adpositions and case markers that are used in the formation of MM patterns are quite diverse. Adpositions/Cases that are used include directional ‘to’, ablative ‘from’, elative ‘out of’, locative ‘at/in’, dative ‘to’, and instrumental ‘with’.

Also, in a language like Dutch, the adpositional strategy is attested. As shown in (77), the adposition typically follows the adjective:

(77)	a.	Marie	heeft	de	zin	hardop		voorgelezen.
		Mary	has	the	sentence	loud-up		read
		‘Mary read the sentence aloud.’
	b.	Jan	vertelde	breeduit		wat	er	gebeurd	was.
		Jan	told	broad-out		what	there	happened	was
		‘Jan explained in detail what had happened.’

Other examples of this A+P pattern are rechtop, as in rechtop lopen (straight-up walk, ‘to walk upright’), and voluit, as in iets voluit schrijven (something full-out write, ‘to write something in full’).

There are good reasons to assume that the sequence A+P forms a syntactic unit, i.e., a constituent; the constituency tests in (78) show this:

(78)	a.	Hardop		las		Marie	de	zin	voor.
		loud-up		read		Mary	the	sentence	prt
		‘Mary read the sentence aloud.’
	b.	Jan	vertelde	[breeduit	en	theatraal]	wat	er	gebeurd	was
		Jan	told	broad-out	and	theatrically	what	there	happened	Was
		‘Jan explained in detail and in a dramatic way what had happened.’

(78a) shows that hardop can be placed in the first position of the main clause, that is, the position preceding the finite verb (las) occupying the second position (the so-called Verb Second phenomenon). (78b) demonstrates the constituency of breeduit by means of the coordination test: breeduit can occur as a conjunct, and therefore as a syntactic unit, in a coordinate structure.

When we consider the modifiability of the pattern hardop, we observe that the gradable adjective hard, as well as the other gradable adjectives in the A+P pattern, can be modified by a degree word, such as the free comparative morpheme minder (79a) or the degree word erg (79b).

(79)	a.	In	die	tijd	werd	er	[minder	hardop]		geklaagd.
		in	that	time	was	there	less	loud-up		complained
		‘In those days, people complained less openly.’
	b.	Er	werd	[erg	breeduit]	gepraat		en	weinig	gezegd.
		there	was	very	broad-out	spoken		and	little	said
		‘There was a lot of talking and little was said.’

The question arises as to which element heads the complex syntactic object A+P. Also, in this case, the degree word genoeg ‘enough’ and the discontinuous pattern zo….mogelijk (so…possible, ‘as …as possible’) come in handy to establish the basic structure of A+P. Consider, first, the distributional behavior of genoeg in (80):

(80)

a.

Dit

soort

dingen

kun

je

niet

[hard

<*genoeg>

op

<genoeg>]

zeggen.

this

sort

things

can

you

not

loud

enough

up

enough

say

‘You can’t say these things loud enough.’

b.

Jan

ging

[zo

breed

<*mogelijk>

uit

<mogelijk>]

staan.

Jan

went

so

broad

possible

out

possible

stand

‘Jan stood as wide as possible.’

The degree word genoeg cannot occur in a position interspersed between the adjective hard and the adposition op. It must follow the A+P-sequence: hardop genoeg. Since the degree word genoeg typically modifies adjectives rather than adpositions, I assume that the adjective constitutes the core element of the A+P pattern. In line with my analysis of (pro)nominal adverbial markers, I propose that the adpositions op, as in hardop, and uit, as in breeduit, are adjuncts that are attached to the adjectival expression. Schematically:

(81)	[aP [aP hard] [PP op]]

With the phrase hardop being an adjectival syntactic object, it immediately follows that genoeg (see (80a)) and mogelijk (see (80b)) can occur at the end of the sequence A+P. Thus, [[A+P] genoeg/mogelijk].

Recall that the morph element typically makes some minimal contribution to the interpretation of the categorial host; that is, the interpretation of hardop and breeduit should be sufficiently close in meaning to the interpretation of the adjectives hard and breed, respectively. The adposition op ‘on/up’, when used intransitively, can have the meaning ‘upwards’, that is, to a high point on a vertical dimension. In the case of MM hardop, this dimension corresponds to ‘sound’. The meaning of the adjectival component hard can be informally described as “with a great deal of volume”. The adpositional component op can also have this “vertical, high-point” reading, that is, “upwards/towards a higher point/value”. Recall that, in the linguistic literature, self in himself has been interpreted as an intensifying or emphatic element. One could say that op in hardop has a similar function: in a way, it emphasizes the high volume that is denoted by the adjectival predicate hard. I assume that uit ‘out’ in breeduit has the same role, not in a vertical dimension but in a horizontal one. The adjective breed refers to the spatial dimension ‘width’. The adposition uit shares this “over a certain length” reading with the adjective breed. Also, in this case, uit could be interpreted as an element that emphasizes the meaning denoted by the adjective.42 Note that the ill-formedness of forms such as *harduit, *breedop follows from our analysis: there must be a match between the dimensional meaning of the adjective and that of the adposition.

8. A Possessive Manner Construction

In Section 4, we saw that complex anaphors featuring the body part noun self have the body part noun either as an adjunct adjoined to the pronoun or as the head of the complex expression. Following Faltz (1985), the English third-person reflexive pronoun himself was taken to instantiate the first pattern, that is, [_DP [_DP him] [_NP self]] (see (33a)), while the English first-person reflexive pronoun myself was taken to instantiate the second pattern, that is, [_DP my [_D’ D [_NP self]]]; see (33b). Recall from Section 6 that manner-adverbial expressions of the type attentively were taken to instantiate the adjunction pattern: the adverbial marking -ly is adjoined to the adjectival host attentive, as in [_aP [_aP attentive] [_nP -ly]]; see (44). From a cross-dependency perspective, the question can be raised as to whether manner-adverbial expressions ever instantiate the second pattern, the one instantiating a possessive relationship between two elements. An example that comes to mind is given in (82):

(82)	Jan	sprak	het	woord	[PP op	[DP Piet’s/mijn	manier]]	uit.
	Jan	pronounced	the	word	at	Piet’s/my	way	prt
	‘Jan pronounced the word in Piet’s/my way.’

In (82), we have a PP, which contains a noun phrase that expresses a possessive relationship with manier as the possessum and Piet’s/mijn as the possessor.

Besides the manner-adverbial expressions in (82), Dutch also has manner-adverbial expressions like (83):

(83)	Jan	sprak	het	woord	[op	z’n	Piet’s]	uit.
	Jan	pronounced	the	word	at	z’n	Piet’s	prt
	‘Jan pronounced the word in a Piet-like way (i.e., the way Piet would pronounce it).’

In (83), op takes the noun phrase z’n Piet’s as its complement. As is clear from the translation, the PP has a manner-like interpretation: ‘in a Piet-like way’. The nominal expression z’n Piet’s is quite puzzling because it appears to display two possessor-elements in a row: z’n, which looks like a weak possessive pronoun—compare: z’n fiets (his_weak bike)—and the proper name Piet, which carries the possessor-marking -s.43

At first sight, one might think that z’n and Piet’s are in a concord relationship, with z’n representing the third-person masculine singular possessive pronoun and Piet being the proper name of a male person. As shown in (84), however, z’n can also co-occur with the name of a female person (Marie) and with a coordinate structure consisting of two proper names (i.e., a plural referent). As shown by (84a), the feminine singular weak possessive pronoun d’r ‘her’ can’t occur in this manner-adverbial construction. The same holds for the plural possessive pronoun hun ‘their’, as exemplified in (84b).

(84)	a.	Ik	sprak	het	woord	[op	z’n/*d’r	Marie’s]	uit.
		I	pronounced	the	word	at	z’n/d’r	Mary-s	prt
		‘I pronounced the word in a Mary-like way (i.e., the way Mary would pronounce it).’
	b.	Zij	dansen	[op	z’n/*hun	Fred	en	Ginger’s].
		they	dance	at	z’n/hun	Fred	and	Ginger-s
		‘They dance in a Fred and Ginger-like way (i.e., the way in which Fred Astaire and Ginger Rogers danced).’

Another remarkable property of the manner expression in (83) is that the possessor noun phrase Piet’s is not followed by an overt noun. At the surface, it looks as if we have an NP ellipsis pattern, that is: [op [z’n Piet’s ø]]. Importantly, however, possessor noun phrases like Piet’s normally do not license NP ellipsis. This is shown in (85):

(85)	*Piet	sprak	[PP op	[XNP Marie’s	manier]]
	Piet	spoke	in	Marie’s	way
	en	Marie	sprak	[pp op	[XNP Piet’s	ø]].
	and	Marie	spoke	in	Piet’s	way
	‘Piet spoke in Marie’s way, and Marie spoke in Piet’s way.’

From the ill-formedness of (85), I conclude that the pattern op z’n Piets does not involve NP ellipsis. Instead, I assume that it instantiates a pronominalization pattern (compare English a small one), where z’n is a pro-nominal—specifically, a pro-φP (see below)—which originates in the head position of a possessive noun phrase. The surface order results from the movement of z’n to [Spec,DP], as in (86b):44

(86)	a.	[PP op [DP D [PosP Piets [Pos’ Pos [φP z’n]]]]]
	b.	[PP op [DP z’n [D’ D [PosP Piets [Pos’ Pos [φP z’n]]]]]]

The pro-form z’n in (86) represents a possessum. I assume this pronominal possessum is a relational (pro)noun; that is, it expresses a two-place relation between the entity denoted by z’n and the entity denoted by the possessor Piet.45 In a way, z’n represents an entity (say, “something possessed,” i.e., his way of behaving/behavior), which is an inalienable part of the possessor Piet, who stands for the ‘whole’ of the ‘part-whole’-relationship. As such, the relationship between Piet’s and z’n in (83) is quite similar to the relationship between the possessor Piet and the inalienable body-part noun neus ‘nose’ in Piet’s neus ‘Piet’s nose’.

I assume that the weak proform z’n has a composite structure, namely ze + ’n (i.e.,/zə/+/ən/); see Barbiers and Bennis (2003). In line with Déchaine and Wiltschko’s pronominal typology (see Section 7.1), I assume that z’n is a pro-φP that consists of two parts: a φ-head (ze-) and an NP-complement (-n). So, we have the following structure: [_φP ze- [_NP -n]]. I propose that the first element is phi-deficient in the sense that it is specified only for the property ‘third person’ but not for the properties ‘gender’ and ‘number’. The second element is an equivalent of the English ‘one’ (i.e., a pro-NP/nP).46 The composite pronoun z’n represents the possessum, and the presence of ze-, which lexicalizes the φ-head, marks a third-person agreement relationship between the possessor Piets and the possessum z’n, which just like the SELF-part in the English reflexive pronouns yourself or myself, is a relational (pro-)noun that is (near-)reflexive: the “intrinsic” property represented by z’n is always some individual’s (e.g., Piet) property. I assume that the possessor Piet in the base structure Piets z’n in (86a) functions as the local antecedent for z’n.

At this point, it is interesting to observe that in certain dialects of Dutch (e.g., Maastricht Dutch), the manner-adverbial pattern in (83) does not feature a weak possessive pronoun but rather a weak reflexive pronoun, that is, the equivalent of Standard Dutch zich ‘himself’.

(87)	Zèk	’t	mer	op	z’ch	Mestreechs.	(Maastricht Dutch; Shepherd, 1946, p. 78)
	say	it	prt	at	refl	Maastricht-s
	‘Say it in a Maastrichts-like way.’

Following Postma (1997) and Barbiers and Bennis (2003), I take the reflexive pronoun zich ‘himself’ to have a composite structure that consists of a possessive pronominal part ze- (representing third person) and a possessive affix -ig. As noted by Barbiers and Bennis, this affix is also found on adjectives such as ernstig ‘serious’ and bloedig ‘bloody’, where it functions as some kind of possessive marker. The adjective bloedig, for example, indicates that the object it belongs to possesses the property [+blood]. Along the same lines, Maastricht Dutch z’ch in (87) indicates that the object it belongs to, namely Maastricht, possesses the property ‘third person’ (represented by ze-).

In summary, it was shown in this section that the Dutch manner-adverbial expression op z’n Piets instantiates a possessive relationship. From a cross-dependency perspective, the relationship between your (possessor) and self (possessum) in the reflexive pronoun yourself is quite similar to the relationship between Piets (possessor) and z’n (possessum) in the manner-adverbial pattern op z’n Piets. While self in yourself remains in situ, the pro-form z’n occupies an ex-situ position as a result of DP-internal movement, as represented in (86b).

9. Bare APs

In the preceding sections, we came across various protection strategies that are used to make an adjectival predicate complex. Adding complexity to an adjectival predicate was needed to avoid an IDI configuration, specifically a configuration in which a verbal predicate (VP) and an adjectival predicate (AP) enter into a direct modification dependency, that is, [VP^AP]. In Dutch, for example, adjectival predicates with a manner-adverbial function can be made complex by adding -jes, as in zachtjes (Section 7.1), or an adpositional element like op, as in hardop (Section 7.3). Importantly, Dutch, but also other languages (see (29)), permits the adverbial use of bare adjectives, as exemplified in (88):47

(88)	a.	…dat	Jan	de	zin	vreemd	uitsprak.	(Dutch)
		that	Jan	the	sentence	strangely	pronounced
		‘…that Jan pronounced the sentence strangely.’
	b.	…dat	Jan	de	deur	voorzichtig	opende.
		that	Jan	the	door	carefully	opened
		‘…that Jan opened the door carefully.’

If these are truly bare APs, one would expect these modification patterns to be ruled out. They would instantiate the illegitimate IDI configuration in (42); that is, *[[VP_PRED] [AP_PRED]]. A way out would be to say that these adjectival predicates do have a complex structure (see (43)), but that complexity is hidden at the sound surface due to the silence of the added morpheme. In other words, the adverbial marking is phonologically unrealized. According to this analysis, vreemd in (88a) would have the structure in (89):48

(89)	[aP [aP vreemd] [Ø]]

One may wonder what the exact nature of the silent morpheme is. In line with the added morphemes discussed so far, it should be an element that makes a minimal contribution to the interpretation of the adjective. Furthermore, it should leave intact the predicative function of the adjectival modifier.

In older varieties of Dutch, the adverb was distinguished from the adjective by the presence of the suffix -e.49 This adverbial -e is still found in present-day Dutch fossilized forms such as gaarne (willing-e, ‘willingly/gladly’), bij lange na (by long-e after, ‘in no way’), and verre van duidelijk (far-e from clear, ‘far from clear’). It is arguably also still present as an augmentative/emphatic element in the italicized expressions in (90); see Verdenius (1939), Van Langendonck (1978, note 6)):

(90)	a.	Kom	hier-e!					(dialectal)
		come	here-e
		‘Come here!’
	b.	‘t	is	al	vel-e	beter!		(dialectal)
		it	is	already	much-e	better
		‘It is already much better!’
	c.	Ik	vind	haar	verdomd-e	aardig!		(colloquial)
		I	find	her	damned-e	kind
		‘I think she is really kind!’

In (90a), augmentative -e is attached to the proximate locative modifier hier, in (90b) to the measure-denoting quantifier veel ‘much’, and in (90c) to the evaluative degree modifier verdomd ‘very/really’. The extra element adds emphatic or affective force to its host (Corver, 2004). Importantly, -e can also be absent, which yields the forms hier, veel, and verdomd at the sound surface.

Interestingly, augmentative -e can also be found on predicative adjectives with a superlative form, both in copular environments (91a) and in manner modificational ones (91b):50

(91)	a.	Deze	zin	is	[‘t	vreemdst(-e)].
		this	sentence	is	‘t	strangest(-e)
		‘This sentence is the strangest.’
	b.	Deze	zin	sprak		Jan	[‘t	vreemdst(-e)]	uit.
		this	sentence	pronounced		Jan	’t	strangest(-e)	prt
		‘This sentence, Jan pronounced in the strangest way.’

Suppose now that augmentative -e makes visible/audible a syntactic position that, in non-emphatic contexts, remains silent.51 In the case of ’t vreemdste in (91), we would have the structure in (92a), while the pattern ’t vreemdst would have the structure in (92b):52

(92)	a.	[’t [aP [aP vreemdst] [-e]]]
	b.	[’t [aP [aP vreemdst] [Ø]]]

While Dutch has the bare adverbial APs as its default strategy, a language like (Standard) English uses bare adverbial APs in a much more restricted way (Larson, 1985; J. R. Ross, n.d.). Some examples illustrating the adverbial use of bare APs are given in (93):

(93)	a.	John will leave the room fast.
	b.	Don’t talk so loud!

Interestingly, the distribution of these bare adverbial APs turns out to be restricted: they can occur only in postverbal positions (Sugioka & Lehr, 1983; C. Ross, 1984; Corver, 2014), as exemplified for fast in (94a). In this respect, they differ from -ly-adverbials, which can occur both postverbally and preverbally, as exemplified in (94b).53

(94)	a.	John will <*fast> leave the room <fast>.
	b.	John will <quickly> leave the room <quickly>.

What underlies this distributional contrast? For my tentative answer to this question, I would like to build on the idea that manner-adverbials, just like reflexive pronouns (see (33)), can instantiate an adjunction configuration—see (50), where the body part noun -ly is adjoined to the aP attentive, yielding attentively—or a possessive-like configuration. The latter does not take the form of a Spec-head structure, as familiar from possessive noun phrases like John’s house (alienable possession) and John’s nose (inalienable possession). It rather takes the form of a so-called construct state construction, a type of possessive pattern that is familiar from studies on Modern Hebrew (Ritter, 1988; Hazout, 2000; Siloni, 2002) and illustrated in (95):

(95)	a.	beyt	ha-mora
		house	the-teacher
		the teacher’s house
	b.	rina	[yefat	mar’e].
		Rina	beautiful	look
		‘Rina is good-looking.’

Both the noun phrase beyt ha-mora and the adjective phrase yefat mar’e display characteristic properties of Construct state constructions. Firstly, the head of the nominal/adjectival construction is initial (beyt/yefat); it occurs in the leftmost position of the phrase. Secondly, the noun/adjective directly precedes a noun phrase, that is, without the mediation of any (dummy) preposition. Thirdly, phonological alternations are found between construct state forms (beyt/yefat) and free state (i.e., non-construct state) forms (bayit ‘house’; yafa ‘beautiful’).

Importantly, as noted in Siloni (2002), the adjectival CS in (95b) is limited to inalienable nouns (e.g., mar’e) and is typically (but not exclusively) found with body parts. This last property makes an interesting connection with the adverbial marker -ly, which historically has its origins as a body noun (see Section 4). Suppose now that English bare APs are hidden Construct state constructions featuring a silent body part noun (say, -ly, where the use of small capitals indicates silence). This silent body part noun is not a nominal expression adjoined to aP but rather a nominal expression that functions as a complement of the adjective. Schematically:

(96)	[aP fast [nP -ly]]

Suppose now that silent -ly abstractly represents the body (part), which has the property ‘fast’ and is relationally associated with the subject of the clause (e.g., John in (93a)). According to this analysis, (93a) could be loosely paraphrased as John will leave the room “fast of body/legs”, and (93b) as Don’t talk “so loud of body/voice”.54

I propose that it is this ‘head+complement’ structure in (96) that blocks the preverbal appearance of superficially bare adverbial APs like fast. Essentially, preverbal fast in (94a) is out for the same reason that preverbal with care (as opposed to carefully) is out:

(97)	a.	John will <*with care> open the box <with care>.
	b.	John will <carefully> open the box {carefully>.

The PP with care in (97a) also instantiates a head-complement structure, namely [_PP with [_nP care]]. The hidden head+complement in (96) and the overt head+complement structure of with care are ruled out in the preverbal position since they violate what Emonds (1976, Chapter 1; 1985, p. 130) calls the (Right) Recursion Restriction. This constraint states that if a language is head-initial, any phrasal modifier ZP to the left of a head X^o must terminate in its head Z (in s-structure).55 Crucially, this account can be successful only if head-complement structures (e.g., fast + [_complement -ly]) are subject to the (Right) Recursion Restriction, and XP+adjunct structures (e.g., quick + [_adjunct -ly]) are not.56

In summary, it was argued in this section that superficially bare manner-adverbial APs, such as Dutch vreemd ‘strange(-ly)’ and English fast, are actually complex Syntactic Objects in the sense that they carry a silent adverbial marker. For Dutch, it was argued that this marker is adjoined to AP/aP and that it can surface in a restricted number of environments (e.g., in superlative APs). For English bare APs, it was proposed that these are hidden (possessive) Construct State patterns in which the adjective is followed by a silent body part noun -ly. It is the hidden head-complement structure of fast that blocks its preverbal occurrence.

10. A Brief Note on Other Licensing Strategies for the Expression of Manner Modification

As noted in Reuland’s (2011, 2018) study on reflexivization, besides the protection strategy, languages can use other licensing strategies in order to resolve the effects of IDI. One such strategy consists of separation, that is, the realization of two arguments as part of different predicates. Schematically:

(98)	DP V1 [V2 PRON]

In (98), the antecedent DP is an argument of V1, while the pronominal element is an argument of V2. In this configuration, the pronoun can get a reflexive reading in certain languages.

This separation strategy can also be found cross-linguistically for the expression of manner modification. Consider, for example, the sentences in (99) and (100); see Loeb-Diehl (2005):

(99)	[om		geh]	[mə mə	faiah].		(Ambrym, Austronesian, East-Oceanic; Paton, 1971, p. 77)
	2sg.pres		work	3sg.pres	be.strong
	‘You work strongly.’ (lit.: you work, it (is) strong)
(100)		[nyonuví	lá	dzi	ha]	[wò	víví].
		girl	the	create	song	it	be.sweet
	‘The girl sang sweetly.’ (lit.: the girl created a song, it (is) sweet)

In these examples, manner modification is expressed by means of an asyndetic coordinate structure. The left conjunct introduces the VP, which denotes an event—that is, ‘you work’ in (99), and ‘the girl sang’ in (100)—and the second conjunct consists of a copular construction whose neuter subject (‘it’) refers to the event expressed by the VP of the first clausal conjunct. The predicative AP assigns a property (‘strong’, ‘sweet’) to the neuter pronominal subject. In a way, this property is indirectly associated with the event expressed by the first conjunct. Schematically:

(101)

[[clause subject VPi]] & [clause iti APmanner]]

Since the VP-predicate of the first conjunct and the AP-predicate of the second conjunct are part of different local domains—namely, the left clausal conjunct and the right clausal conjunct, respectively—there is no violation of IDI.

Another type of strategy that languages can use to license reflexivity is called ‘Bundling’ in Reuland (2011, 2018). This (lexical) operation satisfies the IDI constraint by lexically reducing the argument structure of a two-place predicate (i.e., detransitivization). Specifically, it reduces the internal argument of a two-place predicate and combines the internal theta role (theme) and the external theta role (agent) into a composite agent–theme role. Schematically:57

(102)	Vacc(θ1, θ2) → Rs (V) (θ1,2)
	(where θ1,2 stands for the bundling of θ1 and θ2).

Examples (103)–(104) show how a reflexive reading can be obtained by means of Bundling. Consider, first of all, the sentence in (103), which contains the base transitive verb wash. As shown, wash is interpreted as a two-place predicate with Mary as the agent and Sue as the theme. Consider next the sentence in (104a), which contains the derived reflexive verb wash. As shown by its interpretation in (104b), the reflexive verb also has two theta roles, but they are assigned to the same argument, namely Mary.

(103)	a.	Mary washes Sue.
	b.	∃e [wash(e) & Agent(e,Mary) & Theme(e,Sue)]
(104)	a.	Mary washes.
	b.	∃e [wash(e) & Agent(e,Mary) & Theme(e,Mary)]

In (104), reduction takes place at the level of argument structure (i.e., the valence of the predicate). Two separate theta-roles (Agent, Theme) are reduced to a composite theta-role: [Agent–Theme]. Suppose now that Bundling does not take place only at the level of argument structure but also at the level of predicates; that is, two predicates (Pred₁, Pred₂) can be reduced to a single composite predicate (Pred₁+Pred₂). Possibly, this bundling strategy is what underlies Greek ‘manner-adverb + verb’ complexes of the type in (105a); see Rivero (1992) and Alexiadou (1997). Observe that the complex [kalo+efage] cannot be interrupted by clitics or other adverbs; see (105b,c).

(105)	a.	Kaloefage.	(Modern Greek; Alexiadou, 2013, p. 474)
		well.ate.3sg
		‘He ate well.’
	b.	*Kalo to-efage
		well it-ate.3sg
	c.	*Kalo-sinisthos-efage
		well-usually-ate.3sg

Although a full-fledged discussion of the separation strategy and the bundling strategy falls beyond the scope of this article, I hope to have shown that here, too, we can see certain similarities between the licensing strategies of reflexivization and those of manner modification.

11. Conclusions

This article aimed to increase our understanding of the syntax of manner modification by examining it from a cross-dependency perspective. Specifically, I tried to discover abstract grammatical properties of manner modification by studying them from the perspective of anaphoric dependencies. Crucially, the idea was not to reduce one syntactic dependency (e.g., manner modification) to the other (e.g., the anaphoric dependency). What I did intend to show, though, was that the two grammatical dependencies share certain abstract formal properties and are governed by the same type of principle governing the computational system (C_HL). As for the latter, building on Reuland’s (2011, 2018) IDI constraint and also related principles proposed in the generative-linguistic literature, I proposed that C_HL is unable to distinguish two predicates (e.g., a verbal predicate and an adjectival one) if they are in a local domain. More specifically, an adjectival predicate (e.g., quick) cannot merge directly with a verbal predicate (e.g., walk). The C_HL can only deal with two predicates if their linguistic environment allows them to be distinguished as different occurrences. This formal distinctness can be achieved by means of various licensing strategies, including protection and bundling, which are operations that are also active in the licensing of anaphoric dependencies. The protection strategy was discussed in most detail: it was shown that the various morpho-syntactic implementations of the protection strategy are quite similar for reflexivization and for manner modification: (grammaticalized) body nouns, doubling pronouns, and adpositional material can be used for making the dependent element (i.e., the anaphor; the manner modifier) more complex. It was further proposed that superficially bare manner modifiers (e.g., fast) are actually complex syntactic objects, where the complexity comes from the (hidden) presence of a silent element.

The approach to manner modification put forward in this article puts strict limits on the expected variation space of the manner modification system in human language. This space is defined by the various morphosyntactic means to avoid the “adverbial IDI-configuration”. Importantly, a single language can use more than one type of protection strategy. This is what we saw for Dutch, which can recruit different elements for protecting the predicate, such as the ‘diminutive+-s’ form -jes (e.g., zachtjes, ‘softly’); a zero-morpheme Ø (zacht, ‘softly’), which sometimes surfaces in superlative patterns (’t zachtst(-e), ‘the most softly’); an adpositional element like op (e.g., hardop, ‘aloud’) or uit (e.g., breeduit, ‘stretched out/broadly’); a weak-pronominal element z’n, which surfaces in possessive-like manner-adverbials (e.g., op z’n Piets, ‘in a Piet-like way’); or a reduplicative pattern (e.g., handje contantje ‘cash’). Also, in other languages, we find manner-adverbial forms that fall within this variation space, as exemplified in (106):

(106)	a.	English: quickly (A+marker); fast (bare A); out loud/aloud (P+A);
		higgledy-piggledy; (Redupl.); in a strange way (PP)
	b.	French: rapidement (A+marker); vite (bare A); pêle-mêle (Redupl.);
		d’une manière étrange (PP)
	c.	Turkish: rahat-ça (A+marker, ‘comfortably’); çabuk (bare A; ‘quickly’);
		yavaş yavaş (Redupl.), kirici bir biçimde (PP; ‘in a hurtful way’)

It goes without saying that, in order to obtain a better view of the grammar of manner-modification in human language, a more systematic cross-linguistic study of manner-adverbial patterns is needed, as well as more in-depth studies of the fine-grained structure of manner-adverbial expressions in individual languages. These two lines of empirical research will help us find answers to questions about the intra-linguistic and cross-linguistic variation space of manner modification, such as: What are the limitations of this space? What would we expect languages to prefer in avoiding adverbial IDI configurations, and what would we expect never to find? As for the first question, we expect, for example, that the class of adverbial markers (protectors) in human language(s) consists of elements that are sufficiently close (i.e., proxy) in meaning to the adjectival host. As for the second and third questions, one would expect from the perspective of computational efficiency (Chomsky, 1995) that the simultaneous use of different protection strategies is excluded; that is, if one licensing strategy suffices for protecting the manner-adverbial, why use an additional protection strategy? Possibly, this principle of “minimal protection” excludes forms such as Dutch *breed-uit-jes (broad-out-DIM), English *a-loud-ly, and Turkish *yavaş yavaş-ça (slow-slow-ça).

Based on a restricted set of data (especially from Dutch and English) discussed in this article, I nevertheless hope to have shown that we can gain a deeper understanding of the formal properties of manner modificational dependencies by studying them from the perspective of a grammatical dependency whose formal properties are better-known. The commonalities between anaphoric dependencies, on the one hand, and manner modificational ones, on the other hand, strengthen the idea that linguistic phenomena (e.g., anaphora, manner modification) do not result from construction-/dependency-specific rules but rather from general, abstract principles of human language (e.g., IDI and the typology of licensing strategies). The language-specific dimension of grammatical dependencies is a consequence of the lexical atoms that a natural language chooses to formally shape its grammatical dependencies.