Subextraction in Japanese and subject-object symmetry

Abstract

For a long time Japanese was taken to be a language lacking a subject-object asymmetry in subextraction. Two recent experimental studies have challenged this conclusion suggesting that Japanese complex NP (CNP) subjects are more opaque to subextraction than CNP objects (Jurka 2010; Jurka et al. 2011). Given the significance of this claim for the cross-linguistic landscape of subextraction phenomena, this study further explores the nature of subextraction phenomena in Japanese with three acceptability judgment experiments. We show that overt subextraction (scrambling) out of CNP subjects and objects results in similar acceptability ratings once the relative weight and order of constituents are properly controlled for. Recent experimental work which suggested that Japanese has a subject-object asymmetry to subextraction predicted that wh-in-situ adjuncts should lead to greater degradation for subextraction out of subjects as compared to subextraction out of objects. To test this prediction, we also present novel experimental data on wh-in-situ phrases inside subjects and objects in Japanese. Our results show that the argument/adjunct status of wh-in-situ phrases does not interact with the subject/object status of the CNPs, further invalidating the recent claims. Together these findings support the traditional view that Japanese has no subject-object asymmetry in subextraction, whether overt or covert. Having restored the status of Japanese as a language with no subject-object asymmetry in subextraction, we discuss possible reasons that could account for the absence of such an asymmetry. We suggest that the lack of asymmetry is due to Japanese subjects’ inability to satisfy the EPP on T/C; we further contend that cross-linguistic variation in subject transparency follows from whether or not the feature D comes from subjects.

Introduction

Subject-object asymmetry is a ubiquitous feature of natural language, and it has played a particularly prominent role in studies of long-distance dependencies. In this context, Japanese has long enjoyed the unusual status of a language in which subjects and objects appear to show no difference with respect to subextraction. We define subextraction as an operation under which a constituent is displaced out of a larger phrase in which it is embedded, leaving a gap at the base position (Corver 2017, and references therein):

  1. (1)

    [YP\(\dots\ \alpha _{\mathrm{i}}\ \dots\) [XP … X ti…]]

The literature on Japanese subextraction has unanimously claimed that there is no subject-object asymmetry in subextraction in Japanese (e.g., Ross 1967; Kuno 1973; Saito 1985, 1992; Kikuchi 1987; Nishigauchi 1990; Lasnik and Saito 1992; Watanabe 1992; Takahashi 1994; Ishii 1997; Richards 1997, 2000, 2008; Stepanov 2007). However, Jurka (2010) and Jurka et al. (2011) challenge this long-standing conclusion using data from formal acceptability judgment experiments, which suggest that Japanese complex NP (henceforth CNP) subjects are more opaque to subextraction than Japanese CNP objects. If this is true, Japanese subjects are islands after all. For Jurka’s theoretical stance, this finding was of utmost importance, as it supports the notion that the language faculty includes a universal subject island condition (e.g., Condition on Extraction Domain (CED); Huang 1982; Uriagereka 1999; cf. Stepanov 2007; see Sect. 2 below). Even if we leave this particular theoretical approach aside, Jurka’s results had significant implications for our understanding of the cross-linguistic landscape of subextraction, which has played a central role in the development of current theories of movement operations and their constraints (see Sect. 2).

The primary goal of this paper is to further explore the nature of subextraction phenomena in Japanese in light of the mixed evidence reviewed above. The current study was motivated, in large part, by potential methodological issues in Jurka (2010) and Jurka et al. (2011). As we illustrate below, subextraction out of the subject in Jurka and colleagues’ experiments is confounded with a violation of a well-known psycholinguistic constraint in Japanese, long-before-short preference, a preference to place longer constituents before shorter constituents (Dryer 1980; Hawkins 1994; Yamashita and Chang 2001; Yamashita 2002).

Our own experiments 1 and 2 control for this independent psycholinguistic constraint and show that subextraction out of subjects and objects results in similar acceptability ratings. To take these findings further, experiment 3 presents a novel test of subextraction that explores the acceptability of wh-in-situ phrases inside subjects and objects. We show that the argument vs. adjunct status of wh-in-situ phrases does not interact with the subject vs. object status of the phrase that contains it, even though Jurka’s account predicts that adjunct wh-in-situ phrases inside subjects should lead to greater degradation than those inside objects. Taken together, the results of the three experiments constitute evidence against the claim that Japanese subjects are islands, lending support to the traditional view that there is no subject-object asymmetry in Japanese subextraction. That takes us back to the question of why that could be the case and what accounts for parametric differences in subextraction across languages.

The rest of the paper is structured as follows. Section 2 summarizes two previous accounts of subject island phenomena: the freezing-based account and the spell-out/phase-based accounts. Section 3 reviews Jurka (2010) and Jurka et al. (2011) focusing on their experimental evidence which suggests that subjects are more opaque to subextraction than objects in Japanese, contra all earlier scholarship. Here, we point out a methodological issue in their experimental design, namely that their CNP object sentences violate the long-before-short preference while their CNP subject sentences do not. Sections 4 and 5 present the results of our Experiments 1 and 2 respectively. Together they show that subextraction (subscrambling) out of CNP subjects and objects results in similar acceptability ratings once the long-before-short preference is controlled for. Thus, these experiments support the widespread notion that Japanese does not have a subject-object asymmetry to subextraction. Section 6 presents the results of Experiment 3, which show that the argument vs. adjunct status of wh-in-situ phrases inside CNPs does not interact with the subject vs. object status of the CNP that contains it. Taken together, these three experiments all indicate the transparency of Japanese subjects. Section 7 summarizes cross-linguistic variations in the island status of subjects and discusses the implications of our findings for theories of subject-object asymmetry in subextraction phenomena. We suggest that the total lack of subject island phenomena in Japanese is problematic for both the freezing-based and spell-out/phase-based accounts first presented in Section 2, and thus offer some new considerations on the reasons for the presence/absence of the asymmetry.

Previous accounts of subject island phenomena

Based on informal judgments, it has long been noted in the literature that subextraction from subjects results in a more severe acceptability degradation than subextraction from objects, and that subjects constitute islands for movement (e.g., Ross 1967; Chomsky 1973; Huang 1982). Within the current transformational approach to syntax (Minimalist Program: Chomsky 1995), there are two major theoretical approaches to the subject-object asymmetry in subextraction: freezing accounts and spell-out/phase-based accounts.

Freezing accounts postulate that arguments become opaque to subextraction after undergoing syntactic movement (e.g., Wexler and Culicover 1981; Takahashi 1994; Stepanov 2007; Rizzi 2006, 2010; Rizzi and Shlonsky 2007); that is, the opacity of subjects is attributed to their movement to [Spec, TP]. For instance, Takahashi (1994) accounts for the subject-object asymmetry based on the following assumptions:

  1. (2)
    1. a.

      Chain Uniformity: Chains must be uniform

    2. b.

      Shortest Move: Make the shortest move

    3. c.

      Uniformity Corollary on Adjunction (UCA): Adjunction to a part of a nontrivial chain is not allowed.

Take an example of subextraction of a wh-phrase out of a subject, as in the English example below, (3a). Assuming that the subject DP first moves from [Spec, vP] to [Spec, TP] and forms an A-chain (3b), a subextraction out of the moved subject DP must violate either Shortest Move (2b) or UCA (2c). If a constituent moves out of the subject DP and adjoins to it, such a derivation incurs no violation of Shortest Move. However, this adjunction violates UCA, since this is an adjunction to a part of a nontrivial chain, i.e., the A-chain between the subject DP and its copy (3c). No violation of UCA would result if the extracted constituent skipped the DP adjunction and adjoined to TP, but this derivation violates Shortest Move instead (3d). Thus, a constituent cannot be extracted out of a subject DP that has been raised to [Spec, TP] without violating one of the assumptions in (2). In contrast, the same issue does not arise with a constituent inside an object DP, under the assumption that objects remain in situ. A constituent can move out of an object DP and adjoin to it without violating UCA, as in (4b–c).

  1. (3)
    1. a.

      *Whoi does [a picture of __i ] hang on the wall?

    2. b.

      [TP [DP a picture of who]k [vP [DPa picture of who]k hang on the wall]]

    3. c.

      [TP [DP whoi [DP a picture of whoi]k][vP [DPa picture of who]k hang on the wall]]

    4. d.

      [TP whoi [TP [DP [DP a picture of whoi]k] [vP [DPa picture of who]k hang on the wall]]]

  1. (4)
    1. a.

      Whoi did you see [a picture of __i ]?

    2. b.

      [TP [DP you]k did [vP [DPyou]k see [DP whoi [DP a picture of whoi]]]]

    3. c.

      [CP whoi didj [TP [DP you]kdidj [vP [DPyou]k see [DPwhoi [DP a picture of whoi]]]]]

In spell-out/phase-based accounts, the subject-object asymmetry follows from the way subjects and objects are introduced in syntactic derivation (Nuñes and Uriagereka 2000; Johnson 2003; Rackowski and Richards 2005; Jurka 2010; Müller 2010; among others). For example, Nuñes and Uriagereka (2000) attribute the subject-object asymmetry to the linearization requirement. Following the Linear Correspondence Axiom of Kayne (1994), they argue that when subject merges with v′, its internal constituents cannot establish an asymmetrical c-command relation with v′, and the linear order cannot be determined at PF. To avoid this, the internal constituent of the subject is spelled out before the merge with v′, such that the subject effectively serves as a single word. This obligatory spell-out makes the constituents inside subjects inaccessible to any syntactic operations, including subextraction. However, when objects and transitive verbs merge, no such spell-out is necessary, because asymmetric c-command is inherent in the verb-object structure.

Although individual spell-out/phase-based accounts differ greatly in how they derive subject island effects,Footnote 1 an important common prediction of the spell-out/phase-based approaches is that extraction out of subjects should always cause island violations, hence it should be more degraded than extraction out of objects. In contrast, freezing approaches predict that the islandhood depends on whether a subject has undergone movement, so the difference between subjects and objects is not set in stone.

In summary, the two opposing approaches to subextraction are motivated by the two distinctions explored here, between external and internal arguments and between moved and in situ constituents. While the external/internal distinction lends support to the spell-out/phase-based accounts, the freezing accounts can capture the effects of the moved/in situ distinction. Stepanov (2007) argues the lack of subject-object asymmetry in subextraction in languages such as Japanese provides an important counterargument against the spell-out/phase-based accounts of subextraction phenomena. However, Jurka (2010) and Jurka et al. (2011) challenge the previous judgment reports in Japanese, and present experimental evidence that arguably shows that subjects are more opaque to subextraction than objects even in Japanese. As such, the empirical questions about Japanese subextraction acceptability have significant implications for the current theories of constraints on movement.

In what follows, we will critically examine the empirical evidence presented in Jurka (2010) and Jurka et al. (2011). To anticipate our discussion, we first identify a potential confound in their study, and propose new experiments that address the confound. The proposed new experiments also test the predictions made by the freezing accounts and the spell-out/phrase-based accounts against Japanese data. Our results show that Japanese does not exhibit subject island phenomena. Accordingly, neither the freezing accounts nor spell-out/phase-based accounts are supported by the Japanese data.

Jurka (2010) and Jurka et al. (2011): Japanese subjects as islands

It has been argued that there is no subject-object asymmetry in subextraction in Japanese (e.g., Ross 1967; Kuno 1973; Saito 1985, 1992; Kikuchi 1987; Nishigauchi 1990; Lasnik and Saito 1992; Watanabe 1992; Takahashi 1994; Ishii 1997; Richards 1997, 2000, 2008; Stepanov 2007). Lasnik and Saito (1992) compare subextraction from within CNP subjects (5) and CNP objects (6):

  1. (5)
    figurea
  1. (6)
    figureb

Both (5) and (6) involve subscrambling of the wh-phrase dono hon-o ‘which book-acc’ out of a CNP. The authors judge both examples as marginal due to a CNP Constraint violation (but see Sect. 4 for a discussion of the island status of CNPs in Japanese). However, they contend that there is “no clear contrast” between (5) and (6); hence, there is no subject-object asymmetry in Japanese subextraction (Lasnik and Saito 1992:125).

Stepanov (2007) argues that the lack of subject island phenomena in languages including Japanese serves as evidence against spell-out/phase-based accounts reviewed in Sect. 2. To take Nuñes and Uriagereka’s (2000) approach, for example, subjects in languages without subject-object asymmetry, such as Japanese, must be analyzed as sisters to the verb in order for their subconstituents to remain accessible to syntactic operations. This amounts to claiming that languages like Japanese are non-configurational, a view which has long been rejected (e.g., Kuroda 1980; Saito 1983, 1985; Saito and Hoji 1983; Hoji 1985). Freezing accounts, on the other hand, can accommodate the lack of subject-object asymmetry in subextraction in languages like Japanese by capitalizing on the claim that subjects can remain in situ.

With this background, Jurka (2010) and Jurka et al. (2011) present several arguments challenging this claim. First, they point out that there are non-trivial structural differences between (5) and (6). While (6) involves no additional embedding other than the CNP itself, (5) involves a center-embedded CP, which plausibly incurs a higher processing cost and a lower acceptability judgment. Therefore, the acceptability of the baseline sentences without subextraction may be different for (5) and (6). Jurka and colleagues conducted an acceptability judgment experiment to investigate this issue, using a 2 × 3 experiment design with argument (subject vs. object) and subextraction (no subextraction, scrambling, clefting).

Here, we discuss only the results of their no subextraction and scrambling conditions, as the results with the clefting condition were very similar to those of the scrambling condition. Examples of Jurka et al.’s experimental sentences are shown in (5). These examples involve three levels of embedding in both the CNP subject sentences (7a) and (7c), and the CNP object sentences (7b) and (7d).

  1. (7)
    figurec

The results of this experiment showed no significant difference between the two argument conditions under scrambling, (7c) and (7d), thus confirming the intuition reported in Lasnik and Saito (1992). However, there was a significant difference between the two argument conditions without subextraction: the mean of the CNP subject sentences (7a) was significantly higher than that of the CNP object sentences (7b). The authors interpreted this finding to indicate that subextraction disproportionally affects subjects in Japanese, causing a greater degradation in acceptability with the CNP subject sentences between the no subextraction and scrambling conditions. In other words, Jurka et al.’s findings suggested that subjects are islands in Japanese, a completely novel claim and one that goes well beyond the understanding of Japanese grammar because, if correct, it casts doubt on principled differences between languages such as English and Japanese with respect to island effects.

However, there is a possible confounding factor in this study: the judgments provided by the participants could have been influenced by a psycholinguistic constraint in Japanese known as the long-before-short preference, a preference to place longer/heavier constituents before shorter/lighter constituents (Dryer 1980; Hawkins 1994; Yamashita and Chang 2001; Yamashita 2002). Given this long-before-short preference, it is possible that Japanese speakers judge sentences that do not conform to this preference as less acceptable.

In the subject condition in (7a), the CNP subject is the first element within the embedded clause, which is consistent with the long-before-short preference within the embedded clause. When the entire sentence is taken into consideration, the CNP subject is the second item of the whole sentence, following only the matrix topic phrase sono shoojo-wa ‘the girl-top’. In contrast, in the CNP object condition (7b), the CNP object is the second element in the embedded clause and the third element in the entire sentence. Thus, in terms of the long-before-short preference, the CNP object sentence in (7b) is less optimal than the CNP subject sentence in (7a). Under the subextraction condition, however, subscrambling effectively neutralizes the differences in the weight of the relevant constituents. When the embedded direct object is scrambled to the sentence-initial position as in (7c) and (7d), it splits the CNP in two, resulting in a large reduction in the weight of the CNP. As a result, the CNP object sentences in the scrambling condition in (7d) are more consistent with the long-before-short preference.

In sum, it is conceivable that the results of Jurka (2010) and Jurka et al. (2011) were affected by differences in the relative weights and order of the constituents in their experimental sentences. If so, we predict that the significant difference between the CNP subject and object sentences in the no subextraction condition observed by Jurka et al. (2011) would disappear if the relative weight and order of the constituents were properly controlled. In contrast, if Japanese subjects are indeed islands, we should observe a significant difference in acceptability judgments between sentences with subextraction out of CNP subjects and sentences with subextraction out of CNP objects, even when the aforementioned factors are properly controlled. Furthermore, when combined with the freezing accounts and the spell-out/phase-based accounts, the hypothesis that Japanese subjects are islands predicts that the external/internal distinction and the moved/in situ distinction among arguments should have effects on acceptability of sentences with subextraction out of CNPs in Japanese.

To test these predictions, we conducted two acceptability judgment experiments (Experiments 1 and 2 below). Using sentences with overt subextraction out of a CNP, i.e., subscrambling, these experiments addressed whether (i) the relative weight and order among constituents affect acceptability of Japanese sentences with subextraction out of a CNP, and (ii) the external/internal and the moved/in situ distinction affect acceptability of subextraction out of a CNP in Japanese. To the best of our knowledge, the effects of the long-before-short preference on the acceptability of Japanese sentences have not been tested experimentally before.Footnote 2 The results of the experiments show that (i) whether or not sentences are consistent with the long-before-short preference makes significant differences in their acceptability, (ii) neither the external/internal distinction nor the moved/in situ distinction plays a significant role in the determination of acceptability of Japanese sentences with subextraction out of CNPs, and (iii) the alleged differences between subextraction out of CNP subjects and objects reported in Jurka (2010) and Jurka et al. (2011) are not observed once the relative weight and order of constituents in sentences with a CNP are properly controlled for. Based on these findings, we argue that Japanese subjects and objects are not different in their transparency/opacity with respect to subextraction.

Experiment 1: Long-before-short preference, external/internal and moved/in situ distinctions among arguments, and acceptability of sentences with subextraction

The main goal of Experiment 1 was to determine whether the manipulation of the relative weight and order of constituents affects acceptability of Japanese sentences with a CNP. Given the long-before-short preference among Japanese speakers (Dryer 1980; Hawkins 1994; Yamashita and Chang 2001; Yamashita 2002), acceptability data in Jurka et al. (2011) may not serve as a valid comparison of subjects and objects as islands. The second goal of Experiment 1 was to examine if acceptability of subextraction out of a CNP differs depending on the external/internal status (transitive vs. unaccusative subjects) and moved/in situ status (fronted vs. non-fronted objects) of CNPs in Japanese.

In order to examine the effects of the long-before-short preference and external/internal and moved/in situ distinctions among arguments on acceptability of sentences with and without subextraction, Experiment 1 investigated the acceptability of the following ten types of sentences which belong to three larger categories (see Table 1).

Table 1 Sentence types used in the experimental manipulation

All three sentence types involved embedding and were presented with and without subextraction out of the CNP. In the examples of our stimuli below, the letters -n and -e indicate ‘no extraction’ or ‘extraction’ respectively. Sentences of type A conform to the long-before-short pattern, and B sentences do not.

We will now illustrate our stimuli in a series of examples. Further details on the stimuli are available as a supplementary appendix to the online version of this paper. Examples in (8) show transitive CNP subject sentences (Type 1), with no subextraction (8a) and with subextraction (8b). The scrambled phrase is shown in italics, and the long and short constituents are shown in boxes.

  1. (8)
    figured

In order to clearly indicate the presence and absence of subextraction in our experimental sentences, we used complex sentences with an embedded clause, and a phrase is subextracted from a CNP inside an embedded clause and ends up preceding a matrix element (e.g., the matrix subject) in ‘Type -e’ sentences. This way, the presence of subextraction can be indicated clearly.Footnote 3

Examples in (9) show transitive CNP object sentences (Type 2). These sentences were presented in long-before-short order (Type 2A) and short-before-long order (Type 2B) both with and without subextraction. Again, the scrambled phrase is indicated in italics, and the long and short constituents are in a box:

  1. (9)
    figuree
  1. (10)
    figuref

Examples in (10) illustrate unaccusative CNP subject sentences of Type 3, which were also presented in long-before-short order (Type 3A) and short-before-long order (Type 3B), with and without subextraction.

  1. (11)
    figureg

As summarized in Table 2, the relative order of the relevant constituents was manipulated only with transitive CNP object sentences and unaccusative CNP subject sentences (Type 2 and 3). This was because a possible trade-off between potentially positive effects of conforming to the long-before-short preference and potentially negative effects of fronting the CNP, the long constituent, may be predicted only if the CNP is an internal argument, namely with transitive CNP objects and unaccusative CNP subjects. Fronting of the short object NP in transitive CNP sentences is unmotivated as it goes against the long-before-short preference.

Table 2 Types of experimental sentences in Experiment 1

At this point, we would like to address several considerations behind the design of the experiment and its materials. First, following previous studies, all of our experimental sentences involve a CNP. This is because scrambling out of a simple NP, where a genitive phrase appears outside of its NP, results in severe ungrammaticality. In other words, scrambling is subject to the left-branch condition (e.g., Kishimoto 2006; Agbayani et al. 2015).

  1. (12)
    figureh

Second, all CNPs in our examples involve a complement clause headed by koto ‘fact’, as they have been argued to be non-islands (Ross 1967, 1986; Haig 1976; Fukuda and Sprouse 2017; but see Saito 1985 for a different view).Footnote 4 Under the assumption that CNPs headed by koto ‘fact’ are not islands, any difference in acceptability judgments that we obtain in our experiment may be assumed to be due to the alleged subject-object asymmetry.

Third, we used CNPs to manipulate the order and relative weight of the constituents of our experimental sentences. While the relative heaviness of a constituent can only be determined with respect to the heaviness of its co-occurring constituents, in our experiment, we simplify the issue by treating the long/short distinction as categorical, with all CNPs labeled as long/heavy constituents and simple NPs without a clausal complement as short/light constituents.

Finally, recall that our experiment involves manipulations of the presence and absence of subextraction and the relative order and weight of constituents. In order to clearly indicate the presence or absence of subextraction in our experimental sentences in the subextraction conditions, we used complex sentences including an embedded clause, with a phrase that is subextracted from a CNP inside an embedded clause preceding a matrix element (e.g., the matrix subject). This way, subextraction can be clearly shown. At the same time, we manipulated the relative order and length of constituents within the embedded clause, since the CNP is always inside the embedded clause.

We are now ready to state our predictions. First, the sentence types that involve subextraction are predicted to be less acceptable than their counterparts without subextraction due to the extra processing cost incurred by the displacement and the filler-gap dependency. Thus, stimuli with extraction (e-sentences) should be more degraded than those without (n-sentences). At this point, we cannot say whether A sentences should be better than B sentences; although the former conform to the long-before-short preference, they also involve an extra scrambling of the internal CNP argument. Which ones are judged as relatively more acceptable would depend on the relative size of the two hypothesized effects: the processing cost due to the fronting of CNPs and the effect of the long-before-short preference.Footnote 5 Within the subextraction conditions (e-sentences), subextraction out of CNPs could potentially neutralize the difference between the same sentence types in the two orders, as subextraction shortens the offending long constituent in the short-before-long order.

With respect to the effect of external/internal status of CNPs on subextraction, if external arguments are more opaque to subextraction than internal arguments in Japanese, as predicted by the spell-out/phase-based accounts, subextraction out of sentence-initial transitive CNP subjects should be less acceptable than subextraction out of sentence-initial unaccusative CNP subjects. This follows from the assumptions that (i) unaccusative subjects are internal arguments, (ii) both sentence-initial transitive and unaccusative subjects undergo syntactic movement,Footnote 6 and (iii) they are both consistent with the long-before-short preference. Thus, while no significant difference in acceptability is predicted between transitive sentences with a CNP subject and unaccusative sentences with a CNP subject without subextraction, a significant difference in mean acceptability is predicted between these two sentence types with subextraction, so that transitive sentences with a CNP subject would be judged as less acceptable than unaccusative sentences with a CNP subject, because only the former involves subextraction out of an external argument.

Finally, if moved arguments are more opaque to subextraction than in situ arguments in Japanese, as predicted by the freezing-effects-based accounts, subextraction out of fronted CNP arguments should be judged as less acceptable than subextraction out of those that are in situ.

Several factors further complicate our predictions with respect to the effects of syntactic movement on subextraction. First, there is an issue of whether or not Japanese subjects obligatorily undergo A-movement. If Japanese subjects—in contrast to objects—must undergo A-movement, then CNP subjects are predicted to be more opaque to subextraction than CNP objects. However, a comparison between CNP subjects and objects involves two competing factors: (i) the assumed A-movement undergone by CNP subjects would make them more opaque to subextraction than CNP objects in situ, while (ii) the non-initial position of CNP objects would make them inconsistent with the long-before-short preference whereas the initial CNP subjects are consistent with it. Thus, before we compare sentences with CNP subjects and CNP objects, we first need to know more about the effects of syntactic movement and the relative order and weight of the relevant constituents on acceptability of sentences containing a CNP with and without extraction. To this end, we first need to compare a single CNP argument in a moved vs. in situ minimal pair. Assuming that internal arguments do not undergo obligatory A-movement, we can use sentences with a CNP object and manipulate its position between in situ and moved. In particular, the freezing-effects-based accounts predict that clause-initial CNP internal arguments like transitive objects and unaccusative subjects should be more opaque to subextraction than the same arguments in their in situ positions. But these predictions are still complicated by the long-before-short preference, according to which heavy constituents such as CNPs are more acceptable in sentence initial positions. Therefore, moved internal argument CNPs such as transitive CNP objects and unaccusative CNP subjects might be more opaque to subextraction than when they are in situ, but they are more acceptable in moved positions because of the long-before-short preference. As such, whether these sentences in the long-before-short order or the short-before-long order are judged as relatively more acceptable would depend on the relative size of the two hypothesized effects: the freezing effects due to the fronting of CNPs and the effects of the long-before-short preference.

Participants

Sixty-two university students in Gifu, Japan, participated. Two participants’ results were excluded from the analyses because either they were non-native speakers of Japanese or they did not complete the task as instructed.

Materials

Four verbs were selected for each of the three sentence types: transitive CNP subject, transitive CNP object, and unaccusative CNP subject. For transitive CNP subject sentences, four transitive verbs that subcategorize for an NP subject that denotes a cause (e.g., an event that surprised someone), which can be a CNP, and an NP object that denotes an experiencer (e.g., the person who is surprised by the event) were used: odorokase-ru ‘surprise’, shimpaisase-ru ‘worry, concern’, konwakusase-ru ‘confuse’, and keikaisase-ru ‘alarm’. For transitive CNP object sentences, four transitive verbs that subcategorize for an NP subject that denotes an agent (e.g., the person who leaks the information) and an NP complement that denotes a theme (e.g. the information that is leaked), which can be a CNP, were used: moras-u ‘leak’, hirome-ru ‘spread’, shir-u ‘know’ and baras-u ‘expose’. Finally, for unaccusative CNP subject sentences, four unaccusative verbs that subcategorize for an NP subject that denotes an stimulus (e.g., information), which can be a CNP, and a PP that denotes a goal (e.g., the recipient of information) were used: more-ru ‘leak’, hiroma-ru ‘become spread’, shire-ru ‘become known’ and bare-ru ‘become exposed’. As described earlier, all of the experimental sentences were embedded under one of four verbs that select for a clausal complement: shinji-ru ‘believe’, omo-u ‘think’, yu-u-u ‘say’ and kangae-ru ‘think’. With two conditions for transitive CNP subject sentences (no subextraction vs. subextraction) and four conditions for transitive CNP object and unaccusative CNP-subject sentences (long-before-short vs. short-before-long and no subextraction vs. subextraction), there were forty patterns of target sentences. Four lexicalizations of each of the forty patterns were constructed (40 × 4 = 160) and distributed into four separate lists using a Latin square design so that no participant saw two different conditions from the same lexicalization. Each of the four lists of 40 target sentences was mixed with the same 44 filler sentences of various constructions and acceptability. The order of each of the four lists of 84 sentences was pseudo-randomized with the first and last four sentences being the same filler sentences, consisting of two clearly grammatical sentences and two clearly ungrammatical sentences. This was done to encourage participants to use the full range of the scale when giving their judgments.

Procedure

The participants were instructed to use a 7-point scale to provide their judgments, with 7 being “completely natural” and 1 being “completely unnatural.” The experiment was presented as an off-line paper survey. The participants first read instructions that explained the general purpose of acceptability judgment experiments with examples that illustrated these points and were encouraged to use the full range of the scale to judge the sentences. They first rated four practice sentences to become familiar with the task before judging the experimental sentences.

Data analysis

All experiments in this paper use the same data analysis procedure. First, the raw judgments provided by the participants were transformed into z-scores prior to analysis (e.g., Cowart 1997; Schütze and Sprouse 2013). The z-score transformation converts a participant’s scores to units that represent the number of standard deviations a particular rating is from that participant’s mean rating and corrects for the potential individual biases in treating the scale differently, e.g., using only a subset of the available ratings. The obtained z-scores were analyzed using linear mixed-effects models with word order and subextraction as fixed effects within the unaccusative CNP subject sentences and within the transitive CNP object sentences, and with argument type (unaccusative subject vs. transitive subject) and subextraction as fixed effects within the unaccusative and transitive CNP-subject sentences, and participants and items as random effects in both. We report the results with a model that included the fixed effects and an interaction term as random slopes only for participants and not items, as the fully specified model with random slopes for both participants and items failed to converge (see Barr et al. 2013 for a relevant discussion). The analyses were performed using the lmerTest package (Kuznetsova et al. 2016) in the statistical software R (R Core Team 2015).

Results

Three plots in Fig. 1 summarize the mean z-scores for the three comparisons in Experiment 1: (i) the unaccusative CNP subject sentences in two different orders with and without subextraction out of the CNP (left plot), (ii) the transitive CNP object sentences in two different orders with and without subextraction out of the CNP (central plot), and (iii) the unaccusative CNP subject and transitive CNP subject sentences in the long-before-short order with and without subextraction out of the CNP (right plot). The error bars in all of the plots presented in this paper represent standard error of the means.

Fig. 1
figure1

Mean z-scores for the unaccusative CNP subject sentences (left), the transitive CNP object sentences (center), and the unaccusative and transitive CNP subject sentences (right)

We begin with the results for Type 3 (the unaccusative CNP subject sentences), one of the simpler conditions in this study, (10a–d). Table 3 summarizes the factors that we manipulated and the predictions made by the two hypotheses: the spell-out/phase-based accounts (the spell-out/phase accounts) and the freezing-effects-based accounts (the freezing accounts).Footnote 7

Table 3 Unaccusative CNP subject sentences in two different orders

The long-before-short preference predicts that 3A should be rated relatively more acceptable than 3B while under the freezing-effect based accounts, the advantage of 3A being consistent with the long-before-short preference may be cancelled out by the fact that the CNP subject is moved in 3A but not in 3B. In other words, subextraction out of the CNP subject is expected to bring out an interaction between the long-before-short and the freezing accounts. The spell-out/phase accounts do not predict such an interaction.

Table 4 presents a summary of the statistical analysis. Within the unaccusative CNP subject sentences, both subextraction (β = −1.24, SE = 0.12, p<.01) and word order (β = 0.5, SE = 0.11, p<.01) are significant predictors of the acceptability judgment data, but their interaction is not significant (β = −0.02, SE = 0.16, p = .88).

Table 4 Summary of model estimates, standard errors, t-values, and p-values for the unaccusative CNP subject sentences in Experiment 1

We now move to the results for Type 2 (the transitive CNP object sentences), as in examples (9a–d) above. Table 5 summarizes the factors that we manipulated, and the predictions made by the positional accounts and the freezing accounts.Footnote 8

Table 5 Transitive CNP object sentences in two different orders

Recall that we have the same set of predictions as for the unaccusative CNP subjects with this comparison. The long-before-short preference predicts that 2A should be rated relatively more acceptable than 2B while the freezing-effect based accounts may predict that the advantage of 2A is overridden by the fact that only 2A involves the CNP object movement. In other words, subextraction out of the CNP object is expected to bring out an interaction between the long-before-short and the freezing accounts. The positional accounts do not predict such an interaction.

Table 6 summarizes the statistical analysis. The results with the transitive CNP object sentences are very similar to the results with the unaccusative CNP subject sentences. Both subextraction (β = −1.16, SE = 0.12, p<.01) and word order (β = −0.45, SE = 0.11, p<.01) are significant, but their interaction is not (β = 0.00, SE = 0.17, p = .99).

Table 6 Summary of model estimates, standard errors, t-values, and p-values for the transitive CNP object sentences in Experiment 1

Finally, let us look at the comparison between Type 1A and Type 3A, that is, the transitive and unaccusative CNP subject sentences in the long-before-short order (examples (8a, b) and (10a, c) above). Table 7 summarizes the factors that we manipulated and the predictions made by the positional accounts and the freezing accounts.

Table 7 Unaccusative CNP subject and transitive CNP subject sentences in the long-before-short order

Under the assumption that sentence-initial Japanese subjects obligatorily undergo A-movement, the freezing accounts predict no difference between 1A and 3A. In contrast, the positional accounts predict that only the CNP subject in 1A should be opaque to subextraction under the assumption that transitive subjects are external arguments while unaccusative subjects are internal arguments. Thus, an interaction between the external/internal distinction with CNPs and the extraction/no extraction distinction is expected.Footnote 9

Table 8 presents a summary of the statistical analysis. While subextraction is a significant predictor of the acceptability of the unaccusative and transitive CNP subject sentences (β = −1.34, SE = 0.09, p<.01), neither argument type (β = 0.00, SE = 0.08, p = 0.95) nor the interaction between the two factors is significant (β = 0.07, SE = 0.11, p = .53).

Table 8 Summary of model estimates, standard errors, t-values, and p-values for the unaccusative and transitive CNP subject sentences in Experiment 1

Discussion

The first two comparisons involved CNPs as internal arguments, either as unaccusative subjects (Type 3) or transitive objects (Type 2). The results of the comparisons show that A-sentences (the sentence types that conform to the long-before-short preference) were judged as significantly more acceptable than their counterparts which do not conform to this preference. These results hold regardless of whether or not subextraction takes place out of the CNPs (there was no interaction between the position of the CNPs and presence or absence of extraction). This suggests that in Japanese, moved constituents are not more opaque to subextraction than in situ constituents. In the third comparison, we found no significant difference between unaccusative (Type 3A) and transitive (Type 1A) CNP subjects in terms of subextraction, and no interaction between the status of the CNPs (external or internal argument) and presence or absence of extraction. In other words, transitive CNP subjects, which are external arguments, are not more opaque to subextraction than unaccusative CNP subjects, which are internal arguments. This finding indicates that the external vs. internal argument distinction does not affect the acceptability of subextraction in Japanese.

The finding that Japanese subextraction shows no sensitivity to the external/internal or moved/in situ distinctions among arguments is unexpected if Japanese subjects are islands, as suggested by Jurka (2010) and Jurka et al. (2011). Furthermore, the results of Experiment 1 increase the plausibility of the alternative account of Jurka’s findings that the apparent subject island effect may only reflect Japanese speakers’ preference for placing longer constituents before shorter ones. In our experiment, the long-before-short sentences (A-sentences) all got much higher ratings than the short-before-long sentences (B-sentences), whether or not they are of the basic word order or involve scrambling of the sentential subject/object. This also indicates that the preference for long-before-short word order is so strong that the processing cost of scrambling is negligible. This allows us to concentrate on the sentences that conform to the long-before-short pattern, and that is what we will do in the next experiment, which examines the subject-object asymmetry in Japanese.

Experiment 2: Subject-object asymmetry in subextraction out of CNPs

Experiment 2 examined whether the alleged difference between subextraction out of CNP subjects and objects reported in Jurka (2010) and Jurka et al. (2011) disappears once the relative weight and order of the relevant constituents is properly controlled for. To this end, Experiment 2 investigated the acceptability of two sentences types: (i) transitive sentences with a CNP subject and a simple NP object (transitive CNP subject sentences) and (ii) transitive sentences with a simple subject NP and a CNP object (transitive CNP object sentences); both sentence types were presented with and without subextraction out of the CNP. To control for the long-before-short preference, all of the experimental sentences were constructed with the CNP as close as possible to the sentence-initial position. In other words, what we are doing now is the direct comparison between Type 1A and Type 2A in Experiment 1, and we will do so with the same (minimally-paired) set of lexicalizations. (See Table 9.)

Table 9 Types of experimental sentences in Experiment 2

For the CNP object sentences (henceforth “object sentences”), conforming to the long-before-short pattern requires fronting the CNP object such that it precedes the experiencer subject. However, under the assumption that subjects always undergo A-movement in Japanese, the fronting of the CNP object is a non-issue, as both CNP subjects and fronted CNP objects move and therefore would be subject to the freezing effect.Footnote 10

We also needed to make it clear to our participants that subextraction out of a CNP took place, while keeping the structure of the experimental sentences as simple as possible to avoid higher processing costs incurred by more complex structures (e.g., multiple embedding). To achieve this balance, we began all of our test sentences with the same type of PP adjunct (“according to X”).Footnote 11

Using the structure with X-niyoruro ensured that the subextracted constituent could move to a clause-initial position only across this PP. This made it clear that subextraction took place, while it crossed only one clausal boundary.Footnote 12 Examples of the experimental sentences in the four conditions are given below. Further details on the experiment materials are available as a supplementary appendix to the online version of this paper. The scrambled phrase is in italics and the long and short constituents are in boxes below.

  1. (13)
    figuren
  1. (14)
    figureo

Assuming that the weight of the constituents and their relative linear order were properly controlled for in our experimental sentences, and that the cost of fronting of the CNP objects is negligible (as the results of Experiment 1 suggest), subject and object sentences in the no-subextraction condition should receive comparable ratings. Now, if Japanese transitive subjects are islands, the acceptability of the object sentences under the subextraction condition is predicted to be significantly higher than that of the subject sentences under the same condition. On the other hand, if Japanese subjects are not islands, then no significant difference in the mean acceptability judgments is predicted between the subject and object sentences under both the no-subextraction and subextraction conditions.

Participants

Fifty-three university students in Tokyo, Japan participated in this study. Three participants’ results were excluded from the analyses because either they were non-native speakers of Japanese or they did not complete the task as instructed.

Materials

Experiment 2 had a 2 × 2 design with argument (subject vs. object) and subextraction (subextraction vs. no-subextraction) as conditions. In order to compare sentences consisting of subextraction out of CNP subjects against those with subextraction out of CNP objects, we constructed sentence pairs with subjects and objects that are as close as possible to minimal pairs, by using eight sets of psychological verbs that alternate between subject-experiencer and object-experiencer verbs via inchoative-causative alternations (Akatsuka 1976; Pesetsky 1995; Iwata 1995; Matsumura 1996).

  1. (15)
    figurep

Subject-experiencer verbs such as okor-u ‘become angry’ select a nominative-marked experiencer and a dative or accusative-marked CNP denoting stimulus. Their object-experiencer counterparts, such as okor-ase-ru ‘anger’, select a nominative-marked CNP and an accusative-marked experiencer. Following previous studies on experiencer verbs (e.g., Belletti and Rizzi 1988; Pesetsky 1995; Iwata 1995; Landau 2010), we assume that subject-experiencer verbs have the nominative experiencer as their external argument and the dative/accusative CNP as their complement, while object-experiencer verbs have the nominative CNP as their external argument and the accusative experiencer as their complement. These eight sets of verbs allow us to create pairs of sentences where the same participants reverse their roles, functioning as subjects in one structure but complements in the other. Crucially, one of the arguments is always a CNP.

Two lexicalizations were created for each of the eight sets of experiencer verbs under each of the four conditions (8 verbs × 4 conditions × 2 lexicalizations = 64 sentences). These 64 sentences were divided into four lists using a Latin square design, and the resulting 16 target sentences from each list were mixed with the same 24 fillers of various constructions and acceptability as in Experiment 1 (16 + 24 = 40).Footnote 13 The order of the sentences was pseudo-randomized.

Procedure

Like Experiment 1, Experiment 2 used an acceptability judgment task with a 7-point scale, with 7 being “completely natural” and 1 being “completely unnatural.” The experiment was presented as an off-line paper survey with the same structure as Experiment 1.

Data analysis

Experiment 2 used the same data analysis steps as Experiment 1.

Results

Figure 2 gives mean z-scores from the results of Experiment 2, and Table 10 presents a summary of the statistical analysis.

Fig. 2
figure2

Mean z-scores for Experiment 2

Table 10 Summary of model estimates, standard errors, t-values, and p-values for Experiment 2

Subextraction was a significant factor for predicting the acceptability of the experimental sentences (β = −0.79, SE = 0.11, p<.01), while argument type was not (β = 0.05, SE = 0.09, p = .57). The interaction between the two factors was not significant either (β = −0.12, SE = 0.13, p = .36).

Discussion

Experiment 2 tested whether the alleged subject-object asymmetry in subextraction reported in Jurka (2010) and Jurka et al. (2011) disappears if the relative weight and order of the relevant constituents are properly controlled for. The results show no significant difference between subextraction out of CNP subjects and subextraction out of CNP objects in Japanese once the relative weight and order of the relevant constituents are properly controlled for. Taken together, the results of Experiments 1 and 2 provide strong evidence that the findings in Jurka (2010) and Jurka et al. (2011) are due to the effect of the long-before-short preference independently attested in Japanese. This leads us to conclude that there is no subject-object asymmetry in subextraction in Japanese.Footnote 14

Experiment 3: Wh-in-situ and subject-object asymmetry

So far, we have examined overt displacement where a subconstituent of a subject or object undergoes subextraction to the sentence-initial position. While overt movement is cross-linguistically the most common way to investigate the nature of constraints on extraction, we noted that sentences with scrambling out of CNPs are significantly degraded compared to their counterparts without scrambling due to the extra processing cost incurred by the displacement and the filler-gap dependency, regardless of the status of CNPs as subjects or objects. This leaves us with the possibility that the lack of difference in the mean acceptability judgments between the CNP subject and the CNP object sentences observed in Experiment 2 is due to a “floor effect.” In other words, both types of sentences were below the range of acceptability that Experiment 2 was designed to capture, and because of this limitation, no difference between the two types of sentences was found. We believe this is unlikely, because the actual mean z-score for the CNP subject sentences with scrambling (−0.28) was higher than the mean z-score for the CNP object sentences (−0.36). Nevertheless, it would be informative to compare the opacity of CNP subjects and objects in such a structure where extraction does not cause significant degradation.

With this background, Experiment 3 used wh-in-situ phrases that make the baseline conditions acceptable, much like the overt movement tests that have been used in previous studies on English and some other languages. The consensus in the Japanese syntax literature is that CNP subjects and CNP objects allow wh-in-situ arguments. This consensus is based on the acceptability of examples such as (15):

  1. (16)
    figurer

In (15a) and (15b), the wh-phrase dare ‘who’ is inside an RC modifying the matrix subject and object, respectively, and the sentences are acceptable. This is in clear contrast with scrambling out of an RC, which results in unacceptable sentences (Haig 1976; Saito 1985; see also fn. 5).

  1. (17)
    figures

In addition to sentences with wh-in-situ arguments, Experiment 3 also involved those with wh-in-situ adjuncts. While wh-arguments such as dare ‘who’ and nani ‘what’ are known to be licensed inside CNPs, there is a robust cross-linguistic generalization across wh-in-situ languages that certain adjunct wh-phrases, especially reason adverbial wh-phrases such as naze ‘why’, are not licensed in similar syntactic environments (e.g., Huang 1982; Fukui 1988; Lasnik and Saito 1992; Aoun and Li 1993; Tsai 1994; Kishimoto 2005; Ko 2005).Footnote 15

  1. (18)
    figuret

If we now combine the observation that the reason adverbial wh-phrase naze ‘why’ is island-sensitive with the hypothesis that Japanese subjects are more opaque than objects with respect to (covert) subextraction, a prediction can be made that wh-questions with a CNP with naze ‘why’ should be judged less acceptable inside a CNP subject than inside a CNP object. If argument wh-phrases are not island-sensitive, then we would predict an interaction between the status of CNPs—CNP subject vs. object—and the type of wh-phrase (an argument wh-phrase vs. the reason wh-phrase) with the combination of CNP subject and the reason wh-phrase naze ‘why’ showing the island effect.

In order to test these predictions, Experiment 3 involved the following four conditions: (i) nani ‘what’ inside a CNP subject, (ii) nani ‘what’ inside a CNP object, (iii) naze ‘why’ inside a CNP subject, and (iv) naze ‘why’ inside a CNP object. The experimental sentences in these four conditions were all embedded under a verb that selects for a clausal complement, with the Q-particle licensing the wh-phrase following the matrix verb. The weight and order of the constituents in the experimental sentences were controlled for so that they would be consistent with the long-before-short preference. (See Table 11.)

Table 11 Types of experimental sentences in Experiment 3

Examples of the experimental sentences in the four conditions are given below. Further details on the experiment materials are available as a supplementary appendix to the online version of this paper. The wh-phrase is in italics, and the long and short constituents are shown in boxes.

  1. (19)
    figureu
  1. (20)
    figurev
  1. (21)
    figurew

Participants

Fifty-one university students in Tokyo, Japan participated. The results of two participants were excluded from the analyses because either they were non-native speakers of Japanese or they did not complete the task as instructed.

Materials

Experiment 3 had a 2 × 2 design with argument (subject vs. object) and wh-phrase (nani ‘what’ vs. naze ‘why’) as the conditions. We used the same set of psychological verbs that we used in Experiment 2 to construct experimental sentences for Experiment 3. The four conditions for each of the eight sets of verbs were created and distributed into four lists using a Latin Square design. Each of the four sets of eight experimental sentences was mixed with the same thirty-one fillers with different constructions of varied acceptability as in the previous experiments, and the order of presentation was pseudo-randomized.Footnote 16 Thus, each participant rated a total of 39 sentences.

Procedure

As in Experiments 1 and 2, the task was an acceptability judgment with a 7-point scale, and the procedure was identical to Experiments 1 and 2 (an off-line paper survey).

Data analysis

Experiment 3 used the same data analysis steps as Experiment 1 and 2, with argument type (subject vs. object) and wh-phrase (nani ‘what’ vs. naze ‘why’) as fixed factors and participants and items as random factors.

Results

Figure 3 presents mean z-scores from the results of Experiment 3, and Table 12 summarizes the statistical analysis.

Fig. 3
figure3

Mean z-scores for Experiment 3

Table 12 Summary of model estimates, standard errors, t-values, and p-values for Experiment 3

The results show that wh-phrase is a significant factor predicting the acceptability of the experimental sentences (β = −0.53, SE = 0.18, p<.01). However, neither argument type (β = −0.24, SE = 0.17, p = .16) nor the interaction between argument type and wh-phrase (β = 0.13, SE = 0.24, p = .58) is significant.

Discussion

Experiment 3 was designed to test whether subjects are more opaque than objects with respect to the covert subextraction dependency established for wh-in-situ inside subjects and objects, using the argument wh-phrase nani ‘what’ and the reason adverbial wh-phrase naze ‘why’ (only the adverbial wh-phrase has been claimed to be island-sensitive in Japanese).

The results of Experiment 3 show significant differences in the mean acceptability between the sentences with naze ‘why’ inside a CNP and the ones with nani ‘what’ inside a CNP; these differences are independent of the grammatical function of the CNP (subject vs. object). To our knowledge, this is the first ever experimental finding in support of the long-standing observation that naze ‘why’ exhibits a stricter locality constraint than that of argument wh-phrases.Footnote 17 Further still, we found no significant difference in the acceptability of wh-in-situ inside CNP subjects and CNP objects, and this absence of subject-object contrast was independent of the wh-phrase type. Taken together, the results of Experiment 3 do not support the hypothesis that Japanese subjects are more opaque than objects to covert subextraction. This result fully corroborates our conclusions from the first two experiments: Japanese does not show any difference between subjects and objects with respect to transparency to subextraction, overt or covert.

Implications

Japanese subextraction in the context of cross-linguistic variation

The results of the three experiments together show that, once relative constituent weight and linear order of constituents are properly controlled for, subextraction out of CNP subjects in Japanese is no different from subextraction out of CNP objects. This result holds regardless of the nature of the subextraction, as the same subject/object symmetry is found in overt (Experiments 1 and 2) and covert (Experiment 3) subextraction. These results support our explanation that the apparent subject-island effects reported in Jurka (2010) and Jurka et al. (2011) are due to the independently motivated long-before-short preference in Japanese. Our results also suggest that the transparency/opacity of arguments in Japanese is not affected by the distinction between external and internal arguments on the one hand and the distinction between in situ and moved constituents on the other. The next question is what kind of theory of subject island phenomena best accommodates such languages like Japanese that show total lack of asymmetry between subjects and objects?

As discussed in Sect. 2, there are two major approaches to subject/object asymmetry in subextraction available in the market: the spell-out/phase approaches, which derive the asymmetry from the inherent derivational/structural differences between the subject position and the object position in transitive sentences (e.g., Nuñes and Uriagereka 2000; Johnson 2003; Rackowski and Richards 2005; Jurka 2010; Müller 2010) and the freezing-effect approaches, according to which syntactic movement renders a constituent opaque to subextraction (e.g., Wexler and Culicover 1981; Takahashi 1994; Stepanov 2007; Rizzi 2006; Rizzi and Shlonsky 2007). As we will see below, however, the recent findings from experimental studies as well as the findings from our experiments pose empirical challenges to both approaches.

As mentioned above, Stepanov (2007) argues against spell-out/phase-based accounts such as Nuñes and Uriagereka’s based on the lack of subject-object asymmetry in languages like Japanese, as one must assume that such languages are non-configurational, contra the standard assumption. He claims that freezing accounts can capture the Japanese facts by stipulating that subjects can remain in situ in Japanese.

Recent experimental work that formally measures acceptability judgments has provided mixed findings for Stepanov’s generalization on cross-linguistic variation in subextraction effects (Sprouse et al. 2011, 2015; Polinsky et al. 2013). For example, Sprouse et al. (2015) examined acceptability variation in a variety of island constructions in English and Italian, including subextraction out of subjects and objects. They found a significant interaction between argument types (subjects vs. objects) and extraction types (extraction of an entire argument vs. subextraction out of an argument) with Italian wh-questions, and the same interaction was marginally significant with English wh-questions. These findings suggest that the subject island effect holds in Italian and English, although the strength of this constraint may vary across languages to some extent.Footnote 18 While the subtle variations in island effects do not follow straightforwardly from either freezing accounts or position-based accounts, the fact that both languages demonstrate reliable subject island effects is compatible with Stepanov’s generalization.

Kravtchenko et al. (2009) and Polinsky et al. (2013) used long-distance scrambling of wh-phrases in Russian to investigate whether subextraction acceptability varies as a function of verb types (transitive, unergative, and unaccusative) as well as positions of subjects and objects (raised, pre-verbal vs. in situ, post-verbal). Both studies report that Russian unaccusative subjects are found to be more transparent for extraction than transitive subjects or subjects of unergatives. However, only the object subextraction sentences demonstrate degradation in the movement condition versus the in situ condition. Taken together, these findings suggest that in Russian, the external/internal argument distinction plays a crucial role, while the in situ/moved distinction is irrelevant.Footnote 19

Our findings from subextraction phenomena in Japanese together with the findings from these experimental studies paint a picture of cross-linguistic variation in subject/object asymmetry in subextraction as a cline of transparency/opacity of subjects with several main types of languages emerging: languages like English, where subjects are never transparent (although the island effect is weaker with unaccusative subjects) while objects are always transparent; languages like Russian and German, where only unaccusative subjects are transparent together with objects; languages like Italian, where subjects are transparent like objects as long as they remain in their base position, and a language like Japanese, in which subjects appear to be no different from objects with respect to subextraction.Footnote 20 An obvious question is what determines such cross-linguistic variation and whether more language types may be available.

Such cross-linguistic variation is not straightforwardly accounted for under either of the two existing approaches to subject island phenomena. The freezing-based approaches fail to predict the findings from Russian that unaccusative subjects were more transparent than transitive and unergative subjects whether they were in situ or moved (Kravtchenko et al. 2009; Polinsky et al. 2013). Our finding that moved internal arguments are not more transparent than their counterparts in situ in Japanese also contradicts the freezing-based accounts.Footnote 21 Spell-out/phase-based approaches also face difficulty accounting for the freezing-effects observed with external arguments in German (Jurka 2010). The complete lack of asymmetry between subjects and objects in Japanese with respect to subextraction in our experimental results further challenges the spell-out/phase-based accounts. Thus, our findings from Japanese together with the findings from previous experimental studies with other languages suggest that neither the external vs. internal difference nor the moved vs. in situ difference among arguments is sufficient to account for the cross-linguistic variation in subextraction phenomena.

Below, we explore two grammatical processes that might be considered as holding the key to accounting for the cross-linguistic variation in subextraction: the EPP and Agree. Our brief examination of the current understanding of the EPP and Agree and relevant cross-linguistic data lead us to present a potential account where the EPP is the most promising grammatical feature that holds the key to accounting for the cross-linguistic variation in subextraction. In particular, we show that the conception of the EPP as a well-formedness constraint at an interface between syntax and another module of the grammar (e.g., PF) that can be satisfied by an independently motivated syntactic process (Landau 2007; McFadden and Sundaresan 2018) predicts a wide-range of cross-linguistic variation in how the EPP can or needs to be satisfied. It also suggests that transparency/opacity of subjects in a particular language or in a particular syntactic environment may be linked to how the EPP can or needs to be satisfied in a given language or in a given syntactic environment.

The EPP as the locus of cross-linguistic variation in subject opacity/transparency

Under the conception of the EPP as a well-formedness constraint at an interface between syntax and another module of the grammar that can be satisfied by an independently motivated syntactic process, there are a number of factors that predict that languages can differ in a variety of ways in terms of how the EPP can and needs to be satisfied.

To begin with, the EPP may be considered universal (Jurka 2010)Footnote 22 or not (Kuroda 1985, 1988; Kitagawa 1986; Fukui 1986, 1988; McCloskey 1996). Under the assumption that the EPP is universal, it can be satisfied in many different ways. For languages in which the D feature is analyzed as triggering the syntactic operation that satisfies the EPP, the EPP can be satisfied by a non-pleonastic subject occupying [Spec, TP], either by moving or being base-generated there, or by an expletive, as in English. In another type of languages, the EPP is still linked to a D-feature, but the syntactic operation that checks the D-feature and ends up satisfying the EPP does not have to involve a subject DP but may involve syntactic operations with other constituents that bear the necessary D feature, for example, by verb raising (Alexiadou and Anagnostopoulou 1998; Davies and Dubinsky 2001). Davies and Dubinsky (2001) argue that the English-type languages require subjects to be DPs while subjects in Bulgarian do not have to be DPs because the EPP is satisfied with a D-feature in the former while it is satisfied with a V-feature via non-subjects, i.e., verb raising, in the latter. The association between the EPP and verb (or VP) movement has been explored particularly extensively with respect to verb-initial (V1) languages. If this is on the right track, we can expect V1 languages to lack subject/object asymmetry in subextraction. Some V1 languages corroborate this prediction; in particular, subjects are transparent in Halkomelem (Salish; Gerdts 1988), in Tzotzil (Mayan; Aissen 1996), Chol (Mayan; Coon 2009), and Chamorro (Austronesian; Chung 1991, 1998). In quite a few V1 languages, all arguments, regardless of their structural position, are opaque, which suggests that the opacity may be due to independent reasons. The above observations motivate a generalization in (20).

  1. (22)

    Subjects that provide the D feature for the anchoring of the EPP are islands.

Now, under the hypothesis that transparency/opacity of subjects to subextraction is somehow linked to how the EPP can or needs to be satisfied in a given language or in a given syntactic environment, our finding that Japanese completely lacks subject/object asymmetry in subextraction might be accounted for under different assumptions about the status of the EPP in the language. First, if the EPP is not universal and Japanese lacks the EPP (Kuroda 1985, 1988; Kitagawa 1986; Fukui 1986, 1988), Japanese is clearly outside of (20) and the lack of subject/object asymmetry in subextraction is expected. Under the assumption that the EPP is universal, the lack of subject/object asymmetry in subextraction might be attributed to Japanese being (i) a discourse configurational language or (ii) a Bulgarian-type language. It has been argued that in discourse configurational languages such as Japanese and Hungarian, the syntactic operation that satisfies the EPP might be tied not to a D feature but to features such as topic and focus (É. Kiss 1995; Miyagawa 2010). If this analysis is on the right track, (20) predicts the lack of subject/object asymmetry in subextraction in Japanese. Alternatively, Ueda (2008, 2009) argues that the EPP is satisfied in Japanese by a modal, which makes it a Bulgarian-type language. Again Japanese falls outside of (20).

Let us now consider several alternative accounts. First, since Japanese arguably lacks morphological agreement, lack of agreement seems to be an obvious alternative account for the cross-linguistic variation in the subject/object asymmetry in subextraction. But there is good evidence that the presence/absence of agreement morphology is not a critical ingredient for an account of the asymmetry. First, despite the long-standing claim that Japanese lacks morphological agreement, some researchers have proposed that honorific marking on the verb, as in (21a) below, is a form of agreement (Suzuki 1989; Toribio 1990; Boeckx and Niinuma 2004; but see Bobaljik and Yatsushiro 2006 for arguments against this analysis). However, the presence of subject honorification does not render that subject an island.

  1. (23)
    figurez

So even if honorific marking were indeed evidence of agreement, it does not contribute to the opacity of the subject with the relevant feature. On more solid ground agreement-wise, most Nakh-Dagestanian languages have agreement in gender and number, and their subjects are transparent. For example, in Tsez, the verb agrees in gender and number with the absolutive argument (Tsez is morphologically ergative).Footnote 23 In (22a), the verb agrees with the absolutive subject, which can appear in different positions on the surface (we do not show the data on objects, but they are very similar).

  1. (24)
    figureaa

Regardless of the surface position of the subject, such a subject can host subextraction:

  1. (25)
    figureab

At least two diagnostics show that Tsez NP-splits arise through subextraction: case connectivity and sensitivity to negative islands (Polinsky 2018). Hindi is another language where unmarked subjects (and objects) determine agreement in number and gender only. In Hindi, too, such agreed-with subjects are transparent to subextraction (Alok 2016).Footnote 24

One may argue, however, that the above data are irrelevant for the question under discussion because they do not involve person agreement. Here, it is important to note that a number of researchers have proposed that the morphological expression of D feature can either be reduced to the feature [person] (Longobardi 1994, 2005, 2008), or at least includes this feature (Benmamoun 2000; Roberts 2010; Miyagawa 2010). The connection between finiteness, which is linked to [tense], and [person] is well known and has solid semantic underpinnings. Tenses and participant denotations (say, pronouns) share indexical, anaphoric, and bound variable uses and can neither denote nor name their referents (Partee 1973). Just like [person] turns property denotations into individual denotations (i.e. denotations that can be referred to by pronouns) (Longobardi 2005), [tense] turns predications into propositions; in each case, the end result is a functionally complete entity (Harder 1996; Rothstein 2001).

Assuming such a connection, we can expect that if morphological agreement in person serves to satisfy the EPP on T/C, the noun phrase bearing the [person] feature is going to be frozen to subextraction. This is a refinement of the freezing principle according to which agreement makes a given expression φ-complete (Boeckx 2008; Lohndal 2011). Crucial for our argument is that there is no one-to-one correlation between agreement in person and opacity. For example, in English the D feature on T can be supplied by an expletive as in existentials, and in that case, the agreement with the associate does not render that nominal opaque. Hence the contrast between regular subjects, which are islands, (24a) and associates in existential constructions, which are transparent, (24b) (Lasnik and Park 2003):

  1. (26)
    1. a.

      */?Whati were [protests against \(t _{\mathrm{i}}\)] at the rally?

    2. b.

      Whati were there [protests against \(t _{\mathrm{i}}\)] at the rally?

We tentatively conclude that agreement does not play an integral role in the cross-linguistic variation in the subject/object asymmetry in subextraction.

Another alternative to our approach is to link the cross-linguistic variation in the subject/object asymmetry in subextraction to the DP/NP-language distinction. If the D feature is the crucial ingredient of the cross-linguistic variation of subextraction, as we are arguing, why not make DP/NP distinction the locus of the cross-linguistic variation of subextraction, rather than making the D feature play a mediating role between the EPP and the cross-linguistic variation of subextraction? In fact, it has been pointed out that languages without subject/object asymmetry form a subset of NP, rather than DP-type, languages (Corver 1990, 1992; Bošković 2005, 2008, 2009). If this is on the right track, we may be back to the familiar correlation between the absence of determiners and the possibility for left-branch extraction. Tsez, Hindi, and the two Mayan languages whose subjects are transparent (Chol and Tzotzil, see above) all lack determiners, thus instantiating the NP-type. On the other hand, Halkomelem has clear determiners, but its subjects are transparent. For Japanese, Watanabe (2006) argues that it has DP structure (but see Saito et al. 2008 for arguments against Watanabe’s extended nominal projections). These observations weaken the putative correlation between the lack of subject/object asymmetry and the NP-type. More generally, the DP/NP parametric division, as proposed by Bošković, is associated with a cluster of properties of which several are empirically problematic; for example, polysynthetic languages are predicted to be of the NP-type, but Adyghe has clear determiners (Smeets 1984; Testelets 2009), while only DP-languages are predicted to have clitic doubling, yet such doubling is found in Slovenian, which lacks determiners (Marušič and Žaucer 2010).

To conclude, we have argued that the cross-linguistic variation in the transparency/opacity of subjects is related to cross-linguistic variation in how the EPP on T/C is satisfied. In languages in which the EPP satisfaction is anchored to the D feature, the selection of a subject via movement (as in English) or base-generation (as in Arabic preverbal subjects, cf. Soltan 2007, 2011) is just one of many ways of providing this feature. But importantly, those subjects that provide the D feature for the anchoring of the EPP are islands; on this approach, the opacity of subjects is a just side effect of their being the source of the D feature. Morphological agreement in person is a good surface indication that the subject may have provided the D feature on T/C, but there is no one-to-one correlation between agreement and opacity. Now, if the D feature to which the EPP is anchored comes from sources other than the subject (expletive; V(P)-raising), then there is no subject/object asymmetry in subextraction. Lack of subject/object asymmetry is also predicted if the syntactic operation that satisfies the EPP does not involve a D-feature, as in discourse-configurational languages such as Japanese.

However, the difference between languages with “traditional” EPP (only an agreeing subject can satisfy the requirement) and “inclusive” EPP (any constituent can satisfy the requirement) alone does not account for the entire range of facts. The EPP in Russian is similar to the EPP in Japanese in that it is not tied to subjects (Bailyn 2004, 2012: 317–319);Footnote 25 nevertheless in Russian only unaccusative subjects are transparent (Polinsky et al. 2013). Likewise, German is analyzed either as not having the EPP at all (Wurmbrand 2006) or as having the “inclusive EPP” (Richards 2016:141), yet the transparency of German subjects is similar to that of Russian subjects if only in a more restrictive way, as unaccusative subjects lose transparency when fronted. These observations suggest that there may be other factors inducing subject/object asymmetry in subextraction in languages with inclusive EPP. For example, in languages with an inclusive EPP, subject DPs do not have to move to [Spec, TP] to satisfy the EPP. Therefore, the status and nature of moved subject positions may vary. We expect that, all factors being equal, A-positions should be transparent and A-bar positions should be opaque for subextraction.

Although the account presented in this subsection is promising, there are still potential questions to answer. First of all, it is important to consider if the data reported for other languages have potential confounding factors, the same way as the long-before-short effect in Japanese, which we discussed in this paper. Additionally, it remains to be seen whether alternative accounts can explain the same set of data.Footnote 26 We leave these issues for future research.

Conclusions

Three new acceptability judgment experiments, which controlled for a ‘long-before-short’ preference in Japanese, showed that Japanese subjects are not more opaque to subextraction than objects, with both overt subextraction (i.e. scrambling) and potential ‘covert’ subextraction (wh-in-situ). These results allow us to reinstate Japanese as a language without a subject-object asymmetry in subextraction; prior work questioning this position (Jurka 2010; Jurka et al. 2011) did not take into account a word-order preference which led to a different pattern of results.

After experimentally verifying the lack of a subject-object asymmetry in subextraction in Japanese, we then offered some considerations as to why languages differ with respect to subject transparency/opacity. With respect to Japanese, our conclusions echo what has been proposed by a number of other researchers: Japanese subjects do not satisfy the EPP on T/C. However, we depart from much of the earlier work in viewing the connection between the EPP and subjects as less direct. We consider the EPP a selectional feature (cf. Landau 2007; McFadden and Sundaresan 2018). As a selectional feature, it is parasitic on some other feature that it needs to be anchored to. With respect to functional heads responsible for finiteness, it is anchored to the feature D. On this approach, parametric variation in subject transparency follows from whether or not the feature D comes from the subject or not. If the subject is the source of the feature D to which the EPP is anchored, such a subject is an island; otherwise, it is expected to be transparent if it is in an A-position, as opposed to an A-bar position.

Notes

  1. 1.

    Rackowski and Richards (2005) argue that only constituents that enter into an Agree relation with a phase head (complements) are transparent to subextraction. This follows from the assumption that once a probe has entered into an Agree relation with a particular goal, it can disregard that goal for future Agree relations, i.e., it can enter into an Agree relation with a constituent internal to the former goal. Under their proposal, subextraction out of subjects is not permitted because subjects do not enter an Agree relation with v. For languages that allow for subextraction out of subjects, Rackowski and Richards (2005) suggest that subjects actually enter into an Agree relation with v (585; fn. 15). Under Müller’s (2010) proposal, in order for a phrase to move to a phase edge, an edge feature must be assigned to the phase head. Edge features can be assigned to a phase head freely as long as the head has an undischarged feature and therefore remains active in the derivation. Once a phase head has discharged all its features, it becomes inert and no edge feature can be assigned to it. This means that subextraction from a subject is blocked if the operation that merges a subject is the final operation taking place in a phase. Subextraction out of a moved phrase is blocked for the same reason.

  2. 2.

    Tamaoka et al. (2003) examined the effects of relative weight and order of constituents in processing and comprehension of Japanese sentences. Their results reveal no effects of constituent weight or order on either reading time or comprehension accuracy of Japanese transitive sentences, both at phrase- and sentence-levels.

  3. 3.

    We manipulated the relative order and length of constituents within the embedded clause since the CNP is always inside the embedded clause. While previous studies report the effects of the long-before-short preference in simple sentences (Dryer 1980; Hawkins 1994; Yamashita and Chang 2001; Yamashita 2002), we expected that the effects of the preference would be seen in embedded clauses as well.

  4. 4.

    Ross (1967, 1986) argues that Japanese CNPs with a complement clause headed by koto ‘fact’ are not islands (i) while CNPs with a complement clause headed by to yuu, literally ‘that say,’ are (ii), based on the observation that relativization out of CNPs results in unacceptability only with CNPs with to yuu, as in (ii-b):

    1. (i)
      figurei
    1. (ii)
      figurej

    Haig (1976), on the other hand, argues that scrambling can occur from a CNP with a complement clause headed by by to yuu (iii-a), while scrambling out of a relative clause (RC) is impossible (iii-b).

    1. (iii)
      figurek

    Finally, Saito (1985) sides with Haig (1976) in that scrambling out of a RC (iv-a) is worse than that out of a CNP with a complement clause headed by to yuu (iv-b). Nevertheless, Saito considers both cases to instantiate island violations. Saito further notes that subextraction out of a CNP with a complement clause headed by to yuu is much more degraded if the CNP is definite, for instance, accompanied by a demonstrative.

    1. (iv)
      figurel

    Thus, a consensus in the theoretical literature appears to be that Japanese RCs are islands whereas the island status of the CNPs with koto ‘fact’ and to yuu ‘that say’ has remained controversial. More recently, Fukuda and Sprouse (2017) used a factorial design to systematically compare subextraction out of a CP complement with subextraction out of different structures including RCs and CNPs headed by to yuu. Their study found a significant interaction between the structure type (CNP vs. RC) and subextraction (no scrambling vs. scrambling), supporting the claim that RCs are islands in Japanese. The comparison between CPs and CNPs with to yuu, however, found no significant interaction between the structure type (CPs vs. CNPs with to yuu) and subextraction (no scrambling vs. scrambling), suggesting that CNPs with to yuu are not islands. Taken together with the observation that CNPs headed by koto are more transparent to subextraction than CNPs headed by to yuu, Fukuda and Sprouse’s (2017) results suggest that it is safe to assume that CNPs headed by koto are not islands.

  5. 5.

    The results below will show that the baseline acceptability is actually much higher for A sentences than for B sentences, which indicates that the processing cost of CNP arguments is negligible.

  6. 6.

    But see below and also Sect. 7 for further discussions of this effect.

  7. 7.

    In Table 3, CNPS stands for “CNP subject.”

  8. 8.

    CNPO stands for “CNP object.”

  9. 9.

    A reviewer points out that the spell-out/phase constraint and the freezing constraint may be operating simultaneously; that much can be deduced from the findings reported in Jurka (2010) and Jurka et al. (2011) for German. Under such a scenario, 1A violates both of the constraints (i.e., subextraction out of a moved external argument) while 3A only violates the freezing constraint (i.e., subextraction out of a moved internal argument). We did not include our prediction under such a scenario in the table because it is not clear if and how acceptability judgments reflect simultaneous violations of multiple constraints: should we expect an additive effect? Our results in Experiment 1 and 2 show, however, that in Japanese neither the freezing effect nor the spell-out/phase effect is observed. Thus, we can conclude that a scenario where both effects are present is equally untenable.

  10. 10.

    Again, under the scenario in which both the positional constraint and the freezing constraint are simultaneously operating, (i) violates both of the constrains (i.e., subextraction out of a moved external argument) while (ii) only violates the positional constraint (i.e., subextraction out of a moved internal argument). As stated in fn. 10, we did not discuss the possibility of both constraints operating. First, it is not clear if and how acceptability judgments reflect the difference between a single violation of a constraint and simultaneous multiple violations of multiple constraints, and second, Experiments 1 and 2 show that neither the freezing effect nor the positional effect is observed with Japanese arguments.

  11. 11.

    The translation of this adjunct may be somewhat misleading, suggesting that the relevant PP is a parenthetical; however, these constituents in our stimuli are bona fide PPs and can be used as a signpost for short scrambling.

  12. 12.

    Here one may worry that the sentence-initial PP X-niyoruto ‘according to X’ may be interpreted as a part of the following CNP. Under such parsing of the experimental sentences, the scrambling could be analyzed as taking place inside the CNP, contrary to our intention. However, we believe that the suggested parsing of the PP as a constituent of the CNP is unlikely for several reasons. First, the experimental sentences were presented with punctuation which facilitates the intended parse. In particular, a comma was placed between the PP and the initial element of the CNP in all the experimental sentences, discouraging the parsing of the PP together with the following CNP. Second, all the experimental sentences involved X-niyoruto ‘according to X’ and an evidential marker such as yooda ‘seem’ in the matrix predicate. This is a very common pattern, and the presence of an evidential marker in the matrix predicate strongly suggests that the PP also belongs to the matrix clause. Finally, the CNPs in the experimental sentences all involved -koto ‘fact’, which triggers a presupposition that the proposition expressed by a -koto clause is true (e.g., Kuno 1973). In contrast, the PP Xniyoruto ‘according to X’ explicitly denies such a presupposition. This semantic mismatch between -koto and X-niyoruto makes the hypothesized constituent with the PP the CNP headed by -koto infelicitous. Thus, we believe that speakers are unlikely to have entertained a parse of X-niyoruro according to X as a part of a CNP headed by -koto.

  13. 13.

    These fillers are different from the fillers used in Experiment 1.

  14. 14.

    An anonymous reviewer questions whether null results can be taken as evidence for the null hypothesis (i.e., that there is no meaningful difference between two populations). Sprouse and Almeida (2017) provide a useful guideline for the minimum number of participants that acceptability judgment studies with different rating methods need in order to have sufficient statistical power for null results to be taken as evidence for the null hypothesis (80% or above). Their estimates indicate that, for a study that examines a medium-effect size acceptability contrast with a Likert scale, the minimum number of participants required to reach the 80% power is 37 under a conservative assumption that each participant rated only one item per experimental condition (Sprouse and Almeida 2017: 21, Fig. 3). Since the numbers of participants for the three experiments whose results are reported in this study are 62, 43 and 41, respectively, we contend that the null results that we obtained constitute evidence for the null hypothesis.

  15. 15.

    Why also exhibits unusual behavior in languages with overt wh-movement, for example, in Italian (e.g., Rizzi 1990; Ko 2005).

  16. 16.

    These fillers are different from the fillers used in Experiments 1 and 2.

  17. 17.

    Recall the argument in fn. 5 that Japanese CNPs headed by -koto ‘fact’ and by to yuu ‘that say’ are not islands. Given this claim, the degraded nature of ‘why’-sentences in Experiment 3 cannot be attributed merely to island-sensitivity; even embedding ‘why’ inside a non-island causes degradation. For future research, we would like to see whether naze ‘why’ becomes degraded even inside a simple embedded clause.

    A ban on long-distance interpretations of why is also observed in English sluicing (although it is well-known that English sluicing is island-insensitive). Lasnik (2005) notes that the sluiced wh-phrase why in examples such as (i) only allows the ‘short distance’ reading in (b), suggesting that such why obeys a stricter condition than island effects.

    1. (i)
      figurey

    In sum, although naze ‘why’ inside a non-island CNP is degraded, there is no additional island effect for a subject CNP.

  18. 18.

    Sprouse et al. (2015) found no interaction between argument types and extraction types with Italian relative clauses, while the interaction was significant with English relative clauses. This is compatible with the claim by Rizzi (1982) that subjects are not islands in Italian relative clauses, and Sprouse et al. argue that the dependencies in wh-questions and relative clauses cannot be assumed to be identical. While this conclusion potentially impacts the generality of subject island effects, the potential uniqueness of relativization is irrelevant to the current experiments that focus on scrambling and covert wh-movement.

  19. 19.

    In German, Jurka (2010) and Jurka et al. (2011) report that both the external/internal distinction and the moved/in situ distinction have a significant impact on the acceptability of subextraction out of arguments.

  20. 20.

    The proposed distinction between Italian and Japanese is tentative, as we only examined subextraction out of moved vs. in situ unaccusative subjects in Japanese with Experiment 1, and we do not have data on subextraction out of moved and in situ external arguments (i.e. transitive and unergative subjects) in Japanese.

  21. 21.

    Following Chomsky (2000, 2001), some freezing-based accounts restrict freezing to movement for case and/or agreement, on the assumption that criterial properties such as [focus] or [topic] do not influence minimality considerations (cf. Gallego and Uriagereka 2008). On this particular approach to freezing, the transparency of Japanese subjects may be taken to follow from the absence of phi-feature agreement (Kuroda 1985, 1988; Fukui 1986, 1995, 2006).

  22. 22.

    We would like to underscore that in Jurka’s model, the subject condition that cannot be reduced to freezing effects is universal. That is why the results obtained by Jurka and colleagues for Japanese were of such importance.

  23. 23.

    There are four genders in the singular and two in the plural (indicated as (n)IPL in the glosses below). The verbal exponent of agreement is always a prefix, although agreement is marked only on a subset of vowel-initial verbs (Polinsky and Potsdam 2001; Polinsky 2003).

  24. 24.

    With respect to non-agreeing subjects in Hindi, it has been claimed that they satisfy the EPP on T but that their movement to T is EPP-driven, not Agree-driven (Anand and Nevins 2006); evidence for that comes from the lack of reconstruction for scope, along the arguments developed by Miyagawa (2001) for Japanese. Hindi ergative expressions are islands (Alok 2016), but their island status may be independently motivated by their status as PPs, not DPs, with the ergative exponent ne being a postposition (see Mahajan 1997 for arguments supporting this analysis).

  25. 25.

    There are two main proposals concerning the EPP on T in Russian. According to Bailyn (2004, 2012), a designated A-position in [Spec, T] can host a variety of arguments. According to Slioussar (2011) and Krejci et al. (2018), only agreed-with arguments appear in that position, whereas all other arguments are in a different, higher, A-bar position.

  26. 26.

    We thank an anonymous reviewer for pointing out these potential problems.

References

  1. Agbayani, Brian, Chris Golston, and Toru Ishii. 2015. Syntactic and prosodic scrambling in Japanese. Natural Language and Linguistic Theory 33: 47–77.

    Google Scholar 

  2. Aissen, Judith. 1996. Pied-piping, abstract agreement, and functional projections in Tzotzil. Natural Language and Linguistic Theory 14: 447–491.

    Google Scholar 

  3. Akatsuka, Noriko. 1976. Reflexivization: A transformational approach. In Syntax and semantics 6: Japanese generative grammar, ed. Masayoshi Shibatani, 51–116. New York: Academic Press.

    Google Scholar 

  4. Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, v-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491–539.

    Google Scholar 

  5. Alok, Depak. 2016. The syntax of a split: The case of Hindi and Magahi. Paper presented at RULingXI, Rutgers University.

  6. Anand, Pranav, and Andrew Nevins. 2006. The locus of ergative case assignment: Evidence from scope. In Ergativity: Emerging issues, eds. Alana Johns, Diane Massam, and Juvenal Ndayiragije, 3–26. Dordrecht: Springer.

    Google Scholar 

  7. Aoun, Joseph, and Y.-H. Audrey Li. 1993. Wh-elements in situ: Syntax or LF? Linguistic Inquiry 24: 199–238.

    Google Scholar 

  8. Bailyn, John. 2004. Generalized inversion. Natural Language and Linguistic Theory 22: 1–49.

    Google Scholar 

  9. Bailyn, John. 2012. The syntax of Russian. Cambridge: Cambridge University Press.

    Google Scholar 

  10. Barr, Dale. J., Roger Levy, Christoph Scheepers, and Harry J. Tily. 2013. Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language 68: 255–278.

    Google Scholar 

  11. Belletti, Adriana, and Luigi Rizzi. 1988. Psych-verbs and theta theory. Natural Language and Linguistic Theory 6: 291–352.

    Google Scholar 

  12. Benmamoun, Elabbas. 2000. The feature structure of functional categories: A comparative study of Arabic dialects. Oxford: Oxford University Press.

    Google Scholar 

  13. Bobaljik, Jonathan D., and Kazuko Yatsushiro. 2006. Problems with honorification-as-agreement in Japanese: A reply to Boeckx and Niinuma. Natural Language and Linguistic Theory 24: 355–384.

    Google Scholar 

  14. Boeckx, Cedric. 2008. Understanding minimalist syntax: Lessons from locality in long-distance dependencies. Oxford: Blackwell.

    Google Scholar 

  15. Boeckx, Cedric, and Fumikazu Niinuma. 2004. Conditions on agreement in Japanese. Natural Language and Linguistic Theory 22: 453–480.

    Google Scholar 

  16. Bošković, Željko. 2005. On the locality of left branch extraction and the structure of NP. Studia Linguistica 59: 1–45.

    Google Scholar 

  17. Bošković, Željko. 2008. What will you have, DP or NP? In North East Linguistic Society (NELS) 37, eds. Emily Elfner and Martin Walkow. 101–114. Amherst: University of Massachusetts.

    Google Scholar 

  18. Bošković, Željko. 2009. More on the No-DP analysis of article-less languages. Studia Linguistica 63: 187–203.

    Google Scholar 

  19. Chomsky, Noam. 1973. Conditions on transformations. In A Festschrtft for Morris Halle, eds. Stephen R. Anderson and Paul Kiparsky, 232–286. New York: Holt, Rinehart and Winston.

    Google Scholar 

  20. Chomsky, Noam. 1995. The Minimalist program. Cambridge: MIT Press.

    Google Scholar 

  21. Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by step: Essays on Minimalist syntax in honor of Hward Lasnik, eds. Roger Martin, David Michaels, and Juan Uriagereka, 890–155. Cambridge: MIT Press.

    Google Scholar 

  22. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, ed. Michael Kenstowicz, 1–52. Cambridge: MIT Press.

    Google Scholar 

  23. Chung, Sandra. 1991. Sentential subjects and proper binding in Chamorro. In Interdisciplinary approaches to language: Essays in honor of S.-Y. Kuroda, eds. Carol Georgopoulos and Roberta Ishihara, 75–99. Dordrecht: Kluwer.

    Google Scholar 

  24. Chung, Sandra. 1998. The design of agreement. Chicago: University of Chicago Press.

    Google Scholar 

  25. Coon, Jessica. 2009. Interrogative possessors and the problem with pied-piping in Chol. Linguistic Inquiry 40: 165–175.

    Google Scholar 

  26. Corver, Norbert. 1990. The syntax of left branch extractions. PhD diss., Tilburg University.

  27. Corver, Norbert. 1992. On deriving certain left branch extraction asymmetries: A case study in parametric syntax. In Northeast Linguistic Society (NELS) 22, 67–84. Amherst: GLSA.

    Google Scholar 

  28. Corver, Norbert. 2017. Subextraction. In Wiley Blackwell companion to syntax, 2nd edn., eds. Martin Everaert and Henk C. Van Riemsdijk, 566–600. Oxford: Blackwell. https://doi.org/10.1002/9781118358733.wbsyncom054.

    Google Scholar 

  29. Cowart, Wayne. 1997. Experimental syntax: Applying objective methods to sentence judgments. Thousand Oaks: Sage.

    Google Scholar 

  30. Davies, William D., and Stanley Dubinsky. 2001. Functional architecture and the distribution of subject properties. In Objects and other subjects: Grammatical functions, functional categories and configurationality, eds. William D. Davies and Stanley Dubinsky, 247–279. Dordrecht: Kluwer.

    Google Scholar 

  31. Dryer, Matthew S. 1980. The positional tendencies of sentential noun phrases in universal grammar. The Canadian Journal of Linguistics 25: 123–195.

    Google Scholar 

  32. É. Kiss, Katalin. 1995. NP-movement, operator movement, and scrambling in Hungarian. In Discourse configurational languages, ed. Katalin É. Kiss, 207–243. Oxford: Oxford University Press.

    Google Scholar 

  33. Fukuda, Shin and Jon Sprouse. 2017. Are there islands in Japanese? Ms., University of Hawai‘i at Mānoa and University of Connecticut.

  34. Fukui, Naoki. 1986. A theory of category projection and its applications. PhD diss., MIT.

  35. Fukui, Naoki. 1988. LF extraction of naze: Some theoretical implications. Natural Language and Linguistic Theory 6: 503–526.

    Google Scholar 

  36. Fukui, Naoki. 1995. Theory of projection in syntax. Stanford and Tokyo: CLSI and Kuroshio.

    Google Scholar 

  37. Fukui, Naoki. 2006. Theoretical comparative syntax: Studies in macroparameters. New York: Routledge.

    Google Scholar 

  38. Gallego, Angel, and Juan Uriagereka. 2008. Freezing effects. 18th Colloquium on Generative Grammar, Lisbon.

    Google Scholar 

  39. Gerdts, Donna. 1988. Object and absolutive in Halkomelem Salish. New York: Garland.

    Google Scholar 

  40. Haig, John H. 1976. Shadow pronoun deletion in Japanese. Linguistic Inquiry 7: 363–371.

    Google Scholar 

  41. Harder, Peter. 1996. Functional semantics: A theory of meaning, structure, and tense in English. Berlin: de Gruyter.

    Google Scholar 

  42. Hawkins, John A. 1994. A performance theory of order and constituency. Cambridge: Cambridge University Press.

    Google Scholar 

  43. Hoji, Hajime. 1985. Logical form constraints and the configurational structures in Japanese. PhD diss., University of Washington.

  44. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. PhD diss., MIT.

  45. Ishii, Toru. 1997. An asymmetry in the composition of phrase structure and its consequences. PhD diss., University of California, Irvine.

  46. Iwata, Seizi. 1995. The distinctive character of psych-verbs as causatives. Linguistic Analysis 1–2: 95–120.

    Google Scholar 

  47. Johnson, Kyle. 2003. Towards an etiology of adjunct islands. Nordlyd 31(1): 187–215.

    Google Scholar 

  48. Jurka, Johannes. 2010. The importance of being a complement: The CED effects revisited. PhD diss., University of Maryland.

  49. Jurka, Johannes, Chizuru Nakao, and Akira Omaki. 2011. It’s not the end of the CED as we know it: Revisiting German and Japanese subject islands. In West Coast Conference on Formal Linguistics (WCCFL) 28, eds. Mary Byram Washburn, Katherine McKinney-Bock, Erika Varis, Ann Sawyer, and Barbara Tomaszewicz. 124–132. Somerville: Cascadilla Press.

    Google Scholar 

  50. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge: MIT Press.

    Google Scholar 

  51. Kikuchi, Akira. 1987. Comparative deletion in Japanese. Ms., Yamagata University.

  52. Kishimoto, Hideki. 2005. Wh-in-situ and movement in Sinhara questions. Natural Language and Linguistic Theory 23: 1–51.

    Google Scholar 

  53. Kishimoto, Hideki. 2006. Japanese syntactic nominalization and VP-internal syntax. Lingua 116: 771–810.

    Google Scholar 

  54. Kitagawa, Yoshihisa. 1986. Subjects in Japanese and English. PhD diss., University of Massachusetts.

  55. Ko, Heejeong. 2005. Syntax of why-in-situ: Merge into [Spec, CP] in the overt syntax. Natural Language and Linguistic Theory 23: 867–916.

    Google Scholar 

  56. Kravtchenko, Ekaterina, Maria Polinsky, and Ming Xiang. 2009. Are all subject islands created equal? Poster presented at CUNY, UC Davis.

  57. Krejci, Bonnie, Vera Gribanova, and Boris Harizanov. 2018. AGREE-dependent A-movement and low copy pronunciation in Russian. In Formal Approaches to Slavic Linguistics (FASL) 25: The Third Cornell Meeting 2016, eds. Wayles Browne et al., 119–139. Ann Arbor: Michigan Slavic Publications.

    Google Scholar 

  58. Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge: MIT Press.

    Google Scholar 

  59. Kuznetsova, Alexandra, Per B. Brockhoff, and Rune H. B. Christensen. 2016. lmerTest: Tests in linear mixed effects models. R package version 2.0-30. Available at https://cran.r-project.org/web/packages/lmerTest/lmerTest.pdf.

  60. Kuroda, S.-Y. 1980. Bun no Kozo [The structure of sentences]. In Nichi-ei Hikaku Koza II: Bunpo, ed. Tetsuya Kunihiro, 23–62. Tokyo: Taishukan.

    Google Scholar 

  61. Kuroda, S.-Y. 1985. Whether we agree or not: Rough ideas about the comparative syntax of English and Japanese. Ms., UCSD.

  62. Kuroda, S.-Y. 1988. Whether we agree or not: A comparative syntax of English and Japanese. Linguisticae Investigationes 12: 1–47.

    Google Scholar 

  63. Landau, Idan. 2007. EPP extensions. Linguistic Inquiry 38: 485–523.

    Google Scholar 

  64. Landau, Idan. 2010. The locative syntax of experiencers. Cambridge: MIT Press.

    Google Scholar 

  65. Lasnik, Howard. 2005. How to evade moving violations. Handout from a lecture at LSA Summer Institute, MIT.

  66. Lasnik, Howard, and Myung-Kwan Park. 2003. The EPP and the subject condition under sluicing. Linguistic Inquiry 34: 649–660.

    Google Scholar 

  67. Lasnik, Howard, and Mamoru Saito. 1992. Moveα: Conditions on its application and output. Cambridge: MIT Press.

    Google Scholar 

  68. Lohndal, Terje. 2011. Freezing effects and objects. Journal of Linguistics 47: 163–199.

    Google Scholar 

  69. Longobardi, Giuseppe. 1994. Reference and proper names. Linguistic Inquiry 25: 609–665.

    Google Scholar 

  70. Longobardi, Giuseppe. 2005. Toward a unified grammar of reference. Zeitschrift für Sprachwissenschaft 24: 5–44.

    Google Scholar 

  71. Longobardi, Giuseppe. 2008. Reference to individuals, person, and the variety of mapping parameters. In Essays on nominal determination: From morphology to discourse management, eds. Henrik Høeg Müller and Alex Klinge, 189–211. Amsterdam: Benjamins.

    Google Scholar 

  72. Mahajan, Anoup. 1997. Universal grammar and the typology of ergative languages. In Studies on universal grammar and typological variation, eds. Artemis Alexiadou and Thomas A. Hall, 35–57. Amsterdam: Benjamins.

    Google Scholar 

  73. Marušič, Franc, and Rok Žaucer. 2010. Clitic doubling in a determinerless language with second position clitics. In Formal Studies in Slavic Linguistics, Proceedings of FDSL 7.5, eds. Gerhild Zybatow, Philip Dudchuk, Serge Minor, and Ekaterina Pshehotskaya, 101–115. Frankfurt am Main: Peter Lang.

    Google Scholar 

  74. Matsumura, Hiromi. 1996. On Japanese psych-verbs. In Toronto Working Papers in Linguistics 15, eds. Päivi Koskinen and Carolyn Smallwood, 123–140. University of Toronto.

    Google Scholar 

  75. McCloskey, James. 1996. Subjects and subject positions in Irish. In The syntax of Celtic languages: A comparative perspective, eds. Robert Borsley and Ian Roberts, 241–283. Cambridge: Cambridge University Press.

    Google Scholar 

  76. McFadden, Thomas, and Sandya Sundaresan. 2018. What the EPP and comp-trace effects have in common: Constraining silent elements at the edge. Glossa: A Journal of General Linguistics 3(1): 43. doi:http://doi.org/10.5334/gjgl.419.

    Google Scholar 

  77. Miyagawa, Shigeru. 2001. The EPP, scrambling and wh-in-situ. In Ken Hale: A life in language, ed. Michael Kenstowicz. 293–331. Cambridge: MIT Press.

    Google Scholar 

  78. Miyagawa, Shigeru. 2010. Why agree? Why move?: Unifying agreement-based and discourse-configurational languages. Cambridge: MIT Press.

    Google Scholar 

  79. Müller, Gereon. 2010. On deriving CED effects from the PIC. Linguistic Inquiry 41: 35–82.

    Google Scholar 

  80. Nishigauchi, Taisuke. 1990. Quantification in the theory of grammar. Dordrecht: Kluwer.

    Google Scholar 

  81. Nuñes, Jairo, and Juan Uriagereka. 2000. Cyclicity and extraction domains. Syntax 3: 20–43.

    Google Scholar 

  82. Partee, Barbara H. 1973. Some structural analogies between tenses and pronouns in English. The Journal of Philosophy 70: 601–609.

    Google Scholar 

  83. Pesetsky, David. 1995. Zero syntax: Experiencers and cascades. Cambridge: MIT Press.

    Google Scholar 

  84. Polinsky, Maria. 2003. Non-canonical agreement is canonical. Transactions of the Philological Society 101: 279–312.

    Google Scholar 

  85. Polinsky, Maria. 2018. Freezing and phi-feature agreement: On the role of [PERSON]. In Freezing: Theoretical approaches and empirical domains, eds. Jutta M. Hartmann et al., 284–316. Berlin: de Gruyter.

    Google Scholar 

  86. Polinsky, Maria, Carlos Gómez Gallo, Peter Graff, Ekaterina Kravtchenko, Adam M. Morgan, and Anne Sturgeon. 2013. Subject islands are different. In Experimental syntax and island effects, eds. Jon Sprouse and Norbert Hornstein, 286–309. Cambridge: Cambridge University Press.

    Google Scholar 

  87. Polinsky, Maria, and Eric Potsdam. 2001. Long-distance agreement and topic in Tsez. Natural Language and Linguistic Theory 19: 583–646.

    Google Scholar 

  88. Rackowski, Andrea, and Norvin Richards. 2005. Phase edge and extraction: A Tagalog case study. Linguistic Inquiry 36: 565–599.

    Google Scholar 

  89. R Core Team. 2015. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Available at https://www.R-project.org/.

    Google Scholar 

  90. Richards, Norvin. 1997. What moves where when in which languages? PhD diss., MIT.

  91. Richards, Norvin. 2000. An island effect in Japanese. Journal of East Asian Linguistics 9: 187–205.

    Google Scholar 

  92. Richards, Norvin. 2008. Wh-questions. In The Oxford handbook of Japanese linguistics, eds. Shigeru Miyagawa and Mamoru Saito, 348–371. Oxford: Oxford University Press.

    Google Scholar 

  93. Richards, Norvin. 2016. Contiguity theory. Cambridge: MIT Press.

    Google Scholar 

  94. Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris.

    Google Scholar 

  95. Rizzi, Luigi. 1990. Relativized minimality. Cambridge: MIT Press.

    Google Scholar 

  96. Rizzi, Luigi. 2006. On the form of chains: Criterial positions and ECP Effects. In WH-movement: Moving on, eds. Lisa Cheng and Norbert Corver, 97–134. Cambridge: MIT Press.

    Google Scholar 

  97. Rizzi, Luigi. 2010. On some properties of criterial freezing. In The complementizer phase: Subjects and operators, 17–32. Oxford: Oxford University Press.

    Google Scholar 

  98. Rizzi, Luigi, and Ur Shlonsky. 2007. Strategies of subject extraction. In Interfaces + Recursion = Language?: Chomsky’s minimalism and the view from syntax-semantics, eds. Uli Sauerland and Hans-Martin Gärtner, 115–160. Berlin: de Gruyter.

    Google Scholar 

  99. Roberts, Ian. 2010. Agreement and head movement: Clitics, incorporation, and defective goals. Cambridge: MIT Press.

    Google Scholar 

  100. Ross, John R. 1967. Constraints on variables in syntax. PhD diss., MIT

  101. Ross, John R. 1986. Infinite syntax! Norwood: ABLEX.

    Google Scholar 

  102. Rothstein, Susan. 2001. Predicates and their subjects. Dordrecht: Kluwer.

    Google Scholar 

  103. Saito, Mamoru. 1983. Case and government in Japanese. In West Coast Conference on Formal Linguistics (WCCFL) 2, eds. Michael Barlow, Daniel P. Flickinger, and Nancy Wiegand. 247–259. Stanford: CSLI.

    Google Scholar 

  104. Saito, Mamoru. 1985. Some asymmetries in Japanese and their theoretical implications. PhD diss., MIT

  105. Saito, Mamoru. 1992. Long distance scrambling in Japanese. Journal of East Asian Linguistics 1: 69–118.

    Google Scholar 

  106. Saito, Mamoru, and Hajime Hoji. 1983. Weak crossover and Move α in Japanese. Natural Language and Linguistic Theory 1: 245–259.

    Google Scholar 

  107. Saito, Mamoru, T.-H. Jonah Lin, and Keiko Murasugi. 2008. N’-ellipsis and the structure of noun phrases in Chinese and Japanese. Journal of East Asian Linguistics 17: 247–271.

    Google Scholar 

  108. Schütze, Carson T., and Jon Sprouse. 2013. Judgment data. In Research methods in linguistics, eds. Rob J. Podesva and Devyani Sharma, 27–50. New York: Cambridge University Press.

    Google Scholar 

  109. Shimoyama, Junko. 2001. Wh-costructions in Japanese. PhD diss., University of Massachusetts, Amherst.

  110. Slioussar, Natalia. 2011. Russian and the EPP requirement in the Tense domain. Lingua 121: 2048–2068.

    Google Scholar 

  111. Smeets, Rieks. 1984. Studies in West Circassian phonology and morphology. Leiden: Hakuchi Press.

    Google Scholar 

  112. Soltan, Usama. 2007. On postcyclic Merge and Agree in syntactic derivations: First conjunct agreement in Standard Arabic revisited. In Perspectives on Arabic linguistics, ed. Elabbas Benmamoun. Vol. 19, 191–213. Amsterdam: Benjamins.

    Google Scholar 

  113. Soltan, Usama. 2011. On issues of Arabic syntax: An essay in syntactic argumentation. Brill’s Annual of Afroasiatic Languages and Linguistics 3: 236–280.

    Google Scholar 

  114. Sprouse, Jon, and Diogo Almeida. 2017. Design sensitivity and statistical power in acceptability judgment experiments. Glossa: A Journal of General Linguistics 2(1)14: 1–32.

    Google Scholar 

  115. Sprouse, Jon, Shin Fukuda, Hajime Ono, and Robert Kluender. 2011. Reverse island effects and the backward search for a licensor in multiple wh-questions. Syntax 14: 179–203.

    Google Scholar 

  116. Sprouse, Jon, Ivano Caponigro, Ciro Greco, and Carlo Cecchetto. 2015. Experimental syntax and the variation of island effects in English and Italian. Natural Language and Linguistic Theory 34: 307–344.

    Google Scholar 

  117. Stepanov, Arthur. 2007. The end of the CED? Minimalism and extraction domains. Syntax 10: 80–126.

    Google Scholar 

  118. Suzuki, Tatsuya. 1989. A syntactic analysis of an honorific construction o-ni naru under the DP hypothesis: Towards a unified theory of honorification. In West Coast Conference on Formal Linguistics (WCCFL) 8, eds. E. Jane Fee and Katherine Hunt. 384–398. Stanford: CSLI.

    Google Scholar 

  119. Takahashi, Daiko. 1994. Minimality of movement. PhD diss., University of Connecticut.

  120. Tamaoka, Katsuo, Hiromu Sakai, Jun-ichiro Kawahara, and Yayoi Miyaoka. 2003. The effects of phrase-length order and scrambling in the processing of visually presented Japanese sentences. Journal of Psycholinguistic Research 32: 431–453.

    Google Scholar 

  121. Testelets, Jakov, ed. 2009. Aspekty polisintetizma. Očerki po grammatike adygejskogo jazyka. Moscow: RGGU.

    Google Scholar 

  122. Toribio, Almeida Jacqueline. 1990. Specifier-head agreement in Japanese. In West Coast Conference on Linguistics (WCCFL) 9, ed. Aaron Halpern. 535–548. Stanford: CSLI.

    Google Scholar 

  123. Tsai, W.-T. Dylan. 1994. On nominal islands and LE extraction in Chinese. Natural Language and Linguistic Theory 12: 121–175.

    Google Scholar 

  124. Ueda, Yukiko. 2008. Person restrictions and syntactic structure of Japanese modals. Scientific Approaches to Language 7: 123–150. Center for Language Sciences, Kanda University of International Studies.

    Google Scholar 

  125. Ueda, Yukiko. 2009. The right periphery in the Japanese CP. Scientific Approaches to Language 8: 95–118. Tokyo: Center for Language Sciences, Kanda University of International Studies.

    Google Scholar 

  126. Uriagereka, Juan. 1999. Multiple Spell-Out. In Working minimalism, eds. Samuel D. Epstein and Norbert Hornstein, 251–282. Cambridge: MIT Press.

    Google Scholar 

  127. Watanabe, Akira. 1992. Subjacency and s-structure movement of wh-in-situ. Journal of East Asian Linguistics 1: 255–291.

    Google Scholar 

  128. Watanabe, Akira. 2006. Functional projections of nominals in Japanese: Syntax of classifiers. Natural Language and Linguistic Theory 24: 241–306.

    Google Scholar 

  129. Wexler, Kenneth, and Peter Culicover. 1981. Formal principles of language acquisition. Cambridge: MIT Press.

    Google Scholar 

  130. Wurmbrand, Susi. 2006. Licensing case. Journal of Germanic Linguistics 18: 175–236.

    Google Scholar 

  131. Yamashita, Hiroko. 2002. Scrambled sentences in Japanese: Linguistic properties and motivations for production. Text 22: 597–633.

    Google Scholar 

  132. Yamashita, Hiroko, and Franklin Chang. 2001. Long-before-short preference in the production of a head final language. Cognition 81: B45–B55.

    Google Scholar 

Download references

Acknowledgements

We would like to thank Ken Hiraiwa, Hideki Maki, and Hajime Ono, for allowing us to run our pilot studies and experiments with their students, and Ted Levin, Kamil Deen, and Julie Jiang for reading earlier drafts of this paper and providing us with valuable comments and suggestions. We are grateful to three anonymous reviewers, Caroline Heycock, and Julie Legate for their helpful suggestions and comments that greatly improved the content, organization and presentation of this paper. Many thanks are also due to the audiences at the 10th Workshop on Altaic Formal Linguistics (WAFL10) and the 8th Formal Approaches to Japanese Linguistics (FAJL8), where earlier versions of this study were presented. This work is an output of a research project implemented as part of the Basic Research Program at the National Research University Higher School of Economics (HSE). It was also supported in part by JSPS KAKENHI Grant Number JP17K02823 to Chizuru Nakao and NSF grant BCS-1619857 to Maria Polinsky.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Shin Fukuda.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

The first author of this paper, Akira Omaki, passed away on August 6, 2018, following complications from lymphoma. Akira worked on this project while fighting cancer, until early June 2018, when he had to put his research activities on hold to focus on a new treatment. We feel honored and privileged to have been able to share this piece of research, one of Akira’s last, with him. We hope that those who knew Akira can see his passion, dedication, and professionalism in this paper.

Electronic Supplementary Material

Below are the links to the electronic supplementary material.

(PDF 427 kB)

(PDF 176 kB)

(PDF 163 kB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Omaki, A., Fukuda, S., Nakao, C. et al. Subextraction in Japanese and subject-object symmetry. Nat Lang Linguist Theory 38, 627–669 (2020). https://doi.org/10.1007/s11049-019-09449-8

Download citation

Keywords

  • Subextractions
  • CED effects
  • Subject island effects
  • Scrambling
  • wh-in-situ
  • Japanese
  • Sentence acceptability judgment