2.2. Results

The data were analyzed in the R statistics software, version 3.5.1 (R Core Team 2018) through generalized linear mixed models using the MCMCglmm package (Hadfield 2010) that is suitable for ordinal data. The approach used in this package fits multi-response generalized linear mixed models using Markov chain Monte Carlo techniques, which update the fixed and random effects as a single block, “resulting in better mixing properties and shorter chain lengths than other strategies” (Hadfield 2010: 14). The experimental factors — Topic Function⁴ with two levels (High-Exp and Low-Goal) and Clitic with two levels (Clitic and No-Clitic) — and the interactions between them were entered as fixed effects into the model, using effect coding, that is, the intercept represents the grand median (the initial median, with no effect of the factors) and fixed effects compare the levels of the factor to each other. The median was used as the localization measure since the response variable is ordinal and was treated as such rather than an interval scale. Participants and items were included in the model as random effects. The results are shown in Table 1.

The data from Study 1 revealed a significant effect for the “Function” factor (p < 0.001). In the “dative-Experiencers” sentences higher median overall values were observed when compared to the “dative-Goals” sentences. The range of values for this difference is between 0.626 and 1.444 (CI 95%). There was no significant difference between the levels of “Clitic” factor – Clitic vs. No-clitic sentences (p = 0.240). There were also no significant interactions between the “Function” and “Clitic” factors using α = 0.05 as usual. However, if we consider α = 0.1 the negative interaction corresponds to a median decrease of 0.218 in the responses of the “dative-Experiencers” (in relation to the overall responses). Thus, if there is a significant increase in the median response, this can be explained from the “dative-Goal” level. This can be easily seen in Figure 1 below. The mean (dots) in “dative-Goal” is higher in the “No-Clitic” condition than in the “Clitic” condition and in the “dative-Experiencers” the pattern is the inverse.

Figure 1

Mean and Median acceptability ratings on a scale from 1 to 7 for Study 1.

The contrast between the “dative-Goal” and “dative-Experiencer” structures, shown above, can clearly be seen from Figure 2, below. The distribution of the answers at each of the seven points of the scale reveals that at the extreme points, the pattern of responses for both types of sentences is reversed. At point 1 of the scale (not acceptable), the highest percentage of the total answers is in the “Goal” condition (71.5%) and, inversely, at point 7 of the scale (totally acceptable), the highest percentage is observed in the “Experiencer” condition (70.4%). The rates at point 6 reach 80.2%.

Figure 2

Distribution of acceptability ratings on a 7-point Likert scale in the Study 1 (1 = ‘not acceptable’; ‘7 = totally acceptable’).

In addition to the statistical approach described above for ordinal data, we also fitted a linear mixed model with Gaussian family by using MCMCglmm package. We also used the lmer function with the lme4 package (Bates et al. 2015). This last approach was previously used for acceptability judgment data (e.g., Bader 2018). The results of Table 2 and Table 3 are like those shown in Table 1 (i.e., the results were the same order of size as the earlier analysis and matched the conclusions).

2.3. Discussion

These results confirm that, irrespective of the presence of the clitic, dative Experiencers differ from dative Goals. The latter yield lower ratings. Even though the clitic factor was not statistically significant, an interesting pattern emerges from the data. While the presence of the clitic contributes to lower acceptability rates with dative Goals, dative Experiencers display the opposite pattern.

These observations are in line with the findings by Belletti & Rizzi (1988) on Italian. These authors discussed evidence that dative Experiencers in preverbal position differ from standard left-dislocated objects and argued that they are quirky subjects. One piece of evidence discussed by Belletti & Rizzi (1988) concerns the peculiar behavior of the clitic with these pre-verbal datives. We review this evidence next.

It is a well-known fact that negative QPs cannot be CLLDed (cf. [26b]). When fronted, they must leave a gap in argument position:

Contrasts such as these fall naturally under an approach to CLLD that assumes that CLLDed topics are base-generated directly in place. The dislocated DP introduces an entity and the comment contains an ‘open’ position (a pronominal category) satisfied by the entity referred to by the dislocated DP. Since negative QPs are incapable of picking up a referent, they cannot be CLLDed. Thus, fronting of a nonreferring expression is only possible through movement, in which case it leaves a gap (cf. [26c]).

Interestingly, as originally noted by Belletti & Rizzi (1988), dative Experiencers in preverbal position display a different pattern. In particular, in this case, a non-referential QP is not incompatible with a doubling clitic. This is illustrated by the following examples, which were attested on the web:

A search in Google for the sequence “a ninguém lhe” retrieves a number of examples of this kind and all of them contain stative psychological predicates. This indicates that the status of the clitic that doubles dative Experiencers in preverbal position is different from that of the clitic that doubles left-dislocated Goals.

Belletti & Rizzi (1988) and subsequent works on the topic argue that the case of Experiencers is not structural and is rather “lexicaly” marked, or inherent. If so, then one can think of the clitic in (27) as a case marker rather than a pronominal category. In the case of CLLDed Goals, by contrast, the doubling clitic has the status of a pronominal category. Since, in the case of dative Experiencers, the clitic is (or can be) a mere case marker, no problem arises when it co-occurs with a fronted negative QP. Under this account, the inverse pattern obtained in our test results regarding the effect of the presence of the clitic is no longer surprising, in view of the different status of the clitic in each case. Setting the dative Experiencers case aside, the experimental results indeed show that the presence of the clitic contributes to lower acceptability ratings.

These observations support the thesis that preverbal dative Experiencers do not behave exactly like standard dislocated objects. However, as pointed out by Gutiérrez-Bravo (2006) for similar data in Spanish, it doesn’t necessarily follow from this that these dative Experiencers are quirky subjects.

The term “quirky subject” was originally applied to oblique arguments in Icelandic that behave like surface subjects in every relevant aspect. It was then extended to Italian and Spanish dative Experiencers (Belletti & Rizzi 1988; Masullo 1993). However, Gutiérrez-Bravo (2006) shows that Spanish preverbal dative Experiencers lack the crucial properties that characterize Icelandic quirky subjects. The author discusses three properties that uniquely characterize quirky subjects in Icelandic. These properties were identified by Sigur∂sson (2004) as distinguishing Icelandic quirky subjects from the non-subject preverbal obliques of German (cf. also Zaenen, Maling & Thráinsson 1985). Gutiérrez-Bravo (2006) shows that Spanish has none of these properties, so it is not like Icelandic, but like German, which does not have quirky subjects. In what follows we show that EP patterns with Spanish and not with Icelandic.

The first property concerns person agreement. Icelandic quirky subjects block person agreement between the verb and the nominative argument (Sigur∂sson 2004). Thus, 1st and 2nd person agreement in (28a, b) is impossible:

EP behaves like Spanish (and German) and unlike Icelandic (we refer the reader to Gutierrez-Bravo (2006) for the Spanish data):

The second property concerns Conjunction Reduction, as in (30), where a nominative subject is coreferential with a deleted dative in the second conjunct.

Conjunction reduction with a dative Experiencer verb is impossible in EP (just like in Spanish, as originally pointed out by Masullo 1993):

The third defining property of quirky subjecthood is that quirks can be controllees in infinitival clauses.

As originally noted by Masullo (1993), Spanish preverbal datives do not have this property. Likewise, in the example below, the null subject of the infinitival clause cannot be a dative Experiencer controlled by the matrix subject:

This paradigm shows that preverbal dative Experiencers do not have the behavior of quirky subjects in Icelandic.

Coming back to our experiment, the assumption that preverbal datives are subjects in EP would predict that the status of our test sentences with an intervening dative Experiencer should be no different from that of (2a), repeated here as (34), with a preverbal subject.

However, this prediction is not quite confirmed given that speakers’ ratings of the sentences with pre-verbal dative Experiencers are not exactly optimal, with ratings ranging from 1 to 7 and no detectable difference with respect to preverbal Goals in the middle points of the scale (cf. Figure 2). Therefore, these preverbal datives do not quite pattern with preverbal subjects either.

One other fact that constitutes an argument against a quirky subject analysis of these dative Experiencers is the observation that, within the Romance languages, only the null subject languages have been argued to have quirky dative subjects. As noted by Landau (2009: 89–90), French doesn’t allow “quirky” dative Experiencers: “Preverbal datives in French are always dislocated (as topic or focus) and cannot switch with the nominative DP.”

This fact is surprising on a quirky subject analysis. If the preverbal dative Experiencers that are found in EP, Italian and Spanish were quirky subjects, it is not at all clear why they should be absent from French, a closely related language. Belletti & Rizzi (1988) claim that this is due to the fact that, in French, nominative case cannot be assigned to the right. However, the construction know as Stylistic Inversion displays nominative subjects in postverbal position:

If, on the other hand, preverbal dative Experiencers are topics in all of these languages, the issue doesn’t arise. Since, in a null-subject language, Spec,TP doesn’t need to be overtly filled and the grammatical subject may be post-verbal (cf. [37]), a sentence with a topic Experiencer and a postverbal subject will always have the appearance of a sentence with a quirky subject (cf. [38]). This is not so in French, where the EPP forces overt subject movement to SPec-TP.

Landau (2009) proposes that Experiencers denote mental locations and that constructions with preverbal dative Experiencers are instances of locative inversion, which induces presentational focus on the postverbal DP. However, in English locative inversion constructions, the inverted locative blocks extraction, so a straightforward extension of the locative inversion analysis to our case is of no help. In fact, Landau (2009) argues that locatives in English locative inversion are subjects that must move to topic position because PPs cannot be subjects in English. According to him, the availability of quirky subjects in Italian and Spanish allows for Experiencer datives to raise to Spec-TP and stay there. However, there is actually no independent evidence that PPs can be subjects in EP (on Spanish, see the discussion in Gutiérrez-Bravo, 2006). Since a quirky subject analysis of preverbal dative Experiencers in EP (or Spanish, for that matter) is problematic, for the reasons stated above, we do not adopt this analysis. We note, however, that, in a null subject language, topicalization of a dative Experiencer argument will in reality have the same interpretative effect as locative inversion. The dative Experiencer in topic position is assigned prominence and, since the grammatical subject may stay in situ, to the right of V raised to T, it will be the most deeply embedded constituent and hence be assigned information focus (=presentational focus).

Dative Experiencer arguments differ from dative Goals or direct objects in two ways: they are assigned inherent case and, in the base, they are projected higher than the nominative argument (the Theme). Thus, on second inspection, there is actually no reason to expect that dative Experiencer topics should pattern exactly like lower object topics. In fact, the rejection of a topic analysis of preverbal dative experiencers is grounded on the assumption that all left-dislocated topics are expected to pattern alike with regard to extraction. However, this assumption is factually wrong. Barbosa & De Cat (2019) carried out a study designed to determine how native speakers of French rate examples with different types of CLLDed topics embedded within the domain of wh-movement. Test items contained relative clauses and indirect questions. The syntactic position of the intervenor was manipulated so that it was either a canonical subject (39), a dislocated subject (40), a dislocated object (41) or an adjunct (42) (Barbosa & De Cat 2019: 4):

Stimuli were presented orally, and each test item was preceded by a short context. An ordinal regression analysis revealed a consistent picture, of which we highlight the following conclusions:

These results show that embedding a dislocated subject within a wh-movement domain yields better results than embedding a dislocated object, regardless of the type of structure (i. e., relative clause or indirect question). Unsurprisingly, there is a partition between dislocated subjects, on the one hand, and adjuncts or subjects, on the other. However, the significant split between subjects and objects within the group of structures involving CLLD is something that, to our knowledge, has never been noticed before. This subject/object asymmetry indicates that height of the base position associated with the topic matters in determining whether it may or may not appear within the domain of wh-movement.

These data also argue against an analysis that attempts to explain this subject/object asymmetry by claiming that subject clitics are agreement markers, in which case the fronted DP is a subject rather than a CLLDed Topic (Culberston 2010). The fact that subjects are less disruptive of wh-chains than CLLDed subjects would remain unexplained under an agreement marker analysis of subject clitics.⁷

This subject object asymmetry could be attributed to the particular status of subjects as natural topics. Since subjects are natural topics, subject CLLD is less marked that object CLLD.⁸ This kind of explanation, however, would require the additional proviso that only marked topics block wh-extraction, a non-trivial matter. Here we will pursue an alternative explanation.

A close look at the configurations involved in (40) versus (41) reveals that they differ from each other in the way the scopes of the wh-item and the topic interact. Schematically, (40) corresponds to the structure shown in (44a) and (41), to the structure shown in (44b).

While (44a) involves nested dependencies, (44b) involves intersecting dependencies.

With this much as background let us now reconsider the configurations involved in our EP study. Examples (23a,b) and (23c,d), reproduced here as (45a) and (45b), respectively, also differ from each other in the way the scopes of the wh-item and the topic interact. While (45a), with a dative Experiencers topic, displays nested dependencies (45b), (46a) exhibits intersecting dependencies (46b):

We thus observe that the EP data are consistent with the French data. In both cases, we find a similar pattern: the structures that involve nesting dependencies are rated significantly higher than those that display intersecting dependencies.

In fact, nested dependencies have been argued to be favoured over crossing dependencies quite generally (Bach 1977; Baker 1977; Fodor 1978; Kaplan 1973). The following typical examples in (47) and (48) where wh-movement and Tough-movement are both applied, are illustrative:

Baker (1977: 63) explicitly stated this constraint in terms of processing:

Fodor (1978) formulated the constraint as an anti-ambiguity parsing strategy:

Finally, Pesetsky (1982: 309) reaffirmed this condition in the form of a syntactic constraint on A’-movement:

Pesetsky shows that the generalization in (51) has broad empirical scope.⁹ In order to verify whether the no crossing constraint is indeed operative in these structures, a second grammaticality judgment task was designed. The goal of this study was to test for the two types of dependencies, crossing versus nesting.

3.2. Results

The same statistical approach described for Study 1 was used to analyze the results of this experiment. The data revealed a significant effect for the “Wh-height” factor (p = 0.006). In the “Wh-dative” sentences higher median overall values were observed when compared to the “Wh-subject” sentences. The interval for this difference is between 0.084 and 1.100 (CI 95%). This time, however, there was a significant difference between “Clitic” vs. “No-clitic” sentences (p < 0.001); the “No-clitic” sentences received significantly higher rates compared to those in which the clitic was present (p = 0.012). There were no significant interactions between the “Wh-height” and “Clitic” factors (α = 0.05; p = 0.510). The results are shown in Table 4.

The pattern of responses represented by the means and the medians can be seen in Figure 3. Note that for both “Wh-dative” and “Wh-subject” the mean in the “No-Clitic” condition is higher than in the “Clitic” condition.

Figure 3

Mean and Median acceptability ratings on a scale from 1 to 7 for Study 2.

The overall acceptability rates of the experimental stimuli were low. A higher percentage of responses at the lowest points of the scale (1–3: Wh-dative, 58.6%; Wh-subject, 70%) than in the highest points (5–7: Wh-dative, 26.7%; Wh-subject, 15%) was observed. At the midpoint (4) the response rate was 14.7% for the dative conditions and 15% for the Wh-subject conditions.

Nonetheless, similarly to the previous study, the distribution of the answers at each of the seven points of the scale reveals that at the extreme points, the pattern of responses for both types of sentences is reversed (Figure 4, below). At point 1 of the scale (not acceptable), the highest percentage of the total answers is in the “Wh-subject” condition (65.5%) and, inversely, at point 7 of the scale (totally acceptable), the highest percentage is observed in the “Wh-dative” (64.4%) condition.

Figure 4

Distribution of acceptability ratings on a 7-point Likert scale in the Study 1 (1 = ‘not acceptable’; ‘7 = totally acceptable’).

As in Study 1, we also used a linear mixed model with Gaussian family by using MCMC method (Table 5) and the lmer function with the lme4 package (Table 6). Again, the results were similar to those observed when using the MCMC with ordinal family (cf. Table 4).

3.3. Discussion

The results of this study indicate that there is a statistically significant preference for dative relative clauses over subject relative clauses containing an object topic. We consider these findings particularly striking in light of the well-known observation that, in head initial languages quite generally, subject relative clauses are preferred over object relative clauses on a number of different measures (Ford 1983; Frauenfelder, Segui & Mehler 1980).

Compared to Study 1, the results of this study show a more marked concentration of the ratings at the lowest points of the scale, suggesting greater consistency among participants in their judgments than that seen in Study 1. Yet another result that is different from the previous study is that the clitic factor is statistically significant, with the presence of the clitic contributing to lower ratings in both conditions.

We start by noting that this preference for embedding a topic within the domain of a wh-dative chain over a wh-subject chain is consistent with the French judgments reported by Rizzi (1997: 306):

Recall that, for Rizzi (1997; 2004) topics do not count as intervenors in a wh-movement chain. This asymmetry between wh-subject and wh-dative chains is rather due to the ECP. The ECP requires that traces be properly head-governed. A trace in complement position is properly head-governed (by the verb), but a trace in subject position normally is not, unless C is turned into a governor by agreeing with the wh-subject. This is why the agreeing form of C (qui) must be used in cases of subject extraction in French. Under the assumption that topics are introduced by a Topic head, the representation of an example such as (55b) is as follows (Rizzi 1997: 307):

According to Rizzi, even if C is turned into a governor via agreement, it is too far away to license the subject trace due to the intervening Top head, a standard case of relativized minimality. Therefore, the structure is ruled out as an ECP violation. (55a), by contrast, is a mere subjacency violation.

This ECP-based account predicts that structures comparable to (55b) and (56b) should be fine in a null-subject language such as EP. As is well-known, many null subject languages lack that-trace effects, as shown in (56) for EP:

The grammaticality of examples such as (58) has been attributed to the fact that the null subject languages may avail themselves of a strategy that is unavailable in English or French, namely extraction from post-verbal position, which is head governed by V raised to T (Raposo 1992). In view of this, the ECP-based account predicts that the EP counterparts to (55b) and (56b) should be fine. However, this prediction is not confirmed. Our study shows that structures comparable to (56b) are not judged as acceptable in EP and that the contrast between (56a, b) is replicated in EP. Therefore, the ECP-based account cannot be right.

Additional evidence against an ECP-based account of the contrasts in (55), (56) comes from the observation of French examples with wh-adjuncts and embedded CLLD, such as the following, which are taken from Barbosa & De Cat (2019: 4)

In the acceptability judgement elicitation task carried out by Barbosa & De Cat (2019), native speakers of French judged (59b) worse than (59a). On a scale of 1 to 10, mean acceptance rating for the intervening CLLDed object condition was around 4, whereas mean acceptance rating for the intervening CLLDed subject condition was above 7.5 (Barbosa & De Cat 2019: 8). This difference between subject CLLD and object CLLD is not expected under an ECP based account. Adjunct traces must be properly governed. Therefore, if the Top head is what prevents proper government of the adjunct trace in (59b), (59a) should be just as bad, contrary to fact.

On the other hand, a close look at (59a,b) shows that these examples differ from each other in the way the two dependencies interact. If we assume that the trace of the temporal adjunct occupies a position that is lower than the subject, but higher than the object (Cinque 1999; Laenzlinger 1988), then (59a) involves nesting (60a) and (59b) displays crossing (60b) A-bar dependencies:

A similar account carries over to the contrasts in (55) and (56). (55a, 56a) display nesting dependencies while (55b, 56b) exhibit crossing dependencies. Thus, in all of the multiple dependency constructions examined the illicit examples involve crossing dependencies while the others involve nesting dependencies. We therefore conclude that a restriction against crossing dependencies is operative in these constructions, with one important qualification. The restriction against crossing only applies to configurations in which the topic appears to the right of the wh-phrase, i.e., not in cases like (61), in which the topic precedes the wh-constituent (this holds independently of the presence of the clitic).

(61) involves crossing dependencies and yet it is acceptable. Moreover, the constraint also doesn’t apply to dependencies established between different topics. In general, there are no ordering restrictions on multiple topics, regardless of whether they are CLLDed (62) or simply topicalized (63).

This suggests that the constraint in question should be restricted to apply to chains created by wh-movement. The following descriptive generalization adequately captures the patterns observed:

As argued in Barbosa & De Cat (2019), it is not clear how to derive (64) from any familiar syntactic constraint. On the one hand, the fact that topics are not subject to the no-crossing constraint is consistent with the idea that they are base-generated in place (Anagnostopoulou 1997; De Cat 2007; Dermidache 1992; Raposo 1998).¹⁰

However, if topics are base-generated in place, one cannot appeal to a constraint on movement, such as Pesetsky’s Path Containment Condition or any of its current instantiations, such as the Minimal Link Condition, to rule out the cases that do not fall under (64). Conversely, if we do assume that CLLD involves movement and that (64) is derived from a ban on intersecting movement operations, then (61), (62a) and (63a) are predicted to be ruled out, contrary to fact. In other words, we have no explanation for why the ban on intersecting dependencies applies whenever the wh-constituent precedes the Topic and not when the reverse order obtains. For these reasons, we will explore an account of (64) that doesn’t rely on principles of narrow syntax.

A purely syntactic account of the relevant contrasts faces yet one additional challenge. While full DPs are disruptive of wh-chains in crossing configurations, pronouns are not. In other words, CLLDed pronouns appear to be exempt from (64). In the study by Barbosa & De Cat (2019: 21) on French, 1st and 2nd person pronouns are accepted as intervenors even when the clitic they are associated with is not a subject. Thus, examples such as (65) received a very high acceptability score:

Structurally, (65) is similar to (56b). In our study on EP, we didn’t include pronouns, but a cursory search in Google retrieves a number of examples of subject relative clauses containing a CLLDed pronoun (66a–d):

These examples have precisely the same structure of (54a), which was rejected by our subjects.

One might speculate that the peculiar behavior of pronouns is due to Clitic Doubling. In effect, in French as well as EP, direct and indirect object strong pronouns are necessarily clitic doubled. However, it is not possible to deny that the pronoun is left-dislocated in these examples, given that it appears in the left-periphery. Configurationally, there are no differences between the structures containing CLLDed pronouns and structures containing non-pronominal CLLDed constituents.

Another possibility would be to attempt to derive the exceptional behavior of pronouns from the fact that they posssess the feature Person, which is not present in full fledged DPs. This approach, however, would only work under a minimality account of the deviance of (54a, b). A minimality account, however, would assign the same status to (54a, b) and (54c, d), contrary to fact. This is why we believe that (64) is not amenable to explanation on the basis of syntax alone.

On the other hand, there is some indication that (64) could be fruitfully investigated as deriving from processing constraints. In fact, two aspects of our data that are problematic for a purely syntactic account — the fact that the restriction against intersecting paths applies to wh-movement chains and not to topic-chains; the fact that CLLDed pronouns are exempt — are better understood when (64) is considered from a processing perspective.

Intuitive judgements of acceptability are known to be affected by processing difficulties (Frazier 1985; Gibson 1998; 2000; Kimball 1973; Miller & Chomsky 1963; Vasishth et al. 2010; Yngve 1960). Multiple center-embedded sentences are one example of structures which are judged unacceptable by virtue of a processing overload effect. Sentences with one level of embedding are judged as being grammatical (67a), while additional levels of embedding are rated ungrammatical and unacceptable (67b).

The contrast above shows that processing overload may affect intuitive judgments of acceptability. Processing overload has been attributed to limitations of the computational resources of the language processor. It is now a well-established fact that dependencies of the kind investigated in this paper consume a certain amount of processing resources (Gibson 1998; 2000; Fiebach, Schlesewsky & Friederici 2002; King & Just 1991). In particular, it has been demonstrated that, once a filler is encountered, comprehenders anticipate the location of potential gap sites and attempt to construct dependencies in advance of information about the gap position, a phenomenon that came to be known as active dependency formation (Crain & Fodor 1985; Frazier & Clifton 1989).

In addition, several electrophysiological studies have provided evidence of the memory cost of keeping an open dependency. In a study of Event-Related Potentials (ERPs) registered during the processing of subject and object questions, Kluender & Kutas (1993) found that object questions elicited a larger left anterior negativity (LAN) at the filler and gap positions. Similar findings were obtained by Fiebach, Schlesewsky & Friederici (2002). In this study, ERPs were recorded while participants processed case-unambiguous German subject and object wh-questions with either a long or short distance between the wh-filler and its gap. A sustained LAN was observed for object questions with long filler-gap distance but not for short object questions. The authors interpreted the sustained negativity as reflecting working memory processes required for maintaining the displaced object in memory. A second effect that was found was a broadly distributed positivity elicited for object as opposed to subject wh-questions at the second NP for both short and long object-wh-questions. Parietal positivity, also termed P600, has been observed in response to increased integration difficulty (Kaan et al. 2000), so Fiebach, Schlesewsky & Friederici (2002: 268) interpreted this effect as a reflection of the difficulty of local integration processes associated with the gap position in the sentence.

Building on these results, Felser, Clahsen & Münte (2003) recorded ERPs during the processing of unambiguous German sentences containing different types of filler-gap dependency: topicalization constructions and wh-questions. Both topicalization constructions and wh-questions were found to elicit a LAN prior to the processing of the subcategorizing verb. At the subcategorizing verb, sentences containing a wh-dependency produced a parietal positivity (P600). Topicalization structures did not produce this effect. These results constitute further evidence for the notion of separable parsing processes, with memory-based processes being manifested in terms of LAN, and the relative difficulty of integrating the filler with its subcategorizer manifested as P600. The fact that the size of the observed LAN was not influenced by the type of filler suggests that the working memory cost induced by processing filler-gap dependencies is independent of the type of syntactic dependency involved. Integration cost is influenced by the type of the filler-gap: displaced wh-phrases are more costly than topicalized elements. According to Felser, Clahsen & Münte (2003), integration cost is higher in the wh-movement condition because in addition to semantically integrating the filler with its subcategorizer, an operator variable dependency must be upheld at the same time for the sentence to be assigned the correct interpretation.

These findings are relevant for our purposes, because they constitute evidence that there is a difference in integration cost between topic fillers and wh-fillers. If (64) is related to processing constraints and if the integration costs of topic dependencies are lower than those incurred by displaced wh-phrases, it is no longer surprising that (64) should apply to wh-dependencies and not to topic-(cl)-gap dependencies (note that topics introduce a referent in discourse; wh-operators introduce operator variable dependencies).

The second fact that we considered problematic for a syntactic account of (64) concerns the difference between pronominal and phrasal intervenors. This effect, however, is unsurprising from the point of view of processing. Indeed, a similar asymmetry between pronouns and full DPs has been observed in a variety of phenomena in the literature on language processing. In the case of center-embedded sentences, for example, their intelligibility is increased if the subject of the most embedded relative clause is a pronoun, as in (68a) rather than a full DP, as in (68b) (cf. Bever 1970; Gibson 1998; 2000; Warren & Gibson 2002).

Warren & Gibson (2002) carried out a judgment task evaluating how sentences like (iiib) compared to their counterparts with first and second person pronouns in the most deeply embedded subject position. They found that the latter were rated significantly higher in acceptability than the former. Different accounts of this contrast can be found in the literature on processing complexity, but all of them converge on the idea that pronoun intervenors are less disruptive of structural integration of a non-local dependent than full DPs. Therefore, we take the difference between pronominal and phrasal intervenors as evidence in favor of the idea that structural integration cost is the key factor affecting wh-chains containing intervening topic fillers.

For these reasons, we believe that deriving (64) from the processing integration cost of wh-fillers is a promising line of approach. Viewed from a processing perspective, generalization (64) amounts to the claim that maintaining an incomplete (topic) dependency while processing a wh filler-gap dependency is costly for the human processor. By contrast, if the topic filler-gap dependency is complete by the time the wh-gap is encountered, performing the integration of the wh-gap is easier.

One result of the present study that would receive a natural explanation on these grounds is the variable effect of the presence of the clitic in the different studies. In the case of Study 1, our results indicate that the factor “clitic” was not statistically significant. However, a close look at Figure 1 shows that the mean in the dative-Goal context is higher in the “No-Clitic” condition than in the “Clitic” condition. In Study 2, the factor “clitic” was statistically significant. In both cases, the presence of the clitic contributed to lower ratings.

In a series of experiments on the online processing of CLLD constructions in Spanish, Pablos (2006) provides evidence that left-dislocated topic fillers are reactivated at the clitic, before the verb is encountered. Pablos (2006) acknowledges that, even though the topic is reactivated by the clitic in preverbal position, it is stored and maintained as an incomplete dependent until its subcategorizer (or associated gap) is encountered. If this is true, then by reactivating the representation of the topic, the presence of the clitic immediately before the verbal complex has the effect of bringing the topic to the focus of attention. By hypothesis, this increases the interfering effect of the topic on retrieval and integration of the wh-filler, particularly in those contexts in which the gap associated with the topic and the gap associated with wh-phrase are coarguments (the dative Goal cases of Study 1 and the test items of Study 2).

We conclude by observing that we do not claim to have reached an explanatory account of the patterns obtained. We established an empirical generalization and we formulated a hypothesis thus laying out the ground for further research on the online processing of multiple dependency constructions in the left-periphery.