One of the controversial issues surrounding unaccusativity in Japanese is the surface position of unaccusative subjects. While compelling evidence has been found that the theme subject of an unaccusative verb originates as an internal argument, it is still an open question whether or not the subject moves from its original position to a derived subject position. This article aims to contribute to this debate from a psycholinguistic perspective. We conducted two sentence-processing experiments in which causer adjuncts were used as an indicator of the surface position of the subject. Experiment 1 measured the processing time of unaccusative sentences with causer adjuncts, and it was revealed that sentences in which the causer adjunct preceded the theme subject were processed significantly faster than those in the other order. This result indicates that the canonical surface position of unaccusative subjects is below the adjunct and that subject movement is optional. In experiment 2, we examined whether the movement of unaccusative subjects is linked to givenness, a discourse-pragmatic factor that has been shown to be associated with scrambling. We found that the processing cost of the subject movement was reduced when the moved subject was marked as discourse given, which is consistent with previous studies on scrambled objects. Overall, our study supports the hypothesis that the unaccusative subject need not move to the derived subject position in Japanese.
The Unaccusative Hypothesis states that intransitive verbs are divided into two types: unergative and unaccusative (Perlmutter 1978, Perlmutter & Postal 1984, Burzio 1986, and Levin & Rappaport Hovav 1995). An unergative verb (e.g., run) takes an agent subject in the same way that a transitive verb does. An unaccusative verb (e.g., fall) takes a theme subject, and, crucially, this subject is considered to be underlyingly an internal argument, like the object of a transitive verb. In the sentence The branch fell, for instance, the NP the branch originates as an object and then moves up to the surface subject position (i.e., spec-TP), yielding the subject–verb word order:
(1)
[TP [NPThe branch]1 [VPfell t1]].
The Unaccusative Hypothesis has been intensively examined across many languages (see the references cited above, among many others), and Japanese is no exception. Theoretical studies of the language suggest that the unergative–unaccusative distinction is real in terms of the original position of subjects, based on a range of diagnostics, for instance, floating numeral quantifiers (Miyagawa 1989), quantifier scope (Nakayama & Koizumi 1991 and Yatsushiro 1999), resultatives (Miyagawa 1989 and Tsujimura 1990), and the scope of the adverb meaning ‘again’ (Asami & Bruening 2025); see Fukuda 2017 for a discussion of what counts as a reliable unaccusative diagnostic.
However, there has been a rigorous debate about where the subject is located on the surface in an unaccusative sentence like the following.1
(2)
Eda-ga
branch-nom
ot-i-ta.
be.fallen-inch-pst
‘The branch fell.’
One view argues that the subject does not have to move to the surface subject position—the no-obligatory-movement hypothesis, schematized in (3a) (Kageyama 1993, Nakayama & Koizumi 1991, Yatsushiro 1999, and Fukuda 2017). Another claims that subject movement is obligatory—the obligatory-movement hypothesis, schematized in (3b) (Kishimoto 2010, Kishimoto 2012, and Kishimoto 2017).2
(3)
a.
No-obligatory-movement hypothesis
[TP [VP theme V] T]
b.
Obligatory-movement hypothesis
[TP theme1 [VP t1 V] T]
The goal of this article is to contribute to this debate from a psycholinguistic perspective by reporting two sentence-processing experiments.
Experiment 1 directly addressed the debate by examining the canonical position of unaccusative subjects relative to a particular type of adjunct, namely, the causer adjunct. This adjunct is headed by the postposition -de ‘due to’ and may appear either before or after the subject:
(4)
(Zisin-de)
earthquake-due.to
kabin-ga
vase-nom
(zisin-de)
earthquake-due.to
war-e-ta.
be.broken-inch-pst
‘A vase broke (due to earthquake).’
Several works claim that causers originate hierarchically above internal arguments and below surface subjects in spec-TP (Fujita 1996, Folli & Harley 2007, Alexiadou et al. 2015, and Asami 2024). In conjunction with this view, the no-obligatory-movement hypothesis predicts that an unaccusative sentence with a causer adjunct, like (4), is canonically ordered causer–theme, whereas the obligatory-movement hypothesis predicts the opposite order to be canonical.3 Experiment 1 tested these predictions by measuring the total reading time required to process a sentence. The results showed that it took longer to process the theme–causer order than the causer–theme order. Under the assumption that non-canonical word order (e.g., OSV in Japanese) is more costly to process than its canonical counterpart (i.e., SOV), we interpret the results as suggesting that the causer–theme order is canonical, which is consistent with the no-obligatory-movement hypothesis.
Experiment 2 investigated whether the movement of unaccusative subjects patterns with scrambling, with special attention to a discourse-pragmatic factor: the givenness of a moved phrase. Previous sentence-processing studies have found that the processing cost for non-canonical word orders is reduced, if not completely alleviated, when a sentence-initial, scrambled phrase is discourse old (Kaiser & Trueswell 2004, Imamura et al. 2016, Koizumi & Imamura 2017, and Jeong et al. 2025). Their findings are consistent with Kuno 1978’s claim that a non-canonical word order is a marked option and that it tends to be used when motivated by a discourse constraint called the from-old-to-new information-flow principle: old information tends to come before new information in a sentence. We hypothesize that if movement of unaccusative subjects is analogous to the scrambling of the object in OSV sentences, then the processing cost for the theme–causer order, detected in experiment 1, should be reduced when the sentence-initial theme phrase is marked as discourse old. This is indeed what we found in experiment 2, further supporting the idea that the theme–causer order is the result of an optional movement of the subject.
The overall results support the hypothesis that posits no obligatory movement of the unaccusative subject. We thus conclude that, in Japanese, the unaccusative subject stays in situ by default and that its optional movement is associated with discourse givenness, in a way parallel to the scrambling of the object in OSV sentences.
Section 2 reports on experiment 1, which provides initial evidence for the hypothesis that unaccusative subjects do not obligatorily move into the derived subject position in Japanese. Section 3 presents experiment 2 and shows that the movement of unaccusative subjects is tied to the status of givenness, offering further evidence for the optionality of this movement. Section 4 offers a general discussion, and section 5 concludes the article.
2 Experiment 12.1 Competing hypotheses and predictions
As introduced in the previous section, recent generative hypotheses about unaccusative subjects in Japanese fall into one of two types, the no-obligatory-movement hypothesis and the obligatory-movement hypothesis:
(5)
a.
No-obligatory-movement hypothesis
[TP [VP theme V] T]
b.
Obligatory-movement hypothesis
[TP theme1 [VP t1 V] T]
= (3)
In a head-final language like Japanese, it is necessary to add a “signpost-like” expression in order to tease these hypotheses apart. For this purpose, a causer adjunct headed by the postposition -de ‘due to’ is ideal.
(6)
(Zisin-de)
earthquake-due.to
kabin-ga
vase-nom
(zisin-de)
earthquake-due.to
war-e-ta.
be.broken-inch-pst
‘A vase broke (due to earthquake).’
= (4)
Several studies claim that causers start out sandwiched between the internal-argument position and the surface-subject position, spec-TP: Fujita 1996, Folli & Harley 2007, Alexiadou et al. 2015, and Asami 2024. For convenience, we follow Asami 2024 and assume that a causer adjunct adjoins to BecomeP.4 In conjunction with this syntactic analysis of the causer adjunct, the two hypotheses in (5) make different predictions about the canonical word order of unaccusatives when the causer adjunct is present. The no-obligatory-movement hypothesis claims that the unaccusative subject stays in situ by default; hence, it predicts that the canonical word order is causer–theme, while the other order (i.e., theme–causer) is non-canonical, derived via an optional movement of the theme over the causer. The structures corresponding to the two word orders under the no-obligatory-movement hypothesis are as follows.5
(7)
a.
b.
In contrast, the obligatory-movement hypothesis would regard (7b) as canonical. Kishimoto 2010, Kishimoto 2012, and Kishimoto 2017 propose that an unaccusative subject moves to spec-TP to satisfy an EPP feature on T. Thus, the theme–causer order is predicted to be canonical, while the causer–theme order is expected to be non-canonical, involving an additional movement of the causer over the theme:
(8)
Crucially, the diverging predictions of the two hypotheses are experimentally testable under the following assumption.
(9)
All other things being equal, sentences with complex representations incur a higher cognitive load than their counterparts with simpler representations (e.g., Pritchett & Whitman 1995, Hawkins 2004, Marantz 2005, Koizumi 2023, and Asami 2024).
A higher cognitive load can be reflected in indices such as reaction/reading times (RTs), eye movements, and physiological changes. Here we focus on RTs, as they are the most relevant to the current study.
The assumption in (9) captures the so-called scrambling effect in sentence processing: scrambling increases processing time in a variety of constructions (see, e.g., Koizumi 2015 for a review). For instance, it takes longer to process OSV sentences, such as (10b), than SOV sentences, such as (10a) (Chujo 1983, Tamaoka et al. 2005, and Koizumi & Imamura 2017).
(10)
a.
SOV
Taroo-ga
Taro-nom
keeki-o
cake-acc
tabe-ta.
eat-pst
‘Taro ate the cake.’
b.
OSV
Keeki-o
cake-acc
Taroo-ga
Taro-nom
tabe-ta.
eat-pst
‘Taro ate the cake.’
Under the generative approach, the SOV word order is canonical, while the OSV word order involves scrambling of the object over the subject (see Nemoto 1999 for an early review and Miyagawa 2001, Miyagawa 2003, Miyagawa 2005, Saito 2006, Saito 2010, and Saito 2011 for later developments):6
(11)
a.
[TP subject [VP object V] T]
b.
[TP object1 [TP subject [VP t1 V] T]]
Here, the OSV order exhibits a more complex representation than the SOV order due to an additional TP node and a filler–gap dependency resulting from scrambling. The assumption in (9) therefore correctly predicts that OSV sentences incur a higher processing load than their SOV counterparts.
Let us now return to the two hypotheses under discussion. Given the assumption in (9), the no-obligatory-movement hypothesis predicts that it takes longer to process the theme–causer order than the causer–theme order, while the obligatory-movement hypothesis predicts the reverse pattern:
(12)
Predictions about RTs (“X < Y” represents “RT for X is shorter than RT for Y”)
We conducted experiment 1 to test these predictions.
2.2 Methods2.2.1 Participants
Data from 96 native speakers of Japanese were analyzed. Participants were recruited via Lancers, a crowdsourcing platform in Japan, and were paid 100 Japanese yen (approximately 0.7 US dollars) for their participation.
To ensure data quality, we applied the following exclusion criteria. Participants were eligible if they were native speakers of Japanese and had an approval rating of 95% or higher on Lancers. All participants were required to achieve more than 80% accuracy and have fewer than 10% excluded data points (i.e., RTs shorter than 400 ms or longer than 5000 ms) across all items. In total, we recruited 100 participants, and we excluded four of them from our data analysis, leaving 96 qualifying participants.
2.2.2 Materials and design
The experiment employed a within-subjects design and manipulated word order with two levels: causer–theme and theme–causer. Example items for each condition are provided in (13). We used inanimate nouns for both the nominative phrase and the causer adjunct.7
(13)
a.
Causer–theme
Kazi-de
fire-due.to
ie-ga
house-nom
yak-e-ta.
be.burned.down-inch-pst
‘A house burned down due to fire.’
b.
Theme–causer
Ie-ga
house-nom
kazi-de
fire-due.to
yak-e-ta.
be.burned.down-inch-pst
‘A house burned down due to fire.’
In all, 24 pairs of experimental sentences like those in (13) were created (a total of 48 sentences), using 24 distinct unaccusative verbs:
All of the verbs have lexical-causative counterparts. We take their compatibility with a causer adjunct to indicate that their sole NP argument is an internal argument, since causer adjuncts may not co-occur with an external argument (Schäfer 2025).
Following a Latin-square design, the experimental sentences were divided into two lists so that no participant saw more than one item from the same set of pairs and each participant saw an equal number of causer–theme and theme–causer experimental items. Each list was combined with 72 fillers, yielding two lists of 96 sentences each. The fillers had various syntactic structures. Half of them were semantically and/or (morpho)syntactically anomalous. Each list was presented to 48 participants (96 participants in total).
2.2.3 Procedure
A forced-choice judgment task (Chujo 1983, Tamaoka et al. 2005, Koizumi & Tamaoka 2010, and Asami 2024) was conducted using PCIbex (Zehr & Schwarz 2018). In the experiment, participants were instructed to decide whether a visually presented sentence was a possible or impossible sentence in Japanese. “Possible” and “impossible” here correspond to “acceptable” and “unacceptable.”
Each trial began with a fixation element, “++++++,” displayed at the center of the screen for 1200 ms. A sentence then appeared in full. Participants were instructed to respond as quickly and accurately as possible, deciding whether the sentence was possible in Japanese by pressing “J” on the keyboard for “possible” or “F” for “impossible.” RT and accuracy were recorded for each trial. Participants completed eight practice trials before the actual experiment. The experiment took less than 5 min.
Note that we did not use the self-paced-reading method, despite its popularity in sentence-processing studies, because previous research has rarely detected an effect of word order with this method (see, e.g., Tamaoka & Koizumi 2006 for a relevant discussion; see also Asami & Tomioka 2025 for a follow-up to the present article that circumvented this issue).
2.2.4 Data analysis
We analyzed both accuracy and RTs. For the accuracy analysis, we excluded responses that were too fast (400 ms or less) or too slow (5000 ms or more), which affected 0.6% of the data. For the RT analysis, we further removed incorrect responses, affecting 1.2% of the data. Finally, RTs outside 2.5 standard deviations from individual participants’ means (both high and low ranges) were replaced by boundary values at 2.5 standard deviations, which affected 2.4% of the data. Mean RTs and accuracy by condition are reported in table 1.8
Table 1: RTs and accuracy for each condition.
RT (ms)
Accuracy (%)
Condition
Mean
Standard deviation
Mean
Standard deviation
Causer–theme
1503
563
99.3
8.3
Theme–causer
1663
638
98.2
13.1
For the statistical analysis of accuracy, we fitted logistic mixed-effects models (Jaeger 2008) using the Lme4 package (Bates et al. 2015b) in R (R Core Team 2024). The fixed factor was word order (causer–theme vs. theme–causer), and the random factors were items and participants. Following Bates et al. 2015a, we selected the most parsimonious model. We initially fitted the maximal random-effects model, including by-participant and by-item intercepts and slopes. We then simplified the model by removing random slopes that did not improve model fit. The final model included random intercepts for both participants and items.
For the statistical analysis of RTs, we fitted linear mixed-effects models (Baayen et al. 2008), following the same model-fitting and selection procedures as in the statistical analysis of accuracy. p values were obtained with the LmerTest package (Kuznetsova et al. 2017). The final model included by-participant and by-item intercepts, as well as by-participant and by-item slopes.
2.3 Results2.3.1 Accuracy
A summary of the model is provided in table 2. We found a significant difference between the causer–theme and theme–causer conditions: the former was judged more accurately than the latter.
Table 2: The estimates of the fixed effect in the logistic mixed-effects model of accuracy in experiment 1.
Estimate
Standard error
z value
p value
(Intercept)
5.77
0.56
10.15
<0.001
***
Word order
–0.95
0.42
–2.23
0.025
*
2.3.2 RTs
A summary of the model is provided in table 3. We found a significant main effect of word order such that RTs for the theme–causer condition were longer than those for the causer–theme condition.
Table 3: The estimates of the fixed effect in the linear mixed-effects model of RTs in experiment 1.
Estimate
Standard error
t value
p value
(Intercept)
1505.3
53.07
28.36
<0.001
***
Word order
163.4
28.95
5.64
<0.001
***
2.4 Discussion
The results showed that it took longer to process the theme–causer order than the causer–theme order. This is consistent with the prediction of the no-obligatory-movement hypothesis: the unaccusative subject stays in situ and therefore canonically follows the causer adjunct. In contrast, the results contradict the prediction of the obligatory-movement hypothesis, which holds that the unaccusative subject obligatorily moves to a derived subject position and thus canonically precedes the causer adjunct. Let us discuss two implications of the current findings.
The first implication concerns the EPP feature of T, the driving force behind NP movement into spec-TP. The present results support the view that the existence of this feature is not conditioned by nominative-Case assignment by the T head (Hiraiwa 2001 and Hiraiwa 2005), contra Kishimoto 2010, Kishimoto 2012, and Kishimoto 2017. The findings also suggest that the T head lacks the EPP feature when an unaccusative verb is used in a clause, as if the presence of the EPP feature on T is conditioned by the lexical verb (see Miyagawa & Babyonyshev 2004 and Fukuda 2017 for related discussions). Since previous studies such as Miyagawa 2017 assume that the EPP feature originates in the C head and is transmitted to T via feature inheritance, future research must explore how the C and V heads interact to determine whether T bears the EPP feature. Note that Saito 2011 argues that the C head retains the EPP feature and does not transmit it to the T head. The foregoing discussion is also relevant to this view: it must explain how the unaccusative verb conditions the presence of the EPP feature on C.
Another implication of the current results concerns the base position of causers. In our experiment, we adopted the view that a causer originates above an internal argument, that is, the high-causer hypothesis; we did not consider its alternative, the low-causer hypothesis, which posits that the causer originates hierarchically below the theme (on the latter hypothesis see Belletti & Rizzi 1988, Hasegawa 2001, Hasegawa 2004, Hasegawa 2007, and Cheung & Larson 2015). If the low-causer hypothesis were correct, the basic word order of unaccusatives with causer adjuncts would be theme–causer rather than causer–theme, regardless of whether movement of the theme takes place. As a consequence, it should take longer to process the causer–theme order than the theme–causer order. However, the current results showed the opposite pattern, which is incompatible with the low-causer hypothesis (see also Asami 2024 for another piece of experimental evidence that is incompatible with this hypothesis).
3 Experiment 2
The results of experiment 1 suggest that the theme–causer order of unaccusatives is derived via an optional movement of the theme over the causer. If this movement is indeed optional, does it exhibit the same profile as scrambling, a type of optional movement? Experiment 2 was conducted to address this question, focusing on the status of givenness, which has been shown to be associated with scrambling.
3.1 Optional movement, discourse principles, and predictions
Previous psycholinguistic studies have shown that the processing cost associated with scrambling is alleviated when the scrambled phrase is previously mentioned or discourse given (e.g., Kaiser & Trueswell 2004, Imamura et al. 2016, Koizumi & Imamura 2017, Yano & Koizumi 2018, Yano & Koizumi 2021, and Jeong et al. 2025). For instance, RTs for OSV sentences like (16) are reduced when the scrambled object Kaneda has been mentioned previously compared to when it has not, in a context like (15) (Imamura et al. 2016 and Koizumi & Imamura 2017).
(15)
Gaimusyoo-no
Ministry.of.Foreign.Affairs-gen
zikan-wa
vice.minister-top
Kaneda-da
Kaneda-cop
/
Kuroki-da.
Kuroki-cop
‘It is Kaneda/Kuroki who is the vice minister of the Ministry of Foreign Affairs.’
(16)
Kaneda-o
Kaneda-acc
Kuroki-ga
Kuroki-nom
mukae-ta.
welcome-pst
‘Kuroki welcomed Kaneda.’
These findings are consistent with Kuno 1978’s claim that old information tends to precede new information. This is stated as the from-old-to-new information-flow principle:
(17)
From-old-to-new information-flow principle
In principle, words in a sentence are arranged in such a way that those that represent old, predictable information come first, and those that represent new, unpredictable information come last.
(Kuno 1978: 54, (12))9
Under this principle, the aforementioned reduction in processing cost for non-canonical OSV sentences can be said to occur because the optional movement of the object is more likely to take place when the object represents older information than the subject.
Importantly, the canonical SOV order exhibits a weaker reduction in processing cost than the non-canonical OSV order when following the from-old-to-new information-flow principle (Imamura et al. 2016 and Koizumi & Imamura 2017). This suggests that the status of discourse givenness has less impact on the processing of SOV sentences. This finding follows from the generalization that a marked pattern tends to be used in special contexts whereas an unmarked pattern is less contextually restricted (Kuno 1978, Kuno 1987, Aissen 1992, and Birner & Ward 2009). This tendency is captured by the markedness principle for discourse-rule violations (Kuno 1978 and Kuno 1987):
(18)
Markedness principle for discourse-rule violations
Sentences that involve marked (or intentional) violations of discourse principles are unacceptable. On the other hand, sentences that involve unmarked (or unintentional) violations of discourse principles go unpenalized and are acceptable.
(Kuno 1987: 212, (3.12))10
According to this principle, the canonical SOV order is an unmarked option; hence, it does not need to observe discourse principles like (17). This explains why previous sentence-processing studies have found only a small effect of givenness on this word order.
Based on this background, we make the following predictions. If the conclusion drawn from experiment 1 is correct, then (i) according to the from-old-to-new information-flow principle, in the non-canonical theme–causer order, the sentence-initial phrase should represent older information than the second phrase; (ii) on the other hand, according to the markedness principle for discourse-rule violations, this requirement should be weaker in the canonical causer–theme order. In a sentence-processing experiment, this predicted pattern could appear as an interaction between word order and position of given information: RTs in the non-canonical theme–causer order should be reduced when the first phrase conveys older information than the second, compared to when it does not; in contrast, such RT reduction should be smaller in the canonical causer–theme order.
3.2 Methods3.2.1 Participants
Our analysis included data from 192 native speakers of Japanese. Participant-recruitment and data-exclusion procedures were identical to those used in experiment 1. A total of 200 participants were recruited, and eight were excluded from the data analysis.
3.2.2 Materials and design
The experiment employed a 2 × 2 within-subjects design, crossing word order (theme–causer vs. causer–theme) with givenness position (given first vs. given second). Examples for each condition are provided in (19). We used the demonstrative sono ‘that’ to mark givenness, following Yano 2019. When a reader encounters a phrase with sono ‘that,’ they can infer that it represents discourse-given information. In contrast, a bare noun phrase is interpreted as discourse new in the absence of any context.
(19)
a.
Causergiven–theme
Sono-kazi-de
that-fire-due.to
ie-ga
house-nom
yak-e-ta.
be.burned.down-inch-pst
‘A house burned down due to that fire.’
b.
Causer–themegiven
Kazi-de
fire-due.to
sono-ie-ga
that-house-nom
yak-e-ta.
be.burned.down-inch-pst
‘That house burned down due to fire.’
c.
Themegiven–causer
Sono-ie-ga
that-house-nom
kazi-de
fire-due.to
yak-e-ta.
be.burned.down-inch-pst
‘That house burned down due to fire.’
d.
Theme–causergiven
Ie-ga
house-nom
sono-kazi-de
that-fire-due.to
yak-e-ta.
be.burned.down-inch-pst
‘A house burned down due to that fire.’
To create the items exemplified in (19), we modified the experimental sentences used in experiment 1 by adding the demonstrative sono ‘that’ to either the theme or the causer phrase. As a result, 24 sets of sentences were created, yielding a total of 96 sentences.
Following a Latin-square design, the experimental items were divided into four lists. Each list was then combined with 72 fillers, yielding four lists of 96 sentences each. Each list was assigned to 48 participants (192 participants in total).
3.2.3 Procedure
The procedure was identical to that in experiment 1.
3.2.4 Data analysis
We followed the same data-pre-processing procedures as in experiment 1. For the accuracy analysis, we excluded responses shorter than 400 ms or longer than 5000 ms, which affected 0.6% of the data. For the RT analysis, we also removed incorrect responses, affecting 1.9% of the data. Finally, RTs falling outside 2.5 standard deviations from individual participants’ means (both high and low ranges) were replaced with the corresponding boundary values, affecting 2.2% of the data. Mean RT and accuracy by condition are reported in table 4.11
Table 4: RTs and accuracy for each condition in experiment 2.
RT (ms)
Accuracy (%)
Condition
Mean
Standard deviation
Mean
Standard deviation
Causergiven–theme
1623
574
98.9
10.1
Causer–themegiven
1622
591
98.5
12.0
Themegiven–causer
1743
639
97.9
14.0
Theme–causergiven
1840
720
96.7
17.6
For the statistical analysis of accuracy, we fitted logistic mixed-effects models (Jaeger 2008) using the Lme4 package (Bates et al. 2015b) in R (R Core Team 2024). Fixed factors were the sum-coded (0.5/–0.5) main effects of word order (causer–theme vs. theme–causer) and givenness position (given first vs. given second), as well as their interaction. Random effects included participants and items.
Following Bates et al. 2015a, we selected the most parsimonious model. The model initially included the maximal random-effect structure allowing convergence. We simplified the model by removing random slopes and interactions that did not improve the model. The final model included random intercepts for participants and items.
For the statistical analysis of RTs, we fitted linear mixed-effects models (Baayen et al. 2008), following the same model fitting and selection procedure used in the accuracy analysis. p values were obtained, using the LmerTest package (Kuznetsova et al. 2017). The final model included random intercepts for participants and items.
To analyze the significant interactions, if any, we designed a follow-up analysis using nested contrasts, examining the effect of givenness position at each level of word order. This follow-up analysis involved fitting the same model used in the main analysis but replacing the main effect of givenness position and the interaction between word order and givenness position with two sum-coded nested contrasts: one comparing the given-first and given-second conditions within the causer–theme order and the other comparing the same conditions within the theme–causer order.
3.3 Results3.3.1 Accuracy
Table 5 shows the model summary for the accuracy analysis. There was a significant main effect of word order: causer–theme order sentences were judged more accurately than theme–causer order sentences. Neither the main effect of givenness position nor the interaction between word order and givenness position was significant.
Table 5: The estimates of the fixed effects and their interaction in the logistic mixed-effects model of accuracy in experiment 2.
Estimate
Standard error
z value
p value
(Intercept)
4.51
0.23
18.89
<0.001
***
Word order
0.75
0.23
3.18
0.001
**
Givenness position
0.43
0.23
1.84
0.064
Word order × givenness position
–0.13
0.47
–0.29
0.771
3.3.2 RTs
A bar graph of the RT data is provided in figure 1, and the model summary is presented in table 6. We found significant main effects of word order and givenness position. Specifically, sentences with causer–theme order were processed faster than those with theme–causer order, and given-first sentences were processed faster than given-second sentences. We also found a significant interaction between word order and givenness position.
Figure 1: RTs by condition. Error bars indicate ±1 standard error of mean RTs.
Table 6: The estimates of the fixed effects, their interaction, and effects of givenness position at each level of word order in the linear mixed-effects model of RTs.
Estimate
Standard error
t value
p value
(Intercept)
1710.7
44.47
38.46
<0.001
***
Word order
–171.9
13.89
–12.37
<0.001
***
Givenness position
–48.1
13.89
–12.37
<0.001
***
Word order × givenness position
96.2
27.77
3.465
<0.001
***
Givenness position: causer–theme
<–0.1
19.55
<–0.01
0.997
Givenness position: theme–causer
–96.2
19.72
–4.88
<0.001
***
In the follow-up analysis, we examined the effect of givenness position within each level of word order. This analysis revealed a significant givenness-position effect in the theme–causer order: themegiven–causer sentences were processed faster than theme–causergiven sentences. In contrast, no significant effect of givenness position was observed in the causer–theme order.
3.4 Discussion
We found a significant interaction between word order and givenness position: there was a significant difference between the themegiven–causer and theme–causergiven conditions but not between the causergiven–theme and causer–themegiven conditions. Specifically, the theme–causer order was processed more quickly when the first phrase was marked as given than when the second phrase was. We found no evidence for such a difference in the causer–theme order. The current findings suggest that the status of givenness impacts the processing of the non-canonical theme–causer order but not the canonical causer–theme order. This confirms our predictions: in the non-canonical theme–causer order, there is a preference for the first phrase to represent older information than the second phrase, whereas in the canonical causer–theme order, this preference is weaker.
The results also showed that the causer–theme order was processed significantly faster than the theme–causer order. This finding replicates the result of experiment 1. More importantly, a closer look at the data indicates that the processing time for themegiven–causer sentences (1743 ms) was shorter than for theme–causergiven sentences (1840 ms) but still longer than for causergiven–theme (1623 ms) and causer–themegiven sentences (1622 ms). This suggests that information structure alleviates but does not eliminate the processing cost associated with non-canonical word order. This pattern is consistent with the findings of Koizumi & Imamura 2017. In that study, the processing time for non-canonical OSV sentences was reduced when the sentence-initial object was discourse given compared to when it was not; however, OSV sentences remained slower to process than canonical SOV sentences even when the sentence-initial phrase was previously mentioned. Therefore, we conclude that subject movement in unaccusatives in Japanese is truly optional and that this optionality renders such movement marked and computationally costly to process (cf. Yano & Koizumi 2018 and Yano & Koizumi 2021).
Additionally, we found a main effect of givenness position: sentences were processed more quickly when the first phrase was marked as given than when the second phrase was. At first glance, this might suggest that given information canonically occurs closer to the sentence-initial position. However, this main effect stems from the contrast between themegiven–causer and theme–causergiven conditions, not from the contrast between causergiven–theme and causer–themegiven, as indicated by the interaction discussed above. This clarification is important because, in the current design, phrases marked as given are slightly longer than bare phrases due to the addition of the demonstrative sono ‘that.’ Previous studies have found that longer phrases tend to precede shorter ones in Japanese, a tendency known as the long-before-short preference (e.g., Yamashita 2002, Kondo & Yamashita 2011, Imamura 2015, and Imamura 2019). One might argue that this preference explains the main effect of givenness position. However, our results showed a significant interaction: the effect of givenness was present in the theme–causer order but absent in the causer–theme order. We therefore conclude, considering that the addition of sono ‘that’ increases phrase length by only two morae, that it is unlikely that such a minimal increase is sufficient to trigger the long-before-short preference.
Overall, the current results suggest that the movement of unaccusative subjects is optional, much like the scrambling of objects.
4 General discussion
We have presented two experiments on the processing of unaccusative constructions involving causer adjuncts in Japanese. These experiments were conducted to investigate whether the movement of an unaccusative subject into a derived subject position is obligatory in this language. The overall results are consistent with the hypothesis that the movement in question is optional. In what follows, we first summarize the results and then discuss questions they raise for future research.
Experiment 1 tested whether unaccusative subjects remain in their base-generated position by default (i.e., the no-obligatory-movement hypothesis; Nakayama & Koizumi 1991, Kageyama 1993, Yatsushiro 1999, and Fukuda 2017) or whether they always move to a derived subject position (i.e., the obligatory-movement hypothesis; Kishimoto 2010, Kishimoto 2012, and Kishimoto 2017). If we adopt the view that the causer originates above the theme position and below the surface-subject position (Fujita 1996, Folli & Harley 2007, Alexiadou et al. 2015, and Asami 2024), the no-obligatory-movement hypothesis predicts that the causer–theme order is canonical and the theme–causer order is non-canonical for unaccusative sentences with a causer adjunct, while the obligatory-movement hypothesis predicts the opposite. The results of experiment 1 were that the causer–theme order was processed more quickly than the theme–causer order, suggesting that the unaccusative subject can remain below the causer adjunct. This finding is consistent with the prediction of the no-obligatory-movement hypothesis but not with the prediction of the obligatory-movement hypothesis.
Experiment 2 examined whether the optional movement of unaccusative subjects resulting in the non-canonical theme–causer order is influenced by givenness, in a way parallel to scrambling in OSV sentences. We formulated predictions based on the combination of two discourse principles: the from-old-to-new information-flow principle and the markedness principle for discourse-rule violations (Kuno 1978 and Kuno 1987). According to the former, old information tends to appear earlier in a sentence than new information. The latter states that violations of discourse principles are penalized only when they involve marked structures. Thus, a violation of the information-flow principle should incur a processing cost when it occurs in the marked, non-canonical theme–causer order but not (or not as severely) in the unmarked, canonical causer–theme order. The results of experiment 2 revealed an interaction between word order and givenness position: the theme–causer order was processed more quickly when the first phrase was marked as given than when it was not, whereas no evidence for such a difference was found in the causer–theme order. This pattern confirms our prediction and indicates that the optional movement involved in generating the theme–causer order functions analogously to scrambling in OSV sentences, which is also sensitive to the information-flow principle (Imamura et al. 2016, Koizumi & Imamura 2017, Yano & Koizumi 2018, Yano & Koizumi 2021, and Jeong et al. 2025).
Since the present study focuses solely on the processing of unaccusatives, without considering unergatives, it remains possible that similar results might be obtained with unergative sentences.12 One way to test this possibility, as suggested by a reviewer, would be to use causer adjuncts in unergative sentences and examine how they are processed. However, such adjuncts are generally disallowed with unergative verbs. Consider the sentence in (20), where the unergative verb hasiru ‘run’ is used with the causer adjunct tunami-de ‘due to tsunami.’ Although the intended reading, that a girl ran because a tsunami occurred, is plausible, the sentence is unnatural.
(20)
*Syoojo-ga
girl-nom
tunami-de
tsunami-due.to
hasit-ta.
run-pst
Intended: ‘A girl ran due to tsunami.’
An apparent counterexample to our claim is a sentence like the following.13
(21)
Syoojo-ga
girl-nom
zyoodan-de
joke-due.to
warat-ta.
laugh-pst
‘A girl laughed due to a joke.’
However, it is unclear whether this sentence truly represents an unergative sentence. This is because its most salient interpretation is one in which the subject functions as a theme undergoing a laughing event caused by a joke, rather than as an agent initiating the event. Under this reading, the verb warau ‘laugh’ appears to be used in an unaccusative frame.14 In addition, if we force warau ‘laugh’ to be used as an unergative verb, then the sentence most naturally conveys that the girl laughed jokingly—compare the following.
(22)
Hanako-ga
Hanako-nom
zyoodan-de
joke-de
iyana
bad
koto-o
thing-acc
it-ta.
say-pst
‘Hanako said a bad thing as a joke.’
This type of adverb remains under-studied and lacks a theoretical foundation from which one can derive psycholinguistically testable predictions about subject position.
Taking the above observations together, we contend that it is difficult to test processing differences between unaccusatives and unergatives using causer adjuncts; addressing this issue will require identifying another type of adjunct, which must be explored in future research. Accordingly, we do not intend to make any claims about unergatives based on the current results. Note that not testing unergatives does not undermine the findings reported here because a wide range of linguistic evidence has already supported the unergative–unaccusative distinction in Japanese, including floating numeral quantifiers (Miyagawa 1989), quantifier scope (Nakayama & Koizumi 1991 and Yatsushiro 1999), resultatives (Miyagawa 1989 and Tsujimura 1990), and scope of the adverb meaning ‘again’ (Asami & Bruening 2025).
Before concluding, we would like to highlight two additional questions that the current study raises for future research. First, although we have assumed that the unaccusative subject in the causer–theme order can remain in situ (i.e., in the complement position of the verb), the present results are also compatible with the hypothesis that the subject undergoes short-distance movement to a position below the causer adjunct. Depending on the theory of nominative-Case licensing (e.g., Hiraiwa 2001, Hiraiwa 2005, Koizumi 2008, and Saito 2012), such a movement may be theoretically motivated. However, our current experimental design is not sufficiently fine-tuned to determine the exact position of the subject below the causer adjunct.
Second, equally undetermined is the landing site of the optional movement of the theme subject. In the tree structure in (7b), repeated as (23), we assumed that the moved subject occupies spec-TP for expository purposes.
(23)
However, the question remains whether spec-TP can host an optionally dislocated phrase or whether some position in the CP domain is instead used for such a phrase.
5 Conclusion
This article has presented two sentence-processing experiments that tested whether the subject of an unaccusative verb in Japanese obligatorily moves to a derived subject position. The results of experiments 1 and 2 suggest that this movement is not obligatory and behaves like scrambling in its sensitivity to the status of givenness. We therefore conclude that the no-obligatory-movement hypothesis is supported while the obligatory-movement hypothesis is not. As mentioned in the previous section, we do not intend to determine precisely where an unaccusative subject is placed canonically or which position it moves to when it appears sentence initially. Our results nevertheless allow us to narrow down the domain for an unaccusative subject (i.e., the VP region) and to identify the movement of this subject as an optional movement.
Data availability
Data and code used in the data analyses are available at https://osf.io/2cqkv/.
Acknowledgments
Part of the material in this article was presented at the University of Delaware and at the fifth International Conference on Theoretical East Asian Psycholinguistics. We would like to thank the audiences for their helpful comments and suggestions. We would also like to thank Edson T. Miyamoto and three anonymous reviewers of STAR for their constructive feedback and insightful comments, which have significantly contributed to the improvement of this article. Special thanks to Amanda Payne for editorial assistance. Of course, all remaining errors are our own. This research was supported by funding from the University of Delaware (awarded to the first author).
Ethics and consent
Approval for the experiments reported in this article was obtained from the institutional review board of the University of Delaware (ID number: 2189326-1). All participants in the experiments gave informed consent before data collection.
Competing interests
The authors declare that they have no competing interests.
Interlinear glossing in this article follows the Leipzig Glossing Rules, with the addition of inch = inchoative.
The current study does not consider the view that unaccusative subjects originate as external arguments, because the wide range of diagnostics cited in the main text speaks against it. Kishimoto 1996 represents such a view, but we assume that Kishimoto no longer holds it, as his more recent studies (cited in the main text) acknowledge the syntactic distinction between unergatives and unaccusatives in terms of the base position of their subjects.
Throughout this study, we adopt the default assumption in Japanese syntax that if a phrase XP originates hierarchically above another phrase YP, then XP linearly precedes YP.
Some studies do not assume BecomeP in their analyses of argument structure (e.g., Ramchand 2008). The exact analysis does not matter here, so long as it is compatible with the assumption that a causer adjunct originates hierarchically above an internal argument and below a surface subject.
We assume that the landing site of the NP movement is spec-TP purely for expository purposes, without any theoretical commitment.
We simplify the VP-internal structure for expository purposes.
We decided not to use animate nouns for subjects because evidence from subject honorification and subject-oriented-anaphor binding suggests that animate unaccusative subjects raise, at least to spec-vP/VoiceP, if nothing else occupies that position (Saito 2006 and Takano 2011). We leave it to future research to determine whether the results reported in this article are generalizable to unaccusative sentences with animate subjects.
Filler items had 94.7% accuracy. This indicates that the participants engaged attentively.
The English translation is adopted from Imamura et al. 2016: 9, (12).
Kuno uses the term (un)acceptable here in a broader sense than is standard in generative syntax. Rather than referring strictly to acceptability, the term encompasses discourse felicity (i.e., whether a sentence conforms to pragmatic and information-structural norms).
Filler items had 98.1% accuracy, indicating that participants paid attention during the experiment.
We thank an anonymous reviewer for stimulating the following discussion on unergatives.
We attribute this example to the reviewer.
This point is not unprecedented: previous studies suggest that it is not always straightforward to determine whether a particular intransitive verb is unergative, unaccusative, or ambiguous between the two, especially when it takes an animate subject (see Tsujimura 1994, Hasegawa 2001, and Asami 2025 for relevant discussions).
Aissen, Judith L.1992. Topic and focus in Mayan. Language68.1.43–80. 10.1353/lan.1992.0017.Alexiadou, Artemis & Anagnostopoulou, Elena & Schäfer, Florian. 2015. External arguments in transitivity alternations: a layering approach. Oxford University Press.Asami, Daiki. 2024. Deriving and processing experiencer subject causatives. Glossa9.1. 10.16995/glossa.16435.Asami, Daiki. 2025. Passive head only selects for agentive Voice in Japanese: a reply to Jo and Seo (2023). Journal of East Asian Linguistics34.2.205–240. 10.1007/s10831-025-09294-4.Asami, Daiki & Bruening, Benjamin. 2025. Subjectless readings of again: a response to Bale (2007) and Smith and Yu (2021). Natural Language and Linguistic Theory43.3.1813–1837. 10.1007/s11049-024-09652-2.Asami, Daiki & Tomioka, Satoshi. 2025. Online processing of subject-initial non-canonical sentences: interaction of syntax with information structure. Glossa Psycholinguistics4.1. 10.5070/G6011.42237.Baayen, R. Harald & Davidson, Douglas J. & Bates, Douglas M.2008. Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language59.4.390–412. 10.1016/j.jml.2007.12.005.Bates, Douglas & Kliegl, Reinhold & Vasishth, Shravan & Baayen, R. Harald. 2015a. Parsimonious mixed models. Unpublished paper. Work conducted at: University of Wisconsin–Madison, University of Potsdam, University of Tübingen, and University of Alberta. 10.48550/arXiv.1506.04967.Bates, Douglas & Mächler, Martin & Bolker, Ben & Walker, Steve. 2015b. Fitting linear mixed-effects models using lme4. Journal of Statistical Software67.1. 10.18637/jss.v067.i01.Belletti, Adriana & Rizzi, Luigi. 1988. Psych-verbs and θ-theory. Natural Language and Linguistic Theory6.3.291–352. 10.1007/BF00133902.Birner, Betty J. & Ward, Gregory. 2009. Information structure and syntactic structure. Language and Linguistics Compass3.4.1167–1187. 10.1111/j.1749-818X.2009.00146.x.Burzio, Luigi. 1986. Italian syntax: a government-binding approach. D. Reidel Pubishing Company.Cheung, Candice Chi-Hang & Larson, Richard K.2015. Psych verbs in English and Mandarin. Natural Language and Linguistic Theory33.1.127–189. 10.1007/s11049-014-9259-3.Chujo, Kazumitsu. 1983. Nihongo tanbun-no rikai katei: bunrikai sutoratejii no sougo kankei [Understanding Japanese dialects: the inter-relationships among strategies for sentence comprehension]. Japanese Journal of Psychology54.4.250–256.Folli, Raffaella & Harley, Heidi. 2007. Causation, obligation, and argument structure: on the nature of little v. Linguistic Inquiry38.2.197–238. 10.1162/ling.2007.38.2.197.Fujita, Koji. 1996. Double objects, causatives, and derivational economy. Linguistic Inquiry27.1.146–173.Fukuda, Shin. 2017. Split intransitivity in Japanese is syntactic: evidence for the Unaccusative Hypothesis from sentence acceptability and truth value judgment experiments. Glossa2.1.83. 10.5334/gjgl.268.Hasegawa, Nobuko. 2001. Causatives and the role of v: Agent, Causer, and Experiencer. In: Hasegawa, Nobuko & Inoue, Kazuko (editors). Linguistics and interdisciplinary research: proceedings of the COE international symposium. Kanda University of International Studies. 1–35.Hasegawa, Nobuko. 2004. “Unaccusative transitives” and Burzio’s generalization: reflexive constructions in Japanese. In: Csirmaz, Aniko & Lee, Youngjoo & Walter, Mary Ann (editors). Proceedings of the 1st workshop on Altaic formal linguistics. MIT Working Papers in Linguistics. 300–314.Hasegawa, Nobuko. 2007. The possessor raising construction and the interpretation of the subject. In: Karimi, Simin & Samiian, Vida & Wilkins, Wendy K. (editors). Phrasal and clausal architecture: syntactic derivation and interpretation. John Benjamins Publishing Company. 66–92.Hawkins, John A.2004. Efficiency and complexity in grammars. Oxford University Press.Hiraiwa, Ken. 2001. Multiple Agree and the Defective Intervention Constraint in Japanese. In: Hiraiwa, Ken & Matushansky, Ora & Costa, Albert & Martín-González, Javier & Nathan, Lance & Szczegielniak, Adam (editors). Proceedings of the 1st HUMIT student conference in language research (HUMIT 2000). MIT Working Papers in Linguistics. 67–80.Hiraiwa, Ken. 2005. Dimensions of symmetry in syntax: agreement and clausal architecture. Doctoral thesis. Massachusetts Institute of Technology.Imamura, Satoshi. 2015. The effects of givenness and heaviness on VP-internal scrambling and VP-external scrambling in Japanese. Studies in Pragmatics17.1–16.Imamura, Satoshi. 2019. Word order, heaviness, and animacy. Corpus Pragmatics3.2.123–143. 10.1007/s41701-018-00049-2.Imamura, Satoshi & Sato, Yohei & Koizumi, Masatoshi. 2016. The processing cost of scrambling and topicalization in Japanese. Frontiers in Psychology7.531. /10.3389/fpsyg.2016.00531.Jaeger, T. Florian. 2008. Categorical data analysis: away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language59.4.434–446. 10.1016/j.jml.2007.11.007.Jeong, Hyeonjeong & Kim, Jungho & Yano, Masataka & Cui, Haining & Kiayama, Sachiko & Koizumi, Masatoshi. 2025. The crucial role of the left inferior frontal gyrus (BA44) in synergizing syntactic structure and information structure during sentence comprehension. Brain and Language262.105533. 10.1016/j.bandl.2025.105533.Kageyama, Taro. 1993. Bunpoo to gokeisei [Grammar and word formation]. Hituzi Syobo Publishing.Kaiser, Elsi & Trueswell, John C.2004. The role of discourse context in the processing of a flexible word-order language. Cognition94.2.113–147. 10.1016/j.cognition.2004.01.002.Kishimoto, Hideki. 1996. Split intransitivity in Japanese and the Unaccusative Hypothesis. Language72.2.248–286.Kishimoto, Hideki. 2010. Subjects and constituent structure in Japanese. Linguistics48.3.629–670. 10.1515/ling.2010.020.Kishimoto, Hideki. 2012. Subject honorification and the position of subjects in Japanese. Journal of East Asian Linguistics21.1.1–41. 10.1007/s10831-011-9083-2.Kishimoto, Hideki. 2017. Negative polarity, A-movement, and clause architecture in Japanese. Journal of East Asian Linguistics26.2.109–161. 10.1007/s10831-016-9153-6.Koizumi, Masatoshi. 2008. Nominative objects. In: Miyagawa, Shigeru & Saito, Mamoru (editors). The Oxford handbook of Japanese linguistics. Oxford University Press. 141–164.Koizumi, Masatoshi. 2015. Experimental syntax: word order in sentence processing. In: Nakayama, Mineharu (editor). Handbook of Japanese psycholinguistics. De Gruyter. 387–421.Koizumi, Masatoshi. 2023. Constituent order in language and thought: a case study in field-based psycholinguistics. Cambridge University Press.Koizumi, Masatoshi & Imamura, Satoshi. 2017. Interaction between syntactic structure and information structure in the processing of a head-final language. Journal of Psycholinguistic Research46.1.247–260. 10.1007/s10936-016-9433-3.Koizumi, Masatoshi & Tamaoka, Katsuo. 2010. Psycholinguistic evidence for the VP-internal subject position in Japanese. Linguistic Inquiry41.4.663–680. 10.1162/LING_a_00016.Kondo, Tadahisa & Yamashita, Hiroko. 2011. Why speakers produce scrambled sentences: an analysis of a spoken language corpus in Japanese. In: Yamashita, Hiroko & Hirose, Yuki & Packard, Jerome L. (editors). Processing and producing head-final structures. Springer. 195–215. 10.1007/978-90-481-9213-7_10.Kuno, Susumu. 1978. Danwa-no bunpoo [Grammar of discourse]. Taishūkan Shoten.Kuno, Susumu. 1987. Functional syntax: anaphora, discourse and empathy. University of Chicago Press.Kuznetsova, Alexandra & Brockhoff, Per B. & Christensen, Rune H. B.2017. lmerTest package: tests in linear mixed effects models. Journal of Statistical Software82.13. 10.18637/jss.v082.i13.Levin, Beth & Rappaport Hovav, Malka. 1995. Unaccusativity: at the syntax-lexical semantics interface. MIT Press.Marantz, Alec. 2005. Generative linguistics within the cognitive neuroscience of language. Linguistic Review22.2–4.429–445. 10.1515/tlir.2005.22.2-4.429.Miyagawa, Shigeru. 1989. Structure and case marking in Japanese. Academic Press.Miyagawa, Shigeru. 2001. The EPP, scrambling, and wh-in-situ. In: Kenstowicz, Michael (editor). Ken Hale: a life in language. MIT Press. 293–338.Miyagawa, Shigeru. 2003. A-movement scrambling and options without optionality. In: Karimi, Simin (editor). Word order and scrambling. Blackwell Publishing. 177–200.Miyagawa, Shigeru. 2005. EPP and semantically vacuous scrambling. In: Sabel, Joachim & Saito, Mamoru (editors). The free word order phenomenon. Mouton de Gruyter. 181–220.Miyagawa, Shigeru. 2017. Agreement beyond phi. MIT Press.Miyagawa, Shigeru & Babyonyshev, Maria. 2004. The EPP, unaccusativity, and the resultative constructions in Japanese. Scientific Approaches to Language3.159–185.Nakayama, Mineharu & Koizumi, Masatoshi. 1991. Remarks on Japanese subjects. Lingua85.4.303–319.Nemoto, Naoko. 1999. Scrambling. In: Tsujimura, Natsuko (editor). The handbook of Japanese linguistics. Blackwell Publishing. 121–153.Perlmutter, David M.1978. Impersonal passives and the Unaccusative Hypothesis. In: Jaeger, Jeri J. & Woodbury, Anthony C. & Ackerman, Farrell & Chiarello, Christine & Gensler, Orin D. & Kingston, John & Sweetser, Eve E. & Thompson, Henry & Whistler, Kenneth W. (editors). Annual meeting of the Berkeley Linguistics Society. Berkeley Linguistics Society. 157–190.Perlmutter, David M. & Postal, Paul M.1984. The 1-advancement exclusiveness law. In: Perlmutter, David M. & Rosen, Carol G. (editors). Studies in relational grammar 2. University of Chicago Press. 81–125.Pritchett, Bradley & Whitman, John. 1995. Syntactic representation and interpretive preference. In: Mazuka, Reiko & Nagai, Noriko (editors). Japanese sentence processing. Lawrence Erlbaum Associates. 65–76.R Core Team. 2024. R: a language and environment for statistical computing. Software. R Foundation for Statistical Computing. https://www.R-project.org/.Ramchand, Gillian Catriona. 2008. Verb meaning and the lexicon: a first-phase syntax. Cambridge University Press.Saito, Mamoru. 2006. Optional A-scrambling. In: Takubo, Yukinori & Kinuhata, Tomohide & Grzelak, Szymon & Nagai, Kayo (editors). Japanese/Korean Linguistics16. CSLI Publications. 44–63.Saito, Mamoru. 2010. Semantic and discourse interpretation of the Japanese left periphery. In: Saito, Mamoru & Erteschik-Shir, Nomi & Rochman, Lisa (editors). The sound patterns of syntax. Oxford University Press. 140–173.Saito, Mamoru. 2011. Two notes on feature inheritance: a parametric variation in the distribution of the EPP. Nanzan Linguistics7.43–61.Saito, Mamoru. 2012. Case checking/valuation in Japanese: Move, Agree, or Merge. Nanzan Linguistics8.109–127.Schäfer, Florian. 2025. Anticausatives in transitive guise. Natural Language and Linguistic Theory43.1.421–475. 10.1007/s11049-024-09612-w.Takano, Yuji. 2011. Double complement unaccusatives in Japanese: puzzles and implications. Journal of East Asian Linguistics20.3.229–254. 10.1007/s10831-011-9079-y.Tamaoka, Katsuo & Koizumi, Masatoshi. 2006. Issues on the scrambling effects in the processing of Japanese sentences: reply to Miyamoto and Nakamura (2005) regarding the experimental study by Koizumi and Tamaoka (2004). Gengo Kenkyu129.181–226.Tamaoka, Katsuo & Sakai, Hiromu & Kawahara, Jun-ichiro & Miyaoka, Yayoi & Lim, Hyunjung & Koizumi, Masatoshi. 2005. Priority information used for the processing of Japanese sentences: thematic roles, case particles or grammatical functions?Journal of Psycholinguistic Research34.3.281–332. 10.1007/s10936-005-3641-6.Tsujimura, Natsuko. 1990. Unaccusative nouns and resultatives in Japanese. In: Hoji, Hajime (editor). Japanese/Korean Linguistics1. CSLI Publications. 335–349.Tsujimura, Natsuko. 1994. Unaccusative mismatches and resultatives in Japanese. In: Koizumi, Masatoshi & Ura, Hiroyuki (editors). Formal approaches to Japanese linguistics: proceedings of FAJL 1. MIT Working Papers in Linguistics. 335–354.Yamashita, Hiroko. 2002. Scrambled sentences in Japanese: linguistic properties and motivations for production. Text and Talk22.4.597–633. 10.1515/text.2002.023.Yano, Masataka. 2019. On the nature of the discourse effect on extraction in Japanese. Glossa4.1.90. 10.5334/gjgl.822.Yano, Masataka & Koizumi, Masatoshi. 2018. Processing of non-canonical word orders in (in)felicitous contexts: evidence from event-related brain potentials. Language, Cognition and Neuroscience33.10.1340–1354. 10.1080/23273798.2018.1489066.Yano, Masataka & Koizumi, Masatoshi. 2021. The role of discourse in long-distance dependency formation. Language, Cognition and Neuroscience36.6.711–729. 10.1080/23273798.2021.1883694.Yatsushiro, Kazuko. 1999. Case licensing and VP structure. Doctoral thesis. University of Connecticut.Zehr, Jeremy & Schwarz, Florian. 2018. PennController for Internet based experiments (IBEX). Software. Center for Open Science. 10.17605/OSF.IO/MD832.