Syntactic discontinuous reduplication with antonymic pairs: a case study from Italian

: This article aims at giving a comprehensive account of a so far undescribed reduplicative pattern in Italian named syntactic discontinuous reduplication with antonymic pairs (SDRA). This pattern, characterized by the non-contiguous repetition of the same element within a larger ﬁ xed con ﬁ guration de ﬁ ned by two spatial antonyms, can be schematized as <X i Adv 1 X i Adv 2 >, where Adv 1 and Adv 2 are antonyms (e.g., di qua ‘ here ’ ∼ di là ‘ there ’ ). After describing its formal and functional properties, based on naturally occurring data extracted from the Italian Web 2016 corpus, the SDRA is analyzed as an independent ‘ construction ’ in the Construction Grammar sense. This construction is claimed to convey a general value of ‘ plurality ’ and to have developed a polysemy network of daughter constructions expressing more speci ﬁ c functions such as ‘ distributivity, ’ ‘ related variety, ’ and ‘ dispersion. ’ In addition, we propose considering the SDRA a ‘ multiple source construction, ’ originating from the blending of two independent constructions: syntactic reduplication and irreversible binomials with antonymic adverbs. Finally, we discuss SDRA-like patterns in other typologically different languages (Russian, Modern Hebrew, Mandarin Chinese, German), pointing out similarities and differences, and paving the way to a more systematic study of discontinuous reduplication in a crosslinguistic perspective.


Introduction
Reduplication has been broadly investigated in both theoretical and typological studies (cf., among many others, Hurch 2005; Inkelas and Zoll 2005;Kallergi 2015;Moravcsik 1978;Rubino 2013;Stolz et al. 2011;Urdze 2018). A considerable amount of knowledge has been accumulated in recent years about this phenomenon, in all its manifestations, leading to a number of classifications. The most widely known classification is probably the one distinguishing between partial and full reduplication, but we know of other special types such as echoreduplication or automatic reduplication (cf. e.g., Inkelas 2014;Rubino 2005). This contribution focuses on one of these special types, namely discontinuous reduplication (henceforth DR). DR is defined by Velupillai (2012: 101) as a kind of reduplication "where other morphological material may appear between the reduplicant and the base" (cf. also Rubino 2005: 17). 1 See for instance (1), where the two copies are connected via the linker -ng.
(1) Manila Bay Creoles (Creole, Spanish-lexified; Philippines) buníta 'beautiful' > bunita-ng-buníta 'very beautiful' (Grant 2003: 205) More specifically, this article aims at giving a comprehensive account of a specific instance of DR attested in the Italian language, which we dub here syntactic discontinuous reduplication with antonymic pairs (henceforth SDRA). SDRA is a pattern characterized by the non-contiguous repetition of the same element within a larger fixed configuration defined by a pair of spatial antonyms. See the following examples. 2 (2) I segni erano spariti e, cerca di qua, cerca di là, nessuno ha avuto il coraggio di continuare senza. 'The signs had disappeared and, after searching in all directions [lit. search here, search there], no one had the courage to continue without them.' (3) Una è il "terrorismo islamico" con tutto il suo corredo di Osama, Al Zarkawi, Al Qaida, cellule qua, cellule là, attentati a Sharm El Sheik, Amman, Londra, ovunque. Sempre solo la versione di comodo dei veri terroristi. 'One [fundamental thing] is "Islamic terrorism" with its complement of Osama, Al Zarkawi, Al Qaida, cells everywhere [lit. cells here, cells there], attacks in Sharm El Sheik, Amman, London, everywhere. Always and only the convenient version of the true terrorists.' The SDRA consists in the repetition of an item X (cerca 'search' in [2] and cellule 'cells' in [3]) within a larger configuration built around a pair of spatial, deictic adverbs with opposite meaning. This pattern can be schematized as follows: <X i Adv 1 X i Adv 2 >, where Adv 1 and Adv 2 are opposites. In (2), this schema is instantiated by cerca di qua, cerca di là 'search here, search there', whereas in (3) by cellule qua, cellule là 'cells here, cells there'. Within the SDRA we can therefore recognize the presence of two different elements: (i) a repetitive/reduplicative pattern (X is repeated) and (ii) an antonymic pair of adverbs occurring in the same syntagmatic environment. This article explores in detail the formal and semantic properties of this pattern in Italian by combining theory and corpus data, with the addition of some (still very preliminary) crosslinguistic data. First, in Section 2 we discuss the status of SDRA as an instance of reduplication, against the background of recent work in the field. Then, after describing the corpus-based methodology adopted for the analysis in Section 3, we illustrate the main formal characteristics of SDRA in Italian, focusing on the properties of the repeated element and the syntactic status of the whole pattern (Section 4), with a brief detour into a seemingly related pattern (Section 4.1). Section 5 analyzes the functions and sub-functions that SDRA can express. In Section 6, we propose a Construction Grammar analysis of the SDRA, which is claimed to qualify as a 'multiple source construction' that develops a network of sub-constructions, on the basis of relevant form-meaning associations. Finally, in Section 7, we take a cursory look at other languages to see if SDRA is a language-specific pattern or conversely can be found elsewhere, and with which functions. Section 8 contains some concluding remarks.

Reduplication as a gradient phenomenon
To which extent can we regard SDRA as reduplication? Reduplication is generally considered as " [t]he systematic repetition of phonological material within a word for semantic or grammatical purposes" (Rubino 2005: 11). However, the SDRA clearly goes beyond the word level: it cannot be defined as "repetition of phonological material within a word", but rather as repetition of phonological material within a syntactic domain. Consequently, it is legitimate to question its status as a reduplicative pattern, even more so if we think of its non-contiguous nature. In fact, if we regard reduplication as something that pertains morphology only, it is contradictory to talk about syntactic reduplication. However, there are good reasons to support our terminological choice of using the label reduplication for a syntactic pattern, following other scholars such as Wierzbicka (1986).
Indeed, morphological reduplication and syntactic repetition are related phenomena in that, despite the different domains of application, they both imply the duplication of some kind of linguistic element, which makes it difficult sometimes to draw a line between the two. We all have in mind clear instances of (morphological) reduplication and clear instances of (syntactic) repetition, but at the same time the boundary delimiting these two phenomena is not clear-cut, as already observed, among others, by Gil (2005), Stolz et al. (2011), and, more recently, Freywald and Finkbeiner (2018: 4): "the boundary between reduplication and repetition appears to be quite fuzzy […]. Thus, instead of a clear-cut categorial distinction we rather find a nebulous transition zone." Generally speaking, the absence of clear-cut categories and boundaries leads to regard traditional modules of grammar as poles of a continuum and to treat linguistic facts in terms of prototypicality, or canonicity. The same view has been applied to reduplication. In particular, both a prototypical and a canonical approach to reduplication are described and discussed by Stolz (2018). As for the former, Stolz (2018) posits 8 criteria which define prototypical reduplication (taken from Stolz et al. 2015: 826), summarized in Table 1. As for the canonical approach, which refers to the framework of Canonical Typology (cf. e.g., Corbett 2005), Stolz (2018) postulates 13 criteria that define canonical reduplication, summarized in Table 2. 3 Table : Criteria for prototypical reduplication (adapted from Stolz : ).
i exactly one linguistic sign is involved (as domain) ii the construction is strictly binary iii absolute phonological identity of the constituents applies in quantity and iv quality of their segments v the constituents are syntagmatically adjacent vi without rendering any other structural unit discontinuous vii the construction is different from the singleton units in terms of their semantics and/or functions viii the construction has the status of a syntactic word 3 The difference between the two approaches lies in what follows: the prototype is the best, most representative exemplar of a (fuzzy) category (cf. Croft and Cruse 2004;Rosch 1978), revolving around the prototype itself, which is therefore attested (and, ideally, the most frequent type); the canon is the best example within a theoretical space of logical possibilities, which serves basically as a "tool of measurement", as Stolz (2018: 225) puts it, hence "canonical instances (the best examples, those most closely matching the canon) may well not be the most frequent. They may indeed be extremely rare, or even non-existent" (Corbett 2005: 26). For our current purposes, there is no need to commit to one approach or the other, since our intent is to discuss the status of SDRA within the ongoing debate on the nature and delimitation of reduplicative phenomena.
Let us have a look at these criteria, defining the prototype and the canon, respectively, and compare them with the properties of the SDRA. As for the prototypical approach, the SDRA meets 5 criteria out of 8: more specifically, the SDRA does not conform to criteria v/vi (adjacency) and viii (wordhood of the whole construction). As for the canonical approach, the SDRA meets 11 criteria out of 13: in particular, C3 (adjacency) and C5 (wordhood of constituents) are not met.
Wordhood of the whole construction is not met because the SDRA is larger than a word, as we have already said. Wordhood of constituents is also not met because, as we will see in Section 4, the SDRA applies not only to single words but also to larger units (phrases, even clauses), although in a minority of cases. As concerns adjacency, the SDRA is obviously not composed of two adjacent units because the reduplicative pattern intertwines with the antonymic adverbial pair and, consequently, the first adverb breaks the integrity of the repetition. However, it is worth noting that Stolz (2018) himself justifies this criterion stating that: A' is best recognizable as an image of A if it can be directly compared to the reduplicand. Therefore, the image should be close to the reduplicand in terms of temporal and topological distance. The further away A' is from A in speech, the more their relationship comes to resemble that of anaphor, i.e., A+A' do not form a tight unit with a distinct constructional meaning or function. (Stolz 2018: 227) In this sense, the SDRA can be regarded as a pattern in which the repeated element A' is sufficiently close to A not to be understood as an anaphor. Nevertheless, Stolz (2018: 227) also notes that "immediate adjacency […] is canonical whereas A X A' (with X = intercalation of any size) is non-canonical to the extent that it is doubtful whether the sequence can be classified as reduplication, in the first place" (emphasis ours). In this respect, we may note that, in the SDRA, the distance between the repeated elements is not filled by just any intercalation; rather, the repetition is performed within a fixed syntactic pattern defined by the two adverbs, which makes the whole construction much less occasional and unconstrained than free, discourse-level repetition (see Section 6 for a theoretical proposal that aims at accounting for these properties).
In conclusion, the majority of criteria is met by the SDRA in both cases, which speaks in favor of considering the pattern in question as reduplicative indeed. Overall, it is reasonable to claim that we are in front of a case of non-prototypical and/or non-canonical reduplication.

Methodology
Our analysis of SDRA in Italian is based on a dataset composed of naturally occurring data extracted from the itTenTen16 (or Italian Web 2016) corpus, searched through the Sketch Engine platform (https://www.sketchengine.eu/). itTenTen16 is a very large corpus (4.9 billion words), the largest currently available for the Italian language, and is made of texts automatically collected from the web (formal texts, like Wikipedia entries, and informal texts, like blog comments). These properties (large size and variety of text types, including close-to-spoken texts) make itTenTen16 ideal to search for 'non-mainstream' patterns (in terms of frequency of use) typical of informal registers like SDRA.
To extract the data, we elaborated 8 CQL queries featuring 8 different adverbial pairs, listed in (4) (in alphabetical order).
(4) a. a destra 'to the right' ∼ a sinistra/manca 'to the left' b. di qua 'here, this way' ∼ di là 'there, that way' c. di qui 'here, this way' ∼ di là/lì 'there, that way' The pairs were chosen, based on our intuition, among the more common spatial opposites in Italian that could be compatible with the SDRA, although some (e.g., [4b]) are definitely more common than others (e.g., [4e]). 4 The results were first automatically filtered and then manually cleaned. The final dataset amounts to 319 items, 5 which were annotated according to the following features: (i) the type of adverbial pair (ii) the item being repeated (X) (iii) the complexity of X (one word vs. more than one word) (iv) the lexical category of X (when X is a single word or a phrase) (v) the grammatical features of X (e.g., number, TAM values, etc.) (vi) the function or semantic value of the whole structure The results of the analysis of features (i)-(v) are discussed in Section 4, while we turn to functional properties (vi) in Section 5.

Formal properties of SDRA
As already noted in Section 1, the SDRA consists in the repetition of an item X within a larger configuration built around a pair of spatial adverbs with opposite meaning, and can be schematized as follows: (5) <X i Adv 1 X i Adv 2 > where Adv 1 = opposite of Adv 2 The adverbial pair featuring in the SDRA can vary. As explained in Section 3, we focused on 8 pairs (the ones used in our corpus queries, cf. [4]), which however turned out to have an uneven distribution within the SDRA. Table 3 shows the 8 adverbial pairs ranked according to their frequency.
In our dataset, the adverbial pair X i di qua X i di là is undoubtedly the most frequent covering almost half of the total amount of occurrences. The X i qua X i là and X i di qui X i di là/lì pairs also display a fairly high frequency: 22.6% and 12.5%, respectively. All other adverbial pairs are remarkably less frequent, in particular the three pairs denoting a vertical indexical orientation (X i su X i giù, X i di su X i di giù, X i di sotto X i di sopra) are quite rare in the corpus (6.9% all together). In general, we might say that the pairs featuring the items qua ∼ là (and secondarily qui ∼ lì) are by far the most closely associated with the SDRA.
Turning now to the properties of the repeated element, X, our results show that the vast majority of the items occurring in the SDRA are hapaxes. Out of 319 cases, we have 271 types and 247 hapaxes (amounting to 91% of the types). 6 Only 24 types occur more than once in the corpus and only 6 occur more than twice. The items occurring more than twice are all verbs and also the single most frequent item is a verb, i.e., the inflected form cerca 'search.3SG.PRS.IND,' with 9 occurrences. Because of the high number of hapaxes and the fact that we cannot identify a clear 'leader X' in our dataset, we are probably in front of a productive pattern that can be filled by a wide variety of items.
In the great majority of the cases, X is a one-word-level item (254 out of 319 examples, i.e., 79.6%), like (2) and (3) in Section 1. However, we also find cases in which X corresponds to more than one single (written) word (65 out   a These are clauses. Values in italics indicate sub-values of the higher-level value "Noun (or NP).

Syntactic discontinuous reduplication
Another aspect we considered in annotating our dataset concerns the grammatical features of the repeated element within the SDRA. In particular, we focus on verbs and nouns, since they are by far the most frequent categories (284 out of 319 items, i.e., 89%). As for nominal Xs, proper nouns obviously occur in the singular form, but also common nouns tend to occur more frequently in the singular rather than in the plural (62 vs. 32 occurrences). As for verbs, they occur in a variety of inflected forms, more specifically, in 19 different forms. However, they reveal a striking tendency to occur in the present indicative singular form, especially in the third ( Interestingly, the two most frequent verb forms found within the SDRA -3SG.PRS.IND and 2SG.PRS.INDare homophonous with the second person singular imperative form of those verbs, namely the same kind of form that has been often regarded as a specific (morphomic) stem with its own place within Italian morphological constructions (Rainer 2001;Thornton 2005: 157-160) as well as other lexical constructions (such as the VeV construction: scappa e fuggi lit. run_away and run_away 'quick'; cf. Masini and Thornton 2008). This might be hinting at the fact that X tends to appear in some sort of "underspecified" or "unmarked" form within the SDRA. Indeed, the grammatical features of X are not necessarily anchored to the outside syntactic environment (see also below). Let us take Example (12): here we have a plurality of participants to which gira also refers (people go around here and there and end up where they had started), but the (singular) form of the verb does not reflect this plurality. From the data presented so far, it clearly emerges that Xs are generally (albeit not necessarily) word-level items, and, more specifically, verbs and nouns which tend to appear in some sort of unmarked form. But what is the role of the SDRA within the larger syntactic configurations it appears in?
As a matter of fact, the syntactic status of the SDRA may vary. In some cases, the pattern behaves as a phrase naturally embedded in the argument structure of the clause. See for instance (14), where both tagliano di qua, tagliano di là (lit. cut here, cut there) and mettono su, mettono giù (lit. put up, put down) are basically complex predicates agreeing with the subject tutti 'everybody'; or (15) The SDRA with nouns often appears in verbless sentences, especially in combination with other nouns or clauses (or both), as in (18) This list structure, quite recurrent in our data, strengthens the expressiveness and emphasis the construction already displays (cf. also Section 5).

A related pattern
During the manual checking of the data extracted from the corpus, we encountered a number of examples that conform to what seems to be a related pattern to the SDRA. In this related pattern, schematized in (21), X is not repeated; rather, what we find is the combination of two semantically related items, X and Y.  (22) and (23), where X and Y are co-hyponyms (types of verbal expressions) and antonyms, respectively: Berlusconi e gli juventini si incontrano in albergo la mattina. Sorrisi, strette di mano, complimenti di qua, battute di là 'Berlusconi and the Juventus guys meet at the hotel in the morning. Smiles, handshakes, many compliments and jokes going on [lit. compliments here, jokes there]' (23) Con questo sistema resta il problema della matematica, sottrazioni qui, addizioni lì ed è facile sbagliarsi mentre con le banconote sbagliarsi è quasi impossibile. 'With this system, the math problem remainsa lot of subtractions and additions [lit. subtractions here, additions there] and it's easy to make a mistakewhereas with banknotes it's nearly impossible to make a mistake' Since this pattern does not entail a strict case of repetition, we excluded these cases from our dataset, which was intended to reflect 'core' SDRA cases. However, it is reasonable to regard this pattern as closely related to the SDRA considering their formal and functional resemblance. In this respect, it is worth mentioning the proposal put forward by Inkelas and Zoll (2005: 59-65), who advocate for a unified approach of proper reduplication and constructions featuring semantically related elements, most notably synonyms but also items connected by other types of relationships, such as near-synonymy, subset, opposition. Under this view, "pure reduplication is simply one point along the cline of semantic similarity" (Inkelas and Zoll 2005: 62).

Functions of SDRA
After presenting the main formal characteristics of SDRA, we now focus on the functions expressed by this pattern in Italian (cf. point [vi] in Section 3). 10 Semantically speaking, the SDRA turns out to express a set of closely related functions all associated with the notion of plurality and, more in general, intensification. It is noteworthy that these are functions that proper reduplication very frequently encodes crosslinguistically (cf. Moravcsik 1978, among others). More specifically, we identified four functions: (i) distributivity (ii) related variety (iii) dispersion (iv) additive plurality Before presenting and exemplifying each function, it is interesting to look at their frequency within our dataset. As Table 6 illustrates, the functions are not equally represented. 11 10 For our current purposes, we focus on semantic aspects, leaving a more thorough discoursepragmatic analysiswhich would be indeed highly desirableto future research. 11 The proposed functional classification and the annotation of the data rely on the authors' intuition. Each example in the dataset was annotated by the two authors independently. In case of disagreement, the (few) cases were discussed and resolved.
Distributivity and related variety are the most frequent functions covering, together, approximately 93% of the occurrences. In particular, distributivity is by far the most represented function conveyed by the SDRA.
By distributivity we mean the expression of plurality (either of entities or situations) distributed over space, time, or participants. We could identify two subtypes of distributivity depending on the lexical category of the repeated element. We have distributive plurality with nouns (24) and distributive pluractionality (cf. Mattiola 2019) with verbs (25) and action nouns (26) The second most frequent function of SDRA is what we term here related variety, which expresses a plurality of entities that are different from each other but at the same time share a particular property (thus acting as some sort of co-hyponyms, so to speak). See Examples (27) and (28)  In (27), Natale 'Christmas' refers to family movies released at Christmas, whose titles typically contain the word Natale (Vacanze di Natale 'Christmas holidays', Natale a Miami 'Christmas in Miami', and so on): the SDRA Natale di qua Natale di là refers to a multitude of these farcical comedies, which are united by the very same stereotypes and schemas. Example (28), on the other hand, is interesting because it contains a list exemplifying different kinds of European Days, which already conveys a sense of related variety. Within this structure, the SDRA is inserted at the end, where it acts as a 'general extender'. General extenders are expressions (such as and stuff like that, etcetera) which indicate "additional members of a list, set, or category [and combine] with a named exemplar (or exemplars)" (Overstreet 1999: 11). They are typically found at the end of lists (cf. Masini et al. 2018). See for instance (29) (taken from Overstreet 1999), where and everything refers to everything that belongs to a higher-level category that is inferred from the list of items preceding it (name, address), in this case something like '(any) useful contact information'.
(29) Make sure your bag has a tag with your name and address and everything (Overstreet 1999: 5) The expression giornata europea di qua, giornata europea di là in (28) seems to have precisely this 'general extender' function, by virtue of its position at the end of a list. In addition to these cases, we found several occurrences of what we propose to regard as a subtype of the 'related variety' function. Quite often, SDRA is used to refer to a heterogeneous group of complex, implicit speech acts revolving around the repeated element X. In other words, X represents the salient element (possibly the topic) of a series of speech acts, pronounced by the same speaker or by different speakers, whose content needs to be inferred. The following example is a case in point: here the speaker is complaining about the fact that people cannot do anything better than talking about communists all the time.
(30) sapete dire solo comunisti ….comunisti di qua comunisti di là ….quanta tristezza … 'you are only able to say communists ….communists here communists there ….such a sadness …' The reason why we decided to include these cases in related variety is that, ultimately, the function they express is to represent a plurality of entities (speech acts) that are different from each other but share a given property (the topic, i.e., X), thus falling under the definition of related variety we gave above.
The two less frequent functions we identified are dispersion and additive plurality. By additive plurality, here, we mean the most prototypical meaning of plural, that is, several entities in opposition to a single entity, which is, however, represented by only one example in our dataset (31) Overall, it seems to us that dispersion represents the opposite pole of distributivity in terms of plurality. The difference between these two functions is therefore similar to the difference between paucal and plural values in nominal number (few vs. several).
To conclude this section, it may be useful to add some tentative observations on the connotative and pragmatic side of the SDRA, although this is not the focus of the present contribution. The SDRA is obviously an expressive way to express plurality (in the wide sense intended here), which often brings about other effects. One of these effects is a slightly disparaging value that emerges in some cases (cf. e.g., [26] or [28]). A derogatory connotation is attributed to other types of reduplication (especially echo-reduplication) crosslinguistically (cf. Moravcsik 1978: 323-324;Stolz et al. 2011: 134;Wiese and Polat 2016). In our case, it might be due to an inferential reading of excess. Another effect is, quite obviously, emphasis. This is also reflected by the fact that additive (i.e., standard, unmarked) plurality is not expressed by SDRA, which is clearly linked to more intense pluralities.

A constructionist analysis of SDRA
After describing the main formal and functional characteristics of the SDRA in Italian, we now attempt at giving a theoretical account of this structure. Framing our analysis within Construction Grammar (Fillmore et al. 1988;Goldberg 1995Goldberg , 2006Hilpert 2014; Hoffmann and Trousdale 2013), we may obviously ask: is SDRA a construction? Following Goldberg (2006), a construction is generally defined as follows: Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist. In addition, patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency. (Goldberg 2006: 5, emphasis ours) Indeed, we can say that the SDRA in Italian shows unpredictable and idiosyncratic properties.
At the semantic level, the SDRA has a non-compositional (albeit partially iconic) meaning, since its overall semantics cannot be computed by the mere sum of the meanings of its component parts. Let us take again Examples (2) and (3), repeated here for convenience as (34)  If we interpret the two occurrences of the SDRA in (34) and (35) literally, we obtain a spatial reading in which X occurs/is (depending on its category) first di qua 'here' and then di là 'there'. In other words, when X is a verb as in (34), we would have a reading where the action occurs twice in two opposite locations identified by the two deictic adverbs. When X is a noun in (35), we would have a reading with two referents (denoted by X) found in two opposite locations identified by the two deictic adverbs. This means that in (34) the agent looks for the signs in two indexical locations (cerca di qua, cerca di là), and that in (35) there are terroristic cells placed in two indexical locations (cellule qua, cellule là). In actual fact, this is not the semantic interpretation we get. Rather, both (34) and (35) have a distributive meaning: 'searching repeatedly, in all directions' and 'cells everywhere, in every place', respectively. In (35), this interpretation is even strengthened by the presence of the adverb ovunque 'everywhere' at the end of the list of cities in which a terroristic attack took place. At the formal level, we have some idiosyncratic properties and constraints as well. Firstly, and quite obviously, the two Xs are constrained in that they must be identical. Secondly, the antonymic adverbial pairs are lexically fixed and cannot be reversed. In other words, they must appear exactly as reported in Table 3. A search for the reversed <X i di là X i di qua> pattern within our corpus gives zero results (cf. also below). Thirdly, the whole pattern seems to be endowed with a certain degree of internal cohesion, since it is not found interrupted by other linguistic material. Our intuition is that there might also be a specific prosodic contour associated with the SDRA (cf. footnote 8), which however does not apply to our current data which are taken from a corpus of written Italian. This hypothesis should be tested with data from spoken language.
In sum, the SDRA has more than one idiosyncratic aspect in terms of both form and function. An issue that must be discussed at this point is the predictability "from other constructions recognized to exist," as the quotation from Goldberg (2006: 5) above reminds us.
We have already observed in Section 1 that, within the SDRA, we can recognize two different patterns: a repetitive/reduplicative pattern, and a pattern involving an antonymic pair of adverbs (which we now know to be irreversible). These patterns need to be briefly discussed, since both happen to occur in isolation in Italian, and can reasonably be analyzed as independent constructions of which the SDRA appears to be a mix.
(36) lungo lungo (lit. long long) 'very long' (Grandi 2017: 60) (37) caffè caffè (lit. coffee coffee) 'true coffee' (Wierzbicka 1986: 297) (38) copia copia (lit. copy copy) 'generalized copying' (Thornton 2008: 209) (39) E allora perché dovrei andare a una cena di lavoro, a una cena che sia lavoro, a una cena pallosissima in cui si parlerà di lavoro lavoro lavoro 'So, why should I go to a business dinner, to a dinner which is work, to a deadly boring dinner where we'll be talking about work, work, and work [lit. work work work]' On the other hand, the same antonymic adverbial pairs involved in the SDRA may also occur in isolation linked by the conjunction e 'and', resulting in irreversible binomials (Malkiel 1959;cf. Masini 2006 for a constructionist analysis of these expressions in Italian) meaning 'here and there/hither and thither/everywhere'. Most of these irreversible binomials exhibit assonance and display a rhyming pattern (e.g., di qua e di là 'here and there'), as is often the case with these expressions. They basically express a generalizing function (in the sense of Wälchli 2005: 139, who describes "generalizing co-compounds" as structures whose "parts express the extreme opposite poles of which the whole consists") (42) and a dispersive function (namely, limited, scattered plurality) (43).
(42) Le guardie accorrevano di qua e di là 'Guards were rushing (from) everywhere' (43) Singole poesie sono state pubblicate qua e là su libri o fanzines 'Single poems were published here and there in books or fanzines' Among the pairs considered in the present article, the most frequent binomial expression in the itTenTen16 corpus is, by far, qua e là 'here and there' (32,996 occurrences), followed by su e giù 'up and down ' (12,478) and a destra e a sinistra/ manca 'to the right and to the left' (9,156), whereas di qua e di là 'here and there' (the most frequent in the SDRA) is less common (4,461). 13 The question now is: could the SDRA be in fact predictable from (the unification of) the two mentioned independent constructions, i.e., a reduplicative pattern and antonymic adverbial irreversible binomials? Indeed, this might be the case. However, there are several pieces of evidence supporting the proposal of conceiving the SDRA as an independent construction in Italian.
First, despite being a marginal pattern, the SDRA is reasonably attested in our corpus and, most importantly, productive in terms of type frequency, since it can be filled by a variety of Xs (see Section 4). Second, it displays one function which cannot be traced back to either reduplicative patterns or antonymic adverbial binomials, i.e., related variety. Third, the binomials display the conjunction e 'and' which is not present in the SDRA. Finally, the SDRA displays a marked preference for the di qua ∼ di là pair, whereas this pair is comparatively much less common than other pairs in the corresponding binomial version: if the SDRA were the unification of the reduplicative and the binomial constructions, we would expect the frequency of the pairs within the SDRA to mirror their frequency within the binomials. For all these reasons, we propose to analyze the SDRA as an independent (abstract) construction, namely as a complex sign with its own properties that is part of our linguistic competence.
An advantage of a constructionist analysis to the SDRA is that we may account for the fact that the SDRA displays recurrent form-meaning associations: these can be analyzed quite straightforwardly as a network of related constructions motivated by a more general SDRA mother construction, to which they are linked via polysemous inheritance links. In other words, in our mental representation the SDRA is a polysemous construction that develops different plurality-related senses, each of which is associated with specific formal requirements or tendencies.
These requirements or tendencies were found for all the functions we illustrated in Section 5 (except for additive plurality, for which we have one single occurrence). The SDRA with distributive meaning is typically conveyed by two sets of items yielding the two sub-meanings: distributive plurality is generally expressed through bare nouns that can be either in their singular or plural form, whereas distributive pluractionality through bare verbs in the third or second person singular present indicative form as well as action nominals (cf. Examples [24], [25], and [26] in Section 5). The function of related variety, instead, is mostly associated with bare nouns (especially singular, cf. [27] and [28] in Section 5), but also open to verbs (in no particular form) and other parts of speech, as well as to phrasal items and clauses (due to the 'speech act ' subtype, cf. [30] in Section 5), unlike distributivity. Finally, dispersion is clearly associated with expressions featuring singular nouns preceded by an indefinite determiner and by clauses introduced by chi 'who ' (cf. [32] and [33] in Section 5).
The network of daughter constructions can be schematized as in Figure 1 (where construction is abbreviated as Cxn), where we find the abstract SDRA Cxn licensing three sub-constructions which are more specific in terms of both function and formal requirements, as just described.
Conceiving the SDRA as an independent construction does not mean denying its link with the reduplicative construction and the antonymic adverbial binomial construction. Indeed, the SDRA construction might have actually originated from some sort of blending between the two constructions, with some adjustments (e.g., the loss of the conjunction e 'and'). The SDRA could therefore be regarded as a case of multiple source construction (Van de Velde et al. 2015). Then, once the new construction is established, it acquired a life of its own, and hence new properties, among which the capacity of developing new senses (e.g., related variety).
Under this view, the related <X Adv 1 Y Adv 2 > pattern we discussed in Section 4.1 may also find an explanation: this pattern displays semantically related elements (X and Y) instead of the repetition of the same element X (see [22]: complimenti di qua, battute di là 'many compliments and jokes going on' [lit. compliments here, jokes there]), but if we accept that reduplication proper and other similar constructions featuring semantically related words are actually part of the same abstract mechanism, as proposed by Inkelas and Zoll (2005), then this pattern might be considered as a construction related to the SDRA Cxn. In fact, as suggested by an anonymous reviewer, the very occurrence of <X Adv 1 Y Adv 2 > expressions provides still another argument in favor of the view that the SDRA Cxn represents a construction in its own right, since these systematic 'deviations' from the core SDRA Cxn are only interpretable because the higher-level construction is already established. The exact nature of the synchronic relation between the two constructions (e.g., daughter construction, sister construction, allostruction) depends on the specific properties of the related pattern under discussion, which were not analyzed in detail here. We leave this question open to future research.

A crosslinguistic tour
Given the analysis presented in previous sections, the SDRA would seem to be a rather peculiar construction, which is most likely to be language-specific and, thus, limited to Italian. However, a very preliminary crosslinguistic look revealed the presence of SDRA or SDRA-like patterns in other, typologically different, languages.
For instance, we found an example of what seems to be a case of SDRA in Russian (taken from the ruTenTen11, or Russian Web 2011, corpus, searched via Sketch Engine): The only difference we can point out between this example and the SDRA in Italian is that in (44) we have a reversed order of the antonymic adverbs: in Italian we find the order X i (di) qua X i (di) là 'X i here X i there', whereas in the Russian example the order is X i tuda X i sjuda 'X i there X i here'. Note that, in Russian, the corresponding compound involving the same antonymic pair also displays this order: tuda-sjuda (lit. there-here) 'everywhere. ' We also found an interesting example in (spoken) Modern Hebrew (45), taken from Inbar (2020) (glosses and translation like in the original). similar to the SDRA-related pattern discussed in Section 4.1, where X is not repeated but combined with a semantically related element Y. In this respect, see the following example from the itTenTen16 corpus, which contains two elements (X and Y) very close in meaning to those displayed in (46) (Finkbeiner 2015: 90) Obviously, this pattern shares several properties with the SDRA in Italian (including the possibility to host two non-identical but semantically related nouns, cf. Section 4.1), but at the same time displays some dissimilarities. At the formal level, the N hin, N her construction can host only nouns at the word level (no noun phrases allowed). Syntactically, the pattern as a whole is juxtaposed to a main clause and mostly occurs in the "left outer field" position (Finkbeiner 2015: 96), unlike the Italian SDRA, which displays a more flexible behavior. More marginally, the antonymic adverbs are reversed with respect to Italian: we have X i (di) qua X i (di) là (X i here X i there) in Italian while X i hin X i her (X i there X i here) in German, as in Russian.
At the functional level there are differences too: Finkbeiner (2015: 97) claims that the N hin, N her construction has two semantic readings plus pragmatic values. On the one hand, it has a concessive reading, defined as "a semantic relation between two propositions p and q, which both are entailed, but which normallyaccording to a speaker's previous experience, world knowledge, etc.are expected to be mutually incompatible" (Finkbeiner 2015: 97). Hence, the sentence in (48) can be paraphrased as follows: 'Although we are an equal rights society, the Cartell association remains what it always has been, a men's club.' The other reading of N hin, N her is concessive conditionality: concessive conditionals "relate a set of antecedent conditions to a consequent [and] the consequent is asserted to be true under any of the conditions specified in the antecedent" (König 1986: 231, quoted in Finkbeiner 2015. Thus, Example (49) below can be paraphrased as: 'Whether there was a spelling reform or not: What really interests the true Mainz resident is neither clarified by the Duden nor by anyone else. Is it Määnz or Meenz?' (Finkbeiner 2015: 97 (Finkbeiner 2015: 97) Beside these semantic functions, N hin, N her also has some pragmatic values. At the discourse and interactional levels, it has the function of "construing textual coherence" and "expressing subjectification and stance taking" (Finkbeiner 2015: 104). In particular, the syntactic configuration of the construction "contributes to its pragmatic status as a discourse marker," which has the role of positioning the speaker "with respect to a state of affairs that is part of the discourse universe, creating the expectation that a contrasting attitude will be presented in the subsequent utterance (the juxtaposed main clause)" (Finkbeiner 2015: 103-104). 14 In conclusion, what emerges from this brief crosslinguistic tour is that, while the Italian SDRA is a language-specific device with its own specific properties, other languages present similar patterns that involve the same building blocks (the reduplicative pattern on the one hand, the antonymic adverbial pair on the other) and might share some of its functions. Obviously, these patterns will also have their own specific (especially functional) properties. One hypothesis that might be worth testing is the common origin of these patterns as multiple source constructions, as we proposed for the Italian case in Section 6: our prediction would be that languages that display SDRA-like structures also display (or displayed), as independent constructions, some reduplicative pattern and an antonymic adverbial pair pattern (irreversible binomials or compounds).

Conclusions
In this article, we identified aso far undescribedpattern in Italian that qualifies as an instance of both syntactic reduplication and discontinuous reduplication. Since this pattern is characterized by the non-adjacent repetition of the same item within a larger fixed configuration defined by a pair of antonymic spatial adverbs, we called this structure syntactic discontinuous reduplication with antonymic pairs (SDRA). The SDRA conveys a general value of plurality and develops a network of sub-constructions that express more specific functions such as distributivity, related variety, and dispersion. By virtue of its idiosyncratic properties, we analyzed the SDRA as a construction in the sense of Construction Grammar. We also proposed regarding the SDRA as a multiple source construction that has originated from the blending of two independent constructions: syntactic reduplication on the one hand and irreversible binomials with antonymic adverbs on the other. SDRA-like patterns are found in other languages, too. However, despite their resemblance to the structure analyzed here, they also display their own distinctive features. A broader crosslinguistic investigation would be necessary to assess the actual spreading of SDRA-like patterns in different languages and to test our hypothesis about its 'multiple source' emergence.