Optional subject indexing in spontaneous speech in Modern Persian

1 Introduction

Persian is a pro-drop language with obligatory subject agreement realized by verbal suffixes. In standard Persian, subject agreement for the 3rd person singular is realized by a suffix in present tenses, while it is unmarked in past tenses. In colloquial Persian, however, speakers sometimes use the 3rd person singular pronominal enclitic =eš instead of zero. This use is not endorsed by prescriptive grammars and described as a “new trend” by one of the few existing studies (Rasekh 2014: 25). It could indeed be stigmatized, and even linguistic researchers are inclined to ignore it: in a richly annotated spoken corpus (Adibifar 2016) this subject marker is clearly heard but it does not get transcribed. Descriptive reference grammars do, however, mention this use as a particularity of colloquial or spoken Persian (e.g., Lazard 1957; Paul 2019; Vahidian Kamyar 1964, 2005). The present study is a first usage-based investigation of this non-standard optional subject marker, on the variety spoken in Iran, also known as Farsi.^[1]

The few theoretical studies that have discussed this marker so far focus on its clitichood (clitic vs. affix/agreement marker), aiming to contribute to the ongoing debate on pronominal clitics in Persian and more generally across languages (Jügel and Samvelian 2016; Mahootian and Gebhardt 2018; Rasekh 2014). In this study, I treat subject-marking -eš as an affix; however, while I will present some empirical data to justify my choice, I do not engage with this debate. In other words, my claims and findings do not bear on my view of the affix- or clitichood of -eš.

Up to now, -eš has been viewed as a subject agreement marker resulting from the grammaticalization of the object clitic =eš, as a means of ‘repairing’ or ‘harmonizing’ the verbal agreement paradigm. I show that this account is not empirically justified. On the contrary, all empirical evidence suggests that the optional use of -eš as a sub-standard construction has been part of the indexing system in spoken Persian since its early stages,^[2] while – as all previous studies agree – the written register consistently excludes this use of -eš. I consider this to be a crucial point for the issue at stake. Indeed, it is well-established that spoken and written modalities differ in many respects, in particular in the organization of discourse. What is referred to more specifically by spoken and written modalities here is the difference between spontaneous real time impromptu language production and offline language production, where there is time for choosing between alternative constructions and expressions as well as the possibility for rephrasing (Miller and Fernandez-Vest 2006). Efficient communication does not manifest itself in the same way in these two modalities of language production. Where the written modality prefers (and affords) parsimonious and simple constructions, redundancy is common in spontaneous speech (see, e.g., Degen et al. 2020 and references therein). In a methodologically new approach, I use data from transcribed spoken Persian to uncover patterns of difference between these two modalities. This particular type of data allows for comparing utterances produced spontaneously in real time with their written equivalents. The latter are well-considered sentences produced offline and at a different time by transcribers. These patterns, I argue, show that written Persian prefers parsimonious constructions, whereas in spoken Persian, redundant indexing is common.^[3]

I further show that the use of subject-marking -eš is not a new trend in Modern Persian, and that the historical evidence actually points to the opposite: the evolution of the agreement system in West-Iranian languages (see, e.g., Haig 2017, 2018; Jügel and Samvelian 2016; Lazard 1963) suggests that this use of -eš for indexing the 3rd person singular subject may be a relic of an earlier stage of the agreement system. However, historical and/or regional variation can only explain the existence of zero/-eš alternation, not how it works synchronically. In short, this alternation has not been systematically studied, and, consequently, its triggers remain obscure.

While it is possible to delimit the configuration and, to some extent, the speech context in which -eš can be used to index the 3rd person singular subject, no (inherent or non-inherent) factor is known to require it, i.e., there is no context in which dropping -eš (i.e., zero marking) could be considered ungrammatical or unacceptable. Yet, a basic examination of available spoken corpora easily shows that -eš is far from sporadic in colloquial Persian.

Framing the zero/-eš alternation in terms of Differential Argument Indexing (DAI) – that is, as a case of Differential Argument Marking (DAM, e.g., Seržant and Witzlack-Makarevich 2018), or Differential Subject Marking (DSM, e.g., de Hoop and de Swart 2008) –  it is argued that only a systematic study of subject realization in discourse can shed light on its triggers. These may include lexico-semantic properties of the verb, as well as discourse-related factors, and may involve a degree of sociolinguistic variation. There may be no (more or less) straightforward rule or set of strict constraints or delineated contexts, as for example is the case for Persian’s well-known Differential Object Marking (DOM). Indeed, it is clear that the zero/-eš alternation is not triggered by grammatical rules or ‘hard’ constraints. Nevertheless, there may be a set of ‘soft’ constraints or (statistical) preferences or contexts (dis)favouring the presence of -eš.

The remainder of the article is structured as follows: a brief overview of Differential Argument Indexing is provided in Section 2.1, and an outline of the agreement system in spoken and written Persian with a focus on -eš is given in Section 2.2. I discuss the nature of subject-marking -eš in terms of affix-/clitichood in Section 2.3, and summarize existing claims on the use of -eš as a subject marker in Section 2.4; the research questions are presented in Section 3; the data and methods are described in Section 4; in light of usage data, I underline previous descriptive shortcomings in Section 5; in Section 6, I highlight the importance of the spoken modality in the issue at stake. Section 7 provides a theoretical discussion justifying the analysis proposed in this study against previous accounts.

2 Background

3 Research questions

The aim of this paper is to present a usage-based account of subject-marking -eš in spoken Persian. Firstly, none of the existing studies have based their claims on data from spoken language usage data. Secondly, up to now, subject-marking -eš was either considered solely from an affix-/clitichood point of view, if not ignored, or analysed in terms of grammaticalization. In a nutshell, these studies only considered a few typical examples without any further exploration of other avenues. One of these unexplored avenues is the possibility of viewing the zero/-eš alternation as a case of DAI conditioned by a complex interplay of various factors (see, e.g., Just 2022; Just and Čéplö 2022; Walker 2024; Walker et al. 2024 ).

To this end, I will evaluate previous claims against usage data, explore available historical evidence to test the grammaticalization claim, and consider the hypothesis that the zero/-eš alternation is part of the indexing system of spoken Persian in contrast to written Persian.

It is worth mentioning that the initial goal of this research was to identify the set of factors responsible for triggering (or favouring) -eš. This is clearly a matter of preference and accordingly requires quantitative methods, such as logistic regression or decision trees. Due to the unavailability of richly annotated large-scale corpora (see Section 4.1), this is left for future research – indeed this is a significant methodological challenge, as pointed out in other studies addressing similar research questions (see, e.g., Molochieva et al. 2022).

4 Data collection and methodological challenges

5 Subject-marking -eš in light of usage data

This section presents findings based on a usage study of -eš to index 3rd person singular subject on the verb in spoken Persian. First, previous claims to delimit its use are evaluated: restrictions based on tense in Section 5.1 , and on valency in Section 5.2. Next, in Section 5.3, I present quantitative data comparing the rate of -eš for a set of verbs. Importantly, lexical restrictions and/or preferences are observed in other cases of DAM cross-linguistically (e.g., Walker and Faghiri 2023; Witzlack-Makarevich and Seržant 2018).

6 Relevance of the spoken modality

A crucial point made in previous studies is the persisting strong resistance of formal/written Persian to the use of subject-marking -eš. My data collection confirms this observation: when transcribing spoken data, transcribers frequently correct the speech by dropping extra indexing, according to their personal preference. This is illustrated in (24) and (25), from a TEDx talk.^[25] The utterances are presented as transcribed in written Persian, in a, and as produced in b.

This correction goes beyond the use of subject-marking -eš and includes other uses of this index, as shown in (26) to (28), from the same source.^[26] In the first two examples, the deleted clitic =eš corresponds to an object (P argument) index, and in the third example it marks the possessee.

A similar correction concerns resumptive pronouns in relative clauses. For instance, tu=š ‘in it’, which is frequent in spoken Persian, is regularly replaced with the relative pronoun ānjā, as shown in (29) from another TEDx talk.^[27]

There is another use of -eš that is often corrected in the transcriptions which is not person indexing but, in my view, functions as a tracking device, similar to person indexing, namely for anchoring the utterance in the (previous) speech. In this use, -eš is added to different types of adverbials: e.g., baɁd=eš ‘afterward (lit. after it)’, diruz=eš ‘the day before (lit. its yesterday)’, fardā=š ‘the day after (lit. its tomorrow)’, asl=eš ‘originally (lit. its origin)’, haqiqat=eš/rāst=eš ‘to be truthful/truthfully (lit. its truth)’, vāqeyat=eš ‘in fact (lit. its fact)’. This is a common and frequent use of -eš in colloquial Persian (also see footnote 7).

It is worth noting that these corrections are not limited to the removal of -eš. For instance, as illustrated in the pair of examples below from the same TEDx talk,^[28] redundant NP references are also removed in these transcriptions.

These are non-trivial observations, given that they highlight the complexity of the indexing system in the spoken language in contrast to the written language. And this is by no means limited to the case of Persian. Indeed, indexing some but not all the subjects on the verb brings to mind the well-known cases of clitic doubling. For example, French clitic doubling (including both subject and object) is a prominent feature of colloquial registers, while being rare or even prohibited in careful/formal registers (see, e.g., Liang et al. 2024  and references therein).

In spontaneous speech, including spoken language as well as online chatting, speakers may rely on these devices to put the main information first, and hence make efficient use of their cognitive/memory resources. Indeed, availability-based incremental models of sentence production postulate that, as long as grammar rules do not intervene, the linear order of constituents reflects the order in which they become available for production: there is a general cognitive preference to put constituents activated at an earlier point in time earlier in the utterance (e.g., Kempen and Harbusch 2003).

In Persian, indexing, as a tracking device, goes hand in hand with word order variations and provides the speakers with significant flexibility in information packaging and, therefore, the possibility to reduce speech planning (cognitive) costs. I have selected four examples from the Azad data^[29] (32a–d), to showcase this flexibility. These examples illustrate how in spontaneous speech speakers make use of indexing for more efficient information packaging. Utterances are given as produced with the written standard variants added to the right. For instance, in (32a) the noun budje ‘budget’ is the topic of the sentence. In the Ezafe construction, bozorg šod-an=e budje ‘increasing of the budget’, which is the standard NP construction in Persian (see Samvelian 2007), budje appears in the final position. However, there is the possibility of placing it first, using an alternative order with the help of indexing: budje _i bozorg šod-an=eš _i .

Constructions such as those illustrated in (32) are frequent in spontaneous speech, even in these relatively careful speech contexts. Otherwise, in offline language production, where planning costs are less of an issue, language users voluntarily drop the indexes and use canonical word orders. This contrast is again illustrated by the corrected transcriptions, as illustrated in (33) from the same talk.^[30]

In (33), beyond the usual changes in lexical and morphological levels, we observe different types of corrections, including word order and index drop. In the utterance as produced in (33b), to explain Art Deco the speaker uses a copular clause with a possessive construction (esm=e sabk=eš=e ‘it is the name of its style’), followed by another simple clause (va dar dore=ye dar-vāqeɁ rezā šāh dorost šode be baɁd ‘and is made in fact in the period after Reza Shah’). In the transcription (33a), these clauses are replaced by a relative clause: sabk=i ke baɁd az rezā šāh dorost šode ‘a style which is made after Reza Shah’. In addition, in the transcription the speaker’s hesitation, shown in the late utterance of be baɁd ‘afterwards’, is omitted and rendered with a simplified order baɁd az rezā šāh ‘after Reza Shah’.

7 Optional subject indexing in spoken Persian

In Section 5 and 6, I outlined empirical shortcomings of previous studies and further provided empirical evidence to highlight the importance of the spoken modality in the analysis of -eš. Building on insights from language processing, I showed that the difference between written and spoken Persian is indeed relevant for the issue at stake. This evidence suggests that like many other languages with a standard written form and literary tradition, indexing in spoken Persian is more complex, including variations that are absent from standard written Persian. In the present section, I will further evaluate the alternative view that subject-marking -eš is a means of harmonizing the verbal agreement paradigm.

I focus on Rasekh (2014), which, to my knowledge, proposes the most explicit account of subject-marking -eš in Modern Persian in terms of grammaticalization, claiming that “the grammaticalization of [the] clitic has repaired this ‘defect’ of the paradigm” (2014: 25) – where “defect” refers to the fact that the paradigm has no phonological form for 3SG. In particular, I will show that his account faces at least two serious challenges. First, it does not cover the alternation. In other words, it is not the case that the verbal agreement paradigm is indeed “repaired”. If -eš is to be viewed as an agreement suffix on a par with other agreement suffixes in the same paradigm, then the present state should be considered as transitional and we should expect the use of -eš to be generalized (fully overtaking zero) at some future stage. I, however, do not think that the existing evidence supports this prediction, as I will show that it actually suggests the opposite. In any case, Rasekh (2014) does not provide any historical data or evidence to support this claim. This is even more problematic given the assumption of this “repair” being a new development is also taken for granted. And this is the second issue I will discuss below.

As noted in existing descriptive grammars (e.g., Vahidian Kamyar 1964), it is rather the index drop which is a more recent trend and a particularity of formal/written Persian, whereas -eš has been a feature of Modern Persian since its early days.^[31] Furthermore, it is prominent in some regional dialects, for instance those spoken in Khorasan, including the Mashhadi dialect (Vahidian Kamyar 1964: 36). Importantly, according to Lazard (1963: 257–258), this use of -eš, which he qualifies as pléonastique, is attested already in Early New Persian (8th to 12th centuries), leading Paul (2019: 592) to suggest that “[it] may always have existed, throughout the history of Persian as a possible albeit (until today) non-preferred sub-standard construction”.

The extended usage of 3rd person singular and plural suffixes for representing the subject (both with intransitive and transitive, in the past as well as in the present tense) is conserved well into the modern period, where this use is frequent in the colloquial Persian of Tehran. (Lazard 1963: 257 [my translation])^[32]

Moreover, I follow Haig’s (2018) reasoning and call into question an argumentation that would take historical evidence from written corpora as diachronic justification for considering a feature of spoken language as a new development. Haig refutes object clitic doubling in Persian and argues against van Gelderen (2011), who views clitic object pronouns in Persian as developing into agreement markers, by questioning the assumption that they are new (an “innovation” in van Gelderen’s term), because “[w]e simply lack direct attestation of colloquial spoken Persian prior to the twentieth century, so it is perfectly possible that the kind of sporadic clitic doubling found in modern spoken Persian has been available in colloquial registers for centuries, but never found its way into writing, just as the modern clitic doubling is seldom found in the contemporary written language” (2018: 797).

In addition, putting aside the function of -eš in the Modern Persian agreement system, it is clear that given the diachronic evidence its use cannot strictly speaking be considered “new”, as it appears to be a relic of earlier stages of subject indexing or “des vestiges de cette construction” according to Lazard (1963: 257). Earlier stages of (West)-Iranian languages, namely Middle Iranian languages, including Middle Persian (3rd to 8th century), deployed a quasi-identical paradigm of (second position) pronominal clitics for indexing the P argument in the present tense (34), and the A argument in the past tense (35) (see, e.g., Haig 2018: 798–799).^[33]

Haig (2018) further shows that the same pronominal paradigm has fully developed into subject agreement in Central Kurdish: “In Central Kurdish, the system is still recognizably that of Middle Iranian, but with one very crucial difference: the pronominal clitics that index an A in the past tenses have become fully obligatory” (2018: 799). The following examples are originally provided to illustrate the case of object clitics in the Mukri dialect of Central Kurdish, but they also provide an illustration for A indexing in the past tense, which also shows the similarity with Middle Iranian nicely (compare (35c) to (36a)).

Hence, in terms of the development of the indexing system after the stabilization of the alignment change from ergative (in the past tense) to accusative in Modern Persian (see, e.g., Jügel and Samvelian 2020: 137–138), it seems more feasible to assume that zero marking for the 3rd person singular resulted from simplification of the system. On the one hand, in a six-variable paradigm, the degree of freedom is five, and one value could be dropped. Indeed, this observation should be sufficient by itself to exclude the repairing hypothesis since the system is already (structurally) perfect as it is. However, it does not exclude the “levelling” or “harmonizing” hypothesis (Jügel and Samvelian 2016: 411).

On the other hand, it is a well-established typological observation that zero marking for the 3rd person is common cross-linguistically. It seems reasonable, given the historical evidence, to assume that whereas the zero/-eš alternation continued in colloquial (and/or vernacular) registers, the simplified indexing paradigm, with fixed zero indexing of 3SG, was established as the unified norm promoted as standard Persian with the diffusion of formal written language. This may well be due to an inclination towards a fully transparent (and economical) indexing device – in other words a parsimonious system. The same can be observed with respect to word order. As noted in historical studies (e.g., Khānlari 1994: 18–21 and 266–267), starting from a free word order language, standard Persian has become much more rigid where colloquial Persian shows a greater degree of variation (see, e.g., Faghiri 2016: 30–35).

In addition, there is no evidence to suggest that the zero/-eš alternation would at some point develop into a six-variable paradigm with -eš fully replacing zero. Recall that -eš is a relic of the indexing system in Middle Persian, after the stabilization of the alignment change from ergative to accusative in Modern Persian. This alignment change resulted in (i) a unified behaviour for subject marking in transitive and intransitive predicates, and (ii) a unified behaviour for object marking in past and present tenses. It is reasonable to assume that these unified behaviours are structurally optimal. Now let us consider the two options for 3rd person singular subject marking, i.e., fully non-alternating -eš and fully non-alternating zero, in this respect. Firstly, while zero presents no ambiguity whatsoever, -eš is only unambiguous with intransitive predicates. Secondly, in a non-alternating zero system, there is only one clitic =eš. Hence object indexing operates in exactly the same way for present and past tenses (see Table 1). However, this is no longer the case in a non-alternating -eš system: with present stems it is possible to index both a 3rd person singular subject and object on the verb filling two different (but fixed) slots: mi-xor-ad _A =eš _P ‘s/he eats it’, but with past stems only one slot is available for this combination, mi-xord-eš _A/P ‘s/he ate (it)’. This means that a system with a non-alternating -eš is less systematic/harmonious and displays structural ambiguities. In other words, the prediction that -eš would ultimately be fully generalized (in order to achieve a more harmonious system) is not empirically well-grounded, simply because a system with a non-alternating -eš has fewer structural advantages over a system with a non-alternating zero.

At this point, it is worth mentioning that there is a case of reanalysis (and hence grammaticalization) in the verbal paradigm in colloquial Persian, which, as I will show below, is completely well-founded empirically: subject agreement in the present perfect. The latter is built with the past participle and the copula, as in (37a) and (38a). The past particle is formed with the suffix -e added to the past stem, e.g., rafte ‘gone’, gofte ‘said’, dide ‘seen’. In spoken Persian, the copular clitics are amalgamated with the stem (Paul 2019: 607) and result in one final syllable (37b and 38b). Consequently, in spoken Persian, except for the 3rd person singular, the difference between the present perfect (37b), and simple past tenses (37c), is reduced to the length of the final syllable and/or the position of the stress, which is the last syllable for the present perfect. Stress follows the general rule for the simple past and falls on the final syllable of the verbal stem, and the verbal suffix remains unstressed (see Paul 2019: 585).

For the 3rd person singular, in the spoken language (38b), the copula is fully dropped, and the stress falls on the last vowel, the suffix -e.^[35] The difference between the simple past and the present perfect is hence segmental as well as stress-based. What is further interesting for the issue at stake here is the fact that, even though in the present perfect tense the 3rd person singular bears no person/number marker in colloquial Persian, speakers never use -eš (or any other clitic or affixes for that matter). This is hence a case of grammaticalization in colloquial Persian: the past participle suffix -e is reanalysed as a person/number marker for the 3rd person singular in the present perfect. This undermines the grammaticalization hypothesis in the case of -eš. On the contrary, considering the verbal paradigm more globally, it seems reasonable to postulate that in colloquial Persian -eš (partly) materializes the distinction between the simple past and the present perfect tenses for the 3rd person singular.

Finally, one last piece of evidence to exclude the grammaticalization of -eš as an agreement marker is the fact that it does not behave like other members of the same paradigm. Assuming, for the sake of the argument, that -eš is grammaticalized as a subject agreement marker, it would belong to the same paradigm as the 1st person singular marker -am (39) or the 3rd person plural marker -an (40). However, as illustrated by the ungrammaticality of (41b), -eš does not share all the distributional features of the members of this paradigm. While there may be valid arguments to explain the different combinatory behaviour of eš, the argument against the grammaticalization hypothesis remains: levelling or repairing implies that the expected endpoint (repaired/levelled) result would be a regular homogeneous paradigm, but this is clearly not the case.

In sum, there is no empirical evidence to support an account of -eš in terms of grammaticalization and/or levelling of the verbal agreement paradigm. On the contrary, all the evidence leads to the conclusion that the zero/-eš alternation is a feature of the indexing system in spoken Persian, while absent from written Persian. In other words, optionally indexing subjects on the verb using -eš is a case of DAI and is limited to the colloquial register and/or spontaneous speech.

Written Persian differs from spoken Persian in various syntactic aspects; word order is a well-known feature; another one, highlighted in this study, is argument indexing, in particular, subject indexing. The case of object clitic doubling (see, e.g., (19) in Section 5.2, (26) and (27) in Section 6) is also fairly well-known in the existing literature (see, e.g., Haig 2018: 797; 2020: 100–101). Indeed, in Section 6 we saw that object clitic doubling also exemplifies a case of ‘extra’ indexing in spoken Persian – as opposed to the written language. Interestingly, discrepancies between written (and/or prescriptive) and spoken language in terms of differential indexing have been noticed before (Just and Čéplö 2022). Hence both cases may be viewed as forms of DAI, and the indexing system in Persian could be classified as “split-fluid” in Witzlack-Makarevich and Seržant’s (2018: 28) terms.

8 Conclusions

In this study, I presented a comprehensive overview of subject-marking -eš in colloquial Persian. While using -eš to index the 3rd person singular subject is prohibited by the prescriptive norm, and hence is absent from standard written Persian, it is considered a non-preferred sub-standard construction or a pleonastic use in colloquial Persian by descriptive grammars. Previous research in theoretical linguistics, however, has mostly ignored or neglected this use. The few studies that exist either focus on the clitic/affix properties of the marker or account for it superficially in terms of grammaticalization, making empirically unwarranted claims. My study builds on a sizeable sample of natural data from spoken Persian as well as evidence from historical studies.

I reviewed previous claims and highlighted their shortcomings, and argued against grammaticalization accounts by showing that their hypotheses are not supported by any historical or synchronic evidence. The historical evidence allows us to identify the 3SG subject index -eš as a relic of an earlier stage of Persian’s agreement system, and the zero/-eš alternation as a feature of Modern Persian since its early days. Importantly, my study highlights the importance of the spoken modality in the analysis of -eš, a fact taken for granted previously. To this end, I presented data from spoken Persian transcriptions, made available as subtitles in video recordings in two different contexts. In a nutshell, I compared spoken utterances with their written transcriptions and identified patterns of difference in indexing between the two modalities. I argued that it is possible to explain these patterns on the basis of cognitive principles established by studies in language planning and processing. Spontaneous language production in the spoken modality differs in effect from offline language production in the written modality in terms of utterance planning. Importantly, the spoken modality favours constructions that are efficient with regard to information packaging, and are hence less costly for working memory. The written modality, on the contrary, favours parsimonious constructions regardless of their complexity or cognitive cost. Accordingly, I claimed that extra indexing, namely the zero/-eš alternation, should be considered a feature of the indexing system in spoken Persian, as opposed to written Persian. In this vein, the (redundant) use of -eš to index the subject, similar to object clitic doubling, could be accounted for in terms of differential indexing. Future research, using a robust quantitative methodology, should shed further light on the triggers of these alternations in spoken Persian.