On the Symmetry of V2 in Yiddish and Some of Its Consequences for Extraction

Abstract

This paper examines the distribution of V2 in Yiddish and its effects on extraction. Specifically, we show that both Spec-to-Spec antilocality (Erlewine 2020) and minimality (Rizzi 2006) constrain wh-extractions from embedded clauses in Yiddish. This explains the pattern of that-trace effects in Yiddish, as well as the apparent absence of an escape hatch in certain constructions.

1. Introduction

A notable fact about V2 in Yiddish is its symmetry—the finite verb appears in the second position in both main and embedded clauses (Santorini 1989, Diesing 1990). This differs from the situation in most Germanic languages, which exhibit a main/embedded contrast (Bach 1962, Bierwisch 1963), described crosslinguistically in Holmberg’s (2015) overview. Diesing & Santorini (2020) explore the range of embedded V2 in Yiddish, demonstrating that it is quite pervasive, more so than in Icelandic, another language with symmetric V2. Specifically, embedded V2 in Yiddish is required in all contexts, including those in which it is optional in Icelandic, such as adverbial clauses or indirect questions (Holmberg 2015:357).

The contrast between symmetric and asymmetric V2 languages has led to characterizations in terms of the locus of the finite verb—whether or not it occupies T or C (Holmberg 2015, Pereltsvaig 2016). Thus, a difference in clause structure, even clause size, may matter for V2. In addition to reviewing the evidence for the extent of embedded V2 in Yiddish, we also examine how symmetric V2 in Yiddish affects extraction from complementizerless clauses as well as locality and antilocality in that-trace constructions.

Recent work (such as Bošković 2016, Deal 2019, and Erlewine 2020) has focused on extraction asymmetries in terms of a ban on movement that is “too short”. This constraint is stated in concrete terms in 1 (Erlewine 2020).

1. (1)

   

The data from Yiddish demonstrate that antilocality effects are not merely seen with subject extractions, but in fact apply to any extraction from the highest specifier below the phase head (assuming CP and vP as phases). Furthermore, we show, using the Yiddish data, that extraction from clauses headed by a null C is subject to a minimality constraint, along the lines proposed by Hoge (2001). Effectively, CP headed by a null C cannot serve as a phase, nor can its complement. Our analysis relies on facts concerning embedded V2 in Yiddish, in both declarative and interrogative clauses.

This paper uses data from a variety of sources, collected using different methodologies, including both corpus and native speaker data. The former are from Santorini 1995, utilizing the Penn Parsed Corpus of Yiddish, a syntactically annotated corpus (of about 200,000 words) of Yiddish sentences. The latter are from various published sources, as well as more recently from native speaker consultants.¹

The structure of this paper is as follows. Section 2 presents the initial data on embedded declaratives in Yiddish. Section 3 is concerned with embedded interrogatives. These are of two types, which are selected by different predicates and/or occur in different contexts: Indirect questions, of a propositional type, do not exhibit inversion, whereas true questions do. While the distribution is in some respects similar to that seen in Irish English (Henry 1995, McCloskey 2006), Yiddish allows inversion in a broader range of contexts. Section 4 presents data on complementizerless V2 clauses, which are constrained in distribution, much like their English counterparts. Section 5 presents the extraction facts, from a variety of clause types. Section 6 concludes the paper.

2. V2 in Yiddish: Initial Assumptions and Data

The literature on V2 in Germanic distinguishes between C-V2 and I-V2 languages (Holmberg 2015). Another way of characterizing the distinction is in terms of the domain of the root declarative clause—is it CP or TP (Pereltsvaig 2016)? In either formulation, C-V2 is associated with asymmetric V2 and I-V2 with symmetric V2 (Holmberg 2015).² For expository purposes in this paper, we follow Pereltsvaig’s parametric analysis of V2 in Yiddish. Under her analysis, TP/IP is the domain of the root declarative clause (see Diesing 1990). Giving a fuller description in terms of Pereltsvaig’s parameters, Yiddish has V-to-T raising, unlike English or Mainland Scandinavian. However, like Scandinavian (and unlike Continental West Germanic) Yiddish is head-initial. A further locus of variation is that Nominative case is a weak feature in Yiddish—that is, the subject does not obligatorily raise. However, the EPP feature on T is strong, so some constituent must raise, which, combined with V-to-T raising, gives the pervasive V2 effect. As in Diesing 1990, we assume that Spec, TP can be an A-bar position. This proves a crucial assumption in section 5. With these components in mind, the structure of the Yiddish embedded declarative V2 can be sketched as in 2.³

1. (2)

   

TP is the tense domain of a declarative, and the force of the clause is propositional (in what follows, this is represented by the feature [Force]). The topic phrases XP and YP can be a subject or any other constituent. The initial constituent can be unmarked (as is typical of subjects in the initial position) or a topic or focus (see Prince 1999 for a detailed exposition of these possibilities). The verb raises to T in both the matrix and the embedded clause.

As mentioned above, embedded declarative V2 in Yiddish is quite unrestricted. More specifically, it is not confined to embedded root contexts (Hooper & Thompson 1973, Heycock 2017), nor is it constrained by assertion (Wiklund et al. 2009) or pragmatically derived effects (Julien 2015). Lexical approaches predict that V2 should be blocked when the complement is presupposed—when embedded under factive verbs (but see Julien 2015), but this prediction is not borne out in Yiddish, where embedded V2 occurs quite freely with both nonfactive and factive verbs. The sentential negator nit invariably follows the tensed verb and precedes any nonfinite verbs in Yiddish (Zaretski 1929; see also van der Auwera & Gybels 2014). It thus may serve as a signpost for verb movement to T (this holds for other languages as well, this fact forms a foundational point in the argumentation in Pollock 1989, as well as the analyses of V2 in Scandinavian languages discussed in Holmberg 2015). However, with subject-initial V2, it is not clear whether or not any actual topicalization has taken place, beyond default subject raising (Vikner 1995:67). Thus, evidence for “full V2” must come from examples with nonsubjects in initial position (Santorini 1995).

1. (3)

   

Consistent with this lack of lexical restriction, V2 also occurs obligatorily in Yiddish in noncomplement clauses, such as topicalized clauses, as in 4, and adverbial clauses, as in 5. As noted above, these are among the cases where V2 is optional in Icelandic (Holmberg 2015:357).

1. (4)

   

1. (5)

   

We now turn to embedded questions. As mentioned above, it has been observed in the literature that a distinction should be drawn between V2 clauses with an initial subject and those with a topicalized XP (Travis 1984). The former occurs quite freely in all V2 languages, while the latter can be restricted, particularly in embedded contexts. Following Santorini 1995, we refer to the latter case as full V2. Yiddish allows full V2 in embedded questions (Diesing 1990):

1. (6)

   

Yiddish also allows full V2 in relative clauses, unlike even colloquial Afrikaans, another language with relatively unrestricted embedded V2 (Biberauer 2016), as shown in 7.

1. (7)

   

Thus, the data from Yiddish confirm the predictions made by Pereltsvaig’s parametric approach. Embedded V2 is quite pervasive, and it is found in a wide range of contexts not predicted by assertion-based or lexically-based accounts. The distribution of embedded declarative V2 in Yiddish is broader than that in Afrikaans, which also has a rather permissive V2 profile (Biberauer 2016), since even colloquial Afrikaans does not permit V2 in relative clauses.⁴

3. Embedded Interrogatives

Embedded questions in Yiddish permit not only embedded topicalization as shown above, but also wh-inversion in certain contexts (Santorini 1995, Diesing & Santorini 2020, contra Diesing 1990). That is, Yiddish can show V2 word order (resulting from T-to-C movement, with the finite verb immediately following the wh-phrase) in embedded interroga-tives as well as in matrix questions. The distribution of inversion in embedded interrogatives is not completely free. It is restricted in a manner similar to (but not identical to) that seen in other vernaculars (for example, Belfast English; Henry 1995). Furthermore, for a number of speakers full V2 in these contexts is also marked, requiring discourse licensing (perhaps accounting for discrepancies in elicited judgments and the scarcity of naturally-occurring examples noted in Santorini’s 1995 corpus study).

Predicates that subcategorize for embedded questions fall into two major classes (an idea originating with Baker 1970, and implemented in various ways in subsequent work, including that of McCloskey 2006). Both subcategorize syntactically for wh-questions, but their complements are of different semantic types (Berman 1991, among others). Comple-ments of the ask/wonder class are questions, while complements of the know/tell class are propositions. The distribution of inversion reflects this difference. In Irish English (Henry 1995, McCloskey 2006), inversion in embedded questions is restricted according to the class of the embedding verb—the ask/wonder class freely allows embedded inversion, while the know/tell class does not.

Interestingly, a similar pattern is seen in Yiddish embedded questions. Yiddish freely allows inversion with the ask/wonder class, as in 8a. Inversion is possible in any context that allows the complement to be interpreted as a true question. Such contexts include embedding under modals or negation, as in 8b (the examples are from Santorini 1995, 163b, 164a).⁵

1. (8)

   

Additional examples show that T-to-C movement is also possible with interrogative complements of verbs such as hear and find out. Here, too, paraphrases are possible along the lines illustrated in 8b, where the original subordinate clause turns into a matrix question, and the original matrix clause turns into a parenthetical. The examples in 9 are from Santorini 1995, 168b–e.

1. (9)

   

The distribution of inversion in Yiddish is thus demonstrably broader than that in Irish English, which is subject to lexical restrictions. Inversion in embedded questions is possible beyond the ask/wonder class of verbs.

V2 embedded questions can also occur in noncomplement positions, such as the topic position of a matrix V2 sentence:

1. (10)

   

Thus, T-to-C movement in embedded questions clearly occurs in nonselected contexts. In other words, the pattern observed for wh-clauses is consistent with the generalization of McCloskey (2006), falling under the so-called Kayne-Rizzi-Roberts effect: T-to-C movement occurs in a CP that is not lexically selected at all (in a matrix clause, for instance, or in an adjunct clause or relative clause), or in a CP selected by a functional rather than by a lexical head.⁶

The possibility of inversion in Irish English embedded questions correlates with the possibility of adverbial adjunction (McCloskey 2006:18). Adjunction to embedded questions is most acceptable when inversion takes place, as shown by the contrast between 11 and 12.

1. (11)

   

1. (12)

   

Yiddish permits adjunction of a single adverbial in both declarative and interrogative clauses, resulting in V3 orders, as in 13.

1. (13)

   

While V3 order of this type is most common in root clauses, it also occurs in embedded declaratives, as in 14a (from Santorini 1995:234), and in interrogatives, as in 14b,c (from Diesing & Santorini 2020, 35a,b). Native-speaker consultants accept examples such as 14, given the right discourse context.

1. (14)

   

The question then is: Is this case of adjunction constrained in a way comparable to that seen in the Irish English vernaculars? A closer look at the adjunction to embedded interrogatives in Yiddish reveals that both orders are judged acceptable, but significantly, they are not equivalent semantically. Example 14b is an indirect question, and 14c, with inversion, is a true (embedded) question. The indirect question option is judged more marked. Indeed, Hoge (1999:218) reports such examples (like her 17b) as being unacceptable. Recall that in Irish English, Henry (1995) and McCloskey (2006) report that the acceptability of adjunction to embedded wh-clauses improves with inversion. The data presented here for Yiddish are thus consistent with this observation. The adjunction is most acceptable when T-to-C movement (inversion) has taken place.

Summarizing so far, V2 (including full V2) occurs quite freely in Yiddish embedded clauses. With embedded declarative clauses, there is no restriction to complements of assertions. Embedded V2 is also obligatory in adverbial clauses and embedded interrogatives (Diesing & Santorini 2020). In particular, indirect questions (without inversion) allow both subject-initial V2 and nonsubject topics, the latter requiring special contexts. Inversion is allowed in embedded interrogatives, yielding a true question interpretation; this possibility is determined by the matrix verb (the ask/wonder class) or pragmatic context (negated, modal or interrogative matrix), similarly to Belfast English (Henry 1995). Inversion in embedded interrogatives correlates with the possibility of adjunction of adverbials in embedded contexts (V3), again just as in the case of adjunction in Irish English (McCloskey 2006).

McCloskey (2006) analyzes embedded inversion in Irish English in terms of CP recursion. Though we follow McCloskey in assuming that embedded interrogatives belong to two semantic types (true questions versus indirect questions), we do not adopt his CP recursion approach. Instead, we maintain the analysis of embedded interrogatives in Diesing 1990, 2004, according to which they are nonrecursive CPs. Morpho-syntactically, the two types differ in interrogative force, which we represent featurally as [Force]. We further follow the analysis in Diesing & Santorini 2020 and derive the variation in inversion and [Force] type through features in C. In particular, we assume that the embedded CP is headed by an abstract Q morpheme (Baker 1970, Hagstrom 1998, Constant 2014) selected by the governing verb. The [Force] feature in C determines the interrogative force. In indirect questions, [Force] is [-interrogative] and does not trigger inversion, yielding the representation in 15.

1. (15)

   

The possibility of full V2 in the embedded clause (that is, of YP being a nonsubject topic) depends on discourse conditions. For example, sentences of this type are most acceptable, and most frequent in corpora, when embedded under a matrix sentence with subject-initial V2 (Santorini 1995), as indicated in 15. These restrictions are discussed in more detail in Diesing & Santorini 2020. In the case of embedded true questions, the [Force] feature is [+interrogative] and attracts the finite verb, resulting in inversion, as shown in 16.

1. (16)

   

The structures in 15 and 16 are thus able to represent the full range of embedded interrogatives without recourse to CP recursion. This is desirable, as previous work (such as Iatridou & Kroch 1992) shows no evidence for the additional structure induced by CP recursion.

4. Complementizerless V2

Embedded declaratives in Yiddish are typically introduced by a complementizer such as az ‘that’ or vos ‘that’ for nonfactive and factive complements, respectively. However, embedded V2 without a comple-mentizer is typically (but not exclusively) possible in contexts that typically allow embedded main clause phenomena, such as the declarative complements of verbs see, ask, or think (Heycock 2017). While these complements resemble matrix clauses, they most often occur with subject topics (or a postposed subject with an expletive), and in this regard they resemble ordinary subordinate clauses with complementizers (Santorini 1995). Thus, they are V2 clauses, but full V2 (with a nonsubject topic) occurs only rarely (the examples in 17 are from Santorini 1995, 152a,b).⁷

1. (17)

   

In contrast to the somewhat flexible lexical restrictions on the governing verb, there is a categorical restriction on complementizerless embedded clauses to complement position. Unlike their counterparts with a complementizer, they cannot appear in topic position:

1. (18)

   

Complementizerless adverbial clauses are likewise prohibited, in contrast to their counterparts with complementizers:

1. (19)

   

The restriction of complementizerless declaratives in Yiddish to complement position is similar to complementizer drop in English, as described in Pesetsky 1992 and Bošković & Lasnik 2003.⁸ Pesetsky gives an account of the distribution of complementizerless clauses in English, which assumes the presence of an affixal null C. This null complementizer is licensed by associating with the embedding verb via adjacency. Bošković & Lasnik (2003) refine this analysis by expressing the licensing condition in terms of Morphological Merger, requiring adjacency at PF. Failure of adjacency leads to ungrammaticality, which accounts not only for the restriction of complementizer drop to complement positions, but also for the pattern in 20.

1. (20)

   

While Yiddish requires complementizerless clauses to be in complement position, it does not require PF adjacency to the embedding verb. For example, the matrix main verb can be fronted to form a yes/no question:

1. (21)

   

These restrictions on complementizerless clauses leave one with (in principle, at least) two possibilities, as outlined in Santorini 1995: Complementizerless clauses in Yiddish have a null complementizer (but licensed in some way other than in English), or they have no complementizer, consisting only of the TP-level (root) V2 structure. If the latter, one might expect them to show properties of embedded main clauses (assuming Pereltsvaig’s TP-as-root-clause parameter). Consider-ing facts of distribution, the complementizerless clauses do commonly occur as complements of verbs such as see and think—that is, contexts that typically allow embedded main clauses. However, Santorini’s (1995) corpus study also demonstrates that complementizer-less clauses show properties typical of subordinate clauses; in particular, they do not exhibit a high frequency of full V2.⁹ Thus, there is evidence that favors subordination. We therefore assume the presence of a null comple-mentizer, leaving the explanation of its distribution to future research. We turn next to the properties of embedded V2 and extraction.

5. Extraction and Embedded V2

Extraction facts provide further insight into the nature of C, both null and overt, in Yiddish. In addition to embedded questions, inversion can also occur with matrix extractions from embedded clauses (Diesing 1990). The grammaticality of inversion depends on whether or not the head in C is overt or silent. If the complementizer is overt, inversion is ungrammatical, as shown by the contrast between 22a and 22c. If the complementizer is silent, the converse is true, as shown by the contrast between 22b and 22d (see also Hoge 2001).

1. (22)

   

Inversion can also ameliorate that-trace effects with subject extraction (Diesing 1990):

1. (23)

   

Another mitigating factor is the presence of a nonsubject in Spec (TP) (Diesing 1990:75), which licenses subject extraction (see Culicover 1993 for related discussion concerning subject extractions in English in the presence of clause-initial adverbials):

1. (24)

   

Thus, it is clear that the that-trace effect in Yiddish is not confined to subject extractions. It is rather a left edge effect associated with the embedded Spec (TP) position, or the highest specifier below C (Branigan 2005 draws a similar conclusion, noting that an account based on subject properties such as that of Pesetsky & Torrego 2001 cannot fully explain these data). As noted above, Erlewine (2020) characterizes violations of this sort in terms of the antilocality constraint in 1, repeated here as 25.

1. (25)

   

In other words, individual steps of phrasal movement cannot be too short. Erlewine (2020) focuses his survey on cases of subject extraction, demonstrating that his proposed constraint has considerable cross-linguistic applicability. In the discussion below, we present the details of its applicability to the Yiddish facts.

Turning now to the case of complementizerless embedded clauses, extraction from these sentences is blocked across the board (see also Hoge 2001). Both, subject (26a) and object (26b,c) extractions from complementizerless embedded V2 yield ungrammaticality.

1. (26)

   

The left edge of the embedded complementizerless V2 clause (that is, the specifier of the CP headed by a null C) apparently cannot serve as an escape hatch (or phase edge).

Indeed, extraction is only possible from a complementizerless embedded clause when accompanied by inversion, as unambiguously shown by the contrast between 22b and 22d, where the wh-phrase has passed through the highest specifier of the embedded clause. Note that these cases are arguably embedded questions with inversion, as discussed in section 4, where the interrogative feature [Force] on the embedded C triggers the movement of the finite verb to C.

We represent the various extraction patterns discussed above in 27.

1. (27)

   

The case in 27a, where extraction takes place from the highest specifier below an overt C, clearly violates Erlewine’s antilocality constraint. The wh-phrase crosses only TP itself before landing in the lower Spec, CP, and the result is ungrammatical. In the second case, 27b, extraction of either a subject or an object from a position below a filled Spec (TP) position in a clause with an overt complementizer complies with antilocality. In either case of extraction (subject or object), the wh-phrase must cross the v/VP and the TP. Thus, a well-formed sentence results. The contrast between the configurations in 27a,b is similar to that noted in Rizzi 2006 for the English sentences in 28.

1. (28)

   

Rizzi argues that the presence of the expletive in Spec, TP in 28b allows the subject to skip over that specifier, which makes extraction across the complementizer possible. As pointed out by Erlewine (2020), this analysis is naturally recast in terms of the Spec-to-Spec antilocality constraint.

In the third case in Yiddish, 27c, the extracted phrase moves from the left edge of CP, but in this case there is inversion, rather than a complementizer. The that-trace effect is inoperative, and the result is grammatical. Our claim is that this is a case of extraction from an embedded question with V2, which does not involve any antilocal step of movement. The wh-phrase moves from the embedded Spec, CP to the next phase edge (the embedding Spec, vP), and thereby obeys both minimality and antilocality.

The fourth case, 27d, involves extraction from a declarative with a null C. At first blush, this appears to be a situation in which the movement cannot comply with the antilocality constraint (see Douglas 2017) on the grounds that CP with a null C cannot be a landing site (see also Bošković & Lasnik 2003 on subject extraction from extraposed clauses in English). However, this explanation fails here, since both subject and object extraction yield ungrammaticality. Something else must be going on. Hoge (2001) suggests that the ill-formedness in these cases results from a minimality violation. Diesing (1990, 2004) argues that Spec, TP is a generalized topic position, and as such can (potentially) check A-bar features, and this could lead to an A-bar intervener, blocking movement.¹⁰ However, in itself such an explanation would also predict cases such as those in 24b to be ungrammatical, contrary to fact.

Yet something akin to the minimality explanation seems to be the most promising way of explaining why both subject and object extraction are ruled out in 27d. What seems to be key here is the fact that the specifier of the CP headed by the null complementizer cannot be used as an intermediate landing site for any wh-phrase. The situation is different when C is filled by a finite verb. Diesing & Santorini (2020) attribute this contrast to the fact that when C is deleted (or null), C is no longer a phase head (see also Branigan 2016). Various complements of C can potentially be derived phase heads, but declarative T cannot. Thus, the specifier of a null C cannot serve as an escape hatch.¹¹ This derives the observed minimality violation, while allowing extraction when an overt complementizer is present.¹²

Summarizing this section, Yiddish exhibits both antilocality and minimality constraints in extractions from embedded clauses. The former are seen with both subject and object extractions from az-clauses (with a complementizer). In either case, extraction from the Topic position (Spec, TP) is banned. Conversely, both subject and object extraction from an internal position (below a filled Spec, TP) is possible from an az-clause. In contrast, complementizerless declaratives do not permit extraction, which we attribute to minimality. Extraction is permitted from embedded direct questions, as the extraction proceeds through the phase edge.

6. Conclusion

In this paper, we have examined a variety of V2 structures in Yiddish. Embedded V2 occurs quite freely in Yiddish—more so than in other purportedly symmetric V2 languages such as Icelandic. There are also cases of complementizerless V2 in Yiddish. These cases show a more restricted distribution. The distribution of embedded V2 interacts with wh-movement in interesting ways. While declarative V2 is a result of movement of the finite verb to T and a topic to Spec, TP, V2 also results from wh-movement to Spec, CP and movement of T to C. The latter occurs in both matrix and embedded clauses. Extraction from embedded clauses is constrained by minimality effects, giving rise not only to classic wh-islands, but also to a ban on extraction from comple-mentizerless declaratives. The latter, being nonphasal, do not allow extraction from their topmost specifier. Spec-to-Spec antilocality (Erlewine 2020) constrains movement from the Topic position of embedded clauses with complementizers. Clauses with T-to-C movement (as in embedded direct questions) allow extraction from the left edge, as these are phases.