5 Non-active voice in heritage Greek

In this chapter, we examine the production of verbs which bear non-active (NAct) Voice, as observed in narration tasks by heritage speakers (HSs). In particular, we are interested in HSs of Greek in two different majority language contexts, English and German. These speakers are the outcome of immigration primarily due to political and socio-economic reasons and they belong to the Greek diaspora. Immigration to the United States can be traced back to the 19th century, while immigration to Germany is more recent, mainly after World War II (cf. Chasiotis et al. 2006 for an overview of Greek immigration; Smith 2015). Our speakers are thus HSs of Greek, as they are bilinguals and acquire Greek in a context where English and German are the majority languages. Following recent work on the grammar of HSs, we take these speakers to belong to a continuum of native speakers, as they acquire both languages in a naturalistic way. However, the level of proficiency in the heritage language may vary across their lifespan (Benmamoun et al. 2013; Montrul and Polinsky 2011, 2019; Rothman and Treffers-Daller 2014; Scontras et al. 2015; Wiese et al. 2022). In the first years of their life, HSs are dominant in their heritage language, while upon entering in the educational system the opposite pattern may be observed (Montrul 2016; Polinsky 2018). According to various researchers, the linguistic proficiency maintenance in the heritage language depends on the quantity and the quality of input including the instruction of the language via formal education, the age of onset to bilingualism, the socio-economic status, the well-being, and the generation of immigration of their parents/guardians (Daskalaki et al. 2019; De Houwer 2015; Dosi and Papadopoulou 2019; Flores et al. 2017; Kupisch 2019; Meir and Armor-Lotem 2017; Romanowski 2021; Unsworth et al. 2014).

As heritage languages are usually acquired via oral speech which takes place among the members of the core family and is restricted to conversational informal communication, they are often taken to lack register variation, specifically to lack features that characterize formal communicative situations (Chevalier 2004; Dressler 1991: 101–102). Biber and Conrad (2019: 2) define register as the accumulation of linguistic characteristics that are common in one variety in combination with the situation in which this variety is used. The broad distinction between formal and informal register has been put forth in Halliday (1978: 31–32), who refers to this as the tenor of discourse; Poynton (1985) and Eggins (2004: 100–101) describe each register’s characteristics as follows: in formal communicative situations, monolingual speakers tend to use passive forms of verbs, participles, an abundance of adjectives, and lexical verbs instead of periphrastic constructions as well as complex structures such as subordination. It has been observed that HSs exhibit register leveling in various areas; for instance, German heritage speakers in the United States generalize certain clause types, see, e.g., Pashkova et al. (2022). Russian heritage speakers perform poorly on adjectives, as these characterize the formal register, see Polinsky (2005). Greek heritage speakers have been shown to generalize periphrastic constructions across registers in order to express aspect (Alexiadou and Rizou 2023) and to generalize the use of the imprecise and informal determiner kati ‘some’ (Alexiadou et al. 2022).

Our study investigates the production of novel patterns involving NAct Voice in two groups of HSs, asking the question of how these novel patterns can inform our theories of the morpho-syntax and semantic impact of Voice in Greek. These patterns may be very telling about overgeneralization in inflectional morphology in general, as HSs have been reported to overapply regular morphological rules (Polinsky and Scontras 2020). In view of the fact that NAct morphology in Greek may be involved in creating a variety of interpretations, we might in principle expect it to be generalized across contexts. However, as Greek NAct Voice is associated with the formal register, we do not expect HSs to make productive use of it.

The chapter is structured as follows: Section 5.2 presents an overview of the Greek Voice system, while Section 5.3 explores previous literature on L1 and heritage language acquisition regarding verbs that bear NAct Voice such as passives and deponents. In Section 5.4, the research questions based on certain hypotheses alongside our predictions are presented. Section 5.5 presents the methodology of this study, the design and the profile of our participants. Section 5.6 presents our analysis and Section 5.7 concludes.

Greek has two sets of Voice forms: active and non-active. Voice features have been argued to be next to verbal stem/root, as seen in (1) (Philippaki-Warburton 1998):

NAct Voice in Greek is linked to the [+learned] feature which characterizes words that either come from Ancient Greek or constitute artificial formations of katharevousa (i.e., the artificial form of Greek developed by scholars in the 19th century) (Anastasiadis-Symeonidis and Fliatouras 2003).

NAct Voice is used with passives, reciprocals, reflexives, and certain anticausatives¹ as well as with deponent verbs, see (2) (Alexiadou and Anagnostopoulou 2004; Embick 1997, 1998). While in (2a–c), NAct is the canonical marking of [-] transitivity, deponents, as in (2d), are taken to exhibit a mismatch between morphology, syntax, and semantics, as NAct Voice appears with [+agent] and [+] transitive verbs (Alexiadou 2024; Embick 1997; Grestenberger 2018; Zombolou and Alexiadou 2014).

One possible account of the presence of NAct on the verbs in (2a–c) is as follows. Building on Embick (1998), Alexiadou, Anagnostopoulou, and Schäfer (2015) propose that NAct morphology realizes a Voice projection which lacks a specifier. From this perspective, NAct realizes a structure of the type (3a), where [-D], following Schäfer (2008), signals the absence of a specifier. In the absence of Voice or in the presence of a DP in Spec,VoiceP, Greek verbs are assigned default active Voice, (3b–c). (3a) is basically an unaccusative structure and it is associated with passives, reflexives, and NAct-marked anticausatives. The rule that regulates the spell-out of Voice in the absence of a specifier in this system is given in (4).

In Oikonomou and Alexiadou (2022), (3a) is a special structure not only morphologically, but also semantically: in (3a), Voice forms a single interpretation domain with the vP with which it combines. Thus, Voice may receive a contextual interpretation, but also idiosyncratic and novel interpretations may arise. As we will see, (3a) is also employed in the context of transitive deponent verbs, whereby the external argument in this case is a low agent, introduced below VoiceP, see Grestenberger (2018) and Oikonomou and Alexiadou (2022) for details.

The acquisition of NAct forms is considered to be a late-acquired phenomenon in L1 Greek. The approximate age of L1 NAct acquisition beginning with short passives is around 5 years old (Fotiadou and Tsimpli 2010; Grey 2020; Zombolou et al. 2010). Previous studies on bilingual children show that these do not seem to behave radically differently from monolingual children with respect to reflexives, while they seem to be more vulnerable with passives (Grey 2020; Tsimpli 2006). Furthermore, as observed in Paspali et al. (2022) and Unsworth et al. (2011), Greek bilingual children exhibit lower accuracy with anticausatives bearing NAct Voice.

It has been pointed out that heritage speakers appear to face difficulties especially with inflection in the verbal domain. Several studies have shown that these bilingual speakers are prone to loss of morphological features and thus they simplify the inflectional features, deviating from canonical verbal paradigms (cf. Putnam et al. 2021 for an overview). For instance, Polinsky (2000) observes an incomplete verbal paradigm in American Russian, while Fernández-Dobao and Herschensohn (2021) mention that HSs of Spanish in contact with English overregularize irregular verb stems.

Voice alternations in heritage Greek have not been the focus of extensive investigation. One of the first researchers who noticed some inconsistencies regarding the production of verbs with NAct by HSs is Seaman (1972: 165–166). He reports that Greek HSs in the United States avoid NAct forms and hesitate to use them. Despite the low frequency of verbs produced in NAct Voice identified in his research, one of the examples he includes is that of the deponent verb diigume ‘narrate.NAct’ which HSs changed to active Voice.

Zombolou and Alexiadou (2012, 2014) discuss Greek HSs in Argentina and Australia and show that these speakers generalize active on deponents and so-called marked anticausatives. This observation alongside Paspali et al.’s (2022) and Unsworth et al.’s (2011) findings can be explained on the basis of the Syntactic Complexity Hypothesis (SCH) (Zombolou et al. 2010), according to which, NAct forms are more complex and thus avoided by learners and HSs. This is readily accounted for under the representations in (3), whereby (3a) is more complex than (3b).

Specifically, the majority of the participants in Zombolou and Alexiadou’s study (63%, p < 0.01) changed the morphology of transitive deponents by using active in the place of NAct, (5):

Similar loss of verbal deponency is already observed within the geographical bounds of Greece.

Zombolou and Alexiadou thus concluded that since their participants changed transitive deponents only, this happens due to overregularization. Transitive verbs in Greek canonically bear active and not NAct; NAct, as shown in the previous section, is reserved for intransitive members of an alternation. HSs thus replace NAct with active. As Greeks in Australia changed the morphology of deponent verbs less than Greeks in Argentina, 59% vs. 67% respectively, language interference from English can be excluded. Thus, the authors proposed that we have reanalysis resulting in simplification of the system and loss of morphological variation.

While this study observed the generalization of the active default on marked anticausatives and transitive deponents, other work has observed the opposite generalization: it is the marked, i.e. reflexive morphology that is generalized in, e.g. heritage Spanish (Polinsky 2018). According to Polinsky, this is correlated with low proficiency; (6) is described as odd, as it is not attested in the input, but is produced by low-proficiency heritage speakers of Spanish:

In this study, we aim to explore the following research questions (RQs). The first one (RQ1) is: which verb categories bearing NAct Voice are produced by two groups of HSs of Greek in Germany and in the United States? On the basis of our results, we will further explore whether these categories diverge from monolingual productions and whether we can attribute this to language contact (RQ2). Based on previous findings (Polinsky 2018; Seaman 1972; Zombolou and Alexiadou 2012, 2014), we expect two possible scenaria (RQ3): either to find overgeneralization of the active default or generalization of NAct correlated with low proficiency. RQ4: Given the methodology that we implemented in the present study, the addition of an extra-linguistic factor such as register variation is expected to trigger monolinguals to associate formal registers with NAct Voice verbs and differentiate their narration in informal communication settings by producing less verbs in NAct Voice, unlike HSs who are expected to exhibit register leveling according to previous studies on different phenomena (Alexiadou et al. 2022).

Following Wiese’s (2017) “language situation” setting, we conducted our research by collecting naturalistic data through narration tasks. By implementing this methodology, we are provided with a variety of phenomena allowing us to conduct exploratory research, unlike the data collected from controlled experiments. This methodology is suitable for detecting patterns or even slight changes that are happening in a linguistic system and is appropriate for eliciting both monolingual and bilingual L2 learners or HSs, regardless of the age group they belong to. In addition, the task set-up provides comparable oral and written data in different levels of formality (data sets 2 × 2). From this perspective, we can explore to what extent participants can differentiate their narrations and which linguistic features are present in the different communicative situations. HSs narrated the event in both their heritage and their majority language and the elicitation sessions in the different languages were scheduled from three to five days apart while monolingual participants took part only in their mother tongue. Different elicitation combinations were created to avoid biasing our participants starting each time with the same setting (see Section 5.9: Appendix). The balanced elicitation process consists of 16 elicitation orders, half starting with the heritage language and half with the majority language.

The task consisted of a short video (00:42 minutes) simulating a fictional event. A minor car accident takes place in a parking lot and participants were asked to narrate to different people what happened supposing that they were present at the time the accident occurred. The video presented is inspired by an everyday event which happens in urban cities and, as the pilot study revealed, participants are familiar with this scene and they feel comfortable to talk about such an instance.

For the elicitation of the different communication situations, we simulated two distinct communication settings: the formal and the informal. The elicitation of the formal setting took place in an office with an elicitor formally dressed. Participants were sitting opposite the elicitor at an adequate distance. The elicitor of the formal register used standardized language with honorifics asking participants to testify in detail how the accident happened to the police. The oral narration of the formal register involved leaving a voice mail on the answering machine of the police department, while the written one was to type a testimony on the police’s laptop as they witnessed the incident.

The elicitation of the informal part took place in a different office, where the two interlocutors were sitting close to each other. A different elicitor was responsible for the informal setting who was casually dressed, and very talkative without using any honorifics. A 20-minute chitchat preceded the elicitation of the data to involve participants in a more relaxed and friendly environment so participants could simulate more easily the targeted communication setting. While they were chatting, the elicitor offered treats to the participants in order to make them feel comfortable. After this conversation, the main part of the study took place. In this communication situation, participants had to narrate in a voice message on WhatsApp what happened to a close friend. The written mode was to send a text message to the same close friend and describe the accident again on WhatsApp. The informal elicitor always reminded participants of their appointment one day before via text message in a friendly and familiar way. The whole process was recorded.

For our study, two groups of HSs were recruited, one in the United States (NY and Chicago) and another one in Berlin, Germany. Moreover, one control group was recruited in Athens, Greece, in order to compare the monolingual variety with the heritage ones. Table 5.1 presents the metalinguistic data that are collected in the form of a questionnaire at the end of the elicitation task. Beginning from the upper rows, the table exhibits the number of speakers in every group, their gender, and their mean chronological age by the time of testing. The following three variables have been measured only for HSs, as the control group has been raised monolingually: mean age of onset to bilingualism, and current input in the heritage and in the majority language. The age of onset ranges from 0 to 8 years old meaning that within the groups there are both simultaneous and sequential speakers, while the current input is measured bidirectionally in two scales from 1 to 3, the first one counting how often participants speak Greek to different members of their family and friends and the other one counting how often each of these people speaks Greek to them. The mean score of the self-ratings has been computed out of four questions on reading, writing, comprehending the spoken language, and speaking the language, first calculated for Greek and then for the majority language on a scale from very easy to very difficult (1 to 5) for each question. The input received from media (TV, radio, blogs in Greek/majority language) is measured on a three-point scale (0–2) measuring the frequency from never to often and is listed under the term literacy practices. Another variable that is included in the metalinguistic data is the generation to which our participants’ parents belong. Thus, we can have a clear picture of their baseline information. Additionally, the years and the hours of formal bilingual education have been calculated for the heritage group based on their reports and the equivalent curricula in each country.

Turning to between group differences we ran several non-parametric Mann–Whitney tests: we indeed find some significant differences between the communities. Comparing the two HS groups, we find that there is a significant difference regarding the literacy practices in the heritage language (U = 953.500, Z = −2.673, p < 0.05) and in the majority language (U = 867.000, Ζ = −3.525, p < 0.001) and regarding the current input in the heritage language (U = 680.000, Z = −4.135, p < 0.001) and in the majority language (U = 649.500, Z = −4.348, p < 0.001). Further significant differences are observed for the visits to Greece (U = 678.000, Z = −5.492, p < 0.001). No other significant differences were found for the remaining metalinguistic factors.

In this section, we present an overview of our results. The total tokens per speaker group from the latest version of the RUEG corpus (Wiese et al. 2021) and the tokens of verbs in absolute numbers and frequencies are presented in Table 5.2. Table 5.2 exhibits also the verbs found in ACT and NAct Voice per group and their frequency based on the number of verbs. Among the matches of verbs in NAct Voice are copulas like ime ‘be’, passives prokalume ‘caused’, reflexives like travmatizome ‘get hurt’, reciprocals like sigkruome ‘collide’, and intransitive deponents like erhome ‘come’. The third line of Table 5.2 exhibits the verbs in NAct Voice having excluded the copula ime ‘be’ and the deponent ginome ‘become’. All queries can be found in the Appendix (Section 5.9).

A manual categorization of the different verb categories is presented in Table 5.3, which includes row instances and frequencies derived from normalization per 100 tokens based on the tokens of verbs found in the corpus. The data upon which the manual categorization has been conducted are stored in an OSF repository.²

What we can notice in Table 5.3 is that, as expected, monolinguals produce the most deponent verbs bearing NAct forms followed by the group of heritage speakers in Germany (although in this group less participants were recruited) and finally the group that produced the least deponent verbs is HSs in the United States.

We observed 10 non-target forms in our two heritage speakers’ groups (see examples (7)–(9)), 7 of which were found in the US data. Examples (7)–(9) are cases where NAct Voice appears on anticausative verbs that do not combine with this form. We found six such instances.

Example (10) exhibits an unaccusative verb which remains [-] transitive but is combined with NAct Voice instead of canonical active.

Examples (11)–(12) exhibit the two novel transitive deponent verbs bearing NAct:

Regarding register variation, Table 5.4 presents the distribution of the verbs in NAct in the different formality settings. HSs produce more NAct verbs in the formal register while the opposite and unexpected pattern is found in monolinguals. The behavior of HSs does not align with our expectations in RQ4 as there is a slight differentiation, although it is not statistically significant in either of our HS groups; the exception here is participant USbi07FG who repeated the same non-target form in all four settings, something that was also predicted in RQ3 (second scenario).

Coming back to the non-target forms in Table 5.3, as these were produced by certain speakers only, namely two in the United States and one in Germany, we looked at the metalinguistic data of these speakers in some detail. With respect to the two HSs in the United States, when comparing their data to those of the whole group (Table 5.5), we observe that, at least for participant USbi07FG, her current input, the self-ratings in Greek, the literacy and the education in Greek both measured in years and in hours that she received are comparatively lower than the majority’s average. This supports Polinsky’s correlation of non-canonical marking and low proficiency. Regarding both USbi97FG and USbi07FG, we can observe that one of their parents is a second-generation immigrant, leading to interesting issues concerning the baseline these speakers were exposed to, which, however, cannot be answered in the context of our methodology.

Running the same comparison between the group of heritage speakers recruited in Germany and the participant that produced non-target forms (Table 5.6), we observe almost the same pattern as with participant USbi07FG, namely lower self-rating score and reduced input and education received in Greek compared to the majority of the speakers in this group.³

Turning now to a more detailed discussion of our results, what we found is actually the opposite of what has been observed in previous literature. Specifically, in our case, NAct appears on anticausative verbs that do not combine with NAct, and certain HSs form novel transitive deponent verbs bearing NAct, although this pattern is very restricted. These findings cannot be explained by language interference, as the two languages with which Greek is in contact, (a) lack Voice marking altogether and deponents (English), (b) do not mark the same set of anticausatives (German) with Voice morphology, and (c) while German, like Spanish, has inherent reflexive verbs which could be viewed as deponents (Alexiadou 2019 for discussion and references), the one verb that German HSs recategorize as deponent is not an inherent reflexive in German. We cautiously suggest that the production of non-target forms might be an outcome of low input in the heritage language, as the metalinguistic data show, and further exploration is needed. Regarding the slight tendency of HSs to produce more verbs bearing the NAct Voice in formal settings, we cannot make strong claims, as it is not statistically significant. However, this striking pattern needs further exploration in future studies. Register leveling is not at stake here unlike our predictions and previous literature introduced in Section 5.1.

Nevertheless, the question arises as to what enables such formation of these non-target forms to begin with. Recall that, building on Embick (1997), Alexiadou, Anagnostopoulou, and Schäfer (2015) proposed that NAct morphology realizes a Voice projection which lacks a specifier. In their analysis, NAct realizes a structure where Voice is marked [-D], signaling the absence of a specifier. In the absence of the Voice projection altogether or in the presence of a DP in Spec,VoiceP, Greek verbs are assigned default ACT Voice. The fact that certain HSs prefer the marked construction for anticausatives might indicate a movement toward a more canonical system, where all intransitive variants bear NAct. Basically, the structure in (13a) is an unaccusative structure and it is associated with passives, reflexives, and NAct-marked anticausatives. As this corresponds to a greater number of non-target forms, we suspect that such overgeneralization is a viable explanation:

The rule that regulates the spell-out of Voice in the absence of a specifier in this system is repeated in (14) below:

Unlike Zombolou and Alexiadou (2012), we did not observe a generalization of actives and we identified a correlation with low input and proficiency in Greek. We may speculate that this is so, as the non-active paradigm shows less allomorphy, and thus appears more canonical and preferred by HSs.⁴ Being more canonical, it is especially preferred by speakers with low proficiency. It could very well be that the speakers in Zombolou and Alexiadou’s study were not of low proficiency, but we do not know that. Nevertheless, which verbs will form anticausatives with non-active is highly idiosyncratic, making the overall distribution of non-active in Greek rather non-canonical. Thus, it might very well be that these two pressures lead to variation among participants: certain participants generalize active in order to avoid non-canonicity at the level of verb form choice. In other words, since some anticausatives bear active, all should bear active eliminating the need to choose the idiosyncratic forms. Others, however, identify the non-active form as the marker of intransitivity, thus generalizing this form to all verbs and eliminating allomorphy at the level of the verbal paradigm.

Matters are somewhat more complex when it comes to the creation of new transitive deponent verbs. Although we only found two such instances in our data, their formation is indeed surprising. It has been argued that transitive deponents belong to at least two different classes (Embick 1997): on the one hand, there are transitive deponents which are psych predicates, and specifically subject experiencer verbs; on the other hand, there are transitive deponents which are not psych verbs and in fact seem to mostly behave like canonical transitive verbs. Our speakers create the latter type of deponents. According to Grestenberger (2018), the external argument in this case is introduced lower, i.e. it is a low agent, and Embick’s NAct rule applies, see also Oikonomou and Alexiadou (2022) and Alexiadou (2024):

We would like to apply this analysis to the new deponents created by HSs. Since (15) is a possible structure in Greek, certain HSs may also generalize it: our novel verbs include an agentive Voice and NAct is the result of an agent being introduced lower than Voice. Oikonomou and Alexiadou (2022) suggest that perhaps Greek grammar productively uses this rule to derive novel interpretations of verbs. Voice forms a single interpretation domain with the vP with which it combines. These particular structures guide speakers to derive novel interpretations.

The final question here is why this option too would be chosen by low-proficiency HSs. We speculate that low proficiency correlates with what Scontras et al. (2018) call restructuring of heritage grammars. In our case, HSs restructure their grammar in making extensive use of a productive structure of Greek: it is the minimal domain that yields novel verbal interpretations.

We discussed two changes in the Voice system of certain HSs of Greek. Unlike what has been reported in previous literature, certain HSs generalize NAct on anticausatives and to a certain extent create novel deponent verbs generalizing a productive rule of Greek grammar. Their behavior supports views of NAct Voice as being productively used by speakers to derive novel interpretations of verbs. This non-default overregularization correlates with proficiency.

Corpus queries:

We are indebted to two anonymous reviewers and the editors of this volume for their comments. Our research was funded by the Deutsche Forschungsgemeinschaft grants AL 554/13-1, 394,836,232 and AL 554/15-1, 313,607,803 to Artemis Alexiadou. The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the DGfS Ethics Committee of the Deutsche Gesellschaft für Sprachwissenschaft (German Society for Linguistics). Dates of approval: January 12, 2017, and February 1, 2019.