On the persistence of SVO: the case of Modern Eastern Armenian

2.1 MEA and word order correlations

With respect to word order correlations (or harmonies) highlighted in the typological literature (Dryer 1992; Greenberg 1966), Armenian (Eastern and Western) is highly consistent in most of the operator/operand features associated with typologically SVO languages: it is quite consistently left-branching (Donabedian 2010) and classified as SOV/Po/GN/AN by Hawkins (1983: 286). Relying on comparable pairs of correlations, Der-Houssikian (1978: 227–228) claims, “The evidence shows that whereas neither the SOV nor the SVO holds consistently, Armenian is nevertheless predominantly an SOV language.”^[5] As illustrated in Table 1, adapted from Dum-Tragut (2002: 180) with some additions, the only correlation pair that does not hold in MEA is (N, Rel), since the relative clause follows the head noun.

2.2 The VP level

Apart from branching direction, a bundle of syntactic properties at the VP level has been put forward in studies of the syntax of various languages within the generative framework to distinguish between basic OV and VO orders (see Skopeteas and Fanselow 2010, for instance). Some of these properties have also been evoked in the above-mentioned studies of MEA. A survey of the VP in MEA reveals that it mainly qualifies as an OV language with respect to said properties. The latter will be presented after a brief outline of nominal determination.

Nouns in MEA can be definite, indefinite or non-determined/quantified, i.e., bare. The latter are unspecified for number and have a nonspecific reading, existential (non-specific indefinite), as in (1a), or generic (kind-level), as in (1b).^[6]

Indefiniteness in MEA can also be expressed by the indefinite article mi ‘a(n)’, which is prepositive to the head noun and can be separated from it by adnominal attributes. Definiteness is expressed by a suffixed definite article –ǝ or –n. MEA displays differential object marking: Non-human DOs (indefinite or definite) are always in the nominative case, (2) and (3), which is the default case, i.e., zero marking. Definite human DOs occur in the dative case (4a) and indefinite human DOs occur in the dative case when specific (4b) or in the nominative case otherwise (4c).

4.1 Corpus data: a spoken sample of transitive sentences

As mentioned in the previous section, in their corpus investigation based on a sample of 570 transitive sentences from the EANC, Faghiri and Samvelian (2020b) report that for definite DOs, SVO is by far the most frequent order (79.1 vs. 20.9% SOV). They nevertheless note that their study does not allow for any conclusive generalization on the placement of indefinite and bare DOs, because the latter are quite scarce in their sample (respectively, 17 and 23 vs. 340 definite DOs). This is due to the fact that the periphrastic verbal forms in the sample are limited to cases where the auxiliary immediately follows the lexical verb, i.e., the verb and the auxiliary are adjacent, which has a significant consequence on the composition of the sample by disfavoring bare and indefinite DOs. Recall that in focus-neutral sentences, the auxiliary is placed after indefinite/bare objects, and not after the lexical verb (as it is when the DO is definite). While we can safely assume that for definite DOs, word order variation is not likely to be affected by the presence of the auxiliary, for nondefinite DOs, we cannot rely on Faghiri and Samvelian (2020b).

In order to determine whether the presence of the auxiliary affects the choice of OV versus VO, we studied word order variation in a small-scale spoken corpus, in which we identified and annotated all transitive sentences. We used a task-oriented spoken corpus including recordings of picture-based storytelling from 10 native speakers of MEA (see Khurshudyan 2006). The corpus includes 20 narratives, 2 for each participant, a first version in which the participants are asked to tell a story described by a series of pictures, and a second one, taking place after a lapse of 6–8 hours, in which they are asked to narrate the same story from memory.

We identified 231 finite declarative transitive sentences with SOV or SVO orders that we annotated for the presence and position of the auxiliary and object type (see Table 2).

In Table 3, we focus on sentences that include an auxiliary, where it is placed mainly on the verb (in 68.0% of sentences) or on the object (in 14.1% of sentences). “Other” refers to sentences with a negative auxiliary or sentences in which the auxiliary appears on other constituents, e.g., adverbs.

As expected, finite clausal complements (seven cases in our sample) always appear in SVO order, and, in the two sentences including an auxiliary, the latter is placed on the verb.

In the case of definite DOs, in sentences with an auxiliary, the rate of SVO is 52.2%. We looked more closely into SOV sentences and found that in four sentences, there is an idiomatic relation between the verb and the object, which explains the SOV order, and in one sentence, the verb is focused. In sentences with simple verbs, the rate of SVO is 66.7%. Nevertheless, the difference between the distributions of word order in sentences with periphrastic versus simple verb form (11:12 vs. 9:18) is not statistically significant (χ ² (1) = 0.567, p-value = 0.452).

In the case of indefinite objects, in sentences including an auxiliary (placed on the verb in 61.1% of sentences), the rate of SOV is 38.9%. In sentences without an auxiliary, the rate of SOV is 25%. Likewise, the difference between the two distributions (7:11 vs. 3:9) is not statistically significant (χ ² (1) = 0.156, p-value = 0.693/Fisher’s exact test p-value = 0.694). Note that, as we will see below, the surprisingly/unexpectedly low rate of SOV sentences in this sample is plausibly due to an effect of length or heaviness (see Section 5.3 for more details on this effect), that is, caused by the inherent variation in the length of the DOs (roughly, the number of words that constitute the object NP) that appear in those sentences. In the case of bare objects, in sentences with an auxiliary (placed on the object in 64.7% of sentences) the rate of SOV is 87.5%, and in sentences with a simple verb, SOV occurs at the rate of 75.0%. Here again, the difference (14:2 vs. 15:5) is not significant (χ ² (1) = 0.268, p-value = 0.605/Fisher’s exact test p-value = 0.426). Hence, with respect to our working hypothesis, it seems safe to conclude that there is no substantial difference between sentences with simple and periphrastic verbs, given that, if the presence of the auxiliary affected word order for bare and indefinite objects, it would only reinforce the SOV order.

We took a closer look at sentences with bare or indefinite DOs with respect to the length of the DO. We observe that in 7 out of 10 SOV sentences the indefinite DO is simple (i.e., composed of the head noun and the indefinite article only), while in 16 out of 20 SVO sentences the indefinite DO is not simple (i.e., the head noun is modified by an attributive modifier), as in (18) and (19). Accordingly, if we consider sentences with a simple indefinite DO, SOV occurs at the rate of 63.6%.

In the case of bare DOs, the DO is simple (i.e., one word only) in 21 out of 29 SOV sentences, while in 5 out of 7 SVO sentences it is not simple (i.e., the noun is modified by an attributive modifier), as in (20) and (21). Accordingly, if we consider the 23 sentences with a simple bare DO, SOV occurs at the rate of 91.3%.

The findings of our corpus study present a serious challenge for any assumption of the markedness of the (S)VO order in MEA. Even though there may be a corpus-related bias, the word order distribution in our sample still undermines such an assumption. The rate of SVO order is too high to qualify as a marked option. In particular, the fact that definite DOs are overwhelmingly postverbal is in line with Badikyan’s (1976) and Dum-Tragut’s (2009) claim, as well as with Faghiri and Samvelian’s (2020) observation that the placement of the DO is not random and is triggered by the (in)definiteness of the DO, with indefinite/bare DOs being preferably preverbal, while definite DOs tend to occur post-verbally.

Furthermore, in our corpus data, we observe that long and/or heavy bare and indefinite DOs appear frequently in the SVO order, while simple (that is short or minimal) bare and indefinite DOs are more likely to appear in the SOV order. Nevertheless, our corpus data do not allow us to establish a direct relationship between SVO and length/heaviness of the DO. Hence, in addition to the hypothesis regarding the preferred position of these DOs, we also need to verify the effect of phrasal length on their postposition.

To confirm (or reject) these hypotheses in a controlled manner and to verify that the observed tendencies truly hold in all-focus transitive sentences in MEA, we carried out sentence production experiments. In these experiments, we controlled for the verbal lemma and tense. As for the auxiliary, given that our corpus data do not show any discrepancies that need to be further investigated, our stimuli only include simple verbal forms.

4.2 Experimental data

Our corpus data provided a general idea of the distribution of word order in MEA, where SVO appears at a high rate, especially with definite DOs, while bare and indefinite DOs seem to favor the SOV order when they are not long/heavy. In this section, we tested these hypotheses in two-sentence production experiments. Furthermore, corpus samples naturally contain all types of (declarative) sentences, including all-focus (unmarked) sentences, but also sentences that are pragmatically marked. In our two experiments, we used sentence production protocols designed to collect out-of-the-blue sentences.

Another relevant factor that remains totally uninvestigated in previous studies, as well as in our corpus data, is the effect of animacy. The latter is shown to have an effect on word order preferences (according to the well-known universal ‘animate-before-inanimate’ preference or the ‘animate first’ principle, see e.g., Tomlin 1986). Importantly, studies on word order evolution suggest that SVO is more efficient in cases where both the subject and the object have animate referents and, consequently, are likely to be confused (see Gibson et al. 2013). In other words, animate DOs may favor the SVO order. Given that corpus data is generally not well-balanced with respect to animacy, there might be an undiscovered bias, and it is therefore important to independently check whether there is any interference.

Our first experiment (Experiment 1) uses a cued sentence recall protocol and was carried out in Yerevan. Our second experiment (Experiment 2) uses a constrained sentence production protocol and was carried out via a web-based questionnaire. In Experiment 1 we compare definite and indefinite DOs and in Experiment 2 we compare indefinite and bare DOs.

4.2.1 Experiment 1

The goal of this experiment is to compare definite and indefinite DOs and, at the same time, to study the effect of the animacy of the DO. Animacy is a factor known to have an effect on word order variation crosslinguistically, and importantly, studies on word order change and evolution suggest that SVO is likely to be favored over SOV when the subject and the object both have animate referents (see Gibson et al. 2013). In other words, it is possible that animate DOs favor the SVO order, yet we did not investigate this correlation in our corpus study.

4.2.1.1 Procedure

We used a cued sentence recall protocol, presented as a memory task. Our protocol follows a well-known task used in various experimental studies on word order preferences (e.g., Stallings and MacDonald 2011). Participants were asked to make a sentence with given constituents that appear on a computer screen and to produce it from memory with the help of a cue from the sentence, after a short lapse of time during which, for distraction, they were presented with a basic arithmetic operation. Figure 1 presents the sequencing of different screens as they were presented to participants.

Figure 1:

Sequence of screens as presented to participants in Experiment 1.

Participants were recorded individually. In each session, the experiment was run on one laptop (using the Ibex Farm platform) and recorded via a headset connected to another laptop. The instructions were given verbally and a practice item was presented by the instructor. The participant first filled out a profiling form with information such as their age, language(s), and region. Then they began the trial with a “warm up” session including 3 practice (filler-type) items. Afterward, the participant was asked to start when ready. Each trial included a total of 65 items (20 target items and 45 fillers).

4.2.1.2 Design and material

We manipulated two factors, following a 2 × 2 factorial design: (i) the definiteness of the DO on two levels: indefinite mi astʁ ‘a star’, versus definite astʁ-ə ‘the star’ (see Section 2.1), and (ii) the animacy of the DO on two levels: astʁ ‘star’, versus kin ‘woman’. Table 4 summarizes our four experimental conditions.

Table 4:

Experimental design (Experiment 1).

Condition		Subject	Verb	DO
Cond 1	Indef/Anim	Arpi	saw	a woman
Cond 2	Indef/Inan	Arpi	saw	a star
Cond 3	Def/Anim	Arpi	saw	the woman
Cond 4	Def/Inan	Arpi	saw	the star

We prepared four lists of 20 transitive sentences according to our four experimental conditions, as in (22).

(22)

a.

Arp’i-n	astʁ-ə	tes-av
Arpi-DEF	star-DEF	see-AOR.3SG
‘Arpi saw the star.’

b.

Arp’i-n	mi	astʁ	tes-av
Arpi-DEF	a	star	see-AOR.3SG
‘Arpi saw a star.’

c.

Arp’i-n	knoǯ-ə	tes-av
Arpi-DEF	woman.DAT-DEF	see-AOR.3SG
‘Arpi saw the woman.’

d.

Arp’i-n	mi	kin	tes-av
Arpi-NOM	a	woman	see-AOR.3SG
‘Arpi saw a woman.’

We used the Armenian equivalent of the following verbs (or pair of verbs), each repeated four times: see, find, remember, kill (with animate DOs)/break, beat (with animate DOs)/throw. In 10 items the verb was in the aorist tense (e.g., tesav ‘saw’) and in the other 10 items it was in the conditional/future tense (e.g., ktesni ‘shall see’).

Note that among the three constituents, one is always a proper noun in the nominative case, so the only way to produce a grammatical sentence is to choose the latter as the subject of the sentence. Recall that proper nouns, when used as DOs, must be in the dative case given that MEA displays differential object marking (see Section 2.2).

As illustrated in Figure 2, the subject appeared in the middle of the screen in all the items. The verb appeared at the top (and the DO at the bottom) in 10 items and the reverse in the other 10 items. The cue to produce the sentence was always the subject, hence all the sentences produced by the speakers were expected to begin with the subject. The order of the verb and the DO constituted our dependent variable, with two possible values: (S)VO and (S)OV.

Figure 2:

Screenshot of an item from Experiment 1.

4.2.1.3 Participants

38 native speakers of MEA (living in Armenia) were recorded – including 27 female participants and 11 male participants, with a mean age of 30 years (ranging from 18 to 66 years).^[18] The data were collected during a research stay in Yerevan in October 2019. The recording took place at the Institute of Philology and Ethnography of the Armenian National Academy of Sciences.

4.2.1.4 Results and discussion

Out of a total of 760 (potential) sentences, we gathered a sample of 732 produced sentences – discarding incomplete and/or erroneous sentences, as well as 4 isolated occurrences displaying an unexpected order (OVS, OSV). Table 5 provides the distribution of orders in this sample. We observe that definite DOs strongly favor the SVO order, while indefinite DOs show a clear preference for the SOV order. As for animacy, we only see a small difference between animate and inanimate DOs, with animate DOs slightly favoring SVO. As we can see in Table 6, this difference is larger for definite DOs than for indefinite DOs.

Table 5:

Word order by definiteness and animacy (Experiment 1).

Order	Definite DOs			Indefinite DO			All
	Anim	Inan	All	Anim	Inan	All
SOV	26	37	63	114	113	227	290
SVO	155	146	301	66	75	141	442
Total	181	183	364	180	188	368	732

Table 6:

%SVO by definiteness × animacy (Experiment 1).

Condition	Definite	Indefinite	All
Animate	85.6%	36.7%	61.2%
Inanimate	79.8%	39.9%	59.6%
All	82.7%	38.3%	60.4%

The statistical analyses of the results, using a logistic mixed model – with sum-coded fixed effects and SVO coded as success, including items and participants as random effects – shows a general preference for SVO (Est. = 0.74, p < 0.01), and a strong main effect of definiteness, definite DOs favoring SVO (Est. = 1.42, p < 0.001).^[19] The results of this experiment hence confirm the findings of our corpus study showing that in all-focus transitive sentences, definite DOs strongly favor the SVO order, while indefinite DOs are more likely to occur in the SOV order.

With respect to animacy, the results of this experiment are also very interesting: while the main effect of animacy is not significant (p = 0.23), there is a marginally significant interaction (Est. = 0.20, p = 0.055) indicating that animacy has an effect on the order with definite DOs: animate definite DOs favor SVO more than inanimate definite DOs do. In other words, there exists a tendency to put animate DOs in the postverbal domain, i.e., the direction predicted by Gibson et al. (2013), only with definite DOs, which are more likely to be confused with the subject.

4.2.2 Experiment 2

In this experiment, we focused on the difference between indefinite and bare DOs, as well as on the effect of length. In corpus data, we observe that long/heavy indefinite and bare DOs appear more frequently in the SVO order. The aim of this experiment is, first, to determine the preferred order for simple bare and simple indefinite DOs, and, second, to verify if longer (nondefinite) DOs favor SVO. To this end, we compared simple DOs with “minimally” long DOs, i.e., DOs modified by a single attributive modifier. In their experimental study on the effect of phrasal length on word order in Persian, where definiteness also plays a role on the linear position of the DO, Faghiri and Samvelian (2020a: 18–22)^[20] showed that when an attributive modifier is added to bare or indefinite DOs they are significantly more likely to behave like definite DOs.^[21] Hence, if the results of our experiment show that minimally longer DOs are more likely to appear in the SVO order, we have enough evidence to confirm that phrasal length favors SVO.

4.2.2.1 Procedure

Experiment 2 was carried out on the Ibex Farm platform (Drummond 2013) as a web-based self-administrated questionnaire.^[22] In this experiment, we used an online constrained sentence production task, following the protocol proposed by Faghiri (2016). This protocol proposes a task that is fairly similar to the cued sentence recall production (see Faghiri and Samvelian 2020a: 17–18).

In this task, participants were presented with a preamble and were asked to complete the latter with phrases that appeared on the screen in order to make a sentence. They simultaneously saw a preamble, an incomplete sentence represented by three blank boxes, and a vertical list of phrases (presented in a counterbalanced order, as was the case in Experiment 1). A screenshot of an item (with English translations) is given in Figure 2. The list of options contained four elements, that is, one element more than the number of blanks, in order to prevent participants from guessing the purpose of the experiment and to push them to concentrate on the content of each sentence, so that they produce reasonably natural sentences.

As in Experiment 1, the questionnaire included a profiling form, as well as some practice items, after which the participants were asked to start when ready. Likewise, each trial included a total of 65 items (20 target items and 45 fillers).

4.2.2.2 Design and material

Experiment 2 also follows a 2 × 2 design, with the two following variables: 1) bareness, contrasting bare, astʁ ‘star’, with indefinite, mi astʁ ‘a star’, DOs, and 2) length, contrasting simple/short (one word) DOs with DOs modified by an attributive modifier, such as, p’aylun astʁ ‘bright star’ as in (23).

(23)

Yerekoyan	Arp’i-n	(mi)	(p’aylun)	astʁ	tes-av
evening	Arpi-DEF	a	bright	star	see-AOR.3SG
‘In the evening, Arpi saw (a) (bright) star.’

As illustrated in Figure 3, in this experiment, an adverbial scene-setting phrase constituted the preamble, and the list of four phrases included the subject, two formally identical choices for the DO (DO1 and DO2), and the verb.

Figure 3:

Screenshot of an item from Experiment 2.

We prepared four lists of 20 transitive sentences according to our four experimental conditions, as in Table 7. The stimuli are entirely based on those used in Experiment 1. Apart from the adverbial preamble, all elements are identical. The two choices of DO match the DOs in animate and inanimate conditions. The same set of verbs was used, but, in items where we had different verbs in animate and inanimate conditions in Experiment 1 (kill/break, beat/throw), we kept the one that required an inanimate DO. The verb appeared at the top of the list and the subject at the bottom in 10 items and the reverse in the other 10 items. The DOs appeared in the middle, with the inanimate choice appearing above the animate choice, as in Figure 3.

Table 7:

Experimental design (Experiment 2).

Condition		Preamble	Subject	Verb	DO1	DO2
Cond 1	Bare/Short	In the evening	Arpi	saw	star	woman
Cond 2	Bare/Long	In the evening	Arpi	saw	bright star	old woman
Cond 3	Indef/Short	In the evening	Arpi	saw	a star	a woman
Cond 4	Indef/Long	In the evening	Arpi	saw	a bright star	an old woman

4.2.2.3 Participants

28 native speakers of MEA from Armenia filled out our online questionnaire. We had 19 female and 9 male participants, from 27 to 70 years old.

4.2.2.4 Results and discussion

Out of a total of 560 (potential) sentences, we collected a sample of 525 produced sentences – discarding erroneous sentences as well as sentences with infrequent orders. Table 8 shows the distribution of word orders in this sample. To analyze the effect of our two experimental factors, we put aside the 32 OVS sentences. It is worth noting that neither DO type shows a bias toward this order in this sample.

Table 8:

Word order by bareness and modification/length (Experiment 2).

Order	Bare DOs			Indefinite DO			Total
	Simple	Modified	Total	Simple	Modified	Total
SOV	98	89	187	99	84	183	370
SVO	27	33	60	24	39	63	123
OVS	10	7	17	9	6	15	32
Total	135	129	264	132	129	261	525

We observe a strong general preference for SOV order with a trivial difference between the two types (see Table 9).^[23] There is, however, a more important contrast triggered by modification. As expected, the rate of SOV is lower with modified/longer DOs.

Table 9:

%SOV by bareness × length (Experiment 2).

Condition	Bare	Indefinite	All
Simple	78.4%	80.5%	79.4%
Modified	73%	68.3%	70.6%
All	75.7%	74.4%	75.1%

The statistical analyses of the results, using a logistic mixed model with sum-coded fixed effects and SOV coded as success and including items and participants as random effects,^[24] shows a significant general preference for the SOV order (Est. = 1.56, p < 0.01) and the main effect of length/modification (Est. = −0.40, p < 0.01), but no main effect of the DO type (p = 0.55) nor any interaction between the two factors (p = 0.17).

These results confirm the preference of nondefinite DOs for the SOV order, also observed in our corpus study, indicating furthermore that indefinite and bare DOs are equally likely to favor the SOV order. Interestingly, our results also confirm the observation that non-simple DOs are more likely to appear in the SVO order. In other words, long non-definite DOs tend to behave more like definite DOs.

5.1 The role of areal diffusion

One hypothesis is that the high rate of postverbal (definite) DOs in MEA is an areal diffusion phenomenon, due to contact with Russian, a VO language.^[25] Under this putative view, in the same way that Classical Armenian shifted from (S)VO flexible to (S)OV flexible, probably due to long-standing contact with OV languages, i.e., Iranian languages and Turkish, MEA is now shifting from (S)OV to (S)VO. This assumption is appealing, especially if we consider word order preferences in other OV languages of the area that have been in a similar contact situation with Russian and for which studies on word order variation are available.

The case of Modern Georgian is particularly interesting in this respect. Word order preferences and change have been thoroughly studied in Georgian (see Harris 2000, for an overview). While both OV and VO orders may occur out of the blue, Asatiani and Skopeteas (2012: 128) note that “corpus studies show a clear preference for OV orders in written discourse (Apridonidze 1986: 136–143; Vogt 1971: 222) and same frequency for OV and VO in texts that originate in oral tradition (fairytales).” The situation is thus comparable to MEA, all DOs taken together (i.e., regardless of definiteness): both MEA and Georgian are left-branching languages in most of their constituents but display a very high frequency of VO. The two languages are also comparable in their historical evolution. While Old Georgian (5th–11th centuries) is said to be SVO flexible, Modern Georgian is described as SOV.^[26] Studies on word order variation in Georgian have mainly focused on information structure, especially on the issue of post-verbal focus (Skopeteas and Fanselow 2010), but have not looked into definiteness.^[27] Nonetheless, it is noteworthy that two genetically different languages of the area have followed the same path of evolution.

The case of Azerbaijani, a Turkic language generally considered to be rigid OV, spoken in the Republic of Azerbaijan, in the South Caucasus, but also in greater numbers in Iranian Azerbaijan, is also particularly relevant. Stilo (2014) addresses the occurrence of both pre- and postverbal DOs in the colloquial Azerbaijani of Tabriz (Iran) and Masally (Republic of Azerbaijan) and also in some other non-Turkic languages of the Araxes-Iran Linguistic Area (AILA,^[28] see Stilo 2018), taking into account the factor of definiteness. Stilo’s investigation is based on corpora containing on average ca. 500–600 clauses per (variety of) language. His study reveals a significant difference in the rate of postverbal definite DOs in the two varieties of Azerbaijani. While 100% of indefinite DOs are preverbal in both varieties, 20% of definite DOs occur after the verb (VO) in the variety spoken in Massally, against only 4.6% in the variety spoken in Tabriz.

In order to check whether this difference is coincidental or whether other factors such as areality (or diffusion) are at play, Stilo extends his investigation to non-Turkic languages of the AILA. This investigation shows that, regardless of their genetic affiliation, languages spoken in the north^[29] AILA have an overall significantly higher rate of postverbal DOs than those spoken in the south.^[30] Stilo concludes that there is an isogloss line in the area that separates the AILA languages into a southern zone and a northern zone with regard to postverbal DOs. In other words, the postverbal placement of DOs, and more specifically definite DOs, is an areal phenomenon.

Finally, let us consider Caucasian Urum, an endangered variety of Anatolian Turkish spoken in Georgia. Schröter (2019) examines the effect of focus on word order preferences in this language, in comparison to its substrate language, Turkish, and its contact language, Russian. The results show that both OV and VO can occur in similar discourse contexts in Urum and that focus can occur both pre- and postverbally. This flexibility of Urum is claimed to be a result of its contact with Russian.

To resume, various studies of the languages of the South Caucasus have highlighted the fact that postverbal DOs are a common feature of OV languages of the area. Consequently, it does not seem unreasonable to assume that the high rate of postverbal definite DOs in MEA is also a manifestation of this areal convergence. However, in MEA, contrary to other OV languages of the area, VO is not a marginal option, but the default order for definite DOs. Therefore, areal diffusion, though effective in the reinforcement of the VO order, cannot be the only explanation for its prevalence.

5.2 Word order distribution in Classical Armenian

Our hypothesis is that OV has never been the unmarked order for definite DOs in MEA. While a consistent shift took place from right-branching to left-branching in almost all constituents, historically, word order remained almost/substantially unchanged at the clausal level. Crucially, we will argue that the parameters favoring VO or OV order are analogous in Classical Armenian and MEA.

As previously mentioned, Classical Armenian is generally said to be SVO flexible, or, more precisely, right-branching (cf. Table 1). Note, however, that, like MEA, Classical Armenian displays a high degree of flexibility at the clausal level, so determining which order (VO or OV) was dominant is not a trivial task. Hayrapetyan (1981) investigates this variation in a number of texts from the fifth century, showing that the preference for each ordering may vary considerably depending on 1) whether the text is translated (from the Koine Greek); 2) the period. In 791 sentences from the New Testament (translated from Koine Greek), 619 sentences are in the VO order, and only 172 display the OV order, while in a sample of 732 sentences from texts originally written in Classical Armenian,^[31] OV is slightly prevalent, 390 OV versus 312 VO. Hayrapetyan (1981) reports that the balance changes in favor of VO in the post-Mesropian era (i.e., beginning in the second half of the fifth century):^[32] in a sample of 566 sentences taken from post-Mesropian texts,^[33] VO prevails with 338 occurrences, while 228 sentences display the OV order.

Interestingly, Hayrapetyan (2005) reports comparable rates for OV and VO orders in Classical Armenian and MEA (Table 10). If we abstract from the subject, VO represents 58% and OV 42% in Classical Armenian, which is quite similar to MEA rates, 56% VO and 44% OV.

Hayrapetyan (1981) investigates syntactic, discourse and functional parameters favoring pre- and postverbal orders, such as part of speech (e.g., interrogative pronouns occur preverbally), length (long DOs favor VO order), verb type (e.g., causative and resultative verbs favor VO order), the “construction” of the sentence (e.g., the non-realization or postverbal placement of the subject favors the preverbal placement of the DO), information status (focused DOs are preverbal), etc.

Hayrapetyan (1981) does not mention definiteness per se as a parameter at play in the placement of the DO. In order to check whether definiteness was a relevant parameter in the placement of DOs in Classical Armenian, as it is in MEA, we conducted a small corpus-based investigation of about 50 finite sentences in texts dating from the fifth to seventh centuries.^[34] In this sample, 26% of sentences show OV order, and interestingly, nondefinite DOs appear in the preverbal domain twice as often as definite DOs: out of 29 sentences with a definite DO, only 6 show OV order (21%), while out of 13 sentences with a nondefinite DO, 6 show OV order (46%). The main observation is that non-definite DOs already display a tendency to occur in preverbal position, while definite DOs are placed after the verb.

In sum, despite its high degree of flexibility with respect to the positioning of the DO, Classical Armenian nevertheless displayed a preference for placing definite DOs after the verb. Interestingly, parameters such as length and focus, which Hayrapetyan (1981) mentioned as playing a role in the placement of the DO, are also at work in MEA.

5.3 Cognitive pressures behind the SVO–SOV split in MEA

The persistence of (S)VO from Classical to Modern Eastern Armenian calls for an explanation. We account for this persistence in terms of a general crosslinguistic bias favoring SVO order highlighted in typological and historical studies. Based on an impressive body of works on word order change in various languages,^[35] studies of the dynamics of word order change converge on the view that languages do not change toward all orders equally often. More precisely, while languages drift away from SOV (but not toward it), they tend to drift toward SVO (Li 1977; Givón 1979; Gell-Mann and Ruhlen 2011, among others). This is not to say that the shift from SVO to SOV never happens, but if it does, it is the result of contact (i.e., external pressure),^[36] while change to SVO may occur naturally.

Various accounts have been proposed for this general bias in favor of SVO order. Studies in cognitive science and language processing show that from the point of view of communicative efficiency, that is, with respect to the effort imposed on working memory, SVO provides important advantages that can explain the direction of word order evolution from SOV to SVO (see Gibson et al. 2019: 8–10 for a review).

Our empirical study showed that although the default position of bare and indefinite DOs is preverbal, longer indefinite and bare DOs favor the postverbal position. This is not surprising given general crosslinguistic observations that show that word order variation tends to reduce parsing complexities and cost (or maximize parsing efficiency), by means of minimizing the length between the verb and its dependents, also known as dependency length minimization (e.g., Gibson 1998; Hawkins 1994; Temperley 2008).^[37] Consequently, under this view, generally speaking, a sentence is more efficient in SVO order than in SOV, everything else being equal (cf. Gibson et al. 2019: 8–10 for a review).^[38]

On the other hand, SOV also presents other important cognitive advantages, for instance from the point of view of information theory, by allowing maximization of the predictability of the head verb (Ferrer-i-Cancho 2015). Note that this can explain the initial preference for this basic order. With a simple (short) DO, the pressure for dependency-length minimization is smaller. With long DOs, however, as illustrated by the pair of sentences in Example (24) compared to Example (25), placing a long DO in the postverbal domain results in a less complex sentence, by greatly reducing the distance between the verb and its subject.^[39]

Importantly, as argued by Ferrer-i-Cancho (2014, 2015 and Ferrer-i-Cancho and Gomez-Rodrigez (2021), word order preferences in different languages could reflect a tradeoff between two main pressures: 1) Minimizing the distance between the verb and its dependents, in order to reduce the processing cost, favoring SVO and 2) Maximizing the predictability of the verb, favoring SOV. The evolution of word order in MEA can thus be explained by the interaction between these cognitive constraints.

Classical Armenian was already a highly flexible language with respect to word order at the clausal level, taking advantage of both pre- and postverbal placement of DOs for various purposes, depending on a set of syntactic and functional parameters. Longer DOs (DOs containing modifiers, but also clausal and infinitive complements) favored SVO over SOV as regards the arrangement of constituents. The same holds for definite DOs, which displayed a preference for postverbal position. Preverbal position, on the other hand, was dedicated to short nondefinite DOs.

The preverbal position of bare/indefinite DOs in MEA was probably reinforced due to contact with Iranian languages,^[40] which resulted in an abundant use of VP idioms or so-called ‘complex predicates’ in Armenian. Iranian languages generally speaking, and Persian in particular, are well known for abundantly resorting to noun-verb sequences that display the semantic properties of a lexeme while behaving syntactically as two distinct entities, comparable to a verb and its (bare) direct object. In Classical Armenian, ‘phraseological calques’ of these complex predicates, where Iranian material was replaced with Armenian counterparts, were used to coin new lexical units (see Bolognesi 1993: 264–266; Meyer 2017: 27–31; Schmidt 1983: 104, among others). It is reasonable to assume that these constructions reinforced the preverbal placement of bare/indefinite direct objects in Classical Armenian since there is no clear-cut distinction between complex predicates formed by a bare noun and a verb on the one hand and any syntactic sequence formed by a bare object and a verb on the other hand (Samvelian 2012).^[41]

The regularization of preverbal position for bare/indefinite DOs had its own cognitive advantages. The preverbal nominal element of a complex predicate contributes to a great extent to its argument structure, lexical conceptual meaning, and event type, thereby playing a crucial role in building the meaning of the complex predicate.^[42] Placing the noun before that verb not only accelerates the retrieval of the meaning of the predicate but also increases the predictability of the verb (cf. Ferrer-i-Cancho 2015). Given that the number of nouns that are likely to form a complex predicate is much greater than the number of verbs that are likely to appear in such constructions,^[43] having first access to the noun is a more efficient cue for predicting the verb than the reverse. The same reasoning holds mutatis mutandis for bare objects, which have been argued in several studies to be part of the predicate or the event described by the verb rather than being arguments of the latter.^[44] Like the nominal element of complex predicates, the bare object is an important cue for the interpretation of the whole predicate since it narrows or specifies the meaning of the verb, thereby accelerating access to its meaning. Therefore, its placement before the verb maximizes the predictability of the latter (Ferrer-i-Cancho 2015).

To summarize, Classical Armenian, being SVO flexible, could make use of both pre- and postverbal placement of the DO for various discourse purposes, also depending on functional parameters optimizing processing. Contact with Persian reinforced the use of the preverbal position for bare/indefinite DOs but did not result in the abandonment of the postverbal position. It strengthened the division of labor between the two domains, giving rise to what can, loosely speaking, be called a complementary distribution. The fact that both orders, OV and VO, are efficient in themselves and in interaction with each other explains the persistence of their coexistence from Classical to Modern Eastern Armenian.