‘Again’ separation in Italian

. In Italian, ri-‘again’ can be separated from the constituent it is semantically attached to and challenges the structural account for the ambiguity of ‘again’-type elements. To address this issue, this paper proposes a solution through aspectual agreement and suggests a movement and reconstruction analysis for the separation effect of ri-. It also provides supporting evidence for this analysis through Coordinate Structure Constraint and Relativized Minimality.

This ambiguity surrounding again is commonly accounted for as a structural ambiguity (Morgan 1969;Dowty 1979;von Stechow 1995von Stechow , 1996;;Beck & Johnson 2004;Beck 2005;Bale 2005Bale , 2007;;a.o.).The structural account proposes that again presupposes the truth of its prejacent at a time prior to the utterance time and that its ambiguity arises from its different possible attachment points.In (3), if again is attached to the matrix vP, which contains information about the event of wanting to close the door, the high reading is derived.On the other hand, if again is attached to the embedded vP, which only contains information about the event of closing the door, the low reading is derived.The structural account predicts that in a head-initial language, if 'again' comes before the matrix verb, the high reading is possible but not the low reading, since 'again' can only modify the matrix vP and not the embedded vP.In other words, sentences like (4) are expected to be infelicitous in a context where the low reading is intended (e.g., (2b)). 2   (4) John will again [ vP want to close the door].
Some speakers report that preverbal again appears to require stress.I would like to point out the contrast between (4) and the following example: (ii) John will want, again, to close the door.
Here, emphasis is required on again.The structural account is challenged by (5b), where riis separated from the embedded vP on the surface but still yields the low reading, as if it is attached to the embedded vP.Cardinaletti has termed this phenomenon ri-separation (i.e., 'again' separation).
In order to address the challenge to the structural account posed by 'again' separation, this paper offers an explanation by proposing that riundergoes movement and reconstruction (cf.Xu 2012Xu , 2016;;Liu 2021, to appear).Specifically, I argue that in a construction involving 'again' separation, riis initially base-generated in the embedded vP with an unvalued aspect feature.It then moves to the specifier position of the matrix AspP, where it agrees with the matrix Asp 0 .Finally, it is interpreted at its base-generated position through reconstruction.
Section 2 investigates which verbs allow 'again' separation and establishes a correlation between 'again' separation and restructuring in Italian.Section 3 lays out the movement and reconstruction analysis for 'again' separation and presents evidence in support of this analysis.Section 4 draws comparisons between riand similar elements in English and Mandarin and discusses the crosslinguistic implications of these comparisons.Section 5 concludes.
2. Data.This section provides a more comprehensive data set on 'again' separation.It demonstrates that verbs that allow 'again' separation are restructuring verbs.
2.1.'AGAIN' SEPARATION AS A RESTRUCTURING TEST.In the introduction, we observed an example of a verb that permits 'again' separation.By contrast, in (6), we see a verb that does not allow 'again' separation.In this case, the matrix verb is 'decide', which is immediately preceded by ri-.However, we cannot derive the low reading of riin this example.Instead, it must be the case that the repeated event is the event of deciding, introduced by the matrix verb.Verbs that permit 'again' separation and those that do not are listed in (7), leading us to observe that the former are restructuring verbs.
Based on (7), I propose 'again' separation as a new test for identifying Italian restructuring verbs.
To demonstrate the concept of restructuring verbs and gain a better understanding of their syntactic nature, it is helpful to introduce another phenomenon in Italian: clitic climbing.

CLITIC CLIMBING.
Italian restructuring verbs can be identified using the clitic climbing test, which is also applicable to verbs that allow 'again' separation.Over the past few decades, extensive research (Rizzi 1978(Rizzi , 1982;;Wurmbrand 1998Wurmbrand , 2001Wurmbrand , 2004Wurmbrand , 2015;;Cinque 1999Cinque , 2006;;Matushansky 2006;a.o.) has been conducted on the phenomenon in Romance languages where clitics (e.g., lo, the third person masculine singular clitic object pronoun in Italian) can ascend from a nonfinite embedded clause to the matrix clause when the matrix verb is a modal (8a), aspectual (8b), or motion (8c) verb.Examples illustrating a contrast can be seen in ( 9), which, despite their apparent structural similarity to the examples in (8), do not allow clitic climbing.
(8) (Cinque 2006:11) a Various approaches have been proposed to explain the distinction between ( 8) and ( 9), with Wurmbrand (2004) and many others attributing it to the differing sizes of the complements that the verbs in each group take.According to this view, nonrestructuring embedding verbs take a larger complement that introduces a phrase boundary, which prohibits clitics from moving across it.Conversely, restructuring verbs take a smaller complement that lacks this phrase boundary, enabling clitics to move upwards toward the matrix clause.
In this paper, I make a more specific claim regarding the complement sizes of Italian restructuring verbs and nonrestructuring embedding verbs to support my argument.Specifically, I assume that restructuring verbs can take either AspP or vP complements, while nonrestructuring embedding verbs take CP or TP complements.If this is accurate, the generalization observed in this section can be summarized as (10).
(10) a. Verbs that take AspP or vP complements allow 'again' separation and clitic climbing.b.Verbs that take CP or TP complements disallow 'again' separation or clitic climbing.

Analysis.
Building upon the observation that the size of the embedded clause has an impact on the acceptability of 'again' separation, I suggest that riundergoes movement and reconstruction.In this section, I first elaborate on the specifics of my movement and reconstruction analysis regarding 'again' separation.Subsequently, I present two supporting pieces of evidence in favor of this analysis.
3.1.MOVEMENT AND RECONSTRUCTION.My proposal (cf.Xu 2012Xu , 2016;;Liu 2021, to appear) can be outlined by addressing four primary questions: where riis base-generated, where rimoves to, where riis interpreted, and what the circumstances under which the movement of riis allowed are.
To begin with, where is ribase-generated?While it can be base-generated in other positions, I specifically focus on the cases in which it is externally merged with either the matrix vP or the embedded vP.When it is externally merged with the matrix vP, we obtain the high reading, and when it is externally merged with the embedded vP, we obtain the low reading.
Moving on to where rimoves to, I must first discuss the motivation behind its movement.I propose that riis base-generated with an unvalued aspect feature that must be valued by the closest Asp 0 probe through specifier-head agreement.This requirement drives rito move to the specifier position of the closest AspP, which is situated above vP but below TP and is the same position where adverbs like di nuovo 'again' are located in Cinque's (1999) cartography.
Where is riinterpreted?It is interpreted in its base-generated position through either syntactic or semantic reconstruction (Lechner 1998).
Finally, what are the circumstances under which the movement of riis allowed?To answer this question, we need to add one more data point (11).This data point shows that in an embedding construction, rican also surface low and yield the low reading.Taking this into account, I propose that rialways moves in the cases that I examine, but the distance may differ.If it surfaces high and produces the high reading or surfaces low and produces the low reading, the distance is rather short.In contrast, if it surfaces high and produces the low reading, the distance is longer.
A follow-up question is when the long-distance movement of riis permissible.Recall that according to my proposal, in Italian, restructuring verbs embed either an AspP or a vP, whereas nonrestructuring embedding verbs embed a CP or TP.The long-distance movement of riis allowed when it is base-generated in the embedded vP and is unable to value its feature within the embedded clause.This situation arises when the embedded clause does not have an AspP.This only happenes in one scenario: when the matrix verb is a restructuring verb that embeds a vP.In this scenario, riis initially base-generated with an unvalued aspect feature in a vP embedded by the matrix restructuring verb.Afterward, as it is unable to value this feature within the embedded vP, it moves to the specifier position of the matrix AspP and agrees with the matrix Asp 0 .Ultimately, it undergoes reconstruction and is interpreted in its base-generated position in the embedded vP, resulting in a low reading.Given below are two additional derivation examples.(13a) sketches how high rican yield the high reading, and (13b) sketches how low rican yield the low reading.In both cases, the movement of ritriggered by aspectual agreement is vacuous.In what follows, I present two evidentiary points supporting this movement and reconstruction analysis.That is, Coordinate Structure Constraint (Ross 1967) and Relativized Minimality (Rizzi 1990(Rizzi , 2001(Rizzi , 2004) ) both apply to 'again' separation and provide evidence that riundergoes movement in an 'again' separation construction.
3.3.COORDINATE STRUCTURE CONSTRAINT.The first piece of evidence pertains to the absence of ambiguity of matrix riwhen the embedded predicate comprises a conjunction.( 14) is similar to the baseline example ( 5) with the exception that the embedded predicate contains a conjunction.In this case, high rican only be interpreted as modifying the matrix predicate but not any of the embedded conjuncts.In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct.
However, it remains unclear why across-the-board movement of riis unattainable, as ( 14) does not have the reading 'Gianni will want to jump up again and close the door again' (Raffaella Zanuttini, pers. comm.).I leave this issue for future investigation.
3.4.RELATIVIZED MINIMALITY.The second piece of evidence concerns the effect of intervening adverbs on 'again' separation.To illustrate this, ( 16) is almost identical to the baseline example (5) with the exception that an adverb is inserted between the matrix verb and the embedded verb.In this example, the high reading of ripreceding the matrix verb is available, but the low reading is not.This phenomenon may be explained by Relativized Minimality (17): the movement of an adverb (i.e., ri-) across another adverb (i.e., the inserted adverb 'completely') is prohibited.
(17) Paraphrased Relativized Minimality (Rizzi 1990:1, adapted) Moving X across Y is prohibited if X and Y have certain characteristics in common, such as when X and Y are both adverbials.
Might we find it unexpected that ri-, which appears to be a head, is affected by Relativized Minimality in relation to an adverb, which appears to be a phrase?In the following subsection, I show that ridiffers significantly from other prefixes in Italian, and its phrasal characteristics may be explained by Cardinaletti's (2003) suggestion that it is an incorporated form of synonymous adverbs like di nuovo 'again '. 3.5. INCORPORATION. Cardinaletti's (2003) evidence suggests that riis unique among Italian prefixes.It exhibits a phonological independence from the elements that follow it and is distributed in ways that are not observed in other morphological elements.For instance, while other prefixes in Italian cause the immediately following [s] to become voiced in intervocalic contexts (18), ridoes not ( 19).(20) (Cardinaletti 2003:9) a. rinviare 'postpone' b. riinviare 'send again' Given the above discussion, Cardinaletti proposes that riis the incorporated version of synonymous adverbs such as ancora, di nuovo, and nuovamente, which all convey the meaning of 'again'.This claim is substantiated by two pieces of evidence.6First, rialways appears as the outermost prefix when it is combined with other prefixes (21).
(21) (Cardinaletti 2003:15) a. ridisfare 'undo again' b. * disrifare Second, rican be duplicated but only up to two times ( 22).This aligns with Cinque's (1999) statement that repetitive adverbs in Italian can appear a maximum of two times within a sentence.
(22) (Cardinaletti 2003:13) a. ririfare 'redo again' b. * riririfare If the movement and reconstruction analysis of 'again' separation is on the right track, then Cardinaletti's proposal reinforces the idea that the movement of riin an 'again' separation contruction is most likely phrasal.This, in turn, provides an explanation for the sensitivity of rito Relativized Minimality in relation to adverbs.4. Discussion.In this section, I draw comparisons between Italian riand English again, English re-& Mandarin you 'again'.I also discuss the interesting implications of each comparison.
4.1.COMPARISON WITH ENGLISH again.Recall that in English, again preceding the matrix verb is unable to produce the low reading.(4), copied below in (23b), would not be felicitous in a context where the low reading is intended, as seen in ( 2b), copied below in (23a).If we accept the movement and reconstruction analysis of Italian 'again' separation and its correlation with restructuring, the contrast between the behaviors of English again and Italian rican be attributed to differences in the sizes of embedded clauses in these two languages.It is often remarked that in English, there is little to no evidence of restructuring, and verbs like want take full CP complements (e.g., Landau 1999).Thus, even if English again were base-generated with the same aspect feature and had the same feature checking requirement as Italian ri-, it could not move upwards to the matrix clause when embedded, because the embedded clause is large enough to contain an AspP for the feature to be checked within the embedded clause.4.2.COMPARISON WITH ENGLISH re-.Despite both being a prefix and conveying the meaning of 'again', English reand Italian rihave significant differences in their distribution.Italian riis much more productive than English re-, as it can modify a wider range of verbs, including causative verbs, detransitivized verbs, locative verbs, particle verbs, and verbs in idiomatic expressions (Cardinaletti 2003).On the other hand, English reis generally limited to transitive or unaccusative verbs with an underlying object (Horn 1980).In our English baseline example (4), substituting again with remakes the matrix verb and the sentence completely unacceptable (24).
(24) *John will rewant to close the door.
For more detailed analyses of English re-, I refer the readers to Horn (1980), Keyser &Roeper (1992), andSȃvescu Ciucivara &Wood (2013).4.3.COMPARISON WITH MANDARIN you.Italian is not unique in the exceptional scopal behavior of 'again'-type elements or its correlation with restructuring.In Mandarin, matrix you 'again' can yield a low reading in restructuring contexts as well (25), as reported by Liu (2021, to appear) and Huang (2022).This phenomenon and this correlation with restructuring are also observed with the Cantonese suffix -faan 'again' (Liu & Yip to appear) and the French prefix re-'again' (Vincent Homer, pers.comm.).

Conclusion.
Italian ri-challenges the structural account for the ambiguity of 'again'-type elements, as it can be interpreted as if it is adjoined to a lower position in the embedded clause when it surfaces in the matrix clause.To rescue the structural account, I propose that 'again' separation is correlated with clitic climbing and restructuring and that ri-can move and reconstruct, with evidence from Coordinate Structure Constraint and Relativized Minimality.This analysis suggests that the different behaviors of 'again'-type elements across languages may be attributed to what we already know about the differences in the sizes of embedded clauses in different languages.
(3) a. Matrix vP attachment: High reading John will [ vP [ vP want to [ vP close the door]] again].b.Embedded vP attachment: Low reading John will [ vP want to [ vP [ vP close the door] again]].
want to close the door, and . . .a. High reading 'Gianni wanted to close the door before.' b.Low reading 'Gianni closed the door before.' 2 The example below features the natural usage of English preverbal again:
decide to close the door, and . . .a. High reading only 'Gianni decided to close the door before.' b. * 'Gianni closed the door before.'

(
11) (Nicoletta Loccioni & Raffaella Zanuttini, pers.comm.want to again close the door.' 3.2.DERIVATION EXAMPLES.An example of the syntactic derivation of 'again' separation is provided in want to jump up and close the door, and . . .a. High reading only 'Gianni wanted to jump up and close the door before.' b. * 'Gianni jumped up (but did not close the door) before.' c. * 'Gianni closed the door (but did not jump up) before.'This phenomenon can be accounted for by Coordinate Structure Constraint (15), which prohibits the movement of rifrom an embedded conjunct.(15) Coordinate Structure Constraint (Ross 1967:161) 5 want to close the door completely, and . . .a. High reading only 'Gianni wanted to close the door completely before.'b. * 'Gianni closed the door before.'

(
23) a. Context of repeated closing: Low reading On a certain day, the door to John's room was left open.John had no desire to close the door, but his mother compelled him to do so.She then proceeded to open the door again.It is believed that John will want to, once more, close the door.b. # John will again want [to close the door].
let me believe in love, and . . .a. High reading 'This movie let me believe in love before.' b.Low reading 'I believed in love before.' 11