Short sources, islandhood, and pronominal correlates: New experimental support from German and Spanish for a short source approach to apparent exceptions to the clausemate condition on multiple sluicing

Abstract

This paper focuses on multiple sluicing (MS) utterances in which the antecedent for sluicing is syntactically complex, insofar as it contains at least one embedded clause. Complex antecedent MS (henceforth caMS) is subject to the clausemate condition (CC). Phrased in sententialist terms, the CC bans sluiced wh-phrases from being base-generated in different finite clauses in the elliptic clause. Under specific conditions, it seems that the CC is obviated. We report results from acceptability judgment experiments on German and Spanish that provide further evidence that elliptic wh-clauses in caMS can be monoclausal in nature and that these monoclausal elliptic clauses (so-called short sources) are judged as significantly more acceptable than their isomorphic, biclausal elliptic clause counterparts. We interpret these results as supporting the view that purported obviations to CC are only apparent, following Cortés Rodríguez (2022a) and Cortés Rodríguez & Griffiths (to appear). Because these experiments are novel in that they focus on caMS configurations involving syntactic islands, their results provide new insights into how putative CC-obviation and islandhood interact and demonstrate that not only personal but also relative pronouns can function as correlates in MS.

Headnote

This paper focuses on multiple sluicing (MS) utterances in which the antecedent for sluicing is syntactically complex, insofar as it contains at least one embedded clause. Complex antecedent MS (henceforth caMS) is subject to the clausemate condition (CC). Phrased in sententialist terms, the CC bans sluiced wh-phrases from being base-generated in different finite clauses in the elliptic clause. Under specific conditions, it seems that the CC is obviated. We report results from acceptability judgment experiments on German and Spanish that provide further evidence that elliptic wh-clauses in caMS can be monoclausal in nature and that these monoclausal elliptic clauses (so-called short sources) are judged as significantly more acceptable than their isomorphic, biclausal elliptic clause counterparts. We interpret these results as supporting the view that purported obviations to CC are only apparent, following Cortés Rodríguez (2022a) and Cortés Rodríguez & Griffiths (to appear). Because these experiments are novel in that they focus on caMS configurations involving syntactic islands, their results provide new insights into how putative CC-obviation and islandhood interact and demonstrate that not only personal but also relative pronouns can function as correlates in MS.

(ProQuest: ... denotes formulae omitted.)

1 Introduction: The clausemate condition on multiple sluicing

This paper adopts a sententialist, silent-structure approach to sluicing (Ross 1969), according to which (1) is an elliptic version of (2), where 'ellipsis' refers to the phonological nonrealization of linguistic structure (Ross 1969; Merchant 2001).1 We adopt familiar terminology and notation from the sententialist literature; see (3).

... (1)

... (2)

... (3)

This paper focuses on multiple sluicing (MS), which refers to a sluicing configuration that displays more than one swh in a single elliptic clause (e.g., Takahashi 1994):

... (4)

In particular, we focus on MS configurations that display a biclausal antecedent in which one clause is contained in the other (henceforth complex antecedent multiple sluicing, caMS). Researchers have observed that caMS configurations such as (5),2 in which the correlates occupy different (finite) clauses in the antecedent, are judged as less acceptable than caMS such as (6), in which the correlates occupy the same finite clause in the antecedent (Takahashi 1994; Nishigauchi 1998; Merchant 2001; Rodrigues & Nevins & Vicente 2009; Lasnik 2014; Abels & Dayal 2017; Cortés Rodríguez 2022a). Note that the contrast exemplified by the English sentences in (5) and (6) is also observed in many other languages (Abels & Dayal 2023).

... (5)

... (6)

In the sententialist ellipsis literature, this observation is described by reference to the Clausemate Condition (CC):3

Clausemate condition on MS (Abels & Dayal 2023: 432)

All remnants of sluicing must originate in the same (finite) clause.

Whether caMS configurations such as (7) represent an exception to the CC depends on whether the elliptic clause is syntactically isomorphic or nonisomorphic to its antecedent.4,5 If isomorphic, then (7) represents a true exception to the CC, as the remnants of sluicing originate in different finite clauses in the ellipsis site (see (8)), and yet the sentence is judged as fully acceptable. If the ellipsis site is a nonisomorphic monoclause (henceforth a short source; see (9)), then (7) does not represent an exception to the CC, as the remnants of sluicing originate in the same finite clause, which obeys the CC.

... (7)

... (8)

... (9)

Cortés Rodríguez & Griffiths (to appear) present results from an acceptability judgment experiment on German caMS configurations that are similar in structure to (8) and (9) (insofar as the embedded clause in the antecedent clause of each of their experiment's test sentences is always a finite complement of the matrix verb) that strongly support a 'short source' explanation for why configurations such as (7) are judged acceptable. Furthermore, they provide informally collected acceptability judgment data from English that undermines Barros & Frank's (2023) claim that there exist caMS configurations that are true exceptions to the CC. Put differently, Cortés Rodríguez & Griffiths argue against Barros & Frank's claim that there are acceptable sentences similar to (7) to which a 'short source' analysis cannot be plausibly applied.

This paper functions to extend and strengthen the empirical base for the short source analysis defended by Cortés Rodríguez & Griffiths (to appear). It reports the results of acceptability judgment experiments on caMS configurations in German and Spanish that are structurally dissimilar to those in (7), insofar as the subclause is a strong syntactic island. These results align with Cortés Rodríguez & Griffiths' in that they support a short source approach, and they also provide novel evidence that (relative) pronouns can function as correlates in caMS configurations.

The paper is structured as follows. §2 sketches the current debate regarding whether examples such as (8) and (9) constitute true exceptions to the CC. §3 enters uncharted empirical territory, in that it introduces and presents predictions about a caMS configuration that has received no attention in the previous literature, but whose investigation is likely to be revealing about the nature of the CC. §4 describes two acceptability judgment experiments on German and Spanish and presents their results. In §5, we argue that the results from §4 provide further support for the source short analysis of apparent exceptions to the CC in caMS.

2 Are reported exceptions to the CC only apparent exceptions?

The idea that an elliptic wh-clause can be syntactically nonisomorphic to its antecedent was popularized in the modern sententialist literature by Merchant (2001) and has since been supported by van Craenenbroeck (2004), Rodrigues & Nevins & Vicente (2009), Gribanova (2013), Barros (2014), Barros & Elliott & Thoms (2014), Marušič & Žaucer (2018), Griffiths (2019), Rudin (2019), and Anand & Hardt & McCloskey (2023), among others. Evidence for the existence of nonisomorphic elliptic wh-clauses comes from fully acceptable sluicing configurations in which a nonisomorphic elliptic clause must be present for grammaticality and/or a congruent interpretation to obtain:

... (10)

... (11)

... (12)

Sluicing is known to display variable island-sensitivity (e.g., Chung & Ladusaw & McCloskey 1995; Merchant 2001): the presence of an island-bound correlate in the antecedent clause of a single sluicing configuration sometimes yields a degradation in acceptability (13) and sometimes does not (16). Many researchers have appealed to the existence of nonisomorphic elliptic clauses to explain this contrast (Merchant 2001; Abels 2011; Barros 2014; Barros & Elliott & Thoms 2014; Griffiths 2019). According to this approach, which is now known as the 'island evasion' approach, the examples in (13) are degraded because the only pragmatically congruent elliptic clause available is the isomorphic one, which displays island-violating wh-movement (which lowers acceptability; see (14) and (15)), whereas the examples in (16) are not degraded because a pragmatically congruent nonisomorphic elliptic clause is available, in which no island-violating wh-movement occurs (see (17) and (18)). As mentioned already in §1, the 'island-evading' monoclausal nonisomorphic elliptic clauses that are exemplified by the b-examples in (14), (15), (17), and (18) are known as short sources for ellipsis (Barros & Elliott & Thoms 2014).

... (13)

... (14)

... (15)

... (16)

... (17)

... (18)

As mentioned in §1, single sluicing and MS are treated as the same core phenomenon in the sententialist literature, differing only in the number of wh-remnants each configuration type exhibits. On this view, the existence of nonisomorphic elliptic clauses in single sluicing entails the existence of nonisomorphic elliptic clauses in MS. This entailment has already been exploited to explain variable island-sensitivity in MS (Barros & Elliott & Thoms 2014; Lasnik 2014; Abels & Dayal 2017; 2023). Just like with single sluicing, one can appeal to the availability of a grammatical and contextually congruent short source to explain why caMS configurations such as (19) are acceptable despite having island-bound correlates. Similarly, one can appeal to the unavailability of a suitable short source to explain why the caMS configurations in (21), which also have at least one island-bound correlate, are judged as degraded.

... (19)

... (20)

... (21)

... (22)

... (23)

... (24)

Barros & Frank (2023) propose that a full source (i.e., a biclausal, isomorphic elliptic clause) is available for caMS configurations such as (7) from §1 (repeated below in (25)); recall (8). In other words, they view such configurations as genuine exceptions to the CC. Briefly and approximately stated, they assume that swh2's movement to the left-periphery of the matrix clause cannot proceed successive-cyclically (unlike regular wh-movement) and is, therefore, ordinarily clause-bound (following Grano & Lasnik 2018). They propose that a clause C ceases being a boundary for swh2's movement if C's subject fails to shift attention away from the most salient discourse referent evoked in the clause that embeds C (henceforth, a stable subject). Because C's subject in (25) is a bound pronoun that co-refers with the universally quantified DP introduced in the matrix clause, it is stable, and therefore, C is permeable for swh2's movement. Thus, it yields an acceptable "full source" caMS configuration.6

... (25)

Considering that MS permits short sources (as the discussion above regarding variability to island-sensitivity in MS has already shown), it is also feasible that (25)'s acceptability derives from the fact that its elliptic clause is a short source (see (9)) and therefore do not constitute a true exception to the CC. Cortés Rodríguez & Griffiths (to appear) provide experimental evidence from German that, when morphological case-marking on the wh-remnant is used to force either the short source or full source versions of these sentences (compare the elliptic sources presented in (26)), the short source version receives a significantly higher acceptability rating than its full source counterpart.

... (26)

a. welchem Investor welchem Verkäufer. [full source] which.dat investor which.dat clerk Interpretation: '... [to which investor]i Nadine said that hei should recommend Tanja to which clerk.'

b. welcher Investor welchem Verkäufer. [short source] which.nom investor which.dat clerk Interpretation: '... which investor should recommend Tanja to which clerk.'

Although Barros & Frank do not reject the possibility that caMS configurations exemplified by (26) could involve short sources, they do claim that a short source is unavailable for each of the caMS examples in (27) to (29). If Barros & Frank are indeed correct that no short source is available for any case, then these examples constitute true exceptions to the CC, saving their analysis from being superfluous.

... (27)

... (28)

... (29)

Cortés Rodríguez & Griffiths (to appear) contend that short sources are indeed available for each of the examples in (27) to (29); see (30). Following Abels & Dayal (2023: 436), Cortés Rodríguez & Griffiths emphasize that these short sources are indeed licensable according to recent theories of ellipsis identity.7

... (30)

Furthermore, Cortés Rodríguez & Griffiths report that the caMS sentences in (31) to (33) are judged as strongly degraded compared to their structurally similar counterparts in (27) to (29) (these judgments were informally collected from 8 native speakers: 6 linguists; 2 nonlinguists). The difference between these two sets of sentences is that pragmatically congruous short sources are available for the first set (i.e., (27) to (29)) but not for the second (i.e., (31) to (33)), this comparison provides additional support for Cortés Rodríguez & Griffiths' claim that the relative acceptability of the examples in (27) to (29) is directly correlated with the availability of a short source for each example. This comparison also undermines Barros & Frank's claim that the acceptability of the examples in (27) to (29) is directly connected to the fact that, in each example, the subject of the embedded clause in the antecedent clause fails to introduce a new discourse referent.8

... (31)

... (32)

... (33)

Cortés Rodríguez & Griffiths report that when the full source interpretation of (26) is forced, the mean rating returned is 3.49 (out of 7), and when the short source is forced, the mean rating is 4.13; a difference of 0.64. In a different experiment reported in Cortés Rodríguez (2022a), caMS configurations exemplified by (34), in which the embedded clause's subject is an R-expression that introduces a new discourse referent (henceforth, a shifty subject), were tested. The mean rating for the full source of (34) is 3.24, whereas the short source is 4.68; a difference of 1.44. In both cases, the differences reported are statistically significant.

... (34)

If we assume momentarily that the raw mean results obtained in these different experiments are directly comparable (this a questionable assumption), then it appears that having a stable subject in the embedded clause improves acceptability relative to having a shifty subject. This could be interpreted as showing that the locality constraints on wh-moving swh2 from the embedded clause are indeed suspended in configurations such as (26a), as Barros & Frank propose. Cortés Rodríguez & Griffiths reject this conclusion and instead conclude that, should an amelioration effect indeed be obtained by having a bound pronoun subject in an unambiguously CC-violating caMS configuration, this effect arises because the presence of a bound pronoun subject diminishes the processing burden associated with attempting to recover an ellipsis site in which a grammatical constraint – namely, whatever syntactic locality constraint underlies the CC – is violated.9 The reasoning behind Cortés Rodríguez & Griffiths' conclusion is this: if (26a) did involve a genuine suspension of the syntactic locality constraint underlying the CC, then so should each of the examples in (31) and (33), according to Barros & Frank's theory. As shown already, this situation does not transpire.10

If we assume momentarily that the raw mean results obtained in these different experiments are directly comparable (this a questionable assumption), then it appears that having a stable subject in the embedded clause improves acceptability relative to having a shifty subject. This could be interpreted as showing that the locality constraints on wh-moving swh2 from the embedded clause are indeed suspended in configurations such as (26a), as Barros & Frank propose. Cortés Rodríguez & Griffiths reject this conclusion and instead conclude that, should an amelioration effect indeed be obtained by having a bound pronoun subject in an unambiguously CC-violating caMS configuration, this effect arises because the presence of a bound pronoun subject diminishes the processing burden associated with attempting to recover an ellipsis site in which a grammatical constraint – namely, whatever syntactic locality constraint underlies the CC – is violated.9 The reasoning behind Cortés Rodríguez & Griffiths' conclusion is this: if (26a) did involve a genuine suspension of the syntactic locality constraint underlying the CC, then so should each of the examples in (31) and (33), according to Barros & Frank's theory. As shown already, this situation does not transpire.10

To summarize §2: the short source approach advocated by Cortés Rodríguez & Griffiths states that there are no true exceptions to the CC, and that the availability of a CC-evading nonisomorphic elliptic wh-clause (i.e., a short source) can sometimes give the false impression that the CC has exceptions. There are four main reasons to favor the short source approach over its main competitor, namely Barros & Frank's (2023) analysis. First, the existence of short sources is well established for both single and multiple sluicing. Second, analyses that postulate true CCobviations cannot capture the significant differences in acceptability observed between short and full source configurations when a short or full source interpretation is forced. Third, the variability in CC-sensitivity directly tracks the availability of a short source. Fourth, the variability in CC-sensitivity does not directly track the presence/absence of a shifty subject in the embedded clause of the antecedent, which undermines Barros & Frank's analysis.

3 Entering uncharted territory in the study of caMS

In the previous section, variable island-sensitivity under MS and putative CC-obviations in MS were discussed as independent phenomena. A natural springboard for further exploration of caMS configurations is to investigate what happens when these phenomena interact, as in caMS configurations that fit one of the schemas in (35). In (35a), the subject of the clause-sized island is shifty, whereas in (35b), this subject is stable.

... (35)

As we will henceforth be comparing the schemas in (35) to the caMS configurations discussed in the previous section, it is useful to introduce some shorthand terminology at this juncture. We now have three degrees of freedom to contend with: (i) whether correlate2 is island-bound or contained in a finite complement clause, (ii) whether the subject of the clause in which correlate2 is contained displays a shifty or stable subject, and (iii) whether the ellipsis site is a short source or full source. Hereafter, we use comp and island to describe the nature of the embedded clause, we continue to use shifty and stable to describe the subject of the embedded clause, and we use subscripted F and S to distinguish between full or short source ellipsis sites. Putting these terms together, one yields shorthand descriptions such as a stable-compF (for the full source version of, e.g., (26a)) and shifty-islandS (for short source versions of configurations fitting the schema in (34a)).

Cortés Rodríguez (2022a) reports that, in English, German, and Spanish, there is no significant difference in acceptability between shifty-compF (recall (34a)) and shifty-islandF configurations ((36) below exemplifies the shifty-islandF configuration in German). The fact that acceptability is degraded to the same degree in both configurations suggests that the same syntactic locality constraint on wh-moving swh2 is operative in both cases, despite the fact that (34) involves extracting swh2 from a finite embedded clause and (36) involves extracting swh2 from a strong island (according to Grano & Lasnik 2018; Barros & Frank 2023, this locality constraint is the phrase impenetrability condition, PIC).

... (36)

If the same locality constraint is operative in both shifty-compF and shifty-islandF cases, and if the presence of a stable, bound pronoun subject in stable-compF sentences such as (26a) diminishes the processing burden incurred by violating this locality constraint, then one expects the same amelioration effect to be observed in stable-islandF configurations (i.e. the full source version of (35b)). Thus, the first issue related to the schemas in (35) is whether or not this expectation is met.

Another question relating to the schemas in (35) is whether a short source is available in either schema. Note that obtaining the short source reading for stable-comp configurations such as (26) involves the hearer taking a pronoun as the correlate for swh1. As pointed out to us by Andrew Murphy (pers. comm.), this possibility seems unique to MS, as, in single sluicing, associating the sluiced wh-phrase with a pronominal correlate is impossible (in German, at least):11

... (37)

A pressing question is, therefore, to what extent bound pronouns can function as correlates in caMS configurations. In (26), one observes that a bound subject pronoun in a finite complement clause can function as swh1's correlate in German. But can subject pronouns inside other types of subclause – particularly, clausal islands – also function as correlates? And is this ability restricted to German or observed in other languages?

The experiment reported in the following subsection addresses both of these issues. It does this by first testing whether full and/or short sources are available in German and Spanish for two caMS configurations that exemplify the stable subject schema in (35b). The judgments obtained are then compared to results from previous experiments on caMS (from Cortés Rodríguez 2022a; Cortés Rodríguez & Griffiths to appear).

4 An acceptability judgment experiment on German and Spanish caMS

4.1 The purpose of the experiment and its general setup

As mentioned already in §2, the experiment reported in Cortés Rodríguez & Griffiths (to appear) focused on German stable-comp configurations. There, we exploited the morphological casemarking on the wh-remnant to disambiguate caMS configurations as involving either a short source or full source, and we discovered that the short source versions of our test sentences received significantly higher acceptability ratings than their full source counterparts.

In the experiment reported below, the same experimental setup is applied to two different stable-island configurations. These configurations are exemplified by the English sentences in (38) and (39). In both configurations, CP2 occupies an extraposed position but is interpreted as either a restrictive relative modifier or a complement of the noun salesman. The caMS configurations in (38) and (39), therefore, differ from the canonical stable-comp configuration discussed in the literature (i.e., (25)) in that, in (38) and (39), CP2 is an island for whextraction. Like in our previous experiment, we presented participants with two versions of each caMS configuration, one in which the full source continuation is forced (see the a-examples in (38) and (39)) and one in which the short source continuation is forced (see the b-examples in (38) and (39)).12

... (38)

... (39)

The purpose of this experiment was threefold.

First, it tests to see if (relative) pronouns that are the subjects of clause-sized islands and which are not c-commanded by their antecedent DP (in the surface structure, at least) can function as correlates for MS, given the presence of sufficient cues, such as case-marking and co-reference (see §3). Because these pronouns only function as correlates for swh1 in the short source versions of (38) and (39), we obtain evidence that these pronouns can indeed function as correlates in MS if the short source versions of (38) and (39) receive high mean ratings, as this shows these short source versions are available.

Second, it tests if the processing burden associated with wh-moving swh2 over an island boundary is ameliorated when the island's subject is a bound pronoun (i.e., stable). As mentioned in §3, we expect an amelioration effect to be observed. This prediction is not borne out only if the ratings attributed to our stable-islandF test items and to shifty-islandF sentences (e.g., (36)) are sufficiently similar. We will determine whether this outcome obtains by performing a post hoc cross-experimental comparison. This involves comparing the ratings for our stable-islandF test sentences to the ratings obtained for stable-compF and shifty-islandF caMS configurations in previous experiments (reported in Cortés Rodríguez 2022a; Cortés Rodríguez & Griffiths to appear, respectively). Precisely how this comparison will be conducted is described in §4.2.2.13

Third, by running the experiment on Spanish speakers (in Spanish) as well as German speakers (in German), the experiment broadens the crosslinguistic base with robust and experimentally procured caMS data. Specifically, we aim to show with this experiment that our previous and current experimental findings on caMS are not just quirks of German but reflect broader – perhaps universal – tendencies.

4.2 Subexperiment 1: German

4.2.1 Methods

4.2.1.1 Design and materials

We conducted an acceptability judgment experiment with 24 sentence quadruplets containing caMS configurations. The experiment followed a 2 × 2 within-item and within-subject design. The two independent variables were source and embedding type. The two levels for source were full source and short source. In the full source condition, swh1 displays the same form as the PP in the antecedent's matrix clause (i.e., the italicized PPs in (40) and (41)). In the short source condition, swh1 displays the same case as the personal or relative pronoun in the subject position of the extraposed clause (i.e., the straight-underlined DPs in (40) and (41)). The two levels for embedding type were relative clause (RC) and complement of N (NCC). In the RC condition, the extraposed clause is a restrictive relative clause. In the NCC condition, the extraposed clause is a noun complement clause. Examples of each condition are presented in (40) and (41). Note that the embedding type factor is included to ensure reliability. Put differently, we included two different island types in our experiment to control for the possibility that the effect of modulating source is specific to certain islands (and therefore possibly different across different types of islands).

... (40)

... (41)

In every condition, swh2 was a PP. Although the syntactic category of swh2 (DP versus PP) has no effect on the acceptability of MS in German (Cortés Rodríguez to appear), there is a statistically significant preference for a PP swh2 in Spanish (Cortés Rodríguez 2021, see also Cortés Rodríguez 2024).14 To maintain consistency, we therefore decided to use only PP swh2 in both our German and Spanish subexperiments.

Some researchers have claimed that MS is only licensed when the correlate for swh1 is a universally quantified DP (Nishigauchi 1998; Merchant 2001). This is untrue for English and German; see Cortés Rodríguez (2024) for experimental evidence and additional references. Therefore, the fact that this experiment's test sentences involve no universal quantified DPs as correlates is unproblematic. Abels & Dayal (2017; 2023) recommend using a universally quantified DP as a correlate to exclude the possibility that hearers interpret the ellipsis as an asyndetic coordination of single sluicing elliptic wh-clauses, as in (42). Although exercising caution is commendable, there is no evidence among any of the experimental findings on English, German, and Spanish by Cortés Rodríguez (2024) to suggest that hearers ever retrieve a 'conjunction of single sluices' parse for ellipsis site in the absence of an overt coordinator between the sluiced wh-phrases, regardless of whether the correlate for swh1 is a universally or existentially quantified DP. Therefore, the current experiment runs no risk of eliciting unwanted and confounding 'conjunction of single sluices' parses for the ellipsis site from its participants.

... (42)

Every participant saw a total of 6 items in each condition; thus, 24 critical items in total. Additionally, 72 fillers were included in every list. Fifteen out of these 72 fillers corresponded to the standardized items from Featherston (2009), which range from A-type fillers (most natural) to E-type fillers (least natural). Standardized items were included to ensure proper scale usage. In total, each participant rated 96 experimental tokens. The experimental stimuli were presented in a Latin-square design of 4 lists. All participants saw all fillers, and the order of stimuli was randomized at runtime. The participants undertook a practice round of 5 sentences before starting the experiment in earnest.

4.2.1.2 Participants and procedure

The acceptability judgment test was created using PsychoPy 3 software (Peirce et al. 2019). Thirty-two self-reported native speakers of German were recruited via Prolific (www.prolific.co) and paid £3 for participating. The experiment lasted ~20 minutes. Participants were instructed to read carefully and then rate the naturalness of the test sentences on a 7-point scale, from 1 (very unnatural) to 7 (very natural). Additionally, they were told that there were no right or wrong answers and that they should just follow their intuition. We discarded the data from 3 participants: 1 participant disclosed that their native language was not German, and the other 2 misused the rating scale based on their answers to the standardized items. Accordingly, data from 29 participants (15 female, 12 male, 2 non-binary; Mean age = 33.0, SD = 12.3) were used in the statistical analysis.

4.2.1.3 Predictions

As discussed in §4.1, one of the purposes of this experiment is to replicate and extend the findings reported in Cortés Rodríguez & Griffiths (to appear). Therefore, we predict that the short source versions of our test sentences will receive significantly higher acceptability ratings than their full source counterparts. As mentioned in §3, we expect that the full source versions of our test sentences (i.e., stable-islandF ) receive the same ratings as stable-compF caMS configurations, given that the presence of a bound pronominal subject in both cases provides an amelioration effect for swh2's transgressive wh-movement (or so we claim).

Recall from the previous subsection that we have included embedding type as a dependent variable in this experiment. This decision is based on our own impressionistic judgments that antecedents in caMS configurations involving an extraposed noun complement clause (e.g., (39) and (41)) are harder to parse than caMS configurations involving an extraposed relative clause (e.g., (38) and (40)). We suspect that the source of this disparity is unrelated to ellipsis, and therefore, we predict a main effect for embedding type but no interaction between source and embedded type. Note that these predictions, which we list in (43) for the reader's convenience, apply to both subexperiment 1 (German) and subexperiment 2 (Spanish).

(43) Predictions

a. Short source test items receive significantly higher acceptability ratings than full source test items (i.e., a main effect for source).

b. The full source test items from the current experiment lower acceptability by the same amount as stable-compF caMS configurations do.

c. Test items in the RC condition receive significantly higher acceptability ratings than test items in the NCC condition (i.e., a main effect for embedding type).

d. There is no interaction between the two factors under investigation.

4.2.2 Results of the experiment and post hoc comparison

The mean acceptability ratings for the 4 experimental conditions are listed in Table 1. These means are plotted, along with their 95% confidence intervals, in Figure 1.

In terms of inferential statistics, the data were analyzed in R, version 4.1.2 (R Core Team 2021). We employed a cumulative link mixed effect model (CLMM) to analyze the ordinal data obtained from the Likert scale using the clmm function of the ordinal package (Christensen 2019). The model with the best fit was selected using a manual backward model selection process. This process involves taking a full model that includes all experimental factors and interactions as fixed effects and which also includes random intercepts and slopes for both subjects and items and then simplifying this full model, one factor at a time. The fit of each simplified model m-1 is compared to the fit of its immediate, more complex predecessor m using the anova function. The corresponding formula is provided in the table with the statistical analyses.

The results of the statistical analysis are presented in Table 2. The best fitting model shows a main effect for both source and embedding type but no significant interaction between them. Regarding source, the short source versions of our test items were rated as significantly more acceptable than their full source counterparts. Regarding embedding type, participants rated test sentences with an extraposed relative clause as significantly more acceptable than those with an extraposed noun complement clause.

Formula: rating ~ source + embeddingType + (1 | subject) + (embeddingType | item), threshold= "symmetric".

One sees that the predictions in (43a), (43c), and (43d) are straightforwardly borne out: main effects were observed for source (short source favored) and embedding type (RC favored), whereas no interaction was observed between them.

Although it is possible to perform statistical tests on data collected from different experiments to determine if the prediction in (43b) is borne out, we refrain from doing this here, as the experiments being compared each employed quite different experimental designs, target items, and fillers, and therefore the reliability of statistical results would be questionable. Acknowledging thus that our comparison will be inherently impressionistic, we opt to employ an informal binary heuristic, which states that the prediction in (43b) is borne out if the ratings for the current experiment's test items are numerically closer to the ratings for the stable-compF configurations from Cortés Rodríguez & Griffiths (to appear) than to those for the shifty-islandF configurations from Cortés Rodríguez (2022a). Nonetheless, in an effort to control for variation as much as possible, we z-score all ratings,15 calculate the difference D between the full and short source versions of each caMS configuration being compared (thus further controlling for across-item variation), and then apply our heuristic to these Ds.

Legend: current study; Cortés Rodríguez 2022a; Cortés Rodríguez & Griffiths to appear.

... (44)

According to our heuristic, the prediction in (43b) is borne out, as the figures presented in the final column of Table 3 for the current experiment's test item's differences are numerically closer to the figure obtained for configurations exemplified by (26) than to the figure obtained for configurations exemplified in (44).

4.3 Subexperiment 2: Spanish

4.3.1 Methods

4.3.1.1 Design and materials

The 24 test items used in our Spanish subexperiment were structurally identical to their German counterparts (see §3.2.1), barring some necessary adjustments. In the Spanish RC condition, the relative clause modifying the object of the matrix clause appeared right after the modified noun, meaning that an in-situ or an extraposed parse was available for the relative clause (in both positions, the relative clause is ordinarily an island for wh-extraction). In the NCC condition: the Spanish noun complement clause is headed by a preposition, unlike its German counterpart.

... (45)

... (46)

All items and fillers followed the same distribution as in subexperiment 1. For the standard/control items, we used the 15 filler items from Cortés Rodríguez (2021, 2024), which aim to provide absolute measures for 5 levels of acceptability in Spanish, following the standardized items created for German (Featherston 2009) and English (Gerbrich & Schreier & Featherston 2019).

4.3.1.2 Participants, procedure, and predictions

The procedure of subexperiment 2 replicated the procedure for subexperiment 1. We discarded the data from 3 participants who misused the rating scale based on their answers to the control items. Accordingly, the data from 29 participants (14 female, 15 male; Mean age = 28.6, SD = 8.6) were used in the statistical analysis.

As mentioned already, the predictions for subexperiment 2 were the same as for subexperiment 1 (see (43)).

4.3.2 Data analysis and results

The mean acceptability ratings for the 4 experimental conditions are listed in Table 4. These means are plotted, along with their 95% confidence intervals, in Figure 2.

Regarding inferential statistics, the data were analyzed following the same procedure as in subexperiment 1. The results of the statistical analysis are presented in Table 5. The best fitting model shows a main effect for both source and embedding type but no significant interaction between them. Regarding source, the short source versions of our test items were rated as significantly more acceptable than their full source counterparts. Regarding embedding type, participants rated test sentences with an extraposed relative clause as significantly more acceptable than those with an extraposed noun complement clause. The Spanish subexperiment, therefore, replicated the results of the German subexperiment.

Formula: rating ~ source + embeddingType + (embeddingType | subject) + (1 | item).

Table 6 presents the z-scored mean scores for the full and short source versions of the test sentences from the current subexperiment and from previously run experiments on Spanish caMS configurations.

Legend: current study; Cortés Rodríguez 2022a; Cortés Rodríguez 2022b.

... (47)

... (48)

To determine if the prediction in (43b) is borne out for Spanish, we use the same comparison heuristic as for German (see the text under Table 3). We see that, just as with German, the prediction in (43b) is borne out, as the figures presented in the final column of Table 6 for the current experiment's test item's differences are numerically closer to the figure obtained for configurations exemplified by (48) than to the figure obtained for configurations exemplified by (47).

5 Discussion

5.1 Short sources: (relative) pronouns can function as correlates in MS

Previous research has shown that standard, "good" caMS configurations – i.e., caMS sentences that violate no known grammatical constraints – are rated as 4.68(1.58) in German and 4.22(1.50) in Spanish (Cortés Rodríguez 2022a). The stable-islandS configurations tested in our current study received comparable ratings (a mean rating of 4.86 for German and a mean rating of 4.44 for Spanish) and are significantly more acceptable than their full source counterparts. We interpret these results as demonstrating that this study's stable-islandS configurations are, in relative terms, acceptable, and that this provides novel evidence that (surface-) unbound personal pronouns and relative pronouns can function as correlates of sluiced wh-phrases in MS. This evidence is obtained from the fact that swh1 in the short source variants of our test sentences can only be associated with these pronominal phrases, and not their full DP antecedents (assuming that case-mismatch is avoided whenever possible, see the discussion surrounding (53) below). These results, when added to Cortés Rodríguez & Griffiths' (to appear) results for German showing that bound subject pronouns in finite complement clauses can also function as correlates, demonstrate that the range of elements that can function as correlates in MS is greater than in single sluicing.

Before moving on, we wish to preemptively respond to two potential criticisms of our analysis. One argument that could be voiced against the short source approach to explaining apparent exceptions to the CC is that the approach overgenerates. For instance, it has been reported that the shifty-comp configuration in (49), in which the swh2's correlate in the embedded clause of the antecedent is a bound object pronoun, as unacceptable. Because a short source is readily available for the elliptic clause in (49) (albeit with some modal subtlety lost; see (50)), one might contend that the short source approach incorrectly predicts that (49) should be judged as acceptable.

... (49)

... (50)

There are conceptual and empirical problems with this argument. Conceptually, it requires one to move from the weaker position that short sources are sometimes unavailable for caMS configurations that are apparent exceptions to the CC to the stronger position that short sources are simply never available for any caMS configurations. When one considers the independent evidence for the existence of short sources in caMS configurations is now robust (recall the 'variable island-sensitivity' discussion from §2 and see the 'morphological case marking' data discussed in Abels & Dayal 2017; 2023; Cortés Rodríguez & Griffiths to appear), this stronger position cannot be plausibly maintained. Empirically, the *-mark attributed to (49) is an exaggeration. We informally collected judgments from 9 native speakers of English (4 linguists, 5 non-linguists) on (49) and its 'stable subject' counterpart (namely, Some student claimed that she introduced John to some professor, ...) and found reasonably small differences, albeit with high standard deviation (x̄ = 3.1, s = 1.53 for (49); x̄ = 3.8, s = 2.17 for the stable subject equivalent). Furthermore, Cortés Rodríguez & Griffiths report that the German equivalent of (49) is judged as acceptable. Specifically, they report that the German version of (49) is judged as acceptable when disambiguated as involving a short source (see (51a)) and is judged as degraded when disambiguated as involving a full source (see (51b)). This empirical picture suggests to us that an extraneous factor is responsible for modulating (49)'s acceptability across languages. Our preliminary – and therefore tentative – analysis of this variation runs as follows: (i) bound pronouns must be above a particular threshold of discourse saliency to function as correlates in MS configurations, (ii) bound subject pronouns are inherently more salient than bound object pronouns (Barros & Frank 2023: 660) such that, by themselves, subject but not object pronouns are sufficiently salient to function as correlates, (iii) the accusative case-marking on swh1 in (51a) functions as a cue that directs the hearer's attention to the accusative case-marked object pronoun in the antecedent, which increases the pronoun's saliency enough that it can function as a correlate, and (iv) this saliency-boost does not occur in the English example in (49) because swh1 shows no overt morphological marking of accusative case.

... (51)

If this explanation is on the right track, then it indicates that a variety of conditions must be satisfied for a pronoun to function as a correlate for MS, and that these conditions may differ across languages. Determining precisely what these conditions are and how they differ across languages must remain a task for future research. With this in mind, the (alleged) unexpected absence of a short source interpretation for a particular caMS configuration (such as (49)) does not prima facie invalidate the source short approach, as extraneous restrictions might obtain.

Second, let us address a potential alternative interpretation of our results. Elliott & Murphy (2019) show that, in so-called unconditional sluicing in German, the sluiced wh-phrase must bear the same morphological case marking as its correlate, despite the fact that the only congruous ellipsis site for this configuration is a nonisomorphic short source (compare the possible elliptic clauses in (52a) and (52b)):

... (52)

Seeing as the morphological case-marking on German sluiced wh-phrases is not always a reliable indicator of the structure of the ellipsis site, could it be that the current experiment's German 'short source' test items (and also the 'short source' test items from the experiment reported in Cortés Rodríguez & Griffiths) are not forcing a short source ellipsis site after all? Could it be that the putative 'short version of, e.g., (26) (repeated below in (53)) actually involves the full source, but with exceptional case-marking on swh1 (see (54))? If so, then MS does not uniquely permit pronouns to function as correlates after all: in reality, MS and single sluicing are alike in disallowing pronominal correlates.

(53) Nadine hat [einem Investor]i erzählt, dass eri Tanja einem Verkäufer Nadine has a.dat investor told that he.nom Tanja a.dat clerk nahelegen soll, aber ich weiß nicht, welcher Investor welchem recommend should but I know not which.nom investor which.dat Verkäufer clerk

... (54)

There are a number of reasons to be sceptical of this alternative analysis. First, it involves a false analogy. In the unconditional sluicing case, there is motivation for the sluiced wh-phrase to bear exceptional case – obtaining morphological parallelism between the wh-phrase and its correlate overrides the usual case-assignment algorithm. This same motivation is absent in (54), in which the usual case-assignment algorithm is overridden, and yet morphological parallelism between the wh-phrase and its correlate is not achieved. Second, if (54) were indeed the elliptic clause for (53), then one expects (53) to be harder to process (which translates into lower acceptability) than the full source version of (54),16 in which swh1 does match in case with its correlate einem Investor. One expects greater processing costs in the former configuration because the casemismatch between welcher Investor and einem Investor should impede the process of establishing einem Investor as swh1's correlate rather than facilitate it. We have repeatedly shown through our experiments that this expectation is not borne out since our 'short source' test items are consistently rated significantly higher than their 'full source' counterparts. Third, this alternative analysis is German-specific and, therefore, does not explain why the results we obtained for German are replicated in Spanish, which does not show exceptional case-marking under sluicing (to our knowledge). For these reasons, we discard the alternative analysis outlined above and continue to advocate the conclusion that, under the right conditions, pronouns can function as correlates in caMS configurations.

5.2 Full sources: The amelioration effect of having a stable subject

As mentioned already, we interpret the results of our post hoc, across-experiment comparison as demonstrating that the presence of a bound pronominal subject in a clausal island reduces the processing burden associated with wh-extracting across that island boundary in both German and Spanish. Because the same amelioration effect is observed for stable-compF configurations (see Cortés Rodríguez & Griffiths for German; Cortés Rodríguez 2022b for Spanish), we extrapolate that the effect is observed for any clause-sized locality domain.

There is, of course, an alternative analysis available, which runs as follows: (i) the presence of a bound pronoun in C's subject position renders C permeable to swh2's movement, and (ii) stable-compF configurations receive a lower acceptability rating than short source caMS configurations because the former but not the latter involve recovering a complex, biclausal elliptic clause, which incurs a processing cost that translates into an acceptability penalty. We do not currently advocate this approach because the theories in which this explanation is endorsed – namely, Grano & Lasnik (2018) and Barros & Frank (2023) – make incorrect predictions elsewhere. Regarding Grano & Lasnik, they claim that a clause-level barrier to extraction is suspended only if a c-commanding subject binds C's pronominal subject. Although Grano & Lasnik are primarily concerned with non-island clauses, they extend their claim to clausal islands, noting that an amelioration is observed for standard argumental wh-movement (compare the examples in (55)).

... (55)

As pointed out by Barros & Frank (2023: 672), the technical details of Grano & Lasnik's theory yield the prediction that any clause-sized barrier that obtains between pronominal subject and its binder is suspended, which is not borne out (compare (55a) and (55b)). Barros & Frank also show that, in non-island configurations, an amelioration effect for cross-clausal movement is also obtained when the binder is a non-subject. Seeing as most of our experimental items involve an indirect object in the matrix clause as the c-commanding binder, our experimental results confirm this. This is not predicted by Grano & Lasnik's theory.

... (56)

Recall from §2 that Barros & Frank claim that any non-referential expression, as opposed to only a bound pronoun, can function as a stable subject, thus making subclauses permeable to swh2extraction. Also, recall from §2 that Cortés Rodríguez & Griffiths offer evidence from informally collected judgments on English stable-comp configurations that lack a short source interpretation to suggest that this claim is incorrect. If one extends Barros & Frank's analysis to cover the stable-islandF configurations under discussion here, the same criticism holds – it appears that a clausal adjunct is not rendered permeable to wh-extraction when its subject is expletive there or a negatively quantified DP:

... (57)

Should a theory be developed that amalgamates Grano & Lasnik's and Barros & Frank's (this theory would capture that observation that only bound pronouns can suspend clausal barriers to extraction and that the pronoun can be bound from any c-commanding position), then this theory would represent a serious competitor to the position that we currently endorse. We suspect that online experiments (e.g., self-paced reading, eye-tracking) would be required to empirically distinguish between these competing analyses.

5.3 The main effect of embedding type

As mentioned in §4.2.1, embedding type was included in the experiment reported in §4 merely to ensure reliability. We predicted a main effect for embedding type based on our intuitions, and this prediction was borne out. Because no interaction between embedding type and source was observed for either German or Spanish, we can be confident that the main effect obtained for embedding type's is entirely unrelated to ellipsis. Therefore, offering a viable explanation for why this main effect obtains is not necessary here. If pushed to speculate, we would suggest that the observed difference is related to the fact that the relative clause in the RC condition is an adjunct, whereas the noun complement clause in the NCC condition is a complement. According to some researchers (e.g., Fox & Nissenbaum 1999), adjunct extraposition involves base-generation, whereas argument extraposition involves rightward movement. If so, the NCC condition is syntactically more complex, which could translate into a greater processing cost.

6 Conclusion

In this paper, we reported the results of experiments conducted on complex antecedent multiple sluicing (caMS) constructions in German and Spanish and undertook a comparison of these results with those obtained in prior work (reported in Cortés Rodríguez 2022a,b; Cortés Rodríguez & Griffiths to appear). We interpreted our results and subsequent across-experiment comparison as showing that (i) short elliptic sources corresponding to clausal islands are available and preferred, (ii) both relative and personal pronouns can function as correlates in MS, and (iii) the processing burden associated with island-crossing wh-movement of the second sluiced wh-phrase (swh2) is partly ameliorated by the presence of a bound pronoun in the subject position of the island, just as the burden associated with wh-moving swh2 across a non-island finite clausal boundary is. These results therefore not only support the position endorsed in Cortés Rodríguez (2022a,b) and Cortés Rodríguez & Griffiths (to appear) that apparent clausemate condition obviations are, from a grammatical standpoint, illusory, but complete an empirical paradigm of caMS configurations and present novel evidence that there are usually far more options available for resolving the ellipsis site than previously considered and that the options that are (salient enough to be) available will depend on language-specific factors related to what counts as a cue for ellipsis resolution.

Data availability

All data processing and analysis code are publicly accessible through an OSF repository: https:// doi.org/10.17605/OSF.IO/MWJZ9. The supplementary files with the material and results of Subexperiments 1 and 2 are available in the same OSF repository.

Acknowledgements

We wish to thank the three anonymous reviewers for their insightful comments and recommendations. We would also like to thank all the participants and informants who provided their judgments to our test sentences. All remaining errors are ours.

Competing interests

The authors have no competing interests to declare.

Footnote

1 As will become clear shortly, this paper addresses issues surrounding the extent to which elliptic clauses can deviate in form from their antecedents in multiple sluicing contexts, and it investigates the possibility that exceptions to the clausemate condition on multiple sluicing are only apparent. To our knowledge, multiple sluicing has received no systematic attention in the nonsententialist literature and therefore determining how to even frame the abovedescribed issues in a nonsententialist framework is difficult to ascertain.

2 The "*" judgment attributed to (5) represents the judgment reported in Lasnik (2014: 12); Abels & Dayal (2017: 4); Barros & Frank (2023: 654) (among others) for similar examples exemplifying the same configuration. See Cortés Rodríguez (2022) for experimental results supporting the acceptability contrast between (5) and (6).

3 To our knowledge, CC effects are claimed to be entirely absent in MS configurations in only two languages, SerboCroatian and Indonesian (Lasnik 2014; Sato 2016). For Serbo-Croatian, this absence is observed only for speakers that also allow multiple wh-fronting from separate clauses in nonelliptic multiple questions. For Indonesian MS, Sato (2016) argues that this language shows an absence of CC effects because the second wh-remnant undergoes neither covert nor overt movement; see Sato (2016) for details.

4 In examples in which the internal structure of the elliptic wh-clause is presented, we represent the second sluiced wh-phrase as being phonologically realized in its base position. We do this merely to increase readability. Although we follow the literature in assuming that the second sluiced wh-phrase undergoes movement to left-periphery of the elliptic clause in the languages we are investigating (German and Spanish) (see, e.g., Abels & Dayal 2023, contra Sato 2016 for Indonesian), we remain noncommittal about whether the second wh-phrase is phonologically realized in its base or moved position. Where relevant, the landing position of swh2 is denoted by a Δ.

5 Following Fox & Lasnik (2003: 150), who build on Fiengo & May (1994), we treat an elliptic clause as syntactically isomorphic to its antecedent only if it is structurally identical to the entirety of its antecedent, meaning that variables in the two clauses are bound from parallel positions.

6 In this section, we intentionally set aside caMS configurations whose antecedents contain nonfinite complement clauses. Although it seems likely that the CC in MS also extends to nonfinite CPs (see Barros & Frank 2023: fn.18 for some illuminating examples), more research is required before this can be confirmed.

7 Anand & Hardt & McCloskey (2023) state that a head in an argument domain XP is licensed for ellipsis in sluicing if it has a structure-matching correlate in an antecedent argument domain YP, where 'correlate' refers to a token of the same lexical item and 'structure-matching' refers to occupying the same structural position in XP and YP. According to this licensing condition, both lament in (30b) and each head in write a paper in (30c) are licensed for deletion. Thoms (2015) and others propose that antecedents for ellipsis can be accommodated (cf. Tancredi 1992; Fox 2000; van Craenenbroeck 2013). Assuming that an accommodated antecedent must be mutually entailed by the explicit antecedent (Kotek & Barros 2018; Griffiths 2019, to appear), then some student had a problem with some professor can be accommodated from the explicit antecedent in (27) only in those worlds in which the problem is the student's (which is the only reading available for (27)). Ellipsis is straightforwardly licensed in (30a) from this accommodated antecedent.

8 An anonymous reviewer suggests that the difference in full source availability for (27) and (31) stems from the difference in specificity of a problem and a pile of books: the former "evokes any kind of problem" and therefore fails to introduce a new discourse referent, whereas the latter is specific and, therefore, introduces a new discourse referent. If this is correct, then the contrast between (27) and (31) does not constitute a valid counterexample to B&F's claims. We find this suggestion implausible: a cover-up in (i) is equally as specific a problem in (27)-it evokes any kind of cover-up-and yet no long source interpretation is available for (i), contrary to expectations if the reviewer's suggestion were correct. The reviewer makes a similar suggestion about (33), claiming that somebody in this example introduces a discourse referent and that the full source interpretation becomes available once somebody is replaced with nobody. This paper's second author disagrees with the reviewer's judgment. Furthermore, we point out that (ii) has no full source interpretation despite having a negatively quantified embedded subject DP, which is again unexpected if the reviewer's suggestions were correct.

(i) ?*Every crackpot believes that there is a cover-up happening inside some government department, but I'm unsure which crackpot inside which government department.

(ii) ?*Every critic reckons that nobody has ever died during a certain play, but I can't remember which critic during which play.

As mentioned already in Cortés Rodríguez & Griffiths (to appear), we acknowledge that judgments on English caMS are subtle and clearly subject to idiolectal variation, and therefore view the datapoints in (31) to (33) (and (i) and (ii)) as preliminary evidence. Future experimental work is therefore required to obtain a reliable empirical base for research in this domain.

9 Certain sentences that display grammatical violations can be judged and processed as acceptable, due to a wide range of factors such as flawed retrieval from memory, partial cue-match, the possibility of intermediate analysis in which relevant constraints are satisfied, reanalysis of ambiguous or confusing structures, or purely due to the presence of a syntactic illusion (e.g., Fodor & Ferreira 1998; Fanselow & Frisch 2004; Vasishth et al. 2008; Leivada & Westergaard 2020). We assume that certain grammatical violations inside ellipsis sites fall into this category, following, e.g., Molimpakis (2019).

10 Two additional reasons to disfavor Barros & Frank's (2023) approach are that it fails to predict the presence of an amelioration effect in caMS configurations with bound object pronouns in the embedded clause and that it makes incorrect predictions regarding island-crossing A′-movement (see the data and discussions in §5.1 and §5.2, respectively).

11 For discussion of what types of phrases can function as correlates for single sluicing and why, see Chung & Ladusaw & McCloskey (1995), Ginzburg & Sag (2000), Barker (2013), Barros (2013), AnderBois (2014), among others.

12 Attentive readers will have observed that, in the English examples in (38) and (39), the b-examples are actually ambiguous between a full and short source interpretation, due to the fact that English usually permits P-omission under sluicing. No such ambiguity was present in our German and Spanish test sentences, see §4.2.1 and §4.3.1 for details.

13 Instead of conducting a 2 × 2 experiment and then comparing the results to results obtained in previous experiments (as we do here), one could instead conduct a 2 × 2k (×2k) experiment whose factors are source (full, short), island (island, non-island), and island-type (RC, NCC), where island-type is nested under island (as marked by the subscript "K"). We decided against this option to retain simplicity and sufficient power in our statistical analysis.

14 This preference for PP swh2 has also been observed for English (Richards 2010; Lasnik 2014; Cortés Rodríguez to appear), Dutch (de Vries 2020), French (Gotowski 2022), and Mandarin Chinese (Bai & Cortés Rodríguez & Takahashi 2023) multiple sluicing.

15 Following Gerbrich & Schreier & Featherston's (2019) recommendations, z-scores were calculated using participants' mean ratings (and the standard deviations thereof) for the experiment's standard fillers, as opposed to using participants' mean ratings (and the standard deviations thereof) for the entire dataset.

16 The assumption that ungrammatical sentences and grammatically complex sentences lead to processing difficulties is supported by various studies in psycholinguistics and cognitive science (e.g., Friederici & Pfeifer & Hahne 1993; Vos et al. 2001; Friederici et al. 2006; Hofmeister et al. 2013).

References

References

Abels, Klaus. 2011. Don't repair that island! It ain't broke. Paper presented at the Islands in Contemporary Linguistic Theory conference. University of the Basque Country, Vitoria/Gasteiz.

Abels, Klaus & Dayal, Veneeta. 2017. On the syntax of multiple sluicing. In Lamont, Andrew & Tetzloff, Katerina A. (eds.), North East Linguistics Society (NELS) 47, 1–20. Amherst, MA: GLSA publications.

Abels, Klaus & Dayal, Veneeta. 2023. On the syntax of multiple sluicing and what it tells us about wh-scope taking. Linguistic Inquiry, 429–477. DOI: https://doi.org/10.1162/ling_a_00448

Anand, Pranav & Hardt, Daniel & McCloskey, James. 2023. The domain of matching in sluicing. Linguistic Inquiry, 1–21. DOI: https://doi.org/10.1162/ling_a_00495

AnderBois, Scott. 2014. The semantics of sluicing: Beyond truth conditions. Language 90(4). 887–926. DOI: https://doi.org/10.1353/lan.2014.0110

Bai, Xue & Cortés Rodríguez, Álvaro & Takahashi, Daiko. 2023. An experimental investigation of multiple sluicing in Mandarin Chinese. Languages 8(1). 88. DOI: https://doi.org/10.3390/ languages8010088

Barker, Chris. 2013. Scopability and sluicing. Linguistics and Philosophy 36. 187–223. DOI: https:// doi.org/10.1007/s10988-013-9137-1

Barros, Matthew. 2013. Harmonic sluicing: Which remnant/correlate pairs work and why. In Snider, Todd (ed.), Semantics and Linguistic Theory (SALT) 25, 295–315. DOI: https://doi. org/10.3765/salt.v23i0.2665

Barros, Matthew. 2014. Sluicing and identity in ellipsis. New Brunswick, NJ: Rutgers University dissertation.

Barros, Matthew & Elliott, Patrick D. & Thoms, Gary. 2014. There is no island repair. Ms., Rutgers University, University College London, and University of Edinburgh. Retrieved from http://ling. auf.net/lingbuzz/002100

Barros, Matthew & Frank, Robert. 2023. Attention and locality: On clause-boundedness and its exceptions in multiple sluicing. Linguistic Inquiry, 649–684. DOI: https://doi.org/10.1162/ ling_a_00458

Barros, Matthew & Vicente, Luis. 2016. A remnant condition for ellipsis. In Kim, Kyeong-min & Umbal, Pocholo & Block, Trevor & Chan, Queenie & Cheng, Tanie & Finney, Kelli & Katz, Mara & Nickel-Thompson, Sophie & Shorten, Lisa (eds.), West Coast Conference on Formal Linguistics (WCCFL) 33, 57–66. Somerville, MA: Cascadilla Proceedings Project.

Christensen, Rune Haubo B. 2019. ordinal – Regression models for ordinal data. R package version 2019.12-10. https://cran.r-project.org/web/packages/ordinal/ordinal.pdf

Chung, Sandra & Ladusaw, William A. & McCloskey, James. 1995. Sluicing and logical form. Natural Language Semantics 3(3). 239–282. DOI: https://doi.org/10.1007/BF01248819

Cortés Rodríguez, Álvaro. 2021. Multiple adjacent wh-interrogatives in Spanish. Paper presented Presentation at the II. Encuentro de Lingüística Formal en México. Benemérita Universidad Autónoma de Puebla, Puebla, Mexico. https://doi.org/10.13140/RG.2.2.25419.67363

Cortés Rodríguez, Álvaro. 2022a. Multiple sluicing and islands: A cross-linguistic experimental investigation of the clausemate condition. The Linguistic Review 39(3). 425–455. DOI: https://doi. org/10.1515/tlr-2022-2093

Cortés Rodríguez, Álvaro. 2022b. Multiple sluicing in Spanish: An experimental investigation of the clausemate condition obviation with bound embedded pronouns. Paper presented at the Experimental and Corpus-Based Approaches to Ellipsis (ECBAE) 4, University of Bucharest, Bucharest, Romania. DOI: https://doi.org/10.13140/RG.2.2.35254.16965

Cortés Rodríguez, Álvaro. 2024. An experimental lens on multiple sluicing: Amelioration effects, crosslinguistic variation, and processing. Tübingen: Eberhard Karl University of Tübingen dissertation. DOI: http://doi.org/10.15496/publikation-99065

Cortés Rodríguez, Álvaro. To appear. Which syntactician which kind of ellipsis: An experimental investigation of multiple sluicing. In Konietzko, Andreas & Winkler, Susanne (eds.), Information structure and discourse in generative grammar. Mechanisms and processes (Studies in Generative Grammar 146). Boston/Berlin: de Gruyter Mouton. https://ling.auf.net/lingbuzz/007930

Cortés Rodríguez, Álvaro & Griffiths, James. To appear. An experimental investigation of the clausemate condition in German multiple sluicing. In West Coast Conference on Formal Linguistics (WCCFL) 40. Somerville, MA: Cascadilla Proceedings Project. https://ling.auf.net/ lingbuzz/007931

de Vries, Mark. 2020. An exploration of multiple sluicing in Dutch. Paper presented at the workshop in honor of the defense of Anastasiia Ionova. Leiden University.

Elliott, Patrick D. & Murphy, Andrew. 2019. Unconditional sluicing: An ellipsis identity puzzle. Snippets 35. 3–5. DOI: https://doi.org/10.7358/snip-2019-035-elmu

Fanselow, Gisbert & Frisch, Stefan. 2004. Effects of processing difficulty on judgments of acceptability. In Fanselow, Gisbert & Féry, Caroline & Schlesewsky, Matthias & Vogel, Ralf (eds.), Gradience in grammar, 291–316. Oxford: Oxford University Press. DOI: https://doi.org/10.1093/ acprof:oso/9780199274796.003.0015

Featherston, Sam. 2009. A scale for measuring well-formedness: Why syntax needs boiling and freezing points. In Featherston, Sam & Winkler, Susanne (eds.), The fruits of empirical linguistics, vol. 1, 47–74. Berlin: Mouton de Gruyter. DOI: https://doi.org/10.1515/9783110216141.47

Fiengo, Robert & May, Robert. 1994. Indices & identity. Cambridge, MA: MIT Press.

Fodor, Janet Dean & Ferreira, Fernanda. (eds.) 1998. Reanalysis in sentence processing. Dordrecht: Kluwer Academic. DOI: https://doi.org/10.1007/978-94-015-9070-9

Fox, Danny. 2000. Economy and semantic interpretation. Cambridge, MA: MIT Press.

Fox, Danny & Lasnik, Howard. 2003. Successive-cyclic movement and island repair: The difference between sluicing and VP-ellipsis. Linguistic Inquiry 34(1). 143–154. DOI: https://doi. org/10.1162/002438903763255959

Fox, Danny & Nissenbaum, Jon. 1999. Extraposition and scope: A case for overt QR. In Bird, Sonya & Carnie, Andrew & Haugen, Jason D. & Norquest, Peter (eds.), West Coast Conference on Formal Linguistics (WCCFL) 18. Somerville, MA: Cascadilla Proceedings Project.

Friederici, Angela & Fiebach, Christian J. & Schlesewsky, Matthias & Bornkessel, Ina & von Cramon, D. Yves. 2006. Processing linguistic complexity and grammaticality in the left frontal cortex. Cerebral Cortex 16. 1709–1717. DOI: https://doi.org/10.1093/cercor/bhj106

Friederici, Angela & Pfeifer, Erdmut & Hahne, Anja. 1993. Event-related brain potentials during natural speech processing: Effects of semantic, morphological and syntactic violations. Cognitive Brain Research 1. 183–192. DOI: https://doi.org/10.1016/0926-6410(93)90026-2

Gerbrich, Hannah & Schreier, Vivian & Featherston, Sam. 2019. Standard items for English judgment studies: Syntax and semantics. In Featherston, Sam & Hörnig, Robin & von Wietersheim, Sophie & Winkler, Susanne (eds.), Experiments in focus: Information structure and semantic processing, 305–328. Berlin: de Gruyter. DOI: https://doi.org/10.1515/9783110623093-012

Ginzburg, Jonathan & Sag, Ivan A. 2000. Interrogative investigations: The form, meaning, and use of English interrogatives. Stanford, CA: CSLI Publications.

Gotowski, Megan. 2022. Quoi-sluices in French. Glossa: A Journal of General Linguistics 7(1). 1–44. DOI: https://doi.org/10.16995/glossa.5793

Grano, Thomas & Lasnik, Howard. 2018. How to neutralize a finite clause boundary: Phase theory and the grammar of bound pronouns. Linguistic Inquiry 49(3). 465–499. DOI: https://doi. org/10.1162/ling_a_00279

Gribanova, Vera. 2013. Copular clauses, clefts, and putative sluicing in Uzbek. Language 89(4). 830–882. DOI: https://doi.org/10.1353/lan.2013.0074

Griffiths, James. 2019. A Q-based approach to clausal ellipsis: Deriving the preposition stranding and island sensitivity generalisations without movement. Glossa: A Journal of General Linguistics 4(1). 1–41. DOI: https://doi.org/10.5334/gjgl.653

Griffiths, James. To appear. Questions underpin deletion: A response to Barros and Kotek (2019). In Konietzko, Andreas & Winkler, Susanne (eds.), Information structure and discourse in generative grammar. Mechanisms and processes. Berlin/Boston: Mouton de Gruyter. https://ling.auf.net/ lingbuzz/007569

Hofmeister, Philip & Jaeger, T. Florian & Arnon, Inbal & Sag, Ivan A. & Snider, Neal. 2013. The source ambiguity problem: Distinguishing the effects of grammar and processing on acceptability judgments. Language and Cognitive Processes 28(1–2). 48–87. DOI: https://doi.org/10.1080/016 90965.2011.572401

Kotek, Hadas & Barros, Matthew. 2018. Multiple sluicing, scope, and superiority: Consequences for ellipsis identity. Linguistic Inquiry 49(4). 781–812. DOI: https://doi.org/10.1162/ling_a_00289

Lasnik, Howard. 2014. Multiple sluicing in English? Syntax 17(1). 1–20. DOI: https://doi. org/10.1111/synt.12009

Leivada, Evelina & Westergaard, Marit. 2020. Acceptable ungrammatical sentences, unacceptable grammatical sentences, and the role of the cognitive parser. Frontiers in Psychology 11. 364. DOI: https://doi.org/10.3389/fpsyg.2020.00364

Marušič, Franc Lanko & Žaucer, Rok. 2018. On Sluicing and island repair: Sluicing is neither a ferry nor a bridge. Ms., University of Nova Gorica. Retrieved from https://www2.ung.si/~fmarusic/ pub/mar_zau_sluicing_2018.pdf

Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford University Press. DOI: https://doi.org/10.1093/oso/9780199243730.001.0001

Merchant, Jason. 2008. Variable island repair under ellipsis. In Johnson, Kyle (ed.), Topics in ellipsis, 132–153. Cambridge: Cambridge University Press. DOI: https://doi.org/10.1017/ CBO9780511487033.006

Molimpakis, Emilia. 2019. Accepting preposition-stranding under sluicing cross-linguistically; a noisy-channel approach. London: UCL dissertation.

Nishigauchi, Taisuke. 1998. "Multiple sluicing" in Japanese and the functional nature of the wh-phrase. Journal of East Asian Linguistics 7(2). 121–152. DOI: https://doi. org/10.1023/A:1008246611550

Peirce, Jonathan & Gray, Jeremy R. & Simpson, Sol & MacAskill, Michael & Höchenberger, Richard & Sogo, Hiroyuki & Kastman, Erik & Lindeløv, Jonas Kristoffer. 2019. PsychoPy2: Experiments in behavior made easy. Behavior Research Methods 51(1). 195–203. DOI: https:// doi.org/10.3758/s13428-018-01193-y

R Core Team. 2021. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.r-project.org/.

Richards, Norvin. 2010. Uttering trees. Cambridge, MA: MIT Press. DOI: https://doi.org/10.7551/ mitpress/9780262013765.001.0001

Rodrigues, Cilene & Nevins, Andrew Ira & Vicente, Luis. 2009. Cleaving the interactions between sluicing and P-stranding. In Torck, Danièle & Wetzels, W. Leo (eds.), Romance languages and linguistic theory 2006. Selected papers from "Going Romance", Amsterdam, 7–9 December 2006, 175–198. Amsterdam/Philadelphia: John Benjamins Publishing Company. DOI: https://doi. org/10.1075/cilt.303.11rod

Ross, John Robert. 1969. Guess who? In Blinnick, Robert I. & Davison, Alice & Green, Georgia M. & Morgan, Jerry L. (eds.), Papers from the fifth regional meeting of the Chicago Linguistic Society, 252–286. Chicago: University of Chicago: Chicago Linguistic Society.

Rudin, Deniz. 2019. Head-based syntactic identity in sluicing. Linguistic Inquiry 50(2). 253–283. DOI: https://doi.org/10.1162/ling_a_00308

Sato, Yosuke. 2016. An in-situ syntax of sluicing in Indonesian: Implications for interface economy conditions. Talk presented at CamCoS 5. Cambridge, UK.

Takahashi, Daiko. 1994. Sluicing in Japanese. Journal of East Asian Linguistics 3(3). 265–300. DOI: https://doi.org/10.1007/BF01733066

Tancredi, Christopher. 1992. Deletion, deaccenting and presupposition. Cambridge, MA: MIT dissertation.

Thoms, Gary. 2015. Syntactic identity, parallelism and accommodated antecedents. Lingua 166. 260–293. DOI: https://doi.org/10.1016/j.lingua.2015.04.005

van Craenenbroeck, Jeroen. 2004. Ellipsis in Dutch dialects. LOT, Utrecht: Leiden University.

van Craenenbroeck, Jeroen. 2013. Ellipsis identity and accommodation. Ms., HU Brussels. https://jeroen-vancraenenbroeck.squarespace.com/s/paper-ellipsis-and-accommodation.pdf

Vasishth, Shravan & Brüssow, Sven & Lewis, Richard & Drenhaus, Heiner. 2008. Processing polarity: How the ungrammatical intrudes on the grammatical. Cognitive Science 32(4). 685–712. DOI: https://doi.org/10.1080/03640210802066865

Vos, Sandra & Gunter, Thomas & Kolk, Herman & Mulder, Gijsbertus. 2001. Working memory constraints on syntactic processing: An electrophysiological investigation. Psychophysiology 38. 41–63. DOI: https://doi.org/10.1017/S0048577201980278