On “free” grammatical variation in a mixed lect: Clitic placement in Cypriot Greek

Abstract Variation involving a switch between pre- and post-verbal placement of pronominal object clitics in a single syntactic environment within a language is unexpected. The rationale why this would not be expected is clear: Languages pattern as either proclitic or enclitic with respect to object clitic placement, possibly allowing one or the other option across different syntactic environments. We provide an overview of our research from data collected in Cyprus, related to the development and use of pronominal object clitics for child populations and adult speakers that are bilectal in Cypriot and Standard Modern Greek. While it has been shown that the tested bilectal populations receive exposure to more than one distinct grammar, including mixed grammars with optional choices for clitic placement, an important question remains unaddressed: Is variation really “free” across all speakers or are there universally reliable predictors (such as gender, age, or level of education) that mediate a consistent use of either the standard or the dialect? Combining insights from targeted elicitation tasks administered to different groups, a corpus of spontaneous speech, and an extensive literature review, we show the weakness of such purported predictors and support a claim of free variation.


Introduction
Optionality in linguistic theory can be defined as the co-existence of two or more variants of a given construction within the same grammar (Sorace 2000). The ex-istence of optionality is a challenge both for theoretical approaches to language but also for developmental studies, since the first assumes competition for wellformedness of a given construction and the latter predicts native attainment at a certain stage, which is usually characterized by use of one of the two options (e. g., in parametric models of language acquisition). In different cases where optionality or "free" variation seems to exist, it involves the co-existence of optimal forms, one of which is "more unmarked" than the other in the grammar. Identifying the optimal form may often be as simple as a mere judgment by a native speaker. In other cases, the sociolinguistic factors that characterize a particular developmental path in the early years, and the language input during that time, present a more complex situation, requiring further investigation of the grammatical and extra-grammatical factors leading to this optionality.
Optionality in syntactic theories has been widely discussed in connection with the optional use of the complementizer that for complements, dative alternation, and voice alternations (see Boyd 2007 for recent discussion), for example, identifying as motivating factors the speaker's choice and the nature of syntactic development. Optionality in child language can be ascribed to many causes, such as constraints that delay the appearance of functional categories in child grammars until a particular time in their development (Radford 1996;Sorace 2000). In this period of development, the child will appear to switch between the two options presented to her. Other theoretical approaches may include underspecification of features (Wexler 1994) and incomplete syntactic structure (Rizzi 1994), among many others. A general assumption in language development is that if a child has input from a single grammar, then optionality should be limited. However, it is also the case that if the input comes from different varieties that are closely related, then a single system incorporating elements of the different grammars exemplified in the input appears (Henry 2013). Communities are often nonhomogeneous in terms of "dialectal" features in the language input to the child as well as other sociolinguistic factors that define language use: gender, origin, education, or age.
In the current study, we discuss variation in clitic placement in bilectal speakers of Standard and Cypriot Greek, which is not expected to appear for the syntactic environment of because-clauses investigated here. The phenomenon of clitic placement was chosen as the focus of this study because it shows mixing which has been documented by many studies that use a variety of different elicitation techniques (see next section for an overview). Preverbal clitic placement is predicted, as in the case of most Romance languages which require proclisis by "default" (say, in an indicative declarative clause), but exhibit enclisis in wh-questions. European Portuguese is different in setting the default to enclisis, with proclisis reserved for particular syntactic environments (such as negation, interrogatives, focus, etc.; see e. g. Uriagereka 1995;Raposo and Uriagereka 2005). We also know from child language that clitics are acquired relatively early (around the age of 2; for Greek, see Marinis 2000;Petinou and Terzi 2002). Also, once acquired, children do not make placement errors (cf. Varlokosta et al. 2016). That is, children do not go through a stage in which they would produce proclisis in an indicative declarative clause where enclisis is required or the other way around. (Children may make mistakes in special placements, but that is due to figuring out which syntactic contexts are special; see e. g. Duarte and Matos 2000 for European Portuguese.) This certainly holds for monolingual first language acquisition.
The existence of closely related varieties in the course of child language development may give rise to a variety that involves functionally equivalent variants in adult speakers for whom we observe a mixed, hybrid system, in which elements from different "lects" are merged into a single grammar ). The following section presents an introduction to the linguistic situation in Cyprus and the variation in object clitic placement that has been observed by different studies. We discuss the findings from elicitation studies and spontaneous speech data in Section 2, as well as the validity and relevance of predictors for supporting the argument of "free" variation in Section 3, before a brief conclusion in Section 4.

Variation in object clitic placement
The context of Cyprus was first characterized as diglossic (Newton 1972;Arvaniti 2010) on the basis of the differences in the functional use of Standard Modern Greek (SMG) and Cypriot Greek (CG) as sociolinguistically high-("H") and low-("L") prestige varieties, respectively. CG and SMG have different functional uses in Cyprus. Demotic Greek, or Modern Greek (which linguists refer to as SMG, as used here throughout), is learned through formal schooling and used in most forms of writing and oral formal discourse. CG, however, is acquired at home in a naturalistic setting and used in informal interactions. Given the functional differentiation, diglossia has been argued to characterize this bilectal population.
CG and SMG show differences at all levels of linguistic analysis that are easily identifiable by native speakers. There is often no consensus for the judgments of bilectal speakers, since these tend to be clouded by sociolinguistic attitudes towards using the non-standard variety. CG lacks official codification; indeed, its status as a distinct language or variety is often denied by speakers of CG who ignore the differences between the SMG and CG grammars. All CG speakers have exposure to SMG through education and other mediums -it is in this way that they are competent in both varieties to different degrees. As Rowe and Grohmann (2013) explain, the socio-psychology of speakers in diglossic societies usually centers around the notion of prestige and a negative valuation of the L variety. Since the use of the H variety indicates advanced education, speakers typically want to speak it in order to present themselves as well educated. They further suggest that a "co-overt" prestige of CG has prevented a takeover by SMG or death of CG. In sentences with object clitic placement, speakers are often observed to produce orders that can be argued to be SMG in CG structures. The example below shows an instance of SMG-typical DP focus fronting (which is not a strategy employed in CG) and copula use in a CG cleft structure (which is not part of the SMG grammar  (Tsiplakou 2014: 171) The existence of examples like (1) raises the question whether any of these sociolinguistic criteria, such as prestige and negative evaluation, apply to language use -and consequently language development.
Object clitic placement in CG depends on the position of the verb in different syntactic environments. Pavlou (2018) suggests that there are three possible groups that capture word order variation in object clitic placement according to the syntactic environment: The first is the case where the verb precedes the object clitic, the second when it follows the object clitic, and the third one whether it can either precede or follow. The matrix clause in CG shows a verb-clitic order, as in (2).
(2) a. iden ton b. *ton cl.masc.acc.sg iden. see.past.perf.3sg 'She saw him.' According to Pavlou (2018), the verb moves to C in cases like (2). In contrast, the verb stays low in most other cases, such as clausal negation, negative imperatives, different types of embedded clauses, including those introduced by the subjunctive marker na, and wh-questions. Certain embedded complementizers show optionality with respect to the position of the verb in a CP-recursion syntactic environment (Pavlou 2018). All of the different word orders are the result of the different positions that the verb can occupy in Cypriot Greek (Agouraki 1997;Terzi 1999;Mavrogiorgos 2013). The enclisis in matrix clauses is a good indication that the verb is higher in the structure and it moves there from a lower position unless some element in the syntax blocks this movement.
The syntactically predicted variation does not necessarily predict speakers' productions of sentences with object clitic placement. Several studies have suggested that variation in object clitic placement in CG is present within and across speakers. Within-speaker variation refers to the fact that a speaker may use both proclisis and enclisis in succession, as in (3) Across-speaker variation refers to the preference of certain speakers to produce one of the two options in a more consistent way. For example, in a study comparing CG and SMG, Tsiplakou et al. (2016) found that male speakers that have only completed secondary education produced fewer forms in SMG. Given, first, that not everybody mixes to the same degree and, second, that the mixed lect, which permits both proclisis and enclisis, is highly fluid, it is unclear whether variation is really "free" across all speakers. Certain predictors, such as gender, age, and level of education, have been linked to more consistent use of either the standard or the non-standard variety in specific studies (i. e. linked to specific speaker cohorts). However, the overall consistency and strength of the predictive power of such factors across studies is something that has not yet been examined. This is the aim of the present work. Motivated by the aim of determining whether there are constraints to "free" variation in mixed lects, a cluster of predictors is analyzed from different perspectives. More specifically, this aim is achieved by (i) surveying our research group's studies on elicitation, (ii) summarizing the corpus presented in , (iii) analyzing the literature on the topic, focusing exclusively on the strength and the ecological validity of the most commonly identified predictors, and (iv) discussing all of the above with relevance to "free" variation.  (Grohmann 2011: 196).

The picture from elicitation studies
We begin by reviewing different studies that provide a picture for the acquisition of object clitic placement in Cypriot Greek. Research on the acquisition of object clitics was carried out by the Cyprus Acquisition Team as part of a European research network, COST Action A33 see Grohmann 2011). In this production experiment, the target group consisted of 25 typically developing Greek Cypriot children at the age of 5 (TD5), and a group of younger children aged 3 and 4 years also participated (TD3-4). The children were recruited randomly from several Greek-speaking kindergartens and they were "monolingual" speakers of CG. The COST Action A33 testing tool used, included 19 items, out of which 12 were target structures with because-clauses and the appearance of object clitics (details can be found in Varlokosta et al. 2016). Results are summarized in Figure 1.
In the TD5 group, for which production was very high (95.8 %, the secondhighest of the 11 clitic languages tested; Varlokosta et al. 2016), 10 children mainly used proclisis, 10 children mainly used enclisis, and 4 children mixed the two, raising the question of what the target language is that the children are trying to acquire. In the younger TD3-4 group, out of the total clitic production (91.7 %), there was 100 % post-verbal clitic placement, which is more adult-like performance. In subsequent research, additional data were collected with much higher participant numbers and more age groups (see Grohmann 2014 for summary).
In a follow-up study (Leivada et al. 2010), the same tool was used both in CG and in SMG. For bilectals, born and schooled in Cyprus with both parents from Cyprus, similar performance was found as above. Hellenic Greek monolin-guals, born and schooled in Greece but relocated to Cyprus, with both parents from Greece, however, scored at ceiling in the SMG version -but not in the CG version. Hellenic Cypriots (or binationals), children who were born in either Cyprus or Greece, with one parent being Greek Cypriot and the other Hellenic Greek while now residing in Cyprus, performed at ceiling in the SMG version, but poorly in the CG version. These studies showed that the TD5 group's acquisition of grammar is affected by other factors that contribute to the results observed.
In studies with adults, Papadopoulou et al. (2014) focused on adult performance in an effort to investigate (i) to what extent certain CG-specific lexical choices or syntactic structures affect clitic placement as pre-or post-verbal and (ii) what counts as target placement in children's input when they acquire clitic pronouns in an environment that involves at least two varieties with conflicting clitic placement patterns in some syntactic environments.
157 monolingual adults, born and raised in Cyprus, had participated in the experiment, which consisted of 16 questions (and 4 fillers), 8 in a CG-specific and 8 in an SMG-specific block. This means that CG and SMG items were carefully constructed so that one variety does not exist in the same sentence with another. In this way, each block makes use of verbs and nouns that are as specific to the attested variety as possible in an effort to see to what extent lexical items affect placement. To achieve the above, the design considered verb roots: If the root was not completely different between the two varieties, it was, at least, not identical. The purpose of this strategy was to aim for effects appearing in CG-specific environments vs. SMG-specific environments. Additionally, content words in the test stimuli of one variety were not used in the other variety according to the authors' judgments.
Participants were divided in two groups for the presentation of the data following the order of the blocks. The first order had the block presenting the test items in CG first and then the test items in SMG. The second order involved the exact opposite. The graph in Figure 2 illustrates that participants produce mainly the targeted structure within each language block -post-verbal clitics for CG blocks and pre-verbal clitics for SMG blocks.
Specifically, participants in this experiment produced post-verbal clitics more than 98 % of the time when they are primed with CG, and 65 % and above when primed with SMG. Post-verbal clitics are always produced at lower rates within SMG blocks. The study concluded that clitic placement can be considered a consequence of both syntactic and lexical priming available in (non-)experimental settings.

The picture from spontaneous speech
While an extensive corpus for Cypriot Greek adult speech is not available and more studies examine elicited paradigms, smaller studies have attempted the collection of data in natural conversation, as in . This work investigated patterns of variation in the spontaneous production of five neurotypical, adult speakers of a non-standard variety in terms of three variants, each targeting one level of linguistic analysis: syntax, morphology, and phonology. All participants were native speakers of CG. There were three participants in each session, one participant and two researchers. The two researchers used the home variety that would be typical of a casual conversation with friends and family. Each recording taking place in a familiar location (for example, the participant's house) lasted for approximately 30 minutes and there was no specific topic discussed. In total, five participants and two researchers interacted on five different occasions. The seven participants produced 4,818 utterances while engaged in a conversation in an informal setting with every intelligible unit of speech that was separated by pauses treated as an utterance. An overview is provided in Table 1.
For the analysis of the corpus, three sets of variants were examined, each of which chosen because it belongs to a different level of linguistic analysis: (i) syntax through clitic placement, which varies in indicative declaratives (pre-verbal in SMG, post-verbal in CG); (ii) morphology through the use of the CG diminutive affix -u (vs. -ak in SMG); and (iii) phonology through the CG-specific post-alveolar affricate /tʃ/ which corresponds to palatal /c/ in SMG. The results of the analysis shown in Figure 3 are summarized below.
The study observes the existence of functionally equivalent variants within a single repertoire. Variation is manifested across speakers, as evidenced by the fact that different participants align more with the standard variety than others. Moreover, this variation amounts to a case of language mixing -rather than language fusing -for two reasons: (i) the observed patterns are not stabilized and (ii) intraspeaker variation suggests that speakers do have a choice as to which variant they use.
Our previously obtained results indicate that the linguistic repertoire of the bilectal speakers incorporates elements from different lects across levels of linguistic analysis, resulting in a mixed lect. Precisely because this mixed grammar is not standardized, it may differ with respect to the degree of mixing that is attested from speaker to speaker, since absence of standardization facilitates the emergence of interspeaker and intraspeaker variation (Leivada et al. 2017 and references therein). When translated in grammatical terms, this mixing gives rise to functionally equivalent variants that are the result of bringing two realizations of the same variant into one grammatical system, each of which originating from a different grammar.
The results of the corpus analysis convey the same message as the elicitation studies: Variation is found in all speakers, regardless of gender, age, or level of education. We recognize, however, that the literature reviewed did not primarily focus on the role and strength of predictors: They mostly targeted age of acquisition and patterns of development, while the spontaneous speech study on adults was limited in scope and did not involve a sufficient number of participants from all backgrounds. Thus, we will review and compare in the next section some landmark experimental studies that make reference to such predictors in order to determine whether they converge.
Similarly to the study above, Tsiplakou et al. (2016) collected data with sociolinguistic interviews. There were two interviewers who were speakers of CG and who used the koiné (i. e. a pan-Cypriot variety that lacks local features) throughout, taking care to speak at a relatively informal level. In the interviews, the participants were asked to narrate something exciting or emotionally loaded in order to be spontaneous in their linguistic production. In the study, a total of 57 participants were interviewed (29 males and 28 females), with their ages ranging from 26 to 90 years. The measurements shown in Figure 4 were the percentages of the more Cypriot-spedcific variants, namely (1) [ʃ], (2) [tʃ], (3) Simple Past, and (4) enclisis, out of the total number of occurrences of both CG and SMG variants.
The study reports that the positive differences between the two levels of each factor indicate that subjects used more CG-than SMG-like variants if they had secondary rather than tertiary education backgrounds, if they were familiar with the interviewer, and if they were male.
In sum, the two studies show that CG and SMG variants are both found in Greek Cypriot adult speech with the possibility of some predictors such as education, familiarity, and age playing a role.
On "free" grammatical variation in a mixed lect | 285 Figure 4: Overall production of all tested variants across participants (Tsiplakou et al. 2016: 17).

Determining the strength and ecological validity of predictors: a literature review
This section reviews some landmark papers on the acquisition and development of CG in Cyprus, focusing on specific predictors such as age, gender, and location, starting off with a study that examined the role of syntactic environment. Using different pragmatic markers and complementizers that allowed both pre-verbal and post-verbal clitic placement (e. g., endʒe 'not', oti 'that', giati 'because', epidi 'because', afou 'after'), Pappas (2014) carried out a magnitude estimation experiment, testing 34 bilectal speakers of CG and SMG. The experiment examined the extent to which the score participants assigned to proclisis in one environment (e. g., after giati 'because') was significantly different from the score that was assigned to enclisis in the exact same syntactic environment, following the same item. Using "functional item" as predictor, out of the 10 tested comparisons, only two comparisons showed a highly significant difference in a linear mixed effect analysis. Four comparisons showed a marginally significant difference, while the remaining four comparisons failed to reach statistical significance. It is worth noting that even in the highly significant comparisons, both proclisis and enclisis were accepted to some degree (in part due to the nature of the task employed). We can thus only talk about preferences and not identify an absolute predictor that can be tied to the exclusive use of one variety, through triggering only proclisis or enclisis.
In a meticulously designed study, Tsiplakou et al. (2016) reported a series of sociolinguistic interviews that were conducted by two young, male interviewers who were speaking CG throughout the sessions. The four relevant predictors were age, gender, education, and familiarity with the interviewer. An ANCOVA revealed non-significant effects for all four predictors. Planned contrasts showed a smallsize effect for gender (p=0.17), and medium effects for familiarity and education (p=0.002 and p=0.001, respectively). Overall, Tsiplakou et al. (2016) found evidence of within-subjects coherence, while combinations of predictors (but not the predictors themselves in isolation) were linked to more consistent use of either the dialect or the standard: Specifically, males that have completed secondary education only and were familiar with the interviewer, showed overall higher rates of use of the dialectal forms. Taking into account that none of the predictors in isolation was linked to consistent CG or SMG use, it is unsurprising that Tsiplakou et al. (2016) conclude that the most robust coherence is found at the two ends of the spectrum of the examined factors, while most of the spectrum involves mixed linguistic behavior that incorporates elements from different lects.
The weakness of each predictor in isolation may be taken as support for the claim that variation is largely free, and only the synergistic occurrence of certain factors may lead to a more consistent use of CG or SMG in highly specific conditions. It should also be taken into account that unlike gender or age, familiarity with the researcher is a relatively fluid predictor. It has been argued that sociolinguistic interviews involve insurmountable power asymmetries (Schilling-Estes 2008). In this case, the power dynamics may have been mollified by the young age of the two interviewers together with the fact that they chose to speak exclusively in CG. Put differently, the highly specific constellation of factors that was shown to have the strongest predictive power (for the use of one variety over the other) in this study involved a variable called "familiarity with the researcher", and this variable could have played out in a completely different way, depending on the interviewer's profile and language use. As Milroy and Gordon (2003) suggested, in sociolinguistic interviews, the more powerful conversational role belongs to the interviewer and a senior university researcher would typically be linked to an even more powerful role compared to a young member of the dialectspeaking community: One does not address a senior researcher in the context of an interview in the same way one would address a peer who speaks informally in the same context. If this is indeed the case, the factor "familiarity with the researcher" is at best a fluid one. Such asymmetries and fluctuations in power dynamics could have possibly altered the results obtained -precisely because none of the main, less fluid predictors (age, gender, education) could be linked to robust consistency on their own. The third study that is relevant to the argument presented here is Ioannidou (2009). This work differs from the previous ones both in methodology and in the target group. Ioannidou (2009) used the analysis of classroom talk in a group of 25 students, aged 10-11 years, in order to investigate patterns of codemixing/-switching in bilectal Greek Cypriot children. The results showed heavy mixing and no clear-cut dichotomy between the standard (the official language of instruction) and the dialect (the language of break-time).
Although the great majority of students code-mixed, Ioannidou (2009) did find two exceptions with specific characteristics that include some of the predictors presented in Tsiplakou et al.'s (2016) study. The first exception was a group that consistently used the standard, even when the teacher herself incorporated some dialectal elements when addressing them. She remarks on this group as "consisting of a few high-achieving students […], mostly girls" (Ioannidou 2009: 271). The second exception relates to a group of four boys "with medium to low achievement, varied socio-economic background and a 'lively' presence in the class, who did not make any effort to use standard variants when they were nominated to speak" (Ioannidou 2009: 272). One aspect is interesting with respect to the role of the predictors mentioned here: Once again, female gender seems to be linked to a more likely use of the standard. Indeed, the gender effect is well known among variationist sociolinguists. In her overview of the field, Tagliamonte (2012) highlights the large consensus on this point through arguing that of "all the sociolinguistic principles, the clearest and most consistent one is the contrast between women and men" (Tagliamonte 2012: 32), where men are more associated with the use of dialectal elements.
Focusing on the study under discussion, the first exception in Ioannidou (2009) referred to a group that consisted mainly of girls, but also involved a boy who consistently used the standard. The second exception, which referred to the all-male group that used the dialect, still showed a good degree of free variation: Members of the group did use the standard variants in class, but mainly when they wanted to convince the teachers of something, as Ioannidou (2009) notes. In other words, what sets this group apart was not that they did not mix due to gender or due to a combination of gender, social background, and/or scholarly achievement, but that their mixing seems to be derived from different motivations compared to the other groups. Therefore, we can conclude that in this case, the gender predictor did not actually reveal consistent use of one of the two varieties. Instead, it showed different communication strategies among different groups, meaning that both genders/groups used both varieties, but the use of the standard was simply motivated by different reasons across them. This conclusion does not undermine the role of gender as a factor of influence, as the different motivations are still gender-guided to an important extent. Overall, the difference in motivation strategies may explain the different degrees of use of SMG vs. the non-standard variety.
The findings from Ioannidou (2009) on the sociolinguistic front can be interpreted as lending support to our claim concerning different motivations across groups. More specifically, the first group expressed very positive views overall about the standard and quite negative views about the dialect "in terms of aesthetics, status and appropriateness" (Ioannidou 2009: 272). In contrast, interview data showed that the second group held positive views about the dialect, feeling that this is a marker of identity and also that overall this is the variety that feels easier and more comfortable to them when speaking (Ioannidou 2009: 273).
We interpret this picture by suggesting that free variation exists in the repertoire of both groups, but they prioritize differently due to the presence of "competing motivations". In bi(dia)lectal populations that speak non-standardized varieties, the nature of the attested competition is twofold. Competing grammars (i. e. different grammatical variants that pertain to acrolect vs. basilect vs. mesolects) are complemented by competing motivations that arise through the learner's effort to achieve a trade-off between different needs such as "striving for clarity" or "striving for ease" (von der Gabelentz 1891 [2016] ;Newmeyer 2004;Leivada and Grohmann 2017). For example, formal contexts such as classroom instruction would require the use of the standard; however, the need to accommodate the factor "strive for ease" may result in the use of a dialectal element. In the actual data that Ioannidou (2009) reports, this interference of the dialect into the standard is attested everywhere, without exception, including the productions of the first group which valued highly the standard and even the production of the teacher, who is supposed to be teaching in the standard -a point to which we will briefly return in the discussion of the next study.
Putting the "gender" predictor into a larger context, gender roles are highly susceptible to cultural norms (Haeri 1997). This suggests that gender cannot be a uniformly strong predictor across cultures and studies. Apart from cultural variation, interpersonal variation exists too as both genders show subtle patterns of variability. For example, the term Personal Pattern Variation was coined by Dorian (1994) to describe variation in the East Sutherland Gaelic community; she observed a high degree of interpersonal variation among members of the same community, which could not be explained by any kind of social grouping. Crucially, in the context of Personal Pattern Variation, Dorian (2010) observed that gender-related preferences may be suggestive, but overall variation boils down to "a complex mix that makes any strong association of particular variants with males or females unlikely" (Dorian 2010: 142).
When combining these different sources of variation, it seems a truism that the robustness of certain predictors will be inevitably weakened by differing per-sonal motivations that will be reinforced by cultural differences. For example, speakers differ with respect to how much they value alignment with the standard variety in a way that cannot always be predicted by gender or social background. The attached values may motivate adaptation to whatever variety is deemed standard, even if this is not the target in a given context (e. g., an informal conversation between friends).
However, in certain situations, a different set of values may be prioritized, again in a way that may not be consistent with classical predictor effects. To give a concrete example, in certain occasions, speakers may prioritize the employment of strategies that are heavily marked as dialectal and certainly outside their own regular repertoire. In the bilectal context of Cyprus, this actually happened very recently as the byproduct of a very successful TV sitcom which featured extreme hyperdialectism. The latter refers to instances in which a person or linguistic community perpetuates the use of "erroneous" linguistic forms (e. g. through employing obsolete vocabulary) in order to establish or highlight the difference between their own linguistic identity and that of other groups (Cutillas-Espinosa et al. 2010). Due to the unprecedented popularity of this sitcom, the use of marked dialectal elements quickly became a trademark of modernity even among people that, according to standard predictors, would not be expected to align with the dialect and had explicitly criticized the use of such dialectal elements as a "peasant-like" way to talk (Tsiplakou and Ioannidou 2012).
An anonymous reviewer calls for caution in the use of the hyperdialectism argument, suggesting that the hyperdialectal forms are not part of the speakers' "real" grammars, but are used metalinguistically, for stylization purposes. Indeed, Tsiplakou and Ioannidou (2012) present this use of language as a conscious, premedidated authorical choice in the context of the sitcom. However, the repercussions of this process of reappropriating obsolete forms vastly exceeds the sphere of the performative devices that form part of a sitcom's narrative. A decade after the show's end, one can still observe the use of obsolete characteristics, such as the use of [tilexono] for [tilefono] 'phone'. More importantly, this use is productive even in the younger generations. Ayiomamitou and Yiakoumetti (2017) document a propensity to overapply and even misadapt obsolete phonological and morphological features in young students' written texts. Of course, one can argue that this is again a conscious choice, but first this would be true for any linguistic choice, and second precisely because it is a conscious choice, its repeated deployment means something. Perhaps this is the best evidence that hyperdialectal forms are part of grammar: The use of such marked features is a conscious strategy that reflects "deeply entrenched societal attitudes" (Ayiomamitou and Yiakoumetti 2017: 7). Yiakoumetti (2007) is another study that focused on the classroom environment. In an intervention study that involved exposing students to a textbook that drew on the differences between CG and SMG, 92 students aged 11 and 12 years old, coming from two schools in the Larnaca district (one urban and one rural), were evaluated through written pre-and post-intervention tests, using essays on various topics (language, geography, etc.). The different essays largely revealed the non-robustness of the employed predictors. For example, one of the three postintervention language essays showed a weak effect for location (p=0.043) and gender (p=0.013), but the interaction of these two factors was not consistent between the two locations. In the third language essay, the gender predictor did not reveal a significant difference between girls and boys. In the post-intervention geography essay, gender was again not associated with a statistically significant difference, and neither was location, but the gender × location interaction did reach significance (p=0.016).
The first observation to bring up here is the weakness of statistical effects and the absence of correction for multiple comparisons. The second is the variation in the behavior of predictors across tests and locations. As Yiakoumetti (2007) suggested, in her study gender per se was not significant as a predictor, but there was a combined effect of gender and location: Boys from the rural area were more likely to employ dialectal forms. The observation that "gender and place of residence are not simplex phenomena but interact with each other" : 6, who find a similar effect for clitic placement) goes back to Eckert (1999). As we observed in the context of Tsiplakou et al. (2016), the strongest effects seem to be mediated by highly specific constellations of predictors rather than by the main predictors themselves. Although there is no denying of the repeated participation of certain effects in these constellations (e. g., the gender effect), it is also true that these specific constellations entail that the most robust patterns of alignment with either the standard or the dialect are to be found in only some measures as well as in highly specific and small subsets of participants (e. g., males in rural areas when tested in a language essay but not in a geography essay).
Further complications arise from the fact that it is unclear what counts as "standard" in Cyprus and, especially, who is eventually using it in the classroom setting. A recent study found that bilectal teachers themselves are not fully able to identify dialectal elements that are superimposed on otherwise SMG stimuli (Leivada et al. 2020). To be more specific, in that study, bilectal teachers from Cyprus were put to the test and later compared to monolingual SMG-speaking teachers from Greece on a written variety-judgment task that asked them to tell apart whether an utterance was in the standard or in the dialect. Important differences were found in the accuracy of the two groups across all levels of linguistic analysis, pointing to a sharp discrepancy between what is deemed as "standard" in Cyprus and what the performance in the standard variety really corresponds to according to Standard Modern Greek textbooks (Leivada et al. 2020). This fluidity in defining the standard is coupled with variation when defining (different forms of) the dialect. Overall, this overarching fluidity entails that variation is an essential characteristic of the entire dialect-standard continuum but also that mixing is essentially ubiquitous and largely unconstrained by individual predictors.
In this context, an anonymous reviewer also suggests that we should show why our assessment of the discussed predictors is relevant for or can be transferred to mixed lects. We take this to be the case because the discussed predictors (i. e. age, level of education, gender) pertain to speakers, not to lects/varieties. As properties of speakers, they should apply to all the lects of the continuum that has been associated with the population under study, unless of course one can show that one lect is exclusive to a specific sample of the population, which is something that has not yet been demonstrated so far. Even studies that suggest that specific constellations of predictors are related to the use of more Cypriotlike vs. Standard-like variants (e. g., Tsiplakou et al. 2016) recognize the degree of overlap between the various lects, such that one can talk about a Cypriot Greek koiné with various acrolectal and basilectal registers, but "without identifiably local features" (Tsiplakou et al. 2016: 11).
To provide the complete picture, descriptions of CG in Cyprus range from "peasanty" to "polite", depending on how close or distant one's language is in relation to the basilectal pole of the continuum (Katsoyannou et al. 2006;Tsiplakou et al. 2006). The acrolect refers to the most prestigious variety, while the basilect amounts to the most stigmatized one. Attitudes towards standard forms also vary: People who use the standard are considered to be "educated, attractive, ambitious and intelligent", yet "not found to be more sincere, more friendly, or kinder, or to be more humorous than Cypriot dialect speakers" (Papapavlou 2001: 493). Since these languages exist on a continuum with many intermediate positions available, there is variation in what counts as dialect and standard. As Leivada and Grohmann (2017) note, the standard variety could be SMG or the "Cypriot rendition of Standard Modern Greek" (Davy et al. 1996) or some variety closer to the acrolectal pole, possibly a "koiné" (Terkourafi 2004), or (a recently emerged) "Cypriot Standard Greek" (Arvaniti 2010). Ayiomamitou and Yiakoumetti (2017) briefly mention the predictors "gender" and "class" when they present the semi-structured interviews of eight 10-to 11-year-old students (4 boys, 4 girls) who were interviewed individually. A semistructured format was employed allowing for flexibility in the development of an informal interaction with the interviewer. It is reported that no significant effects were detected (p>0.24 (1,115) for all potential predictors). Therefore, these factors were not further discussed. One of the most important findings of that study is the presence of basilectal CG across many student scripts, which even included the production of obsolete dialectal features and forms that had only recently been "revitalized" by receiving extensive media coverage (i. e. the aforementioned hyperdialectism discussed by Tsiplakou and Ioannidou 2012). This abrupt shift in their repertoires towards (dated forms of) the dialect seems to not be mediated by predictors such as age, gender, location, or social class.

Discussion
The previous sections summarized the findings of elicitation experiments, spontaneous data, and other studies related to the acquisition of object clitics in Cypriot Greek. All of the studies seem to point to the same fact -variation appears in all data collected for both children and adults. In some of these studies, variation may be guided or defined by particular factors: age of schooling (Grohmann 2011), priming effects (Papadopoulou et al. 2014), or other predictors such as gender (Tsiplakou et al. 2016).
The picture that emerges is that variation is free in the sense that predictors such as age, location, and sociolinguistic background do not function as proxies for the use of either the dialect or the standard -rather mixing is ubiquitous across lects, groups, and registers. An argument for mixing instead of code-switching is justified by the fact that the occurrence of both standard and dialectal variants is often found within the same utterance or experimental unit (e. g. a verb and a clitic) in spontaneous production and elicitation tasks respectively. Following the criteria presented in Auer (1999) for distinguishing mixing from switching, the absence of meaning in the alternations together with the fact that hybridity affects units of any size and the fact that speakers do not show preference for one language at a time suggest that the observed variation amounts to mixing.
As an important note, the behavior of these predictors has not been, strictly speaking, predicted in the studies in which they are found. Something that has not been explicitly acknowledged so far in any of the relevant studies is that in the literature on variation and development in bilectal populations in Cyprus, there is a total absence of pre-registered studies. The reported results are observations formulated after obtaining the datasets, not (dis)confirmations of actual predictions which were put on paper prior to any testing having taken place. At times, a hypothesis is indeed presented. For instance, in Tsiplakou et al. (2016), the tested hypothesis was that the different variants that speakers have at their disposal are used for various indexical purposes, thus breaking interspeaker consistency in rates of occurrence of particular variants. In Papadopoulou et al. (2014), the hy-pothesis was that the input is the main factor that induces the use of pre-or postverbal clitics in environments that allow for optionality.
The formulation of such initial hypotheses is useful, but it does not distinguish outcomes that result from predictions from those that result from postdictions (Nosek et al. 2018). Only preregistration can do that. This entails that the specific behavior of factors/predictors as well as the significance of specific constellations were not predicted in the strict sense of the term. Of course, this is not to suggest that exploratory studies involving data mining of novel datasets are not useful and informative. In fact, given how understudied the language population under discussion was until very recently, and perhaps still is to some degree, exploratory studies were the only way to proceed. Nevertheless, it should be acknowledged that if data mining is used as the primary way to verify hypotheses, then the predictive power of the study is weak. A pre-registered confirmation study with a new dataset is subsequently needed in order to verify the robustness of the hypotheses.
In multidialectal settings, sociolinguistic attitudes and the consequent alignment of the speakers with the acrolectal or the basilectal pole is largely a matter of personal preference. Dodson (1985) suggests that every bilingual person has a preferred language -with the term preference used as a psychological term with neutral connotations; this is true of bi-or multilectal speakers as well. One potential difference between bilinguals that speak standard varieties and bilectals that speak one or more non-standard varieties boils down to facility, which refers to one's ability to actively switch to another language (Pavlou 2007). Not always having the possibility to receive formal education in the non-standard variety or equal opportunities of using both varieties to the same degree, bilectals might have only passive knowledge of or limited access to one of their varieties in some registers. This suggests that predictors which may be thought of mediating alignment to either the standard or the non-standard variety -and thus constrain free variation across speakers -are themselves constrained by a number of independent factors such as facility of access, interference of the dominant variety, competing motivations, and fluid sociolinguistic values.
None of this denies the fact that this bilectal population can of course codeswitch, moving towards more/less acrolectal registers, depending on the context. Such adaptations would indeed translate to a coherent preference for one variety instead of another at a time (i. e. in a given context) and this would be a situation of a discourse-induced switching (Auer 1999). The crucial point here is that we are investigating preferences in the absence of any switching, zooming into the variation that exists in the mixed lect that is used as a koiné. The conclusion that emerges with respect to this koiné is that variation is "free" in the sense that (i) we fail to find speakers that do not mix at all, and (ii) the analyzed predictors are not predicting the consistent use of variants from either the standard or the dialect, but not both. The second point is a consequence of the first one: If a specific group of speakers (e. g., females with higher education that live in rural areas) had been found to speak a version of the koiné that does not show any structural hybridity, one would have effectively demonstrated that variation is constrained by certain predictors.
A potential limitation of the present analysis has to do with the fact that we compare studies with different types of participants (children vs. adults) which also employ different methodologies (experimental research, sociolinguistic interviews, and classroom ethnography). This inevitably comes with the territory, due to the scarcity of studies that make mention of the predictors analyzed. If more studies existed, one would be able to perform a focused analysis, targeting a more coherent subset of the relevant literature. At present, this is not the case.

Conclusion
This paper supports the idea that "free" variation or optionality in certain contexts -which are defined by competing factors in language development and other fluid sociolinguistic criteria -is not directly connected to the sociolinguistic factors explained in the literature. Rather, these predictors are weak for predicting predominant use of one variety or another in a bilectal context. A review of different studies examining the placement of pronominal object clitics in Cypriot Greek as acquired by different populations and tested with different experimental methods is provided for the purpose of identifying the reasons that optionality may appear. To support the notion of "free" variation, those reasons must be proven inadequate for predicting the appearance of different variants, thus leaving the question of how variation exists to remain unanswered.