Syntactic change in diachrony versus contact-induced change: two sides of the same coin?

3.1 A brief comparison

Syntactic change, and language change more broadly, might well constitute one of the most baffling facts for our linguistic theories. Syntactic acquisition simply amounts to the “transmission of syntactic systems from one generation to the next” (Willis 2017: 491), which implies that any given system must have been acquirable, and in fact fully acquired, by previous generations. We therefore need to explain why and how the transmission of a given system substantially breaks down at one of its given iterations. A variety of factors have been proposed over time to account for the observed changes: these may be reduced to factors endogenous to the (part of the) system that undergoes change (spontaneous innovation) and to a heterogeneous host of exogenous factors, which can all be ultimately traced back to some fundamental mutations affecting the acquisition process and its original conditions. These include, among others, specific input-related alterations (e.g. whether the input is timely or delayed; and whether it is sufficient or restricted), as well as the possibility that the grammatical system undergoing change exists side by side with one (or more) additional systems.^[2] While the latter class of factors is clearly relevant to research in the field of contact linguistics and, more recently, of heritage linguistics, diachronic strands of research have honed in on the investigation of endogenous factors.

This considerable shift in focus becomes apparent as soon as we seek to compare the main insights into CID and CIC as presented in the relevant literature. Universals of language change (Haspelmath 1999; Hopper and Traugott 1993; Lehmann 1993; Meillet 1912) and the directionality of change (Kiparsky 1968; Hopper 1991; Keller 1994; and more recently Newmeyer 1998; Campbell 2001; Traugott 2001; Haspelmath 1999, 2004) are central to CID-studies. Generally speaking, the key process on which many descriptions of CID concentrate is simplification (or loss) (see Haspelmath 1999 for an overview), which is very often expressed in terms of grammaticalization. Grammaticalization, like its opposite, degrammaticalization (Norde 2009), takes place along a path; the identification of this path makes diachronic change relatively predictable. Theoretically oriented work on CID also focuses mostly on the grammar-internal causes and mechanisms of language change (Roberts and Roussou 2003; Roberts 2007; Roeper 1993; Lightfoot 1991; Kroch 1994; Lightfoot and Westergaard 2007; Westergaard 2008, 2011; Lohndal and Putnam 2024, a.o.).

In contrast, typological studies (see Koptjevskaja-Tamm 2010 for an overview), especially those focusing on language change in multilingual contexts (second language acquisition, heritage linguistics, creole linguistics), have highlighted the fact that CIC seems to be much more unpredictable than CID, since the former, unlike the latter, does not seem to follow a similar specific path. This is commonly assumed to stem from the observation that CIC results from the interplay between grammatical and facilitating factors that are external to the grammar. For instance, Aikhenvald (2006) provides a detailed list of factors that have an impact on lexical or grammatical borrowing in contact situations. Among the grammatical factors are pragmatic salience, tendency to achieve word-for-word inter-translatability, the existence of a perceivable gap in a paradigm, typological naturalness, pre-existing structural similarity, the existence of a lookalike; among extra-grammatical factors, she lists the degree of knowledge of each other’s languages (lingualism), the kind of contact languages are in, the attitudes of the speakers toward the languages, whether contact is balanced (i.e. the languages have an equal social status) or displacive, complete or incomplete language acquisition, and many others.

This complex picture for CIC is considered by some to significantly hinder our collective predictive capabilities, to the extent that, while simplification is commonly seen as the hallmark of diachronic change, both simplification and complexification are attested in contact settings.^[3] Any restrictive, formal model of grammar is capable of accounting for grammatical representations that ‘shrink’ and ‘expand’; however, what is desired, if attainable, is a systematic account pertaining to when and why these changes may occur. More acutely, the formal challenge is to deduce whether or not properties of composition of linguistic representations or underlying cognitive factors promote or hinder these instances of ‘shrinking’ and ‘expanding’ in both CIC- and CID-contexts (Aboh 2015; D’Alessandro and Frasson 2023; Pakendorf 2022).

3.2 A methodological challenge

At this juncture, it is worth mentioning that the very assumption that CID and CIC should be considered separate processes (as one could rightfully infer from the dichotomic literature on syntactic change) is not obvious. Languages, by default, are constantly undergoing some degree of change: this is true even when looking at the language development of a single speaker across their lifespan (see Sankoff 2018 for a general overview; see also D’Alessandro et al. 2025; Putnam and Natvig 2025 for reflections that also appeal to the notion of language contact across the lifespan). Furthermore, no language is an island, and in fact every language is usually in contact with many others; contact can well last for several centuries. The most remarkable case in point is provided by linguistic areas (Sprachbünde, Trubetzkoy 1930; for an overview, see van Gijn and Wahlström 2023), but this observation applies more generally; in fact, as noted by Bowern (2008), first language acquisition itself could rightfully be viewed as an instance of language contact, that between the child and the speakers that provide them with linguistic input. As a result, operationally disentangling CID from CIC is virtually impossible.

If we consider what we can call the extreme cases of CID and CIC, for instance by examining languages that have been in isolation for many centuries, like Icelandic on one side, and languages emerging from very intense and time-restricted contact, like creoles on the other, we can consider CIC as a separate phenomenon from CID. However, what we cannot control, even in these cases, are the socio-historical factors that might enhance or cause the change. Ideally, the same situation needs to be replicated in many possible contexts, through changing the dyads involved, and through the parallel observation of the language change in isolation. This is possible only in unique circumstances, i.e. when multiple instances of contact take place between the same language dyads at the same time and under the same socio-historical circumstances. One such instance is the case of Italo-Romance heritage languages in the Americas: after WW2, a massive emigration from Italy to North and South America took place. The emigrants were mostly monolingual speakers of one Italo-Romance variety (but not Italian), they left at the same time and relocated to a far-away land, severing their contacts with Italy. This has given a unique open-air laboratory to control for some features in contact with multiple Romance varieties (see Andriani et al. 2022a for a detailed description of the sociolinguistic conditions of the various communities).

The microcontact methodology (D’Alessandro 2015, 2021), which consists in isolating one specific feature and checking its development in contact with multiple other languages of the same family, has been designed precisely with the aim to disentangle various possible co-occurring factors affecting the outcomes of change; in particular, we employ it to ascertain whether an attested instance of syntactic change should be attributed to contact or not. The basic idea is to consider nearly identical grammars (ideally, genetically related languages) differing in only one feature (call it X) in a given domain, and observe the development of that feature in a given grammar (call it A) in contact with all other grammars of the set, and in that grammar in isolation. If the feature X of grammar A shows divergent paths of change in contact with different grammars, as well as in isolation, we can attribute its change to contact (factoring out all the possible sociolinguistic elements that could have an impact on this change). If the development is identical in contact with all several different grammars, as well as in isolation, we attribute this change to an endogenous cause. This approach is shown schematically in Figure 1.

Figure 1:

The microcontact methodology.

In what follows, we will primarily examine data elicited with the microcontact methodology as applied within the Romance family. Two notes are in order: firstly, as remarked above, languages are hardly ever isolated. Despite never having been completely isolated, Romance languages exhibited a good degree of isolation in the past; more importantly, they display some macroparameters (following Ledgeway’s 2012 classification) which are shared by all or most languages in the family. We take these common features to be an excellent source of information concerning endogenous change. As an example consider auxiliary selection. Latin did not have auxiliary selection in the way that Romance languages have it: it used be for all periphrastic forms of the verb (chiefly: passives); have was instead not used as an auxiliary. On the contrary, all present-day Romance languages have some form of have as auxiliary (with the exception of Portuguese, which has ter ‘to have’, but from the original Latin tenere ‘to hold’): given its presence across Romance, we regard this as an internal development.

Secondly, as it often happens with heritage language data, it was not always possible to find exact minimal pairs, due to the enormous variability in HL. Furthermore, despite all our efforts, we were not always successful when trying to minimize the differences between speakers and their social and linguistic background. The profile of some heritage languages is such that it does not allow to find large an uniform groups of speakers (see also Rothman and Treffers-Daller 2014; Leivada et al. 2019; D’Alessandro et al. 2021). Despite all the limitations of this domain of research, some solid generalizations can be drawn; we will discuss those, with the disclaimer that the data are at times fragmented or incomplete.

In this paper we investigate both the validity of this approach and the robustness of its coverage with respect to some morphosyntactic phenomena.

3.3 A new heuristic

Kupisch and Polinsky (2022) take a similar comparative stance to assess whether CIC and CID are underlyingly the same: concretely, they examine the grammaticalization path of demonstratives and numerals into articles both in heritage languages, one of the possible change in contact scenarios, and in diachrony. Based on their survey, they conclude that CIC and CID are essentially the same, as the syntactic patterns of change found in contact are the same as those that are found in diachronic change. Further, they argue that our knowledge of diachronic change (and, more specifically, of grammaticalization patterns) enables us to make predictions about syntactic change in contact, the only difference possibly being the pace of the relevant change, as CIC amplifies the effects of CID (to paraphrase them). More directly, syntactic change in contact can essentially be equated to accelerated diachronic change.

The thrust of our argument is that, although the position staked out by Kupisch and Polinsky (2022) holds for a number of cases, the overall picture is admittedly more complex. As a result, the proposal requires refinement. Concretely, we show that there are cases in which CIC cannot be straightforwardly reduced to accelerated diachronic change and, in turn, cannot be fully predicted on the basis of the attested outcomes of diachronic change. Instead, we show that different types of grammatical elements (that is to say: features) react differently to the pressure of change, and therefore they need to be considered separately when examining change. We substantiate this proposal by reviewing different syntactic phenomena: pronouns and demonstratives, Differential Object Marking (‘DOM’), and subject clitics. While the former two phenomena are purely structural, grammatical phenomena, the latter two can be classified as being ‘interface phenomena’, in connection with certain discourse-pragmatic requirements that they need to obey. We show here that Kupisch and Polinsky’s conclusions hold for the first domain, but not for the second one, where the outcomes of syntactic change in contact and of syntactic change in diachrony exhibit divergent paths of development (see also Sorace 2011 for similar considerations on the different nature of phenomena yielding different results in contact/L2 situations).

We relate this difference across domains to different third-factor strategies handling the relevant grammatical elements at play. Specifically, we propose that changes that affect ϕ-features hinge on a cognitive bias towards monotonic computations (Terenghi 2023b) which ensures that their change in contact converges with their change in diachrony, allowing for full predictability of such changes. In contrast, changes that affect discourse-related features, and in particular elements that are related to topicality, or linking with discourse, show important divergences across contact and diachrony. It is therefore possible to predict language change for purely syntactic features; however, this predictability is hindered in contact situations involving different kinds of features, such as semantic or discourse-related features. In case these types of features are involved, the output of CIC might be radically different from that of spontaneous CID, which we take to be monodirectional and predictable, when clarity of initial conditions exists. In a sense, this article is an integration of what Sorace (2011) and all subsequent work on the Interface Hypothesis, up to the recent Hoot and Leal (2023) have already pointed out about contact. Our contribution consists in checking the claims of “predictability” against what is a clearer source of predictable change, i.e. endogenous diachronic change.

4.1 Pronouns: no change

In diachrony, the organization of pronominal paradigms is mainly stable. This has been perhaps most famously stated by Nichols (1992), who concluded, on the basis of investigations carried out over a typologically diverse and balanced linguistic sample (n = 167), that the encoding of the inclusive/exclusive distinction (that is, both its presence and its absence) in pronominal paradigms is remarkably stable genetically: “pronouns and pronoun categories tend to be conservative in families” (Nichols 1992: 123–124). Stated differently, pronominal paradigms that originally displayed three distinct person categories (the speaker, the hearer, the other(s)) typically retain their ternary partition through the diachronic development of the given language; likewise, paradigms that originally displayed four different person categories (i.e., that make the inclusive-exclusive distinction in the first person) also typically maintain this quaternary partition through time.

As a concrete example, consider Table 1. As Table 1 shows, in spite of the apparent degree of morphological differentiation across the two varieties, Campidanese Sardinian fully preserves the ternary partition of the Latin pronominal system: that is, in both varieties, three persons are contrastively encoded.^[4]

Likewise, pronominal paradigms are generally held to be largely stable in contact settings (Heine and Kuteva 2005; Matras 2009; a.o.). In these contexts, the occasional (lexical or structural) borrowings may give rise to a new category or to the loss of an old one (that is, to change from ternary to quaternary systems or from quaternary to ternary systems, respectively), but such changes are considerably rare, as discussed by Siewierska (2004: chapter 7.3).

The stability of the indexical oppositions encoded in pronominal paradigms in contact settings can be shown by considering creole languages. Data relative to some of these varieties are collected in the Atlas of Pidgin and Creole Language Structures (APiCS; Michaelis et al. 2013). Relevant for the present discussion is APiCS’ feature 15, Inclusive/exclusive distinction in independent personal pronouns (Haspelmath et al. 2013), which surveys the structure of the pronominal paradigms of 75 creole languages. For our purposes, these should be compared with the pronominal paradigms of the respective lexifiers (i.e., the variety that provided the lexicon) and substrates. Out of 75 creoles, 64 straightforwardly retained the structure of the pronominal paradigms of their lexifiers (a ternary system in 61 cases, as shown in Table 2 for Portuguese-based Korlai; a quaternary one in the remaining 3 cases) and 10 more retained the distinctions made in their substrates (5 creoles display the inclusive/exclusive distinction, although their lexifier does not; 5 creoles do not display the inclusive/exclusive distinction, although their lexifier does). Only one creole (Sranan) introduced the inclusive/exclusive distinction from scratch, that is, without having it in its constitutive pool of features.

In summary, and again barring orthogonal lexical or morphological differences,^[5] the structure of pronominal paradigms in contact contexts such as in creole languages is quite stable, with an overall preference for the preservation of the distinctions originally encoded in the lexifier and, to a lesser extent, the adaptation of the lexifier’s paradigm to accommodate the paradigmatic structure of a substrate variety. Only very rarely does the pronominal system change by drifting away from its pool of features.

The foregoing showed that, in general terms, the indexical oppositions encoded in pronominal paradigms tend to be stable both across time and, albeit slightly less so, in contact settings. In the latter case, however, departures from the original paradigms are commonly explained as an effect of substrate influence. With this disclaimer in mind, it can be safely concluded that there is little to no room for change in pronominal paradigms and, in this sense, diachrony and contact behave in a similar fashion.^[6]

4.2 Demonstratives: CID and CIC have the same outputs

Let us now turn to a different class of indexical elements; namely, demonstrative systems. Although these are traditionally understood as spatial indexicals (Bühler 1934; see also Lyons 1977; Levinson 1983: chapter 2; Fillmore [1971] 1997; Diessel 2012; a.o.), in what follows we regard them as ultimately akin to person indexicals (such a personal pronouns) by virtue of their being derived by the same set of primitives, i.e. person features (Bjorkman et al. 2019; Harbour 2016; Terenghi 2023b).^[7] More specifically, this section reviews the patterns of change attested by ternary demonstrative systems, that is: demonstrative systems that convey three-way person oppositions, contrastively encoding proximity to the speaker (‘this near me’), proximity to the addressee (‘that near you’) and non-proximity to either (‘that far (from us)’).

These systems can be unstable in diachrony and in contact (heritage varieties, creoles) alike and, crucially, show the same patterns of change across the two domains (Terenghi 2023a, 2023b). Concretely, change in this domain amounts to simplification: both in diachrony (Table 4a) and in contact (Table 4b), ternary demonstrative systems tend to lose the contrastive hearer-oriented semantics (‘that near you’).

As both cases in Table 3 show, the contrastive semantics of the hearer-oriented demonstrative form (‘that near you’) is lost and that deictic domain is ultimately merged with the non-participant oriented one (‘that far’). This new, extended deictic domain, which substantially carries non-speaker-oriented semantics (‘that not near me’), is spelled out by the original non-participant oriented form, as exemplified by innovative Occitan varieties and by Batavia Creole in Table 4a and 4b respectively.^[8]

Table 3:

Demonstrative systems in diachrony (3a) and in contact (3b).

(a) Occitan demonstratives (Ledgeway and Smith 2016)
	Near 1	Near 2	Far
Occitan (old)	aqueste	aiceste	aquel
Occitan (new)	aqueste
(b) Portuguese-based Batavia Creole (Maurer 2013)
	Near 1	Near 2	Far
Portuguese	este	esse	aquele
Batavia Creole	iste

Comparable simplification patterns can also be identified when investigating heritage languages. Evidence for this comes from a study on demonstrative systems in heritage Sicilian and Abruzzese (Italo-Romance) varieties (Terenghi 2022). The homeland counterparts (Sicilian and Abruzzese in Italy) of the heritage varieties included in that study display a three-way deictic opposition in the demonstrative system. The heritage varieties at issue were investigated in Argentina, in Quebec, and in the United States, thus with different majority/dominant languages: respectively, Spanish, with a three-way demonstrative system comparable to the one of the homeland varieties; French, with a (reduced) two-way demonstrative system, comparable to the simplified systems discussed in Table 3; and English, with a two-demonstrative system as well. By testing heritage speakers on both production and comprehension, Terenghi concluded that the demonstrative systems of the heritage varieties under investigation are undergoing a reorganization and, more specifically, are moving towards the loss of the contrastive hearer-related semantics and therefore to a two-way opposition. However, and interestingly, this change cannot be traced back to transfer from the dominant languages, as the patterns attested in the heritage varieties do not consistently align with those shown by the respective dominant languages.

Taken together, these case studies show that the indexical oppositions encoded in demonstrative systems tend to be unstable, both in diachrony and in contact, and more specifically to undergo simplification whereby original three-way distinctions are reduced to two-way distinctions. Importantly, the same patterns of simplification are attested across CID and CIC, corroborating our conclusion that diachrony and contact behave in a similar way in the domain of indexicals.

4.3 Interim summary

In sum, we can conclude that CID and CIC display parallel outcomes in both pronominal and demonstrative systems. While in the former case no change is attested, in the latter comparable patterns of change are attested across the different domains (CID and CIC). This conclusion is in line with the idea that CID and CIC are two faces of the same coin, with CIC amounting perhaps to accelerated CID, as argued for by Kupisch and Polinsky (2022). However, we maintain in this paper that this conclusion is skewed by the specific phenomenon we are examining: as discussed in opening this section, personal pronouns and demonstratives can be regarded as bundles of ϕ features, thus as essentially syntactic items. As such, if they undergo change, they do so in line with the cognitive bias towards monotonic computations discussed in Section 3, which generally affects and constrains the outcomes of language acquisition. In the specific case of pronominal and demonstrative systems, the only possible outcome is the simplification of the system (attested for the latter) and, more concretely, the loss of the hearer-oriented semantics, as argued by Terenghi (2023b); this holds whether the languages examined are considered in isolation (but see again the discussion in Section 3.2) or in contact.

As we will show in the remainder of this paper, if we extend our discussion to phenomena which involve more knowledge of discourse (interface phenomena), the matter is considerably less straightforward and significant mismatches can be identified between the outcomes of CID and CIC.

6.1 Subject clitics and ϕ-features: CIC equals CID

The diachronic development of SCls represents an instance of the subject agreement cycle, in which full pronominal expressions are reanalyzed as agreement markers and eventually fall out of use, first identified by Givón (1976). More recently, van Gelderen (2009, 2011) has provided a minimalist account for the subject agreement cycle in terms of Feature Economy:

Concretely, the interpretable ϕ-features of full pronouns are reanalyzed as uninterpretable ϕ-features during acquisition, leading to their realization as agreement markers; these may subsequently undergo loss altogether:

This formalization is consistent with the diachronic literature on SCls in northern Italo-Romance varieties. Concretely, and as already argued in the foregoing section, SCls in northern Italo-Romance varieties can be regarded as syntactically akin to agreement markers, unlike full pronouns.^[13] Further, the diachronic emergence of SCls following the reanalysis of erstwhile independent pronouns has been carefully traced in a large body of research for a diverse set of Italo-Romance varieties (Poletto 1995; Renzi 1992; Parry 1993; Vanelli 1987; Spiess 1956; Vanelli 1984; among many others). Given our focus, it is worth pointing out here that Poletto (1995) shows that northeastern Italo-Romance SCls originated from a set of independent pronouns between the 16th/17th century.

Against this rich background, substantially less research has focused on the final part of the cycle, namely the diachronic loss of SCls. Before proceeding, a disclaimer is in order: cliticization of erstwhile free subject pronouns targeted different pronouns to different extents, and possibly not reaching a full paradigm of SCls. This state of affairs was discussed, for example, for the Venetan dialect of Verona by Pescarini (2022). As such, present-day incomplete SCl paradigms need not be the diachronic result of the erosion of original larger paradigms; they might instead have never been complete SCl paradigms. This notwithstanding, there is evidence that SCl paradigms underwent a reduction process in diachrony across different northern Italo-Romance varieties. Importantly, the erosion of SCl paradigms is consistent with the implicational hierarchy in (5): that is, the SCls for 1sg, 1pl, and 2pl are more prone to falling out of use over time than the SCls for 3pl, 3sg, and even more so: than 2sg. Direct evidence for this comes from, among others, Milanese (Vai 2018: 167), Genoese (Scala 2021: 10), and central Venetan varieties (Poletto 1993: 156, 171). Despite different timelines (some paradigms underwent reduction in the 18th century, others in the 19th century), we can identify the new, reduced paradigms as the outcomes of CID.

A similar erosion of SCl paradigms is also attested more recently in other varieties, for example across Piedmontese varieties (Scala 2021: 12) or in Florentine (Renzi 1992: 95). Furthermore, clitics for 1sg, 1pl, and 2pl are commonly reported to be optional across various northern Italo-Romance varieties, again in line with the diachronic facts discussed above. Crucially for our discussion, the attested optionality and the eventual loss of SCls are commonly linked to the influence of Standard Italian, which itself lacks SCls (see e.g. remarks in Goria 2004: 159 for Turinese and Astigiano).

In sum, when it comes to the forms of the paradigm, CIC and CID affect the clitic paradigm in the same way, as in both cases the reduction of the paradigm is compatible with the implicational hierarchy uncovered by Renzi and Vanelli (1983). Tying this back to our heuristic for change, this means that the ϕ-features that underlie the derivation of the different SCl forms are affected by change, as some person and/or number distinctions in the paradigm are lost;^[14] but that they are affected by change in one and the same way, as the patterns of reduction of the paradigms are parallel across CID and CIC. This parallelism is predicted by our proposal, since we are dealing exclusively with syntactic features. However, when we turn to observe the distribution of SCls, which is in part determined by discourse factors, the entire picture changes drastically: we explore this in the next section.

6.2 Subject clitics and discourse: CIC differs from CID

Granted that SCl paradigms come in different sizes across Italo-Romance varieties (possibly due to the diachronic or contact-induced erosion of the relevant paradigms discussed in the previous section), in SCl-languages SCls are expected to occur in some specific syntactic environments and under specific discourse conditions (see the introduction to Section 6). In what follows, we illustrate three changes that affect the distribution of SCls in northern Italo-Romance. All these developments have emerged in contact (with Standard Italian, Brazilian Portuguese, and Argentinian Spanish respectively) and crucially find no correspondence in diachrony.

The first effect of contact on the distribution of SCls is a general restructuring of the conditions of occurrence of SCls, leading to these forms being absent in various syntactic and pragmatic environments in which their occurrence is expected, and vice versa present in environments in which their occurrence is not expected. Importantly, this change cannot be simplistically traced back to transfer from SCl-less Standard Italian, but it should be regarded as an instant of CIC nonetheless, as it stems from the multilingual competence of the relevant speakers. For example, Casalicchio and Frasson (2018) investigated change-in-progress in the distribution of SCls in present-day central Venetan varieties. In central Venetan varieties, SCls double both full pronouns and topicalized referential DPs in preverbal position (Benincà 1982, 1983; Poletto 2000).

The results of Casalicchio and Frasson’s acceptability study indicate that, while older speakers substantially adhere to this grammar, younger speakers tend to use SCls according to a syntacticized version thereof. Specifically, the innovative grammar seems to be driven by a general simplification of the interface requirements on the use of SCls: SCls appear to be prone to omission in contexts in which their occurrence is regulated by syntax-pragmatic interface factors, and concretely when they double a topicalized DP subject; on the contrary, SCls appear to be more freely accepted by younger speakers in contexts in which a preverbal subject is missing (atmospheric verbs and postverbal subjects with unaccusative verbs). That is, the innovative grammar seems to rely exclusively on syntactic rules (insert a SCl if the preverbal subject position is not already filled), at the expenses of the formerly available interface conditions (topicality for the syntax-pragmatic interface; argument structure in the case of atmospheric and unaccusative verbs for the syntax-semantics interface).

The second outcome of CIC that we can observe regards the distribution of SCls and null subjects. As discussed by Frasson (2021) for heritage Venetan spoken in Brazil and by Frasson et al. (2021) for heritage Friulian spoken in Argentina and Brazil, heritage subject clitics do not disappear, unlike their homeland counterparts (see above), but rather come to display a full pronominal behavior. This conclusion is based on the fact that these erstwhile agreement markers do not pass Rizzi’s (1986) tests for clitichood any more: for instance, they can be dropped under coordination in contexts (9) in which this would not be possible in the baseline (8):

As discussed in the foregoing, in the baseline SCls are agreement markers and as such an integral part of the verbal morphological system. Therefore, whenever two clauses are coordinated, SCls must be present in both conjuncts, as shown in (8). This is not the case for full pronouns, which are instead fully referential elements and not inflection markers. Full pronouns can thus be dropped under coordination even in languages that are not pro-drop, provided that the subject of the two clauses is the same. The fact that in the Brazilian Venetan example in (9) the subject is only overtly realized in the first conjunct and is missing in the second one plainly suggests that these elements degrammaticalized in heritage varieties and reverted to full pronominal status. This conclusion could not be foreseen on the sole bases of diachronic change, where the subject agreement cycle proceeds instead to completion.

Interestingly, Frasson et al. (2021) show that these pronominalized SCls tend to be dropped in topic continuation contexts, as partly shown by (9) and better exemplified in (10):

(10) displays a clause introducing a new referent (the godfather): this is the new topic in the wider discourse, and as such this element is also marked with a 3sg SCl. The sentence continues and we can see that in (10a) the topic shifts with respect to the previous clause: the new topic is the speaker, which is marked by the SCl for 1sg. However, in the absence of further topic shifts (that is, under topic continuation, as in (10b)), we see that the SCl is dropped. Frasson et al. (2021: 26) found this pattern, whereby SCls are produced in topic shift and dropped in topic continuation, to be significant in Argentina; a similar effect of topicality, albeit a non-statistically significant one, was found in Brazil.

These facts are surprising for two reasons: the first one is that, when Brazilian Portuguese is the contact language, this effect cannot be attributed to transfer from the contact language. Brazilian Portuguese is in fact a partial null-subject language. More interestingly, heritage speakers are known to have difficulties mapping meaning onto silence (the Silent Problem, first identified by Laleko and Polinsky 2017): accordingly, heritage speakers of null-subject languages have been found to disprefer null subjects (see Polinsky 2018: 6.5 for an overview). Therefore, the fact that a heritage language starts developing a null subject is all the more surprising (but see Carvalho and Child 2011, Pérez 2009, Souza et al. 2018 for some attested cases of emergent pro-drop). Focusing now on SCls, this evolution is also interesting as it diverges slightly from the baseline Veneto. As shown by Benincà (1994), Veneto is a language where SCL are only licensed with overt subjects if the subjects are topics, i.e. almost exactly the same contexts in which their heritage counterpart drops them.

A third and final context in which the development of SCls in diachrony diverges from that in contact is the case of 3rd person SCl. Heritage Venetan and Trentino varieties spoken in Brazil display an interesting case of complexification in relation to these new auxiliary forms, discussed in D’Alessandro and Frasson (2023). Not only do SCls not get lost, as we would expect in diachrony, but original patterns of diatopic variation are reinterpreted yielding novel specialized auxiliary + SCl elements. Auxiliaries in Veneto come in three fashions: l’è, z’è, è. These three variants are geographically distributed (Benincà 2007; Bentley et al. 2015; Poletto 1993), and consist in two cases of a SCl (l, z) incorporated into the 3rd person singular form of the auxiliary be. In Brazil, however, these “complex” auxiliaries specialized, as shown in (11), from D’Alessandro and Frasson (2023):

(11) illustrates the specialization patterns: l’è, in (11a), is used by preference with postverbal subjects and anti-agreement effects; zè, in (11b), is used by preference with preverbal subjects and no anti-agreement effects; finally, è, in (11c), is used by preference with preverbal third person plural subjects.^[15]

What happened in Heritage Venetan in Brazil is that these auxiliaries with incorporated clitics came to be part of the same “pool of variants” (Labov 1972) for a group of speakers originating from various parts of Veneto. These speakers did not select one form and abandon all others to simplify the system, but created a distribution rule, as illustrated, assigning one auxiliary + SCl form per construction. For instance, l’è is used almost exclusively with postverbal subjects and is associated to anti-agreement effects. While these forms are not strictly speaking purely SCl, they involve SCl in a such a way that integrates them in a highly complex syntactic system not found in the varieties spoken in Italy.

In conclusion, we hope to have shown that when only ϕ-features are involved, SCl in contact do not preset major differences from SCl that evolve spontaneously. However, when discourse or more complex syntactic computation is involved, the output of CIC suddenly differs substantially from CID, proving that a generalization along the lines of “CIC is accelerated CID” only holds for “core” parts of grammar, not those concerning the interfaces.