Gradualness of Grammaticalization and Abrupt Change Reconciled: Evidence from Microvariation in Romance

Abstract

Grammaticalization has long been understood as a process that takes place gradually, but within it, discrete and abrupt changes take place. This tension has been reconciled by claiming that the semblance of a gradual process is given by different parts of a construction undergoing changes at different points in time. Focusing on synchronic microvariation as gradience, this article discusses cases of clitic loss in four Romance varieties (Brazilian Portuguese, Raeto-Romance, some northeastern Italo-Romance varieties, and French), and identifies common patterns in the cells of the paradigms that are most vulnerable to the process of loss. Relating the grammatical and semantic properties of these cells to established typological hierarchies, the paper explores how general cognitive principles can account for the key properties of gradualness and gradience and, ultimately, language change.

1. Introduction

The idea that grammaticalization is a gradual process has long been at the forefront of conversations about language change, especially within more functional perspectives (Lichtenberk 1991; Hopper and Traugott 2003; Lehmann 2004, to name but a few). More recently, the edited volume by Traugott and Trousdale (2010) brought under the spotlight the relation between gradualness, gradience and grammaticalization: recasting the discussion in the context of a broader understanding of diachrony and synchrony as ‘two perspectives on the same thing’ (Lehmann 2004, p. 2), gradualness is seen as a manifestation of diachronic change, while gradience is the corresponding effect on the synchronic axis.

Reviewing Aarts’ (2004, 2007) distinction between two types of gradience, subsective and intersective, conceived, in broad terms, as gradience within and between categories, respectively, the authors conclude that such a sharp division is unwarranted, and the two can, in fact, be subsumed under subsective gradience. In Traugott and Trousdale’s understanding, then, gradience corresponds to a cline within a category ranging from more to less typical members of that category. This is spelt out more explicitly in Rosemeyer (2014, p. 78), who refines the definition of gradience as referring ‘to the fact that a construction displays structured variation regarding the use of lexical elements, with a cline from less to more typical elements’. What is of interest here is the specification that synchronic variation is structured, a distinction in terms of (proto)typicality, that is, between more core and more peripheral members, and the nature of the cline with respect to which they are measured.

Gradualness, according to Traugott and Trousdale (2010, pp. 25–26), is not to be interpreted as ‘an issue of indeterminacy, vagueness, or undecidability’, nor as ‘drift […] or monotonic incremental change’, nor ‘equated with slow progression over long periods of time’, nor as ‘imperceptible change’: ‘[g]radualness refers to the fact that most change involves (a series of) micro-changes’ (Traugott and Trousdale 2010, p. 23), micro-steps that are discrete and therefore abrupt. They further address the tension between the two conceptions of grammaticalization, as a gradual process on the one hand, and as consisting of discrete changes, involving abrupt shifts, on the other, concluding that ‘because different parts of a construction may undergo changes at different points in time, the change to the construction as a whole may appear to be gradual’ (Traugott and Trousdale 2010, pp. 20–21).

Diachronic gradualness and synchronic gradience are deeply intertwined because, ultimately, they are both expressions of the effects of the micro-changes that characterise language change: ‘some of these micro-steps may give rise to gradient systems at any synchronic “slice” in the development of a particular language’ (Traugott and Trousdale 2010, p. 39). It has long been believed that diachronic changes are manifested in synchronic variation (among others, (Andersen 2001, p. 228), and since change progresses ‘in an orderly, grammatically conditioned step-by-step fashion’ (Andersen 2001, p. 226), its synchronic manifestation is systematic (cf. Rosemeyer’s ‘structured variation’ mentioned above). Synchronic variation, in turn, gives rise to ‘change that has a determinate direction’ (Andersen 2001, p. 225). This two-way relation between synchrony and diachrony, that is, synchronic variation as both the outcome of and the trigger for diachronic change, adds a further dimension to an investigation of synchronic data, allowing the linguist to see reflected in synchrony the unfolding of language change.

Saussure’s polarised conception of synchrony and diachrony has long been abandoned, and although Traugott and Trousdale’s volume may be one of the first to directly and explicitly address the relation between the two, we find that typological cross-linguistic generalisations, such as Moravcsik’s (1974) verb agreement hierarchy, were long before then thought of and recast in diachronic terms as an indication of the order in which, for example, verbs develop agreement markings with their arguments (Givón 1976, p. 156; reformulated more overtly in Givón 2017, p. 90).

Selecting as case studies a set of closely related languages and investigating the synchronic microvariation that they display in the expression of pronominal object clitics, this article sets out to explore the linguistic factors that can be identified as responsible for the uneven progress from one stage of development to the next both within and across languages. More broadly, it explores how the key properties of both gradience and gradualness, highlighted above, can be derived. Unlike genetically unrelated languages, linguistic systems that share the majority of their characteristics and minimally differ from one another in a limited and identifiable number of properties, offer the opportunity to observe the variation manifested in a given phenomenon or structure without the interference of ‘background noise’ (cf., for example, Poletto 2012). The patterns with which morphological defectiveness expresses itself across these closely related languages are related to the Referential Hierarchy (in its original formulation by Silverstein 1976) and the Accessibility Hierarchy (Keenan and Comrie 1977), which have been shown to affect the morpho-syntactic organisation of a variety of languages (e.g., to do with number distinctions, differential case marking of transitive [±animate] subjects, and hierarchical verb agreement, as discussed in Comrie 1981, and references therein).

Taking a textbook example of grammaticalization as a starting point, this article explores specific instances in which the progress from one stage of development to the next has been uneven across the category affected by the phenomenon, in order to try and model the gradualness of grammaticalization in a system that is, ultimately, based on discrete categories and distinctions.

The grammaticalization process that we focus on is the developmental trajectory of pronominal forms as portrayed in the following cline (as, for example, expressed in (Fuß 2005, p. 4):

(1)	independent pronoun → weak pronoun → clitic pronoun → affixal (agglutinative) agreement marker → fused agreement marker → ø

This developmental path represents a specific instantiation of a more general tendency which is the line at the very core of grammaticalization: the shift from lexical and semantically contentful to grammatical and semantically empty to, eventually, complete disappearance (cf. Hopper and Traugott’s (2003, p. 7) cline content item > grammatical word > clitic > inflectional affix (>zero)). Once the final stages at the right edge of the cline have been reached, languages often reintroduce a content element to replace the one that has disappeared, suggesting a more circular or spiral-like path than a linear progression (as already observed in the early twentieth century by scholars such as (von der Gabelentz 1901, p. 256; Meillet 1921, p. 140).

Focusing on the segment of the cline that represents the progress from clitic pronoun to zero, the investigation of the chosen case studies (Raeto Romance, Brazilian Portuguese, colloquial French, and a group of neighbouring Italo-Romance varieties spoken in northeastern Italy, which have all been discussed in Paoli 2014) reveals a gradual and uneven progression of pronominal weakening that results in some cells of the clitic paradigm being affected before others. The identification of recurring patterns across the set of case studies allows us to explore the broader direction behind the phenomenon, which comes out of potentially universal hierarchies.

After a brief consideration of the emergence of pronominal forms from Latin to Romance in Section 2, the article turns to the four case studies, discussed individually in Section 3.1, Section 3.2, Section 3.3 and Section 3.4, and comparatively, in Section 4. Some reflections and concluding remarks follow in Section 5.

2. The Emergence of Pronominal Clitics: From Latin to Romance

Before delving into the four case studies of clitic loss, it is helpful to briefly consider an earlier stage along the cline in (1), namely, the segment that represents the shift from independent pronoun to clitic; that is, the development of pronominal clitics. Latin did not have a clitic series of pronouns, and the development of pronominal clitics is considered one of the distinguishing morphosyntactic features that differentiate the Romance languages from Latin. The process that led to the birth of these atonic pronouns seems to have progressed unevenly (for a more detailed discussion of this stage, see Paoli 2014, pp. 167–69), reflecting perhaps also the change that was to come, the process of clitic loss.

Pronouns in their traditional conception were limited, in Latin, to the first and second persons, along with the reflexive form for the third person (as discussed in Wanner 1987, p. 67). Even then, their use was highly restricted, because Latin was, fundamentally, a null object (as well as a null-subject) language (cf., among many others, Johnson 1991; Mulder 1991; Luraghi 1997): conditions related to both syntactic and discourse restrictions operated on licensing the omission of objects. For the third person, however, when the relevant discourse and syntactic conditions allowed for it, the referents were lexicalised by demonstratives instead. While not systematically argued or independently substantiated, there are indications that in Latin two distinct prosodic versions of the same pronoun existed: often, a single Latin pronominal form evolved into two separate elements in Romance1. At the pragmatic level, too, the same form could be used contrastively or simply refer to an expected/active referent, subsuming the two distinct roles that would be unpacked into the Romance tonic and clitic pronouns, respectively.

Given these facts, it is reasonable to infer that the Latin pronominal forms exhibited certain traits that would eventually define the characteristics of clitics in Romance. Wanner (1987, p. 68) highlights that the ‘true’ pronouns (referring to the first, second and reflexive third persons) had the capacity to exhibit clitic-like behaviour from their earliest usage. The demonstratives that were used as a third-person pronoun, on the other hand, were not witnessed in their clitic modern form until much later (i.e., around 800AD, in the early medieval parody of the Lex Salica; Wanner 1987, p. 68).

It therefore seems plausible to claim (as in Paoli 2014) that the (proto-)clitic use of first- and second-person forms preceded the same development of the third person, suggesting a hierarchy of the emergence of pronominal clitics based on [Person] features (‘>’ is to be understood as diachronic precedence):

(2)	1st and 2nd > 3rd

The gradualness of the weakening of the tonic pronouns to produce clitics is hence due to the process targeting different cells of the paradigm at different times. Although not the specific focus of this article, it is interesting to note that this process of weakening reveals a hierarchy based on Person. The contrast between first and second person, on the one hand, and third on the other, which is manifested in a range of widely differing linguistic phenomena, is well-attested cross-linguistically: as discussed in Silverstein (1976), Siewierska (2004), and related sources, third-person pronouns typically set themselves apart from their first- and second-person counterparts, which inherently carry indexical qualities, and whose reference depends on the identity of the speaker and listener. Benveniste (1971, p. 217) argues against grouping the third person with the first and second, considering it a ‘non-person’; Forchheimer (1953, pp. 5–6) identifies several morphological patterns highlighting how the unmarked nature of the third person allows for its distinct treatment compared to the first and second; Erteschik-Shir (2007) differentiates between the two in terms of discourse status, in that first and second person are ‘stage topics’ or ‘permanently available topics’, i.e., pronouns whose referents are always available in the discourse, unlike third person. Be whatever it may, what we see is that some more general, perhaps even universal, properties related to both structure and discourse impact the progress of an item along a grammaticalization cline. We return to this in Section 4.

Once all the cells of the paradigm had been affected, a complete series of clitic pronouns arose; following (Benincà 1995), I assume that this had taken place by the Middle Ages, at which point all (or most of) the Romance languages had developed a complete pronominal clitic series. This consideration is the basis on which the rest of this article is built, that is, that those modern varieties that do not have (fully) complete clitic paradigms have undergone (or are undergoing) a process of clitic loss.

3. Clitic Loss: Case Studies from Romance

Although generally considered a typical property of Romance, clitic pronouns are not found in all modern Romance languages: partially or totally defective clitic paradigms feature in the Raeto-Romance varieties spoken in the Grisons region (Graubünden) of Switzerland (Haiman and Benincà 1992), in Brazilian Portuguese (Galves 2000; Lucchesi and Lobo 1996; among others), and in a group of neighbouring northeastern Italian varieties (Paoli 2009, 2014), and the ability to omit an object clitic is established in colloquial French (Lambrecht and Lemoine 1996, p. 297; Larjavaara 2000, p. 63). Let us explore them in turn.

3.1. Brazilian Portuguese

Brazilian Portuguese (BP henceforth) allows pronominal objects of transitive verbs to remain unexpressed in constructions in which, in other Romance languages, a pronoun would be needed (cf., Cyrino and Reich 2002):

(3)	a.	Você	tem	que	lavar	ø	antes
		you	you-have	that	wash	ø	before
		de	pôr	ø!
		of	put	ø
		‘You must wash (it) before putting (it) in!’ [talking to someone who is about to put rice into a pan] (from Cyrino 2017, p. 276)
	b.	Os	bolinhos	não	estão	aqui	porque
		the	cupcakes	not	they-are	here	because
		a	Maria	comeu	ø.
		the	Maria	ate	ø
		‘The cupcakes are not here because Maria ate (them)’ (from Cyrino 2017, p. 277)

Typologically speaking, null-object languages allow the omission of an object when it can be recovered from the context, either because discourse-old or because physically present in the space of speaker and hearer, and hence accessible. At the same time, BP also has a defective clitic paradigm in that the third-person forms have disappeared. The accusative clitics o ‘he/it.acc.masc.sg’, os ‘they.acc.masc.pl’, a ‘she/it.acc.fem.sg’ and as ‘they.acc.fem.pl’ do not exist in the spoken language; they are only found in written and formal contexts and are only acquired by children in school (Galves 2000, p. 147). Accusative clitic forms exist, however, for first and second person:

(4)	a.	Me	chocou	profundamente.
		I.acc.cl=	it-shocked	deeply
		‘It deeply shocked me (from Duarte 2012, p. 317)
	b.	Te	vi	no	cinema	ontem.
		you.acc.cl=	I-saw	in-the	cinema	yesterday
		‘I saw you at the cinema yesterday’ (from Duarte 2012, p. 317)

In those contexts in which a third-person clitic pronoun would be otherwise expected, two strategies are available to BP speakers: they can either resort to a nominative tonic pronoun (used with accusative function) or leave the object unexpressed (as in (3) above). The choice is subject to the properties of the referent that acts as an antecedent: [+animate] referents must be expressed by a tonic pronoun (5a), while [-animate] referents can be left unexpressed (5b) (cf., Duarte 1989).

(5)	a.	Amanhã,	o meu	filho	viajará	para	São Paulo.
		tomorrow	the my	son	he-will-travel	to	São Paulo
		Eu	deixei	ele/*ø	ir	sozinho.
		I	will-let	he.ton/ø	go	alone
		‘My son is leaving for São Paulo tomorrow. I am letting him go on his own’. (from Lucchesi and Lobo 1996, p. 308)
	b.	Eu	comprei	o	dicionário	e	emprestei
		I	I-bought	the	dictionary	and	I-lent
		ø/*ele	ao	João.
		ø/*it.ton	to-the	João
		‘I bought the dictionary and lent it to João’. (from Lucchesi and Lobo 1996, p. 308)

A null object must be at least one of [-animate] or [-specific]: in the following examples in which both antecedents are [+animate], while the definite object o preso ‘the prisoner’ cannot be null, the non-specific presos ‘prisoners’ indeed can.

(6)	a.	O	policial	insultou	o	preso	antes
		the	policeman	he-insulted	the	prisoner	before
		de	torturar	ele/*ø.
		of	torture	he.ton/*ø
		‘The policeman insulted the prisoner before torturing him’ (from Vasconcellos Lopes and Cyrino 2005, p. 345)
	b.	O	policial	insulta	presos	antes	de
		the	policeman	he-insults	prisoners	before	of
		torturar	ø/?eles.
		torture	ø/?they. ton
		‘The policeman insulted prisoners before torturing them’ (from Vasconcellos Lopes and Cyrino 2005, p. 345)

In the dative, a third person is preferably expressed by para followed by a tonic form: compare (7a) to its European Portuguese (EP) counterpart (7b), which uses the clitic lhe.2

(7)	a.	Vou	perguntar	para	ela.			(BP)
		I-go	ask	to	she.ton
	b.	Vou-	lhe	preguntar.				(EP)
		I-go=	she.cl	ask
		‘I will ask her’. (from Azevedo 2005, p. 236)

The shift has occurred gradually over the past few centuries. In a study examining plays from the 16th to the 20th centuries, Cyrino (1997) identifies a significant decline in the use of a third-person direct object in clitic form: from 89% in the 16th century to 21% in the 20th century. Initially, the null object was linked to a proposition, that is, a non-referential antecedent in the third person. Subsequently, this evolved to null objects referring to a [-animate] element, specifically a third-person, inanimate referent. Notably, animacy has played a fundamental role in dictating the sequence in which different cells of the paradigm are affected by this process of loss.

Cyrino et al. (2000) note that a dual evolution is underway in BP, as it is concurrently transitioning into a non-null subject language while also becoming a null-object language. Although the authors do not explicitly connect these processes, it is evident that both are being driven by the same set of features, albeit manifesting in opposing directions. Items categorized as [+human] and [+specific] tend to be phonetically realized as subjects first, while expletives such as ‘there’, which are [-human] and [-specific], remain null. Conversely, [-human] and/or [-specific] objects consistently remain unexpressed, that is, null. The authors identify this with the so-called Referential Hierarchy (Silverstein 1976; Comrie 1981; among others); we take up the discussion of this in Section 4.

3.2. Raeto-Romance

A number of closely related Raeto-Romance (R-R henceforth) varieties have varying degrees of defective clitic paradigms. As already discussed by Benincà and Poletto (2005, p. 228), the varieties spoken in Brigels and Camischollas have no pronominal clitics at all3: the accusative object in all the following examples is expressed with a form that is clearly tonic because it occupies a post-verbal position (8a); the possibility of it being enclitic on the verb is excluded by the fact that it can be separated from it by a variety of phrases (8b and 8c)4.

(8)	a.	Ša	ti	vol	el.
		if	you	want	it. acc.ton
		‘If you want it.’ (accusative, AIS VI:i, 1110)
	b.	Yu	amfla	netur	el.
		I	find	nowhere	it. acc.ton
		‘I cannot find it anywhere.’ (accusative, AIS VIII:ii, 1597)
	c.	I	an	caciau	giodor	el.
		They	have	chased	away	he. acc.ton
		‘They have chased him away.’ (accusative, AIS VIII:ii, 1667)

These R-R varieties also completely lack a dative series, as well as a partitive (cf., Italian ne ‘of it’) and a locative clitic (cf., Italian ci ‘there’), which have been replaced, respectively, by the preposition ad ‘to’ followed by a tonic pronoun (9a), the preposition kun ‘with’ followed by a tonic pronoun (9b), and a null object (9c):

(9)	a.	Gi	kuai	ad	el.
		say	that	to	he. acc.ton
		‘Tell him that.’ (dative, AIS VI:i, 1110)
	b.	Koy	figesas	cun	el?
		what	would-you-do	with	it. acc.ton
		‘What would you do with it?’ (partitive, AIS VIIII, 1638)
	c.	K	eu	ø	mondi.
		that	I	ø	go
		‘That I go (there).’ (locative, AIS VII, 1345)

There are also R-R varieties that present a defective clitic paradigm, having retained an accusative series (10a, cf. the pre-verbal position), but completely lost the dative (10b), partitive (10c) and locative ones (10d). The process of clitic loss is still underway, and its gradual progression can be observed synchronically. An example of this is the variety spoken in Zernez:

(10)	a.	Ža	tü	il	vowšt.
		if	you	it. acc.cl=	want
		‘If you want it.’ (accusative, AIS VI:i, 1110)
	b.	Di	at	el.
		say	to	he. acc.ton
		‘Tell him.’ (dative, AIS VIII:ii, 1659)
	c.	Ce	fesat	vus	kun	el?
		what	would-you-do	you	with	it. acc.ton
		‘What would you do with it?’ (partitive, AIS VI:i, 1113)
	d.	Voleys	ka	yow	ø	geya?
		you-want	that	I	ø	go
		‘Do you want that I go (there)?’ (locative, AIS VIII:ii, 1638)

Based on this microvariation in the distributional patterns of accusative, dative, partitive and locative forms across this group of varieties, Benincà and Poletto (2005, pp. 227–28) draw a number of synchronic generalizations, and express them by means of an implicational hierarchy:

(11)	a.	If a Romance language has clitics, it has direct object clitics.
	b.	If a Romance language has dative clitics, it has direct object clitics.
	c.	If a Romance language has partitive or locative clitics, it has dative clitics.

Given the premises laid out in Section 2, these synchronic generalizations can be interpreted as the reflection of a process of diachronic change (as suggested, for example, by Andersen 2001, p. 228), as indeed was done by Paoli (2014): the gradualness of grammaticalization is reflected in the synchronic microvariation across Case exhibited by the realisation of these pronominal forms in this group of closely related varieties. We can conclude that, here, the gradualness of grammaticalization is determined by Case: the partitive and locative clitics are affected first by the process of clitic loss, followed by the dative, followed, in turn, by the accusative paradigm. The hierarchy at play is therefore the following (‘>’ indicates diachronic precedence):

(12)	partitive/locative > dative > accusative

Some diachronic evidence adds some further finer distinctions to this Case hierarchy. A collection of texts dating from the 17th century (with some 18th-century amendments (Decurtins 1880–1883, p. 255)) from the R-R variety Surselvan, which, in its modern form, completely lacks a pronominal clitic series and uses instead the reintroduced tonic series, documents the presence of pronominal clitics. This suggests that, indeed, the current situation is the result of a gradual process of loss that has taken place over the last several hundred years and has followed a specific path. A close investigation of these texts, as in Paoli (2014, p. 172), confirms evidence of a Case hierarchy at work; in addition to the dative series having more cells that can be realised as null than the accusative series, it also offers evidence of the subsequent step, in which the original tonic forms are being reintroduced (recall that the cline representing the progression of pronominal loss goes from the clitic form, to zero, to the possible reintroduction of a tonic form). It also shows a further differentiation stemming from Person and animacy distinctions. Taking the third person as a point of reference, while the accusative paradigm can have a null object [-animate] referent in the plural, in the dative series, neither clitic nor null objects are found, and we only see the reintroduction of a tonic form, which represents a more advanced stage along the cline of grammaticalization in (1).

Furthermore, the same texts also show evidence of a further hierarchy based on Person and animacy features. As discussed in Paoli (2014, pp. 169–76), gaps in the paradigm are mostly found with third-person singular [-animate] referents, suggesting that this is the first form to be lost. These are followed by third-person singular [+animate], realised by tonic forms, in turn, followed by third-person plural. The process has not affected first- and second-person singular to the same extent; these are mainly realised by clitic forms, with a few null attestations. The scarcity of data prevents us from drawing any definite conclusions for first- and second-person plural. If considering Case, we see that in the dative paradigm, more cells can be realised as null compared to the accusative paradigm.

Considering the evidence from these 17th-century texts, we obtain the following partial hierarchies of loss for R-R based on Case (13a), Person and animacy features (13b), (‘>’ means, as above, diachronic precedence of loss):

(13)	a.	dative > accusative
	b.	3rd singular [-anim > +anim] > 3rd plural [-anim > +anim] > 1st, 2nd singular

A final, additional reflection comes from the rather fluid situation witnessed in these 17th-century texts, in which clitic form, tonic form and null realisation can be used interchangeably. In (14), from Decurtins (1880–1883, p. 274, lines 5–6), the speaker refers to themselves by means of a clitic form first, followed in turn by a null object (under coordination reduction) and a tonic form. The pragmatics do not warrant the use of a tonic form (i.e., the tonic form mei does not express a contrast nor does it introduce a new referent), suggesting that all three realisations seem to be in free variation.

(14)	Ti	mi	has	giù	ed	ø	has	cha
	you	I.cl=	you-have	had	and	ø	you-have	not
	saviu		tener;	ed	ussa	enqueres	ti	da
	been.able.to		keep	and	now	you-try	you	to
	pigiar	mei.
	keep	I.acc.ton
	‘You had me and were not able to keep [me]; and now you are trying to detain me’.

The pronominal usage witnessed in these texts represents a snapshot of the change (that was still in progress in the 17th century) that Surselvan underwent. In addition to hierarchies based on both Case, Person and animacy, we see a situation of variation not dissimilar to the intermediate stage in cases of semantic change, in which both old and new meanings coexist; the interchangeability of the three realisations suggests that, at this stage, there are no firm boundaries between them, and although they remain morphologically (and possibly syntactically) distinct, their pragmatics distinctive features are blurred. Let us now turn to the northeastern Italo-Romance varieties, in which the interaction of Case, Person, Number and animacy paints a similarly complex picture.

3.3. Northeastern Italo-Romance

In a group of neighbouring Italo-Romance varieties spoken in the Dolomitic area in North-East Italy (described and analysed in Paoli 2009, 2014), the interaction of Case, Person, Number and animacy results in defective paradigms in both the accusative and dative clitic series.

In the accusative paradigm, there are clitic forms for first- and second-person singular and second-person plural5:

(15)	a.	A:	Me	sientes-tu?
			I.acc.cl=	you-hear=SCL
		‘Can you hear me?’
		B:	Sì,	te	sient	benon!
			yes	you.acc.cl=	I-hear	well
		‘Yes, I can hear you loud and clear!’ (from De Lorenzo Tobolo 1977, p. 347)
	b.	Tasè	che	duci	ve	sienti!
		shut-up	that	all	you.acc.cl=	they-hear
		‘Shut up, everybody can hear you!’ (from De Lorenzo Tobolo 1977, p. 362)

For the third person, there are three possibilities, depending on whether the antecedent is animate or not: a clitic pronoun or, marginally, a tonic pronoun, for a [+animate] referent, or no pronoun at all if the referent is [-animate]. Although a [+animate] referent can be left unexpressed, neither a clitic nor a tonic pronoun can be used with a [-animate] referent:

(16)	a.	A:	As-t	vist	Rosa?
			have-you-SCL	seen	R
		‘Have you seen Rosa?’
		B1:	Sì,	l’ei	vista	ngeri.
			yes	she.acc.cl=I-have	seen	yesterday
		B2:	Sì,	ø ei	vistu6	ngeri.
			yes	ø I-have	seen	yesterday
		‘Yes, I saw her yesterday’ (from Paoli 2014, pp. 148–49)
	b.	A Maria	piasi	la pasta	e ø mangia	ogni dì.
		to M	it-pleases	the pasta	and ø she-eats	every day
	b’.	*A Maria	piasi	la pasta	e la	mangia
		to M	it-pleases	the pasta	and she.acc.cl=	she-eats
		ogni dì. every day ‘Maria likes pasta and she eats (it) every day’ (from Paoli 2014, p. 149)

A similar situation is found in the dative series, albeit with a higher level of deficiency in the paradigm; the only clitic forms that are still found and in use are for the second person, both singular and plural, and, marginally, for the first-person singular. For all the other cells, either a tonic form preceded by the preposition a ‘to’ or a null pronoun is used, the former being the most frequent choice. A summary overview of the accusative and dative clitic series is provided in

Table 1. Accusative and Dative clitic series.

	Accusative	Dative
1st sg	me	me/ø
2nd sg	te	te
3rd sg	la/lu [+anim]; ø [±anim]	ø
1st pl	ø	ø
2nd pl	ve	ve
3rd pl	?li [+anim]; ø [-anim]	ø

:

Against the premises laid out in Section 2, the incomplete clitic paradigms, with the dative series being more deficient than the accusative one, suggest that a process of clitic loss is ongoing, and that it is not uniformly targeting accusative and dative paradigms; once again, we see that there are some hierarchies at work, based on Case and Person. As for Case, the more numerous gaps in the dative paradigm suggest that the dative series of clitics has been affected first; the process of loss extended only subsequently to the accusative paradigm, producing a synchronic asymmetry that is very similar to the situation observed by Benincà and Poletto (2005) for the R-R varieties.

Drawing on the very restricted and, admittedly, not completely reliable diachronic evidence that exists for these northeastern Italo-Romance varieties, there is support for the interpretation of the synchronic incomplete paradigms as the result of a process of clitic loss. Despite these local varieties having been officially introduced in the drawing up of legal documents as early as 1631, instances of clitic pronouns in these texts cannot be found, as the legal writing style favours the repetition of the whole noun phrase rather than its replacement with a pronominal form. There is, however, a religious text translated possibly from Italian (or even Latin) that dates from around 1870, the Passion according to Saint Matthew (Tagliavini 1932), that offers a view of some atonic form usage. The text displays clitic forms for third-person singular and plural in the accusative paradigm, for both [+animate], (17a), and [-animate], (17b), referents; third-person singular dative clitics are used occasionally (but inconsistently, cf. 17b, 17c), suggesting a fluid situation in which the form was probably falling out of use and was being replaced by a null realisation. The third-person plural dative is only found as a null realisation (17c), while for the accusative, there is a clitic (17d). The first-person plural dative is realised as a tonic pronoun preceded by a preposition (17e).

(17)	a.	Allora	duggi	i discepui	i	lu	ha
		then	all	the disciples	SCL	he.acc.cl=	they-have
		impianteu	vilò
		abandoned	there
		‘Then, the Disciples abandoned him there.’ (Ch. XXVI, 56)
	b.	E	Gesù	el	gh’	rispondi:	T
		and	Jesus	SCL	he.dat.cl=	he-answers	SCL
		l’	has	dit	tu.
		it.acc.cl=	you-have	said	you
		‘And Jesus replied: it is you who says it.’ (Ch. XXVI, 64)
	c.	Giuda Scariott	l’	è	du	a ciattà	i Capi
		Judas Iscariot	SCL	he-is	gone	to find	the chiefs
		dei	pré,	e	l’	ø	ha dittu …
		of-the	priests	and	SCL	ø	he-has said…
		‘Judas Iscariot went to find the High Priests and said [to them] …’ (Ch. XXVI, 14)
	d.	I	se	ha	divisu	li me	massari,
		SCL	reflex.	they-have	divided	the my	clothes
		e	i	li	ha	tiredi	alla sorti
		and	SCL	they.acc.cl=	they-have	thrown	to-the fate
		‘They shared my garments and cast lots for them.’ (Ch. XXVII, 35)
	e.	Giuda	l’	ha	purteu	ndoi	i trenta sodi
		Judas	SCL	he-has	taken	back	the thirty coins
		ai	Prenzipi	dei	Sazerdoti	ma	lueri
		to-the	Princes	of-the	Priests	but	they
		i	ø	ha	respondù	Che	importa
		SCL	ø	they-have	answered	what	matters
		a	nujetar?
		to	we.ton
		‘Judas returned the thirty pieces of silver to the High Priests, but they answered [to him]: What does it matter to us?’ (Ch. XXVII, 3–5)

Although the evidence afforded by these texts is not definitive, compounded with the synchronic facts, it forms a coherent picture; the process of clitic loss is ongoing, and, once again, we see that there are some hierarchies at work, based on Case, Person and animacy. As for Case, both synchronic and diachronic data indicate that the dative series of clitics was affected first; the process of decay extended only subsequently to the accusative paradigm, yielding a synchronic pattern that is very similar to what reported by Benincà and Poletto (2005) for the R-R varieties. Furthermore, we see a Person hierarchy at play, since first-person plural, followed by third person, are the cells that are most open to clitic loss. The complete absence of the clitic corresponding to the first-person plural suggests that this is the form that was targeted first by the process of clitic loss; this is, from a cross-linguistic perspective, an unusual fact (see Farrell 1990, pp. 329ff), as typically a null-object pronoun has a default interpretation of third person. Interpreting this fact from the perspective of a process of loss means that it is usually the third-person pronoun that is the first one to disappear. In view of this generalization, Paoli (2009, 2014) considers the first-person plural to be an outlier and discusses possible reasons why the first-person plural cell should be the one to show the highest vulnerability (Paoli 2009, pp. 78–79; 2014, pp. 165–66). The tentative explanation that is put forward there is based on the incompatibility between the semantic load of the composite form that is found in these varieties, a reflex of Latin NOS + ALTER (an exclusive ‘we’, which includes the speaker but excludes the hearer) and the weakness of a clitic form. Although no firm explanations can be reached, partly because of the scarcity of relevant evidence, a plausible conclusion that can be drawn is that whatever is happening to the first-person plural is not the result of the same process that is affecting the other cells of the paradigm. By removing the first-person plural pronouns from the equation, the picture that emerges is not only more familiar, but also more congruous and meaningful: the process of decay affects the third person first, and within it, [-animate] referents are more readily expressed by a null form.

Summing up, the defective clitic paradigms in these northeastern Italo-Romance varieties reveal that two separate hierarchies determine the order in which cells are affected: the first one, based on Case, is represented in (18a); the second, based on Person, in (18b). Diachronic precedence of loss is, once again, expressed as ‘>’:

(18)	a.	dative > accusative
	b.	3rd sg [−anim > +anim] > 3rd pl [−anim > +anim] > 1st, 2nd sg > 2nd pl

Let us now turn to the final case study, colloquial French.

3.4. Colloquial French

Although French is not typologically classified as a null-object language, corpus-based research carried out in the last forty years or so (cf., Fónagy 1985; Lambrecht and Lemoine 2005; Larjavaara 2000, among others) has revealed systematic usages in which referential direct objects are not lexically realised:

(19)	a.	A:	J’ai	un truc	pour toi	si ça	t’intéresse
			I-have	a thing	for you	if this	interests you
		‘I have something for you if you are interested’
		B:	C’est	quoi?
			it-is	what
		‘What is it?’
		A:	Je crois	que	t’aimes	bien,	toi,
			I-think	that	you-love	well	you
			ce genre	de truc.	J’ai ø	trouvé	hier.
			this type	of thing	I-have ø	found	yesterday
		‘I think that you like this sort of thing. I found (it) yesterday’ (from Lambrecht and Lemoine 2005, p. 39)
	b.	Un jour,	je me disais,	je mettrais	une petite	annonce	dans
		one day	I-told myself	I-will-put	a small	ad	in
		Le Provençal:	[…]Mais	je ø	renvoyais	toujours	à plus tard.
		LP	but	I ø	I-put-off	always	until later
		‘One day, I told myself that I would put a classified ad in Le Provençal […] But I kept putting (it) off until later’. (from Larjavaara 2000, p. 63)

Despite facing significant social stigma, this construction remains entirely viable in both casual written and spoken French. Its usage is governed by pragmatic, discursive and stylistic considerations, as discussed in works such as Cummins and Roberge (2005) and Lambrecht and Lemoine (2005). In their investigation of null objects in French, Cummins and Roberge (2005, p. 52) observe that the null object in instances such as those in (19) in which there is no overt clitic is always definite and referential; Lambrecht and Lemoine (2005, p. 14) note that the ‘null instantiation of object pronouns is denoting specific discourse entities’. Null objects are, therefore, always definite, referential, specific, discourse-accessible and recoverable from the context; that is, they must be ‘both active in the minds of the interlocutors at the time of utterance and […] must have the status of an established discourse topic’ (Lambrecht and Lemoine 2005, p. 42).

In addition, the null realization of an object is also dependent on further features of the referent, that is, Person and animacy: only third-person [-human] referents can be realised as null objects.

An interesting parallel emerges when considering the minimal pair in (20):

(20)	a.	J’	ø	aime.
		I	ø	I-love
		‘I love (it)’.
	b.	Je	l’	aime.
		I	s/he.acc.cl=	I-love
		‘I love him/her’.

In any given context, (20a) can only refer to an inanimate referent that is discourse-accessible and recoverable from the context (for example, referring to a dress that a friend is trying on and showing the speaker, asking them for their views), while the object clitic l’ in (20b) can only be used to refer to an animate antecedent (as in a conversation that is asking the speaker’s opinion on a common friend).

The picture that emerges is, by now, very familiar; similarly to what we have already seen for Brazilian Portuguese and the northeastern Italo-Romance varieties, in colloquial French, too, it is third-person, [-animate] referents that can be left unexpressed. Assuming the very initial stages of a possible process of clitic loss, it can be seen clearly that the development is driven by a hierarchy based on animacy and probably Person.

4. Comparative Discussion

The investigation of the course of clitic loss across Brazilian Portuguese, northeastern Italo-Romance and colloquial French has brought to light strikingly similar patterns. Third-person cells are those that repeatedly appear to be the most vulnerable to clitic weakening and loss, with [-animate] referents preceding [+animate] ones in the process. The dative series, too, as witnessed in R-R and northeastern Italo-Romance, is affected by loss before the accusative series. Of the two trends, the hierarchy based on Person and animacy is the common denominator across all the varieties investigated. The hierarchy based on Case, on the other hand, is not immediately apparent; considering French as a manifestation of the incipient stage of the process of loss, we can speculate that a Case hierarchy may linearly follow (in terms of time) a Person and animacy one, or perhaps that they are to be interpreted as individual strands of a complex rope in which a number of hierarchies are simultaneously at play (much on the same line, for example, as Givón 1976, p. 52, or 2017, p. 91, who identifies a number of semantic hierarchies that are subsumed under a grammatical one based on grammatical function, closely reminiscent of Keenan and Comrie’s 1977 Noun Phrase Accessibility Hierarchy); the choice between these two possibly being determined by other properties of the language.

It was mentioned earlier that Cyrino et al. (2000) identify two unrelated changes that are concomitantly taking place in BP, a shift towards becoming a non-null subject and a null-object language, and convincingly make a case for both processes operating along the Referential Hierarchy (Silverstein 1976). Reflecting on the animacy hierarchy, Comrie (1981, p. 192) identifies the link between the natural world, grammatical features and related discourse properties: ‘the animacy hierarchy […] reflects a natural human interaction among several parameters, which include animacy in the strict sense, but also definiteness […] and various means of making an entity more individuated—such as giving it a name of its own, and thereby making it more likely as a topic of conversation’, that is, more accessible. Keenan and Comrie’s (1977) Noun Phrase Accessibility Hierarchy, although formulated to capture the limitations on the availability of a noun phrase to being relativised, offers a ranking of grammatical roles that finds some correspondence in many other linguistic phenomena. As already argued by Paoli (2014, p. 179), these types of hierarchies, referential ones in particular, are an effective way to outline limitations on the morpho-syntactic behaviour of arguments. The limitations observed across languages regarding the positioning of noun phrases, and clitic and tonic pronouns emphasize the significance of referentiality in pronoun usage, whether expressed overtly or left null in languages that allow such possibilities. Similarly to what has already been noted by Comrie and Givón, referentiality emerges from a combination of factors: the nature of the nominal phrase, its argument status, Case, Person, animacy, ability to reference real-world entities, and specificity. These features interact to establish a scale against which linguistic elements are assessed, creating, specifically, what is known as the Referential Hierarchy. At one end of the scale cluster are non-specific, non-referential, inanimate elements; at the other, specific, referential, human ones. Due to their inherent human and definite nature, first- and second-person pronouns rank higher than third person and are deemed ‘marked’. Among third-person pronouns, those categorized as [-human] naturally hold a lower rank than those marked as [+human], representing the most ‘unmarked’ forms. It seems to make intuitive sense that those linguistic forms that are least marked should be more accessible to a process of loss. Conversely, those forms that are richer in this respect, would be more resilient and disappear at a later stage (if at all). Some evidence that can lend support to this reasoning and the cognitive relevance of hierarchies in general, can be derived from preliminary neurolinguistic studies that suggest that referential hierarchies may hold the key to universal aspects of human cognition by playing an important role in language processing (e.g., Philipp et al. 2008; Wang et al. 2009). It is these general cognitive operations that, we claim, are ultimately responsible and determine the degree of affectedness, and, consequently, the order in which different parts of a construction undergo a given change.

In the Introduction, the notions of gradience and gradualness were discussed, and some keynote features were highlighted. For gradience, these were the structured nature of synchronic variation (Rosemeyer 2014, p. 78), its systematicity (Andersen 2001, p. 26), and the determinate direction of the change that it feeds into (Andersen 2001, p. 225). As for gradualness, it was defined as change that involves a series of micro-changes (Traugott and Trousdale 2010, p. 23) that progress in an orderly and grammatically conditioned fashion (Andersen 2001, p. 226). The link between these two notions explains how gradual change, which operates in a step-by-step manner, is manifested in synchronic variation that is systematic and structured. With respect to gradience, the concept of ‘typicality’, that is, a distinction between more core and more peripheral members of a given category, was also underlined; considering the instances of clitic loss under discussion, and the observations about first and second person, on the one hand, and third on the other, in Benveniste (1971, p. 217), Forchheimer (1953, pp. 5–6) and Erteschik-Shir (2007), we would conclude that while first- and second-person pronouns are more typical exponents of the category pronoun, third would be more peripheral. This distinction would then be relevant to the fact that third-person pronouns are more vulnerable to the process of loss; the least marked, or less typical, members of the category would be lost before the more marked, or more typical, ones. The understanding that general principles encoded in the Referential Hierarchy and the Noun Phrase Accessibility Hierarchy are the drivers of change, providing clear step-by-step paths of direction of change, explains the structure and systematicity of the synchronic manifestation of that change, and possibly even why less typical members of a category are more vulnerable to a process of loss.

In terms of gradience as the representation of microvariation within and between languages, the choice of genetically related languages has enabled a finer-grained observation of the modalities of action of the different grammatical and semantic properties that make up nominal elements such as Person, animacy, Case and specificity. The absence of so-called ‘background noise’ may have revealed an order of applicability, so to speak, that affects the way Case interacts with the other features with more clarity than would have been possible by adopting a classical typological approach, comparing genetically distant languages. The recurring patterns in the uneven realisation and spreading of the process of clitic loss unearthed in this article highlight the power of general (possibly even universal) principles, expressed in terms of both syntactic and semantic features whose operation can be captured in the form of an ordered scale. Irrespective of how exactly hierarchies are formulated and labelled, by representing properties that transcend the specific, both in terms of language-specific and phenomenon-specific, they can explain morphosyntactic patterns and the way they are actualised across space and time.

5. Conclusions

Starting from Traugott and Trousdale’s explicit connection between gradualness of change and synchronic gradience, this study set off to explore what may lie behind cases of uneven change. Taking as its object of research a series of cases of clitic loss in a set of Romance varieties, the investigation has brought to light a common denominator, that is, the role played by Person, animacy, specificity and Case in the process of loss. All these are properties that find an expression in the Referential Hierarchy and the Noun Phrase Accessibility Hierarchy, which may be possibly collapsed into the grammatical function hierarchy subject > direct object > oblique (Givón 2017, p. 91). In addition to being responsible for the process of loss itself, as already claimed by Paoli (2014), this article has made a case for the same general aspects of language to be also responsible for the gradualness of change, its determinate direction, and an explanation for the systematicity and structure of its synchronic manifestation. This is particularly evident when comparing diachronic change in closely related languages.

Although the link between the synchronic deficiency of the clitic paradigms of the Romance varieties investigated here and the gradual process of loss is quite clear in this particular case, Traugott and Trousdale conclude that not all instances of synchronic gradience are the result of grammaticalization (Traugott and Trousdale 2010, p. 39). Further research on a wider range of phenomena in a larger number of languages (both typologically distinct and genetically related) is needed in order to explore criteria that may be used to establish whether synchronic gradience is the result of grammaticalization or not. There is also a need for a deeper understanding of the nature and properties of hierarchies in general. However, it seems that the application of hierarchies and, by reflex, of general cognitive principles, may be a promising way to understand the essence of language change.