Circumventing the ‘That-Trace’ Effect: Different Strategies between Germanic and Romance

: In our paper, we deal with the Germanic–Romance language contact, focusing on Cimbrian, a Germanic minority language spoken in Northern Italy. Speciﬁcally, we focus on the violation of the well-known that-trace ﬁlter, as it appears to be an interesting case of the superﬁcial convergence that we ascribe to the status of T, which is either too rich (model language) or too weak (replica language) to represent a viable landing site for subject extraction.


Introduction
Cimbrian 1 is a German(ic) minority language which has long been in contact with Romance varieties in Northeast Italy. It belongs to the group of Southern Bavarian-Austrian dialects from which it has been isolated since the 11th Century (see Bidese 2004); however, it has preserved some morpho(phono)logical features that are common with its medieval cognates (see, among others, Bidese 2010). The three varieties of Cimbrian currently spoken are: (i) the Luserna variety (Province of Trentino-Alto Adige/Südtirol); (ii) the one in the so-called area of the Tredici Comuni (lit. "Thirteen Municipalities") in the Province of Verona (Veneto Region)-where Cimbrian is essentially alive in the village of Giazza/Ljetzan only); (iii) the one in the so-called area of the Sette Comuni (lit. "Seven Municipalities"), close to Asiago, in the Province of Vicenza (Veneto Region)-where only a few Cimbrian speakers are found in the village of Roana/Robaan. The data discussed in this work have been collected in Luserna, i.e., the only variety actively spoken in a community of ca. 300 inhabitants.
The most remarkable aspect in the syntax of Cimbrian at the sentence level is that it has lost the core property of the verb second (V2) phenomenon, i.e., the well-known linear restriction, found in the Continental West-Germanic and Scandinavian languages (see Bidese 2008). However, it has retained facets of its Germanic origin in maintaining the mandatory V-to-C (/FinP, assuming a cartographic perspective on the C-layer) movement in root clauses with residual effects on both subject inversion and the root-embedded word order asymmetry (limited to a specific class of lexical complementizers) (for more details, see Bidese et al. 2020).
Moreover, Cimbrian resembles other "well-behaved" Germanic languages, as it is a non-pro-drop language (cf. 1a, below) (see Bidese 2008;Bidese and Tomaselli 2018), different from all the Romance varieties with which it has been in contact for centuries. In fact, Cimbrian has maintained a lexical expletive with both weather verbs and as a correlate of a postponed subject clause. In this paper, we deal with the unexpected fact that Cimbrian, in fact, aligns with Romance, when it comes to the that-t effect. Pro-drop languages, such as Italian, are known to bleed the that-t effect due to free subject inversion: this property has been assumed to be a corollary of the null subject parameter since Rizzi (1982). We will show that Cimbrian displays a special kind of subject inversion, which combines properties that typologically belong to both German and Italian, circumventing the that-t effect in an original way, with an interesting overlapping with complementizer agreement in Bavarian. We argue that what all these varieties have in common is a "weakened" status of [Spec, TP]-even if this is due to the opposite value of the null subject parameter in each language.

V2 and Subject Inversion in Cimbrian
Cimbrian is taken to be a so-called "relaxed" V2 language (see Bidese et al. 2012a andBidese et al. 2012b), i.e., a language with obligatory V-to-C movement, not displaying the expected linear restriction that is usually found in the Germanic languages, with the well-known exception of Modern English: this means that more than one constituent can occur in front of the finite verb, and the subject shows up among them in Cimbrian. In fact, independent of how many phrases are fronted, the unmarked positions for DP subjects and full pronouns always correspond to the preverbal one, cf. the italicized phrases in (1a,b): Even if the finite verb does not show up as a second constituent (cf. 1 and 2), there is strong evidence to assume that it moves into the left periphery of the sentence, i.e., C, as it is generally taken for Germanic varieties starting from den Besten (1983): in fact, Cimbrian clitic subject pronouns are realized on the right of the finite verb, in the so-called Wackernagel position, and encliticized onto the finite verb ( (4) and (5) feature a special element, i.e., the segment -ta, an allomorph of the enclitic particle -da, which plays a crucial role in our story and will be focused on in the next section: In (4) and (5), the subject always shows up postverbally; however, it receives a presentative reading when it occurs with an unaccusative verb; in transitive constructions, it can be interpreted either as a contrastive focus (5a) or as a new information focus (5b) (see Bidese and Tomaselli 2018, for more details).
The inversion with the whole verbal complex (=VP DP) is a well-known phenomenon in Italian (cf. Rizzi 2004or Belletti 2004, among many others), whereas in the Germanic languages, it is marginal and usually analyzed in a different way: for instance, English postverbal subjects associated with the expletive there (such as There arrived a man) are assumed to be VP-internal, provided that there is a definiteness effect on the subject. On the contrary, postverbal subjects in Italian can even be extraposed, as in Ha comprato il giornale il nonno (for a detailed analysis on the DP subject position in Cimbrian, we refer the reader to Bidese and Tomaselli 2018).
Interestingly, the examples (4) and (5) resemble their Italian counterparts, except for the mandatory presence of the enclitic expletive particle -da; the next subsection is devoted to the morphosyntactic properties of this element of the Cimbrian syntax.

The Role of the Enclitic Particle-da
As we pointed out in Section 2.1, the DP subject can occur in a pre-verbal position (cf. 1a); or, it inverts with the whole VP (cf. 4) cooccurring with the enclitic expletive -da. Generally, this holds true in the case of unaccusative verbs (6a,b); however, it can be generalized to transitive constructions as well. See (7), where (7b) displays a context of contrastive reading.

A Hybrid Subordination System
As regards the embedded declarative clauses in Cimbrian, both traditional grammars (see Panieri et al. 2006) and formal analyses (see Grewendorf and Poletto 2009;Padovan 2011;Bidese et al. 2012a;Bidese et al. 2013;Bidese and Tomaselli 2016) point out that the complementation system is hybrid in the sense that it consists of two different comple-mentizers corresponding to the English that: an autochthonous one, az, which hosts the unstressed pronominal forms and the enclitic expletive -da, triggering the embedded word order (see 8), and a Romance loanword, ke, which cannot host any enclitic elements and manifests the same word order as in the main clauses (cf. 9). Typically, nonfactive (e.g., volitional verbs such as bölln, 'want') and non-assertive verbs (sperarn, 'hope') and negated forms (bridge verbs such as net gloam, 'think that not V') select az: in this case, Vfin undergoes a shorter movement than it does in the main clauses, as it follows Neg and sentential adverbials (see Padovan 2011). Moreover, az hosts morpho-phonologically reduced pronominal forms and the enclitic expletive subject -da (see Kolmer 2005;Bidese et al. 2012b;Bidese and Tomaselli 2018;Bidese et al. 2020).
The second complementizer is ke: clearly, it is a loanword from Italian/Romance. It is selected by assertive verbs (bizzan, 'know'), perception verbs (seng, 'see/observe'), and weak assertive verbs (pensarn, 'think'). Ke has been assumed to be a subordinator rather than a full-fledged clause-typer, as it is in the case of az. In fact, contrary to az, it does not affect the word order. Furthermore, ke-introduced clauses typically display root phenomena such as postverbal negation and subject-verb inversion. Finally, ke cannot host either clitics or the expletive -da.
The dichotomy non-factive matrix verb-az-Neg Vfin [Mood: SBJV] vs. factive/assertive matrix verb-ke-Vfin [Mood: IND] Neg reminds us of the well-known difference between non-assertive and assertive complement clauses in the Scandinavian languages, where exactly the same word order patterns are found (Neg V vs. V Neg); although-in contrast to Cimbrian-the complementizer (at) and mood (indicative) are used in both contexts (see Bidese 2017).
In fact, Norwegian and Swedish clearly show a V > Neg vs. Neg > V asymmetry between assertive and non-assertive complements. While V > Neg introduces clauses with V2 structure (cf. 10a and 11a), Neg > V does not (cf. 10b and 11b) (the data are taken from Wiklund et al. 2009 After presenting the Cimbrian subordination system, it is interesting to note that, in embedded clauses introduced by az, the expletive particle -da is still mandatory, regardless Languages 2021, 6, 84 5 of 13 of the position of the DP subject (see 12a and 12b). On the contrary, in embedded clauses introduced by ke, -da is present only when the DP subject shows up postverbally, exactly as in the case of main clauses (cf. 13a vs. The data in (12) und (13) show that the expletive -da is obligatory whenever the DP subject does not leave the v-layer in a postverbal position in the main clauses and in the subordinates introduced by ke; it also does not leave the v-layer in a preverbal position in instances where az is the complementizer. For our purposes, it is not substantial to determine whether the DP subject is VP-internal or extraposed. 3 What is crucial is that the DP is not raised to [Spec, FinP], which is always the case in the embedded clause introduced by az (see Bidese and Tomaselli 2018).

Absence of That-Trace Effect in Cimbrian
A well-known fact regarding the that-trace effect (henceforth that-t) is that it correlates with the null subject parameter (see Rizzi 1982 and much subsequent work): a prototypical pro-drop language, e.g., Italian, does not display that-t effects, whereas English-as a prototypical non-pro-drop language-does. Because of this correlation with the Null Subject Parameter, we expect the presence of the overt complementizer to violate the that-t filter. However, bearing in mind the parallel between Cimbrian and Italian with respect to low (i.e., non-raised) DP subjects (6-7-12-13), we might surmise that Cimbrian allows subject extraction from the postverbal position in the embedded clause, independent of the verbal class (with both unaccusative and transitive verbs). In other words, there should be no that-t effect in the same fashion as in Italian. This is indeed confirmed by our data (cf. 16a) with the complementizer az. Notice, however, that the absence of the that-t effect is not affected by the choice of the complementizer (cf. Section 2.3); i.e., it is independent from whether or not the embedded clause is introduced by az or ke: As already pointed out, it is not relevant from which position ber is extracted: what is important here is that [Spec, TP] is not targeted by successive-cyclic movement.
As a matter of fact, you might recall that Cimbrian does not allow the verb-subject inversion found in German, where the DP-Subject is merged in [Spec, TP], as shown in (2) and repeated here as (17) As we have seen in Section 2.3, ke does not trigger the matrix/embedded asymmetry, as is the case with az. This is also confirmed by the presence of the expletive 'z (corresponding to the German Vorfeld es), while the enclitic expletive -da has to be taken as the signal of the lower (/vP internal) position of the extracted subject (see Bidese and Tomaselli 2018).

The Proposal
Pesetsky (2017) distinguishes between linear accounts and structure-based accounts of the that-t effect: the former accounts bar the extraction from a position right-adjacent to COMP; the latter accounts focus on the particular extraction sites and ban movement that are too local. In particular, quoting Pesetsky, "Most of structure-based accounts build on the independent observation that successful A-bar-movement appears to proceed successivecyclically through the edges of domains such as CP. Complementizer-trace effects are attributed [ . . . ] to some consequences of movement from the subject position to a position in the complementizer system that is affected by the presence, absence of featural content of the complementizer that such movement crosses".
Our proposal clearly aligns with the "extraction site" of accounts: in fact-as we have already emphasized-the low position from which the subject is extracted is the necessary condition for the that-t effect to be circumvented. However, this does not suffice; in fact, the presence of the expletive particle -da plays a fundamental role in characterizing the featural content of the COMP. In comparative terms, Cimbrian -da corresponds to the so-called "expletive pro" 4 in German, which is assumed in (i) impersonal passive constructions (see 19a,b); (ii) unaccusative verbs (20a,b); and (iii) passive (21a,b): Furthermore, the expletive -da provides morphological evidence to the hypothesis that a subject extraction from an embedded clause, introduced by a lexical complementizer, implies a copy in the vP-internal position. In other words, -da manifests the morphological exponent of the Agree relation between the probe C and the low DP subject (see Bidese and Tomaselli 2018;Bidese et al. 2020). We argue that C probes the low subject, as T is assumed to be deficient: in fact, it does not act as a probe, nor does it attract the finite verb; hence, it is not the locus of nominative case assignment. Consequently, [Spec, TP] is not projected in Cimbrian as the absence of a German-like verb-subject inversion confirms (cf. 3); for further details, we refer the reader to Bidese et al. (2020).
In Italian, T is endowed with tense features, which are rich enough to both incorporate the null subject pronoun and satisfy the EPP feature (see also Holmberg 2010, p. 99). Consequently, [Spec, TP] is not projected in Italian, as the rich verb morphology can be assumed to contain the features that characterize a pro (see Biberauer and Roberts 2010). In particular, Biberauer (2010, p. 164) proposes a refinement of this assumption, decomposing the tense feature in two sub-feature sets, both connected with verbal morphology, i.e., its nominal strand [D] (as in person and number) and its verbal strand [V] (as in tense, aspect, and mood). Thus, for different reasons, [Spec, TP] is inactive both in Italian and in Cimbrian and cannot function as an intermediate position for the extracted elements: in both languages, we expect a (definite) nominal subject to be extracted directly from either a vP internal or an extraposed position.
If [Spec, TP] is not involved in subject extraction or in NOM assignment, the lexical complementizer does not act as an intervener: To sum up, Cimbrian -da provides morphological evidence for the inactivation of [Spec, TP]; whereas, in Standard Italian, this follows directly from the positive value of the null subject parameter 5 .

How Bavarian Circumvents the That-T Effect: Morphological Agreement in C
Among German dialects and Germanic languages in general, there is a great deal of variation with respect to that-t effects (see Featherston 2005;Bayer and Salzmann 2013;Weiß 2017;Schippers et al. 2020). In particular, Bavarian is known to have both strategies to extract the wh-subject from an embedded clause, i.e., by either having the C-head hosting the finite verb, as in German (see 23a, 24a), or violating the that-t filter, as in Italian (23b, 24b): As the examples above show, Bavarian appears to be a partial pro-drop language (see Weiß and Volodina 2018), as only the second person singular and the first and second person plural allow for covert subjects; however, it disallows subject extraposition to the right of the verbal complex: The import of the Bavarian data-with particular reference to the violation of the that-t effect-prompts us to further elaborate on some of the previous assumptions: First, we emphasize that the that-t violation seems not to be affected by the VO or OV word orders: On the one hand, the main difference concerns the position of the DP subject with respect to the past participle (which is extraposed in one case and vP-internal in the other); on the other hand, the common aspect is the countercyclic extraction of the wh-subject directly out of the low position, with an intermediate copy in [Spec, TP] that cannot be assigned a case by C. This latter aspect brings the relation between C and [Spec, TP] to the fore. In our view, it is the morphology in C 6 that is responsible for case absorption/case discharging. Notice, however, that this does not imply that the morphological agreement in C is morphologically overt. In fact, contrary to Cimbrian, Bavarian seems to rely on a strategy based on the D feature absorption, which is somehow reminiscent of the Italian strategy. In fact, we presume that Bavarian is essentially characterized by a full paradigm of COMP agreement morphology, in which the first and third person singular agreement are in fact phonetically empty, and Cimbrian -da represents the overt counterpart of the third person (along the lines of Tomaselli and Bidese 2019).
It is reasonable to assume that, in both Cimbrian and Bavarian, an Agree relation does not take place between C and T: in other words, the Cimbrian -da and Bavarian COMP agreement represent the morphological manifestation of CP keeping its features (in the sense of Ouali 2008) and not entering an Agree relation with [Spec, TP].
34. When the agreement between COMP and [Spec, TP] holds, i.e., through the presence of a third person subject pronoun encliticized onto the lexical complementizer, then subject extraction is ruled out in both Cimbrian (28) and Bavarian (29). In fact, the subject pronoun does not represent COMP agreement morphology: As a matter of fact, COMP agreement hinges on the V-to-C movement in Germanic varieties. In the next section, we will deal with the que/qui alternation in French, which represents the Romance counterpart of COMP agreement activated by subject extraction, independent of the T-to-C movement.

The que/qui Alternation in French: A Reappraisal
That the French complementizer que alternates with qui in the contexts of subject extraction has been a well-known grammatical feature in Romance linguistics since Kayne (1975) and Rizzi (1982) and has been recently revived in Mackenzie (2018) Substituting que for qui in (30) would make the sentence ungrammatical. This occurrence of qui is often referred to as "special qui" to distinguish it from the wh-pronoun qui. In fact, classic analyses assume that qui is the nominative counterpart of the interrogative/relative pronoun qui that occurs after prepositions: 31. À qui veux-tu que je le dise? to whom want-you that I it tell 'Who do you want me to tell?' (Mackenzie 2018, p. 33) On the contrary, there is strong evidence that the special qui behaves syntactically similar to que, i.e., like a C 0 . Mackenzie follows the lead of Koopman and Sportiche (2014) in assuming that subject extraction from sentences introduced by qui is necessary, since case assignment cannot occur clause-internally, and the extracted subject has its case feature checked by an external probe. However, Mackenzie refutes Koopman and Sportiche's assumption that qui is a relative pronoun. Mackenzie takes qui to be unequivocally a complementizer, albeit with a more complex morphological structure than que. In fact, there is diachronic evidence that qui can be decomposed to qu(e) + i, where -i is taken to be a reduced form of the pronoun il. Il was originally inserted along with the complementizer as a resumptive pronoun; eventually, it underwent a process of grammaticalization and lost its person-and number-features, thus becoming a morphologically bound morpheme (see Mackenzie 2018, p. 32).
A strong assumption holds that qui-clauses (see 32c) are reminiscent of infinitival complements with subject extraction and ECM-like case assignment, as seen in (32a,b): 32.
Un homme que je crois [qu+i t est malade] The subject cannot be case-marked in (32a). However, when the subject is whextracted, ungrammaticality dissolves (cf. 32b, c).
In (33), we reproduced the syntactic representation of Mackenzie's analysis: abstracting away from the cartographic details of it, the most relevant aspect is that the wh-subject cannot be case-marked (either in vP or in [Spec, TP]) and has to enter a probe-goal relation, with a higher probe in the superordinate clause.

33.
Languages 2021, 6, x FOR PEER REVIEW 11 of 15 On the contrary, there is strong evidence that the special qui behaves syntactically similar to que, i.e., like a C 0 . Mackenzie follows the lead of Koopman and Sportiche (2014) in assuming that subject extraction from sentences introduced by qui is necessary, since case assignment cannot occur clause-internally, and the extracted subject has its case feature checked by an external probe. However, Mackenzie refutes Koopman and Sportiche's assumption that qui is a relative pronoun. Mackenzie takes qui to be unequivocally a complementizer, albeit with a more complex morphological structure than que. In fact, there is diachronic evidence that qui can be decomposed to qu(e) + i, where -i is taken to be a reduced form of the pronoun il. Il was originally inserted along with the complementizer as a resumptive pronoun; eventually, it underwent a process of grammaticalization and lost its person-and number-features, thus becoming a morphologically bound morpheme (see Mackenzie 2018, p. 32).
A strong assumption holds that qui-clauses (see 32c) are reminiscent of infinitival complements with subject extraction and ECM-like case assignment, as seen in ( The subject cannot be case-marked in (32a). However, when the subject is whextracted, ungrammaticality dissolves (cf. 32b, c).
In (33), we reproduced the syntactic representation of Mackenzie's analysis: abstracting away from the cartographic details of it, the most relevant aspect is that the wh-subject cannot be case-marked (either in vP or in [Spec, TP]) and has to enter a probegoal relation, with a higher probe in the superordinate clause.

33.
According to Mackenzie's proposal, the French data confirm what we have hitherto assumed: (i) Irrespective of [Spec, TP] being the locus of cyclic wh-extraction, that position does not enter a probe-goal relation with the phase head. (ii) The nominative case is 'absorbed'-i.e., in Mackenzie's terminology, "discharged"by -i, which overtly manifests Agree.
According to Mackenzie's proposal, the French data confirm what we have hitherto assumed: (i) Irrespective of [Spec, TP] being the locus of cyclic wh-extraction, that position does not enter a probe-goal relation with the phase head. (ii) The nominative case is 'absorbed'-i.e., in Mackenzie's terminology, "discharged"by -i, which overtly manifests Agree.
In (34), we introduce the French way of circumventing the that-t effect, compared with the ones we discussed in Sections 3 and 4: Cimbrian, Bavarian, and French, all converge in inhibiting the Agree relation between the phasal head C/Fin and [Spec, TP]. The strategy that all the three languages share is represented by a morphologically complex C-head with the following property: The Cimbrian -da, the Bavarian C-Agr, and the French -i absorb the case feature, i.e., the yield/product of the Agree relation, before it can be transferred to T, preventing [Spec, TP] from entering the Agree relation with C as a suitable goal. Regardless of the peculiar differences in every language, Mackenzie's assumption further confirms the idea that subject extraction through a lexical COMP is only possible when case is not licensed in the T layer.
Our proposal finds further confirmation in the Italo-Romance data presented in Brandi and Cordin (1989) In Florentine, wh-extraction is only possible with an expletive proclitic subject (see e' in 35), which shows that there cannot be agreement between T and the internal argument, along with the absence of past participle agreement; In other words, wh-subject extraction is necessary, since the subject cannot be case-marked in the embedded clause, in the same fashion as in French (see 32c). Whenever an Agree relation is established between T and the internal argument via a clitic endowed with phi-features (see le in 35), along with the agreement of the past participle, the extraction is blocked, and the sentence is ungrammatical: It goes without saying that subject proclitics realize agreement (third person plural) in the T domain (see Tomaselli and Bidese 2019).

Conclusions
In this paper, we have dealt with the phenomenon of that-t effect in a comparative perspective, focusing on the Germanic/Romance divide. All the languages considered here display a long wh-subject extraction across an overt COMP. In other words, they all violate/circumvent the that-t effect. Despite this superficial similarity in terms of E-language, all these languages make use of different strategies that boil down to the same mechanism in terms of I-Language: the inhibition of the Agree relation between a probe, endowed with phi-features, and [Spec, TP] as goal. In Cimbrian, this relation is inhibited because [Spec, TP] is assumed not to be projected (see Bidese and Tomaselli 2018;Bidese et al. 2020), and the particle -da manifests the relation between C-the phasal head endowed with phi-features-and the low subject position without involving T. Contrarily, in a pro-drop language such as Italian, it is T which plays a major role: the incorporation of the [D] features into the T head ensures that the Case feature of T is discharged in the richly articulated morphological layer of the probe itself. In turn, [Spec, TP] does not play any role in case assignment. The same incorporation of the [D] features occurs in C, in a V2 language like Bavarian: in this variety, the case feature of C is absorbed/discharged directly on the C-head due to the phenomenon of COMP-agreement (even when the Agr exponent is Ø). In a similar way, the morphological exponent -i-in the French complex head qu-i-absorbs the case feature in C, resulting in the inactivation of [Spec, TP].
The general conclusion we can draw is that each of the four languages described manifests a specific facet of the same operation: [Spec, TP] ends up not being involved in the Agree relation with the probing head.
In a null subject language, the that-t effects can be circumvented by incorporating [D] in T; on the contrary, in non-pro-drop languages, [D] is incorporated in C, with the result of rendering [Spec, TP] inactive.

Author Contributions:
The overall paper is the result of joint work but each author provided their own individual contribution: conceptualization A.T. and E.B.; methodology, A.T., A.P. and E.B.; formal analysis A.T., A.P. and E.B.; investigation and data curation E.B.; writing E.B. and A.P.; supervision A.T. For the concerns of the Italian Academia, A.P. takes responsibility of Sections 3 and 4; E.B. of Sections 1 and 2. A.T. for Sections 5 and 6. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the PRIN project "Models of language variation and change: new evidence from language contact" of the Italian Ministry of Education, University and Research, grant nr. 2017K3NHHY (PI: Maria Rita Manzini).

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The most part of the data discussed are taken from previous literature; novel data concerning the that-t effect have been collected both with dedicated fieldwork and from the crowdsourcing platform, www.vinko.it (accessed on 15 March 2021) (data available in a publicly accessible repository)