Obligatory subjects and the SVO/SOV debate

1 Introduction

In their contribution, Haider & Szucsich (henceforth H & S) discuss a number of properties that distinguish SVO from SOV languages and focus on the classification of Slavic languages within this typology. They show that Slavic languages seem to lack the properties that define typical SVO languages. They then conclude that Slavic languages seem to belong to a third type of languages, in which the position of the head is not constrained directionally.

H & S review eight such properties. These are illustrated in the following table, taken from their contribution:

In this contribution, I will focus on the second property and point out some issues with H & S’s typology. I will first raise the question with respect to the context/exact nature of the criterion H & S use to diagnose the obligatory presence of a preverbal subject. I will show that indeed a generalization emerges, albeit a different one from what the authors propose. I will then discuss an argument brought in the literature on Russian in favor of the obligatory presence of subjects. Finally, I will place Greek in the context of this discussion and offer an explanation with respect to the lack of impersonal passivization in Greek and across languages.

2 SVO and strict impersonal passives

In H & S’s discussion, impersonal passivization is a context which discriminates between SVO and SOV languages: in the former group of languages the presence of an expletive is obligatory, while in the latter this is not the case. Let me briefly summarize their exposition. According to H & S, SOV languages are known to not require an obligatory subject in otherwise subjectless clauses. To this end, German is contrasted with SVO Norwegian and Swedish, as shown in (1), their (4):

According to the authors, English is the only Germanic language in which intransitive verbs cannot be passivized, the reason being that there is no suitable expletive subject available for satisfying the mandatory subject position. While English has both it and there expletives, none of them seems to be selected for this impersonal structure, since, as Haider (2019) argues, they have been recruited for other purposes.

While Slavic languages are a sub-type of pro-drop languages, H & S and Haider (2019) argue that the obligatory subject requirement is independent of the pro-drop nature of the language under discussion. Specifically, H & S show that this applies to Romance null-subject languages as well, irrespectively of their pro-drop nature: passives of unergatives are out, as these languages lack an expletive subject. This is illustrated with Spanish, Italian and French examples in (2), their example (5). For H & S (2d) is in, as French employs a pronoun as an expletive subject. The Spanish and Italian examples are out, as these languages are pro-drop, and lack a null expletive. That is, this structure would require a null expletive, but pro-drop languages lack null expletives, as argued for in detail in Alexiadou and Anagnostopoulou (1998), a view the authors also endorse:

The data in (2) are thus taken as evidence suggesting that the null subject property does not cancel the obligation of a lexical filler for the structural subject position of an [S[VO]] clause.

First, note that the French examples H & S cite include a PP, thus (2d) is not really a so-called strict impersonal passive, i.e., a passive of a plain unergative, as discussed in Pitteroff and Schäfer (2019). I will come back to the PP issue. Second, limiting ourselves to the presence of an expletive in strict impersonal passives, we note that Icelandic is a Germanic SVO language, which, however, also lacks expletives in passives of plain unergatives, as shown in (3), Pitteroff & Schäfer’s (69b). (3) is similar to (1c). Haider (2019) notes that and argues that the absence of an expletive in the Icelandic example (3) might be due to the fact that another maximal projection has been fronted, namely the adverb. To this end, Haider (2019) suggests that contexts such as the ones in (4) are more instructive, where an overt expletive is present in Icelandic but not in German. This contrast suggests that Icelandic is an SVO language. If contexts such as the ones in (4) are more informative, then the presentation of the paradigm in (1) is slightly misleading. Thus, the point seems to be about the right contexts that diagnose the presence of expletive subjects across languages, and whether XP fronting is independently available to satisfy the EPP requirement. This seems to be the case in Icelandic (Holmberg 2000), but not in German, (4b):

But is impersonal passivization the best testing ground for the availability of true expletive subjects? For Haider (2019), the presence of expletive subjects is the hallmark property of SVO languages. However, this particular structure is subject to crosslinguistic variation, discussed in Pitteroff and Schäfer (2019), who note that strict impersonal passives are also out in French. This kind of variation is not expected under H & S’s view, as French is a strict SVO language:

Pitteroff and Schäfer (2019) point out that when impersonal passives are allowed in French, these cases involve either VP-internal object DPs or argument or adjunct PPs, as in the example H & S cite. This suggests that the availability of strict impersonal passives does not seem to correlate with a) the availability of a dedicated expletive for the subject position, as French and Icelandic both have this type of expletive, b) the SVO versus SOV nature of the language, as strict impersonal passives are out in a canonical SVO language, French, but possible in another canonical SVO language, namely Icelandic. French differs from Icelandic in that it does not allow any category to become an expletive, in Holmberg’s (2000) terms.

Interestingly, Pitteroff and Schäfer (2019) suggest that in Russian strict impersonal passives are out too, but importantly their acceptability depends on the presence of an argumental PP, see (6), their (76c). It is not clear how to capture the PP effect in H & S’s system for both French and Russian.

The picture that emerges from Pitteroff & Schäfer’s study is that English, French and Russian lack strict impersonal passives, while Norwegian, Dutch, German and Icelandic have strict impersonal passives. This picture does not match the SVO versus SOV dichotomy proposed by H & S. The explanation Pitteroff & Schäfer offer is that impersonal passives fail because the EPP remains unchecked or T’s phi features remain unvalued or both. By contrast, Pitteroff & Schäfer argue, if a language has a dedicated expletive (Norwegian, overt expletive, and Russian, albeit covert) or is not an EPP language (German and Icelandic), then this language will allow strict impersonal passives. This conclusion is partially in line with H & S’s typology, as German is also viewed as a non-EPP language by them, while Norwegian is an EPP language.

However, it now seems that being an SVO language and having an expletive subject are two independent properties, contra Haider (2019). In other words, the question is: does the presence of an expletive subject and the availability of impersonal passivization correlate with the SVO nature of a language or is the EPP requirement independent of the SVO nature of the language? For H & S the answer to the first question is positive. But if Icelandic is an SVO, non-EPP language, suggesting that EPP and SVO do not correlate, this is not captured by H & S’s typology. The other important issue is why in French and Russian impersonal passives are better if argumental PPs are present, which is again not immediately captured by H & S: Pitteroff and Schäfer (2019) argue this suggests that the overt element in Spec,TP in French or its covert counterpart in Russian is not a real expletive and needs to associate with an argument (see their paper for details). This in turn means that there is an obligatory subject in Russian.

There is a further point to be made about the obligatory presence of subjects. H & S cite McCloskey (2001), who argued that Irish truly lacks expletive subjects. Perlmutter and Moore (2002) contrast Russian to Irish by discussing McCloskey’s contexts. According to Perlmutter and Moore (2002: 643), Russian differs from Irish in various respects, including the presence of personal-impersonal alternations, which are absent in Irish, but robust in Russian, the availability of a definiteness effect, which is absent in Irish but can appear in certain Russian impersonals and obligatory raising, which is absent in Irish, but robust in Russian. I will turn to this last point now.

As discussed in Perlmutter and Moore (2002: 636), Russian has several raising predicates, e.g., načinat’/načat’ ‘begin’, perestavat’/perestat’ ‘stop’, prekraščat’/prekratit’ ‘stop, cease’, and prodolžat’l/prodolžit’ ‘continue’. These verbs take infinitival complements. Importantly, when these predicates take complements, raising is obligatory, as shown in (7), their (71). (7b-c) without raising are ungrammatical. This contrasts with H & S’s conclusion that Russian lacks an obligatory subject. If it lacked an obligatory subject, raising should not occur, as is the case in other pro-drop languages, such as Greek, see Alexiadou and Anagnostopoulou (2021). Such languages show long distance agree, and the subject may remain in its base position in the embedded clause. It may also appear preverbally, but then it is interpreted as a topic, see Alexiadou and Anagnostopoulou (1998):

Perlmutter and Moore (2002: 636) further note that if impersonal clauses have an expletive subject, this will raise, yielding a grammatical result when the impersonal is embedded under a raising predicate, see (8), their (72). Typically, another maximal projection will raise to ‘shield’, as the authors state, the verb from initial position, making this very similar to the cases of fronting in Icelandic:

Perlmutter and Moore (2002) provide several arguments in favor of the presence of an expletive in Russian impersonals, contra H & S’s conclusions, and see also Pitteroff & Schäfer for further arguments for covert expletives in Russian.

Thus, a lot seems to depend on the availability of expletive subjects, which as Haider (2019) stresses is a characteristic of SVO languages. However, the battery of tests that Alexiadou and Anagnostopoulou (1998) used for Greek and McCloskey (2001) for Irish, i.e., languages that have been argued to lack expletive subjects, are not discussed for Russian or other Slavic languages. If these apply to Russian as they apply to Greek and/or Irish, then we might be able to say something more about the nature of the subject position in Russian. Moreover, Alexiadou and Anagnostopoulou (1998) as well as McCloskey (2001) substantiate their claim about the absence of null expletives by discussing particular environments in full pro-drop languages, namely VS(O) orders which go hand in hand with lack of definiteness effects. It should be mentioned here that the status of Russian within the pro-drop typology is rather controversial: for instance, according to Franks (1995), Russian almost obligatorily requires overt theta-marked pronominal subjects, but allows null expletives. Pelekis (2018) argues that Russian does have an expletive element similar to English it, namely èto ’this’.

For full pro-drop languages, Alexiadou and Anagnostopoulou (1998) concluded that while Greek type pro-drop languages lack an overt subject, the EPP is satisfied by verb movement. Thus, clauses are not really subject-less. However, Greek is not a typical VSO language, unlike Irish. Greek has in fact been argued to have two unmarked word orders: SVO and VSO. In a recent paper, Oikonomou and Alexiadou (2021) show that the SVO order is actually ambiguous between what they call a default-SVO and a Clitic Left Dislocation one. Importantly, there is a clear-cut contrast between the two. According to Oikonomou and Alexiadou (2021), the default SVO order is analyzed as involving a subject being in the T-domain but carrying discourse (δ)-features in the sense of Miyagawa (2009), instead of inflectional φ-features. This is in line with the arguments in Roussou and Tsimpli (2006) and Spyropoulos and Revithiadou (2009) that preverbal subjects are in the T-domain as well as the analysis in Philippaki-Warburton (1985), Alexiadou and Anagnostopoulou (1998), and Kotzoglou (2013) against φ-driven DP movement to Spec,TP. This suggests that like in Slavic, SVO is an acceptable word order, associated with discourse features.

Note now that Greek, like Italian and Spanish, also lacks strict impersonal passives, (9):

As mentioned, Alexiadou and Anagnostopoulou (1998) provided several arguments that VS(O) orders lack covert expletives. This would suggest that (9) is out, as there is no expletive pro, a claim mentioned above by H & S for Spanish and Italian.

Given this, but also the idea that the EPP is satisfied via V-movement in Greek, it is not clear why (9) should be blocked. According to Alexiadou and Anagnostopoulou (2021), in Greek type pro-drop languages, Tense has interpretable but unvalued φ-features. These may receive a value in one of two ways: either by entering Agree with an overt DP subject which has valued φ-features, or, when an overt DP subject is not present, by a Topic (Frascarelli 2007). Passives of unergatives in Greek are illicit not because of an EPP-failure, as V-movement takes care of this. Rather passives of unergatives are out as the phi features of T cannot be valued. Greek lacks a mechanism of default feature valuation of the type employed in e.g., German, see Pitteroff and Schäfer (2019) for details.

Thus, it appears that there are several parametric generalizations about subjects and impersonal passivization that can be made and point to a clustering of properties that does not obligatorily correlate with the VO versus OV nature of the language under discussion. The same seems to hold for some of the other properties H & S discuss: for instance, Greek seems to allow extraction out of DP (Horrocks and Stavrou 1987), unlike English, and to not respect superiority (Sinopoulou 2008), again unlike English. However, it is uncontroversial that Greek is a VO language.