On Italian spatial prepositions and measure phrases: reconciling the data with theoretical accounts

3.1 New data: new items and types

The goal of this section is to identify a larger set of Italian spatial prepositions beyond the lists given in (7) and (8). To reach this goal, we adopted the following methodology. We completed a corpus study by querying the PAISÀ corpus of contemporary Italian (Baroni and Bernardini 2016; Lyding et al. 2014). Following Ganfi and Piunno’s (2017) approach, we collected tri-grams realizing the P + N + P template (e.g., a+N + di, in + N + di, di + N + a), with specific queries including contracted prepositions (e.g., al/ai + N + di). The choice of collecting contracted prepositions via dedicated queries ensured that all possible contracted forms were identified. We verified the spatial senses of each token by checking their dictionary entries, or those of the attested noun, in Gabrielli (2015). For instance, we treated ai piedi di as a spatial preposition by verifying that one of the definitions for the noun piedi includes its use in this preposition. We also verified that either a singular or plural noun/nominal form was attested (e.g., that al piede di was not attested).

Overall, di and a emerged as the most frequent prepositions that would act as heads introducing Ground NPs, in e.g., the structures P + N + a, P + N + di. In, a and su were the most common “markers”, i.e., simple prepositions preceding nouns (e.g., nei pressi di ‘in the vicinities of’, sulla testa di ‘on the head of’ and alla testa di ‘at the head of’). Even if rare, per and tra/fra could act as markers (e.g., per il centro di ‘across the center of’, and the emerging tra i sobborghi di ‘through the suburbs of’). We identified 68 items displaying these properties, although a higher number may probably be identified (cf. Aurnague and Vieu 2015; Fagard et al. 2020).^[5]

We identified two sub-sets of systematically understudied prepositions: the former lacking initial contracted prepositions (e.g., in cima a), the latter including these prepositions (e.g., nei pressi di, sulla testa di). We sampled their overall frequencies and determined their diachronic emergence as follows. Ganfi and Piunno (2017) reports that the structures a + N + di, a−Det + N + di are attested in the OVI (Opera del Vocabolario Italiano),^[6] a corpus of written Old Italian. The earliest tokens are attested in the 13th century, though absolute frequencies appear low. These two structures increase in frequency in the La Repubblica corpus of contemporary written Italian (Baroni et al. 2004), thus becoming the most commonly attested prepositional structures.

To verify that this pattern holds for the sub-type of spatial prepositions, we sampled the diachronic emergence of two pairs of prepositions realizing the contracted/uncontracted alternation. The first is a destra di/alla destra di ‘to the right of’; the second, a sinistra di/alla sinistra di ‘to the left of’. We also used the OVI and the La Repubblica corpora to ensure the integrity of the results. We found 7 tokens for a destra di and 1 for alla destra di in the OVI; no tokens for the alla sinistra/a sinistra di pair. In the Repubblica corpus, we found 175 (0.46 per million words/pmw)/310 (0.81 pmw) tokens for a sinistra/alla sinistra di, and we found 151 tokens (0.45 pmw)/313 (0.82 pmw) for a destra di/alla destra di. We acknowledge that the two corpora are heterogeneous in temporal coverage and size (380 M tokens for the La Repubblica Corpus, 1980–current; 30 M for the OVI corpus, 9th–15th century). Nevertheless, the data support the view that complex uncontracted and contracted prepositions represent a (relatively) novel category in Italian.

Overall, we conjecture that prepositions of the spatial type mostly emerged in Italian with low but increasing frequency over the 19th, 20th and 21st centuries. We therefore use the label locutional to describe this type of preposition, in order to be consistent with previous descriptive works on this matter (e.g., Ganfi and Piunno 2017; Hoffmann 2005). We acknowledge that simple, complex and locutional prepositions share precise structural properties that we fully discuss and account for in the next sections. However, the use of this label permits us to describe the properties of such multi-morphemic prepositions in a distinctive manner, while also paving the way for our unified account.

Let us summarize our corpus results. We offer two lists of respectively 16 and 30 items as examples of locutional prepositions that we retrieved as described above in (13) and (14). Our query retrieved a higher number of items for each type (27 and 41 respectively, i.e., 68 items altogether), but we opt to offer this sample for two reasons. First, for the lists in (13) and (14), we could confirm that Gabrielli (2015) had entries reporting their definitions and status. Other prepositions appeared to be rare or newly coined, and thus to have uncertain status in this dictionary (11 uncontracted, 11 contracted). We believe that our generalizations may extend to all prepositions that may eventually be found in PAISÀ or other corpora, or in spontaneous speech. However, we opt to take a conservative stance on this mini-lexicon, to avoid incorrect generalizations. Second, we decided to use this sample of 46 prepositions in our elicitation tasks (see Sections 3.2 and 3.3), so the number of prepositions attested below suffices to test their distribution with MPs:

Two observations are important at this stage. First, the group in (14) includes several prepositions involving nominal items taking fixed plural form (e.g., piedi in ai piedi di, dintorni in nei dintorni di). The group in (13) only includes items that may still display clear nominal origins (e.g., fondo in in fondo a), but only occur in the singular form. However, both groups support the view that these Italian prepositions have a rich grammatical structure, which confirms the validity of the descriptive template P + N + P. Therefore, this group of corpus data shows that Italian includes a richer inventory of spatial prepositions than previously assumed and that the contracted/uncontracted alternation appears in both P’s forming prepositions.

Second, for certain items (e.g., fondo ‘bottom’, cima ‘top’, mezzo ‘middle’) one can observe a systematic relation between (un)contracted and governing prepositions. For instance, one can have uncontracted in mezzo a ‘between, in the middle’, but contracted al mezzo di, nel mezzo di ‘in the middle of’. The presence of a contracted first P invariably determined that the second P was di, irrespective of the item realizing the first P. Uncontracted and contracted pairs may offer distinct, nuanced contributions to the senses of prepositions, as the senses associated to in mezzo a and nel mezzo di suggest. For our purposes, however, the crucial aspect is that contracted prepositions seem to select di as a governing head preposition (cf. also Franco et al. 2021; Piunno and Ganfi 2020).

Third, the data also indirectly show that simple, complex and locutional prepositions are distinct but structurally connected types. One can conceive of complex prepositions as the combination of simple prepositions as heads and spatial nouns, without a marker preposition (e.g., dietro a). This view can be represented via the ∅ _P  + N + P template. The marker preposition is silent or “empty”, and the head preposition can undergo ellipsis, leaving the noun as a remnant or adverbial-like element (Garzonio and Rossi 2016; among others). One can then treat locutional prepositions as the combination of a preposition as a possibly contracted marker with a spatial noun and a simple preposition as a head (e.g., a sinistra di/alla sinistra di), corresponding to the P + N + P template. Locutional prepositions thus include both marker and head prepositions. Simple prepositions, by contrast, (e.g., in, a) can instantiate the template ∅ _P  + ∅ _N + P, i.e., prepositions involving silent spatial nouns and markers. Thus, these types seem to be reducible to a general template, which can be realized via different forms or sub-types. The next section addresses the question of whether these types share common syntactic properties in sentences.

3.2 New data: syntactic validation

The goal of this section is to provide evidence confirming that the newly attested prepositions display the same syntactic distribution as standard prepositions, given their structural homogeneity. The data are based on four standard tests individuating spatial prepositions: Basic Locative Constructions, answers to where-questions, Ground NP ellipsis and locative inversion (cf. Hagège 2010: Ch. 2; Svenonius 2010; Ursini 2014, 2015). The data come from an elicitation task in which informants (N = 50) had to judge whether the sentences were grammatical or not. We used an elicitation task because the corpus data only included declarative sentences. This sentence type offers important evidence on the distribution of PPs, but does not directly illustrate their status as distinct phrasal constituents. We aimed to pursue other forms of syntactic validation of the “new” prepositions via this complementary methodology, to address this constituency issue.

The task worked as follows. Informants were asked to evaluate sentences describing spatial relations in fictional contexts. Our method was based on De Clercq and Haegeman’s (2018) study of definite articles in the Ghent dialect of Flemish. Participants evaluated sentences on a Likert scale, ranging from “1” (“unacceptable”) to “5” (“perfect”). The study considered answers in the “3–5” range as acceptable, and those in the “1–2” range as unacceptable. The study also reported intra-speaker variation via a score index on the fourth gloss line. Given 20 participants’ answers, a sentence receiving 16 answers with a “5” value, 2 with a “4” value and 2 with a “3” value would have the following score index: 1⁰ 2⁰ 3² 4² 5¹⁶. That is, in this case, 16 participants (super-script) answered that the sentence was fully acceptable (value “5”). Score lines also indicate the mean value: in this case, 4.7.

For our goals, we adapted the task as follows. We also asked participants to choose values on a 1–5 Likert scale. Since the contexts would always match the sentences’ content, the sentences were always true. For instance, the sentence in (15), Mario va in fondo al vicolo, was paired with a context in which an individual named Mario reached the end of an alley. The sentence in (31), Mario va dieci metri nella caverna, was paired with a context in which Mario entered a cavern and covered a distance of 10 m. However, while participants accepted (15), they mostly rejected (31), though both sentences were syntactically well-formed and true. If informants found sentences unacceptable, they would reject them because of their possible morphosemantic mismatches. Therefore, (31) may have been rejected because the preposition in and the MP ten meters lack features licensing semantically accessible interpretations: it would be uninterpretable in context.

Differently from De Clercq and Haegeman (2018), we considered scores between 1.0 and 2.0 as marking uninterpretable sentences; between 2.0 and 3.0 as marginal/marked sentences; between 3.0 and 4.0 as interpretable; and 5.0 scores as near-ideal. For each sentence, informants were also invited to offer comments regarding their (un)interpretability, below each tested sentence. When they did so, we evaluated the answers to see whether they supported our predictions regarding the (un)interpretability of the sentences. Our more restrictive use of values was intended to detect whether subtle distributional differences (e.g., the verb type with which prepositions combined) affected the overall acceptability and interpretability of the sentences. Furthermore, following Schütze and Sprouse (2013) we randomized sentence types to make sure that participants did not receive cues regarding the intention of the experiment. In this and the next section, we present examples from the tasks, also including a fourth “results” line, clarifying the role of the results for our analysis where needed.

The elicitation task involved 72 to 74 prepositions: 26 “old” prepositions, 46 “new” prepositions from the corpus study, plus up to 2 ill-formed prepositions (e.g., *cima instead of in cima). These prepositions occurred in 5 different construction types (BLCs, where questions, locative inversion, Ground NP ellipsis, distribution with MPs), and with 2 verb types (directional/locative, but see the comments on locative inversion in the next section). For 50 informants, we had 32,700 token sentences representing 654 sentence (sub)-types. We only discuss a representative sample of sentence types, as a full-fledged discussion of all patterns and combinations would take us too far afield.

Let us move to the experiment. The first test involves the occurrence of prepositions in Basic Locative Constructions (BLCs), defined as sentences that can act as full answers to where-questions (Levinson and Wilkins 2006: Ch. 1). The literature usually focuses on full-fledged locative constructions, i.e., sentences only denoting locative spatial relations. To keep the description of our data compact, however, we also use this label to discuss sentences denoting motion-based relations. BLCs offer evidence on the role of PPs and their spatial contribution in declarative sentences (cf. also Hagège 2010: Ch. 2). They thus allow us to discuss the role of locutional prepositions as heads of spatial PPs, and to verify that it patterns with that of simple and complex PPs.

The second test involves fragment answers to where-questions (cf. Jackendoff 1972; Svenonius 2010; cf. also Merchant 2001: Ch. 3). Answers license semantically congruent question–answer pairs only if they match the category that a wh-word selects. This entails that where-questions individuate PPs as fragment answers, and thus prepositions as their heads, regardless of their type. The third test, locative inversion, involves a fronting operation that targets spatial prepositions and the PPs of which they are heads, moving these PPs in sentence-initial position (Cennamo and Lenci 2019; den Dikken 2006; Svenonius 2010). For the most part, this form of fronting is licensed when prepositions carry locative meaning, whence the “locative” label. It thus offers key evidence that PPs and locutional prepositions as heads of these PPs have the same distributional properties as simple and complex prepositions. We show a sample of our results in (15)–(26):

Consider (15)–(18) first. Standard BLCs can include any preposition type heading a PP, including locutional prepositions (cf. in fondo a and nei dintorni di in (15)–(16)). However, certain prepositions require the presence of di when a pronoun or a spatial indexical is the Ground NP (respectively loro ‘them’ and qui/li ‘here/there’ (cf. verso, tra and su in (17)). This pattern seems to realize a form of “goal marking”, the explicit marking of the goal NP via a dedicated preposition (e.g., Kittilä 2008, 2014; Luraghi 2017). It has also been discussed in Franco (2020), arguing that su, giù, tra and fra are better thought of as a sub-type of complex prepositions. For instance, tra (di) can be considered a complex preposition with an optional di head. An element appears to involve the univerbation (i.e., fusion) of a marker with a spatial noun (e.g., from Latin in-tra to Italian tra: cf. also Fagard 2010: Ch. 3; Luraghi 2011).

This pattern entails that only a, da, di and in are full-fledged simple prepositions, i.e., prepositions potentially acting as heads of a complex preposition. These prepositions cannot combine with di in this type of BLC, as otherwise the sentence will be ungrammatical (cf. (18)). The pattern also entails that su, per, tra and fra are complex prepositions that may (or may not) combine with MPs. Simple prepositions thus seem to mostly mediate between Ground NP and a unit composed of a spatial noun and a spatial marker that may undergo univerbation. In the case of di, this form of mediation may involve the absence of a distinct spatial sense. This preposition can be considered a spatial preposition if it governs spatial nouns and Ground NPs, thus carrying an abstract relational sense (e.g., a ‘part-of’ relation: Franco et al. 2021). Nevertheless, its inclusion in the list in (27) signals that di can also be considered a simple preposition like a, da and in. We exemplify this new partition in (27)–(28):^[7]

Consider now (19)–(22). The fragment answers in (21)–(22) show that either a simple or a complex preposition can head this PP (respectively nel and dietro a). The same fact holds for the new locutional prepositions, whether they belong to the contracted or uncontracted type (respectively in cima a, ai piedi di in (19)–(20)). The presence of a marker preceding a noun, contracted or not (e.g., in for in cima, ai for ai piedi) was deemed necessary. Answers involving the omission of this marker were thus considered ungrammatical, as we show in (20) (i.e., we have *piedi della montagna). Participants invariably considered markers and nouns as forming a single unit (e.g., ai piedi) when forming PPs becoming part of question–answer pairs.

Each preposition type can also head a PP in inverted position, as (23)–(26) show. Thus, uncontracted in cima a and contracted locutional ai piedi di (respectively (23)–(24)), and simple in and complex di fronte (respectively (25)–(26)) are possible prepositions that can undergo inversion. Note that the omission of the marker preposition would also render a sentence ungrammatical, as the example in (24) shows (i.e., piedi della macchina could not be an inverted PP). Overall, the examples in (15)–(26) confirm that both types of full-fledged locutional prepositions display the three key syntactic properties of prepositions. They can be heads of PPs in declarative sentences (BLCs), in interrogative sentences (specifically, fragment answers), and they can occur in inverted, topical positions.

The fourth test involves the aforementioned Ground NP ellipsis. As we mentioned in the introduction, Ground NP ellipsis permits us to distinguish simple from complex prepositions because complex prepositions will always involve a remnant constituent when ellipsis occurs. Once we factor in the locutional type, the test can be used to confirm the role of noun and marker as forming a distinct sub-constituent within prepositions. As we have shown via (5)–(6), only complex prepositions can undergo ellipsis in Italian (e.g., the complement NP of a cannot be elided, the complement NP of dietro a can). Unsurprisingly, locutional prepositions follow a similar pattern:

While in cima a in (29) involves in and the spatial noun cima jointly forming a remnant, nei pressi di in (30) involves the contracted plural preposition ne-i ‘in-the.pl’ forming a remnant with the plural noun pressi. As in the case of question–answer pairs and locative inversion, omission of the marker preposition leads to ungrammaticality (cf. (30) and the results associated with “bare” pressi). Hence, remnants of elided locutional prepositions must involve a combination of a possibly contracted marker and a spatial noun occurring in singular or plural (but fixed) form.^[8] Thus, the ellipsis data also confirm that two macro-units can be identified in Italian prepositions. One unit is a marker plus spatial noun unit (i.e., (P + N), e.g., in cima), possibly undergoing univerbation (e.g., Latin dentro from de plus intra). The second unit is a simple preposition mediating between this univerbed unit and a Ground NP (i.e., [P + N] + P, e.g., a in [in cima] a).

Overall, these results show that Italian has four grammatical types of prepositions: simple (e.g., a), complex (e.g., dietro a), uncontracted locutional (e.g., in cima a) and contracted locutional prepositions (e.g., nel centro di). At the same time, these results confirm that the four types are tightly connected, as they involve the presence or absence of spatial nouns and markers determining the increasing complexity of each type.^[9] The results also invite the conclusion that simple prepositions only include a quartet of items: a, da, di and in. Per, tra/fra, su display the properties of complex prepositions (e.g., taking an optional di in context), even though they cannot be traced to their nominal roots in a direct manner. Hence, Italian spatial prepositions display a homogeneous syntactic distribution, irrespective of their increasing morphological complexity. The subsequent question is whether there is a correlation between structural type and distribution with MPs, and if this correlation hinges on the properties of one or more “P units”.

3.3 New data: distribution with MPs

The goal of this section is to offer evidence on whether the region/projective alternation corresponds to the contracted/uncontracted alternation. The informants participating in the first half of the study were subsequently asked to evaluate sentences including items from each morphological sub-type in their distribution with MPs. Our findings paint a nuanced picture. First, simple prepositions do not combine with MPs. However, they seem to prevent this distribution by virtue of their denoting “general” geometric relations (cf. a, in, a, da). Second, most but not all complex prepositions can be classified as projective prepositions (14/22 of the items; 8/22 are region prepositions). Third, uncontracted prepositions are almost evenly divided between region and projective prepositions (8/15 for region, 7/15 projective items). Fourth, contracted prepositions are mostly classifiable as region prepositions (16/25 items; 9/25 are projective items).

We present a sample of the findings in (31)–(54). We include examples of prepositions combining with verbs of directed motion (i.e., andare ‘to go’, scendere ‘to go down’) and location (e.g., trovarsi ‘to be located’, sedere ‘to sit’, essere ‘to be’). We clarify the role of verbs on a case-by-case basis and organize our discussion around the “new” taxonomy of prepositions. Consider (31)–(34) first.

As (31)–(32) show, the simple preposition nel cannot combine with the MP dieci metri, whether it combines with a directional or locative verb (respectively va ‘goes’ and si trova ‘is located’). The same pattern holds for a, which cannot involve a measurement of the distance between Figure and Ground, even when the Ground acts as a goal (cf. (33)–(34)). Note here that intra-speaker variation is observed, as some speakers did not reject these examples. For instance, six speakers found (31) marked (i.e., they evaluated it as a “2”), one found it a good example (i.e., a “4”), and three rated it as a perfect example (i.e., a “5”).^[10] In general, however, informants observed that these prepositions denoted relations entailing a near-null distance between Figure and Ground. The senses of MPs and prepositions did not “match”, even if each item could independently capture properties of the relevant spatial relations.

A different picture holds for newly defined complex prepositions, e.g., per and fra. These prepositions can combine with MPs when they co-occur with figures also referring to locations (Talmy 2000: Ch. 4). When a location acting as a figure covers or occupies an extended stretch of space with respect to the Ground, the corresponding sentence can include an MP (cf. (35)–(36)). As the detailed scores show, however, informants do not necessarily agree on the acceptability of these sentences. As for (35), two participants found it unacceptable (i.e., “1”); 11 participants found it perfectly acceptable (i.e., “5”). With a mean value of 3.64, this sentence can be considered acceptable, though not perfectly so. These and the other “new” complex prepositions, however, trigger perfectly acceptable values when combined with verbs of direction/motion (cf. (36)–(38)):

“Old” complex prepositions display a potentially simpler interaction with MPs. Some prepositions resist distribution with MPs regardless of the verb type they combine with (e.g., accanto a in (39)–(40)), whereas others license this type of distribution with any verb type (e.g., dietro a/dietro in (41)–(42)). The presence of optional a as a head does not seem to affect this pattern, as mentioned in the introduction:

Locutional prepositions also offer a similarly complex picture. Uncontracted locutional prepositions generally combine with either directional or locative verb types, too (cf. in fondo a in (43)–(44)). They may or may not combine with MPs, though most informants generally observed that the acceptability of such constructions hinged on whether they referred to “locations” or “directions” of some sort (cf. a sinistra di in (45)–(46)). The same pattern, however, holds for contracted prepositions. While al lato di ‘to the side of’ can combine with MPs (cf. (47)–(48)), alla fine di ‘at the end of’ cannot do so, irrespective of the verb type it combines with (cf. (49)–(50)):

From these data, we can conclude that the clear-cut predictions that Cartography makes for prepositions, the PPs they head and their distribution with MPs do not seem not to be entirely borne out. MPs do not necessarily combine with PPs only if they have uncontracted prepositions as heads. Furthermore, it seems clear that the head preposition does not play a role in this pattern. Whether a (e.g., dietro a) or di (e.g., al lato di) are heads of prepositions, their presence does not seem to affect their distribution relative to MPs. This fact strongly suggests that the spatial noun and the marker should determine the licensing or blocking of MP phrases. This holds independently of whether they form a single univerbed unit (e.g., per, dietro) or a phrase potentially acting as a remnant (e.g., di fronte, al lato).

The final piece of the puzzle that we must address pertains to the distinct roles of markers and spatial nouns in this pattern. Our finding was that in locutional prepositions, the marker combining with a spatial noun (e.g., al centro vs. nel centro) did not affect the distribution of prepositions/PPs with MPs. The spatial noun seems to determine this distribution via its contribution to the overall PP in virtue of the type of location to which it refers. We offer examples involving the pairs alla sinistra di/sulla sinistra di ‘to/on the left of’, and al centro di/nel centro di ‘at the center/in the center of’. For the sake of simplicity, we illustrate this pattern by only using the locative verb si trova ‘is located’, in (51)–(54):

As these examples show, the presence of either a or su (for sinistra), or a and nel (for centro ‘center’) does not seem to play a role in the licensing of MPs with PPs. Our general finding was that informants usually pointed to the role of the location or direction under discussion. Therefore, spatial nouns (e.g., sinistra, centro) seem to be the core licensors of MPs. We can thus conclude that irrespective of the simple prepositions constituting prepositions, the distribution of the corresponding PPs with MPs seems to hinge on a single factor. Only the lexical content of the spatial nouns included in each complex and locutional preposition seems to determine the region/projective status of a PP, and its distribution with MPs. This seems to be the case independently of whether these nouns are lexicalized as univerbed parts of prepositions (e.g., per, dietro), or as distinct nouns displaying noun morphology (e.g., fondo in in fondo a, piedi in ai piedi di, and so on).

Some further observations are helpful to understand our results. First, for the simple prepositions a, di, in and su, almost a quarter of informants (N = 12) commented that these prepositions only introduced a “general” spatial relation. Five participants explicitly mentioned the absence of “nouns” and/or “adverbs” being incompatible with the presence of MPs. Without such elements, it was unclear what “direction” was being measured. Simple prepositions thus correspond to the geometrical types: they are prepositions denoting geometrical, general configurations (Aurnague and Vieu 2015; Herskovits 1986). Second, the definite article and its ability to mark a matching NP as carrying definite, specific features do not seem to play a role in the distribution of prepositions and PPs with MPs (cf. also Matushansky and Zwarts 2019). The status of a noun as denoting a measurable scale (or not denoting such a scale) seems to hinge on its lexical content only, as our examples suggest.

A third finding is that even if per, tra/fra and other prepositions seem to favor distribution with MPs and verbs of motion/direction, the role of verbs in these patterns was often marginal. This suggests that nuanced distinctions between prepositions denoting paths or projections proposed in the literature (e.g., Zwarts 2005, 2008; Zwarts and Winter 2000) are not crucial, according to our data. One may argue that prepositions such as a sinistra di ‘to the left of’ may be ambiguous between path-denoting senses and projection-denoting senses (cf. again (45)–(46)). MPs, then, can measure either type of spatial entity, when present. Given our analysis of each item’s contribution to their distribution with MPs, this entails that spatial nouns should be the source of this lexical and semantic ambiguity in context.

From these results and observations, we can offer the following semantic taxonomy of Italian spatial prepositions. The items in the lists in (55)–(57) are partitioned according to their morphological types:

Overall, the corpus- and elicitation-based data offer a coherent picture of the properties of Italian spatial prepositions. From this coherent picture, we can make four generalizations:

First, “bare” simple prepositions are mapped onto the geometric type.

Second, spatial nouns and nominal-like elements moving to the prepositional domain via grammaticalization uniquely determine which semantic type is assigned to complex and locutional prepositions. This seems to be the case independently of whether these items may undergo univerbation with markers (e.g., per, di-etro, in-torno),^[11] or whether they occur as distinct nouns in locutional prepositions (e.g., cima, pressi).

Third, nouns and other nominal-like elements encode information about the type of spatial entity at a grammatical and lexical level. If nouns can project information about the type of referent they denote to the prepositions they are part of, semantic content must be governed by some grammatical feature.

Fourth, this projection mechanism appears to reach the PP and sentential level, and it seems to govern the distribution relative to MPs. A mechanism other than semantic interpretation seems to determine this pattern. That is, the sense of nouns such as cima must determine the sense of prepositions such as in cima a. It must also determine the sense of the phrases they form (e.g., in cima alla montagna) and the sentences they become part of, irrespective of subtle intra-speaker variation. Overall, an account of our data must model the properties of preposition types in which grammatical structure determines semantic interpretation via one mechanism.

We thus have reached our first goal, having shown that previous accounts make incorrect predictions regarding the MP data. We have shown that in Italian spatial nouns act as the unique locus of variation with respect to these data. We have obtained such a novel, nuanced result via a thorough testing of the conditions under which MPs can combine with PPs. Therefore, a new account must be offered to model these novel data.

4.1 Basic assumptions

Our account of Italian prepositions uses Lexical Syntax treatment of adpositions (Hale and Keyser 2002: Ch. 4). We chose this framework because its flexible approach to categories permits us to offer a potentially more coarse-grained, but less restrictive analysis of the data. Two reasons play a key role in our choice. First, by using Lexical Syntax we can abstract away from more iconic and therefore more problematic labels such as AxPart, RelN, Kase, Place and similar labels found in Cartography. In this way, we can also avoid unnecessary assumptions about the valence and features of each category. Second, we can implement an analysis of the data that explicitly represents the features associated with each category. We can thus account for how PPs and MPs can combine insofar as their constituting categories carry matching features. Lexical Syntax’s key assumptions for our account can be summarized as follows:

First, language-specific categories (e.g., verbs, nouns in Italian) can project at least one of four language-general head types. Language-general types, in turn, are defined in terms of their valence, i.e., the number of arguments they take. Depending on the valence of a lexical item in a syntactic context, an item can instantiate a 0-place, 1-place, or 2-place head type. Heads can thus combine or merge with 0, 1 or 2 phrasal arguments, forming a corresponding phrase. A 0-place head represents a “bare” argument (i.e., a phrase). A 1-place head represents an affix or other head merging only with a complement. A 2-place head type represents a “relational” head, i.e., a head merging with a specifier and a complement. Although the framework proposes a second 2-place head type, we can ignore this distinction without loss of precision (cf. Mateu 2002). As we fully show in Section 4.2, spatial nouns and Ground NPs (e.g., fronte, la macchina) project 0-place heads. Markers (e.g., di in di fronte) project 1-place heads. Governing Ps project 2-place heads (e.g., a in di fronte a).

Second, prepositions can correspond to either 1-place or 2-place heads. When prepositions take a Ground NP and another argument (a second NP, e.g., the boy in the park), they act as 2-place heads. When they take only a Ground NP, they act as 1-place heads (e.g., at school). Hale and Keyser’s (2002) treatment of prepositions and their morphosyntactic structures does not include an analysis of complex prepositions. However, it hinges on the assumption that complex prepositions can involve multiple prepositions arranged into hierarchical structures (the “P within P” hypothesis: Hale and Keyser 2002: Ch. 8). Here we explore an alternative interpretation of this hypothesis, which in part resembles Acedo-Matellán’s (2016) account of verbal prefixes and adpositions. In a nutshell, Acedo-Matellán (2016: Ch. 2) proposes that spatial adpositions may include an abstract 2-place head possibly relating Figure and Ground to each other (e.g., to in into). A second 0-place head (a root) determines the content of this relation (e.g., in in into). Section 3 shows that spatial NPs in complex prepositions seem to cover this content-determining role, at least in Italian spatial prepositions.

Third, each projected category can be further specified for the grammatical features it carries. These features can thus determine the selection of the arguments that a head combines or merges with, and the features that the corresponding phrase displays. Hale and Keyser’s (2002) framework does not offer fully explicit rules to account for how features are represented, and how merge can also combine features into complex feature structures. It thus raises the question of how to represent the fact that spatial nouns in prepositions guide the licensing of MPs, once their features project at a sentential level. For this reason, we extend Hale and Keyser’s (2002) framework with a formal treatment of features (Adger 2010). Each head H is represented as projecting a category (e.g., Preposition P, Verb V, Determiner D, Noun N) and the features associated with this category (e.g., tense for verbs), which can also have values (e.g., tense:past). Categories and features with values are represented via sequences of sub-scripts, e.g., H_<P,proj>. To avoid making our account too cumbersome, we only spell out a full feature-based analysis for prepositions and MPs. We thus leave aside a full-fledged representation of verbs’ nuanced contribution to sentences, since it does not play a crucial role in our argument.^[12]

4.2 The account

We use the following notation to represent each category. Prepositions in their “simple” distribution correspond to 2-place head types, represented with the label P (cf. Bottari 1985a, 1985b; Emonds 1985). Prepositions in their role as markers of spatial nouns correspond to 1-place head types, represented with the label P’ (P prime). The resulting argument is a P’P, the “P within P” indirectly predicted by Hale and Keyser’s (2002) hypothesis. We thus treat a 1-place P as a head marking a prepositional (and spatial) role for spatial nouns (cf. Pantcheva 2010 for a similar proposal). We then assume that simple (i.e., 2-place) Ps are ambiguous: they can carry d(irectional) or l(ocative) features, with verbs selecting either feature in BLCs.

Spatial nouns are 0-place heads (i.e., NPs) merging as the complements of P’ items, and carrying the feature ±degree. That is, spatial nouns can carry a feature determining that the location types in their denotations form an ordered scale of entities, or degree (cf. Matushansky and Zwarts 2019; Morzycki 2015: Ch. 2). When merging with 1-place Ps, the deg(ree) feature is merged with the s(patial) feature of Ps to form a P’P carrying one of two complex features: r(egion) or p(rojective). We treat P’Ps as the category determining the corresponding sense type alternation, inherited from the value assigned to their spatial noun and inherited by the PP. Thus, the feature equations projective = [+spatial;+degree] and region = [+spatial;−degree] hold. Since number features in NPs do not play a role, we omit them from our structures, thus only marking each NP as carrying region r or projective p features.

We assume that P’s project a phrase by taking these bare NPs as complements. We also follow Franco (2016) and assume that diachronic changes have formed conflated P’Ps (e.g., in-torno ‘around’) or items involving univerbation (e.g., dietro ‘behind’ from Latin de retrum ‘from back’). We have three reasons to conclude that these apparently different prepositional “fragments” can be represented using a single structure, as discussed in Section 3. First, these items can become remnants in ellipsis patterns, as distinct units: P’Ps as specifiers of another P. Second, irrespective of the processes forming these items (conflation, univerbation, synchronic merge), this account assigns them a single syntactic position and function: determining the semantic type of a preposition. Third, our uniform account of PPs and their licensing patterns builds on the contribution of spatial nouns/NPs and the P’Ps embedding them.

On the basis of these assumptions, we can reconstruct each attested morphological type via a unified account that predicts the semantic type of each vocabulary item. This is because only nouns can contribute either an r or a p feature to their prepositions. The projection mechanism takes care of projecting these features at a phrasal and then sentential level so that feature-matching with MPs’ features can occur. We can consequently predict that simple prepositions lack this distinction altogether, since no specifier is realized, and that complex prepositions either denote a region or projective type, respectively associated to −d or +d features. We offer the structures for each type in (58)–(61), before discussing this prediction. Note that in (59a)–(59b) we offer the structures assigned to prepositions involving optional a heads (e.g., dietro a and dietro). In (59c), we represent the structures assigned to prepositions blocking heads (e.g., verso), in (59d), those involving optional di (e.g., verso di). The feature b(ounded) marks bounded readings for prepositions (cf. again Tortora 2008); the feature a(nimate) stands for prepositions selecting Ground NPs marked as such:

We assume that simple prepositions involve null markers (e.g., alla macchina ‘at the car’ in (58); cf. Folli 2002; Tortora 2008; Garzonio and Rossi 2016). We simplify the representation of contracted prepositions by treating definite articles and their features as an integral part of prepositions’ structure (i.e., we have <P,±d,def>; cf. Asbury 2008: Ch. 4). This apparent simplification allows us to capture the fact that definite articles only seem to contribute a def(inite) feature to prepositions, specifying that a given Ground and region/projection thereof are under discussion (cf. also Matushansky and Zwarts 2019). Note that contracted prepositions without markers would not include a P’P specifier and would thus be ill-formed. Our account predicts that sentences including such PPs would thus be ungrammatical, as the data confirm (e.g., (20), (24), (30)). Furthermore, since these prepositions either have a directional or locative sense, we use a “±” value for the abstract feature d(irection) to represent this ambiguity (i.e., we have –d = +l). The corresponding PP merges with a verb to be disambiguated.

Crucially, this analysis entails that features project from heads and complements to a phrase, in P’Ps and PPs alike. The proper treatment of projection requires a formal apparatus that would go beyond the scope of this study. Here we only use the core principle. Features belonging to each merged category are unified into a single feature structure via feature unification (Shieber 1986: Ch. 2). For instance, a P’ carrying a feature s(patial) merging with a spatial noun carrying a feature −deg(ree) forms a P’P carrying the r(egion) feature. From <P’,s> and <NP,−d> we have <P’P,r>. Therefore, feature unification amounts to the set union of the feature set of each merged constituent (cf. also Adger 2010: 230–234; Sag 2012: 89–94).^[13]

Since feature unification involves the formation of a new feature set, this set is associated with the mother category of the unified daughters (e.g., P’P from P’ and NP). Thus, unification also determines the projection of features from the daughter nodes to the mother node. This analysis captures the fact that P’Ps’ features (i.e., r, p) or the lack thereof (i.e., ∅) determine the features and thus the morphological type of a preposition. It can also capture the fact that features can project higher in the clausal spine, thereby determining licensing phenomena in an apparently non-local manner. We show why this is the case in the rest of this section by tackling the data.

For complex prepositions, we analyze their P’Ps as involving the conflation (i.e., univerbation) of P’ heads and nouns as a single category with unified features (e.g., dietro in (59a)). Hence, they can also carry features specifying whether they are projective or region prepositions. We conjecture that conflation mirrors the systematic merge of their feature structures (cf. Hale and Keyser 2002: Ch. 3 for discussion). Complex prepositions including an optional head involve the same structure but different feature values, as the pairs in (59a)/(59b) and (59c)/(59d) show. This account can also be extended to verso and the other complex prepositions taking di when the Ground is an animate complement or a deictic pronoun (e.g., per, tra, su and giù). The ±a(nimate) features in (59c)–(59d) act as mnemonic indexes to mark this subtle distinction. Thus, different types of prepositions and PPs have a homogeneous structure but can realize different features and their corresponding values.

Locutional prepositions involve the merge of an NP (i.e., a 0-place head) carrying either a +deg or a −deg feature. Once this NP merges with a P’, the P’P inherits these features via projection, in the (layered) guise of either the r or the p feature. In the P’P in cima ‘on top’ in (60), the P’ in marks the noun cima as part of a preposition, as opposed to the definite NP la cima ‘the top’. The same mechanism is at work in (61), which includes the contracted preposition nel (i.e., nel mezzo).

As a result, our account predicts that if prepositions as 2-place heads (i.e., Ps) capture geometric relations, then as 1-place heads (i.e., P’s) they map an NP’s referent from objects to locations. In other words, while cima refers to a ‘top’ part of an object, in cima refers to the location that this part, and the figure(s) located at it, occupies. In our account, the role of in and other simple prepositions merging as 1-place markers is that of marking spatial nouns into constituting elements of locutional prepositions. They do not determine the merging of PPs with MPs, as our data have shown.

Overall, our account can represent the morphosyntactic structures assigned to prepositions in a unified manner, via a feature projection mechanism. The account also shows that simple prepositions lack a specifier and only denote geometric relations, since no P’P introducing a direction or projection is present. Complex and locutional prepositions, by contrast, feature some core distinctions in how they realize the other semantic types. Complex prepositions may involve mono-morphemic (e.g., per) or univerbed items (e.g., dietro) in specifier position and may denote projections. Locutional prepositions include spatial nouns distinct from prepositions, which nevertheless determine the type assigned to the “whole” preposition, once part of the P’P.

We can then offer further evidence in support of this account and its predictions by showing how it can be used to model the syntactic data. We concentrate on BLCs and sentences including MPs, leaving a full-fledged account of the other patterns aside (but see Ursini and Long 2020; Ursini and Tse 2021; for further details). Consider the structures in (62):

We assume that a silent Deg head can merge with a PP and an MP as its complement and specifier, respectively. It only licenses a grammatical and interpretable structure when the features of these arguments match (cf. Morzycki 2015; Svenonius 2010). As (62a) shows, a P’P carries either an r or p feature, cyclically projected at a PP level. The Deg head identifies this feature with the feature that an MP carries. A mismatch can cause the Degree Phrase to be uninterpretable and the whole sentence to also be uninterpretable. BLCs, here modeled as simple VPs (Levinson and Wilkins 2006: Ch. 1) can thus include Degree Phrases that determine overall (un)interpretability.^[14] If a Degree phrase including an MP and a projective preposition are merged into a BLC, a sentence is interpretable (cf. dietro a in (62b)). The status of prepositions as belonging to the projective types crucially hinges on the features of their spatial nouns, and how they project at a sentential level.

The opposite holds when a region preposition is merged, (cf. in cima a in (62c)). A geometric preposition similarly prevents its merge with an MP, but that is because it lacks any features that can be matched with this phrase (cf. a in (62d)). Locutional in cima a carries the r feature and simple a lack either r or p features. Therefore, in cima a and a cannot merge with dieci metri, an MP carrying a p feature. Dietro a, instead, can do so: it carries the p feature. We can thus show that the MP test distinguishes between region, geometric and projective prepositions, since MPs can only merge with PPs carrying the matching features via their P’Ps. We also show that this distinction is orthogonal to the locative/directional distinction: these values belong to a distinct feature set (pace Tortora 2008).

Overall, our account can capture the observed distribution of MPs with prepositions. Even if grammatical and semantic types are not isomorphically connected, our account can aptly capture their relations and contribution to sentences in a systematic manner. Thus, the model reconstructs the idea of prepositions’ discriminating properties hinging on their grammatical (morphosyntactic) features, as proposed in Bottari (1985a, 1985b) (cf. also Fagard et al. 2020). However, the main innovation is that we pinpoint spatial nouns as the preposition “part” that carries the features determining the distribution of PPs with MPs, via a projection mechanism. Specifically, we show that only spatial nouns in their various realizations within prepositions seem to govern these patterns. We can now turn to the discussion.

4.3 Discussion

We believe that four key results emerge from our account.

First, we have shown that grammatical and semantic types do not stand in an isomorphic relation. Thus, the introduction of the RelN and AxPart heads and the distinct prepositional structures they entail can lead to unwelcome predictions regarding the distribution of prepositions with MPs. When we take a wider set of prepositions into consideration, it becomes clear that a unified account of nominal-like items becomes necessary. Our account shows that spatial nouns carrying deg(ree) features determine the semantic status of the preposition embedding this category. When this category merges or even undergoes univerbation with a simple preposition as a marker (e.g., respectively di fronte, dietro), a P’P is formed that can then determine the distribution of the full-fledged PP with an MP (e.g., dieci metri dietro alla macchina). Hence, the different semantic types (geometric, projective, region) emerge from the presence and features of this constituent, which then project at a phrasal and sentential level. Previous cartographic accounts face problems because they either do not analyze such data (e.g., Folli 2008; Tortora 2008), or introduce distinct structures (Franco 2016).

Second, our account is consistent with Aurnague and Vieu’s (2015) hypothesis that spatial (bare) NPs (their “Internal Location Nouns”), once they become part of a PP, determine a preposition’s lexical content. Other works offering an analysis of this NP sub-type also arrive at similar conclusions (Jackendoff 1991; Lang 1990; Levinson 1994). Our account is also consistent with Fagard et al.’s (2020) analysis of Romance prepositions. This work proposes that in complex prepositions, the lexical content of NPs is analyzed as also being the core source of productivity and variation in Romance languages. Simple prepositions within complex prepositions (e.g., in and a in in cima a) are thus analyzed either as markers of nouns (our P’ for in) or as relational elements (our P for a). Acedo-Matellán’s (2016) analysis, also based on Lexical Syntax, resembles our analysis though it does not tackle complex prepositions. Thus, our account is consistent with typological analyses of Romance prepositions, non-generative and generative alike. It however improves on these proposals by showing how grammatical types can determine the semantic types of the spatial relations prepositions denote.

Third, we offer evidence showing that locutional prepositions are a distinct morphological type but have the same syntactic distribution as other preposition types. Previous works on Italian spatial prepositions of distinct theoretical declinations do not directly address this type (e.g., Ganfi and Piunno 2017; Garzonio and Rossi 2016; Iacobini and Masini 2006; Piunno and Ganfi 2020). Therefore, their potential extension to relevant data appears non-trivial and, for the cartographic proposals, it would lead to the paradoxes originally motivating our study. The data presented in this paper can thus offer evidence for an evaluation of our proposal in light of previous proposals.

Fourth, we show that the semantic properties of Italian prepositions require an explicit morphosyntactic representation that captures their distribution with MPs. In our analysis, the semantic patterns involving MPs correlate with the properties of spatial nouns. Simple (un)contracted prepositions, as markers and heads in complex and locutional prepositions, do not affect this distribution. We have proposed that this is the case because spatial nouns carry morphological features that govern this semantic behavior and that they project at a sentential level. This view is consistent with previous Cartography approaches offering compositional treatments of prepositions that hinge on the interpretation of each proposed head (e.g., Svenonius 2008). This view also aims to show that previous works only analyzing these patterns from a semantic perspective may lack a fully compositional view of these data (e.g., Zwarts and Winter 2000). It shows, however, that novel proposals addressing the distinct contribution of each element in a preposition may indeed offer such compositional accounts (e.g., Matushansky and Zwarts 2019). We leave a full-fledged integration of compositional accounts for future work.

We also wish to acknowledge that Franco’s (2016) account correctly individuates the diachronic constraints of having operated on Italian prepositions. This work suggests that nei pressi di may be the result of a degrammaticalization process that operated on presso a. The work also suggests that from an inherently functional item, the preposition presso a, the preposition-like and lexical item nei pressi di emerged during the 20th century. It thus proposes that Italian prepositions operate as a weakly productive, synchronically unstable system in which items may still display distributional properties hinting at their lexical origins. This result is also consistent with the vast literature on the grammaticalization of spatial nouns to the adpositional domain (e.g., Ganfi and Piunno 2017; Heine and Kuteva 2007; Lehmann 1985: 138–144; Svorou 1994). We also leave this topic for future explorations, given its complexity.