2. Previous Proposals on French Prepositions

2.1. Syntactic accounts: Preposition types and the role of ILNs

Reference grammars traditionally distinguish between simple and complex prepositions (prépositions simples and prépositions complexes: Price Reference Price2008: 520–545). Simple prepositions are mono-morphemic, though they can be disyllabic (e.g., à ‘at/to’, parmi ‘among’). Complex prepositions feature an ILN that may precede and possibly follow simple prepositions (e.g., côté preceding de and following à in à côté de ‘beside’). We list simple prepositions in (5); lists of complex prepositions vary, depending on how authors analyze their structure. We thus offer a non-exhaustive list in (6) (Price Reference Price2008: 545):

Simple prepositions can be polysemous (e.g., sur ‘up’, ‘on’, or ‘above’) and can cover both directional and locative meanings. Thus, à can describe a figure moving ‘to’ a ground (directional) but also being ‘at’ the ground (locative: Vandeloise Reference Vandeloise1988). Complex prepositions, instead, involve multi-morphemic structures. Some complex prepositions include de as a head (e.g., hors de), while others block its presence, for instance derrière ‘behind’ but not *derrière de. Exceptions are prepositions such as jusqu’à ‘next to’, which include à as a head conflating with an ILN (Le Pesant Reference Le Pesant, Neveu, Blumenthal and Querler2011). The preposition à may also precede an ILN and conflate with it (e.g., au-tour lit. ‘at.the-round’). Note that de does not have spatial meanings unless it occurs with verbs capturing elative motion (e.g., sortir de ‘coming out of’). Its inclusion in (5) aims to represent this fact, but also its role in complex prepositions, fully discussed in section 3.

Much theoretical work offers evidence for another class of prepositions, locutions prépositionnelles (‘locutional prepositions’ or ‘prepositional locutions’) (e.g., Vandeloise Reference Vandeloise1987, Reference Vandeloise1988; Borillo Reference Borillo1988, Reference Borillo1998, Reference Borillo2000, Reference Borillo2001; Melis Reference Melis2003). Locutional prepositions seem to be a cross-linguistically attested, although understudied, type of preposition (see Di Meola Reference Di Meola2000 on German, Hoffmann Reference Hoffmann2005 on English). This type includes a first simple preposition acting as a “marker” of an ILN and a second preposition following the ILN. Markers can be inflected or uninflected (respectively au bord de ‘at the edge of’, à côté de ‘at the side of’); “final” prepositions can include à or de (e.g., face à ‘against’, dos à ‘at the back of’).

Two corpus-based studies offering an extensive list of prepositions are Le Pesant (Reference Le Pesant, Neveu, Blumenthal and Querler2011, Reference Le Pesant, Neveu, Toke, Durand, Klingler, Prévost and Teston-Bonnard2012). In these, out of a list of 209 items, 23 are simple prepositions, which can be further divided into “direct” and “indirect” prepositions. Simple direct prepositions coincide with the list in (5) plus vers, devant, contre, dans, and derrière. Their label captures the fact that they directly combine with a ground NP. Simple indirect prepositions include auprès de, autour de, hors de, loin de, près de, and face à, which are prepositions requiring de as a head introducing a ground NP. Let us note that the labels “simple” and “complex” in reference grammars, and the labels “simple direct” and “simple indirect” in Le Pesant (Reference Le Pesant, Neveu, Blumenthal and Querler2011, Reference Le Pesant, Neveu, Toke, Durand, Klingler, Prévost and Teston-Bonnard2012) follow different criteria of identification for a sub-set of prepositions. We return to this point in footnote 1.

Previous work on the syntactic properties of prepositions has explored their distribution. For instance, Melis (Reference Melis2003) discusses how prepositions can introduce complements that are not “common” NPs (e.g., pronouns, place names). They show that prepositions and their corresponding PPs form the core element of sentence types known as Basic Locative Constructions (BLCs) (Levinson and Wilkins Reference Levinson and Wilkins2006: Ch. 1). BLCs are defined as sentences that can act as full answers to where-questions, thus carrying the same role of fragment answers in a language ((1)–(4)). Although Melis (Reference Melis2003) does not address their distribution in interrogative sentences, he analyzes the distribution of simple and complex prepositions in BLCs. Melis also shows that locutional prepositions can head PPs that are the spatial complements of verbs, describing the location/direction of the figure (as in (7), from Melis Reference Melis2003: 109). We propose a concise but illustrative list of locutional prepositions in (8):

At the same time, Melis (Reference Melis2003) also observes that some simple and complex prepositions may also combine with an indexical or pronominal ground (as in par in (9)). Additionally, they can combine with de when it introduces a ground NP (e.g., parmi des chaises ‘among some chairs’), which thereby acts as an intermediate head combining the preposition and the ground NP.Footnote ¹ Melis proposes that some prepositions can be reclassified as “complex” because of their distribution with pronouns and indexicals (e.g., par, pour, parmi, sur, sous). Other prepositions can be so reclassified since they cannot occur as elements mediating between an ILN and a ground NP (e.g., chez, entre, devant, derrière). A similar argument is made in Fagard (Reference Fagard2010: Ch. 2), thereby leading to the list of simple prepositions offered in (10):

Overall, “novel” simple prepositions correspond to those items that can act as heads of complex prepositions or as markers of ILNs (e.g., à in à droite de). By “novel”, we mean that the classification proposed in Fagard (Reference Fagard2010) represents a new manner of classifying French prepositions, when compared to previous classifications from reference grammars (e.g., Price Reference Price2008). Simple and complex prepositions may display distinct semantic properties, which we discuss in more detail in section 2.2, after we address the semantic accounts of these preposition classes.

From a diachronic perspective, several studies have also investigated the gradual emergence of complex and locutional prepositions (e.g., Fagard Reference Fagard2006, Reference Fagard, Fagard, Prevost, Combettes and Bertrand2008, Reference Fagard2009a, Reference Fagardb, Reference Fagard2010, Reference Fagard2012; Fagard and De Mulder Reference Fagard and De Mulder2007, Reference Fagard, Mulder, Neveu, Toke, Durand, Klingler, Mondada and Prévost2010; Fagard and Sarda Reference Fagard, Sarda, François, Gilbert, Guimier and Krause2009). A common trait of these studies is that they offer five criteria distinguishing locutional prepositions from other types. A first criterion is frequency: locutional prepositions such as au cœur de ‘at the heart of’ appeared as as expressions with a rigid internal order of items but not necessarily conveying spatial meanings in Medieval French. Over time, these sequences were increasingly associated with spatial meanings (e.g., ‘in the innermost part of’, for au cœur de). Thus, they underwent a process of lexicalization: the formation of a vocabulary item with a rigid, synchronically stable internal order of morphemes, retaining some internal complexity (e.g., parmi, chez).Footnote ²

A second criterion is the presence of an inflected preposition/marker. Locutional prepositions that include an inflected preposition represent “less fixed” or novel items (e.g., à l'intérieur de). In these constructions, the loss of an article signals an older incorporation into the prepositional system (e.g., en bas de ‘at the bottom of’).

The third criterion actually consists of a cluster of criteria. Locutional prepositions resist the modification and pluralization of ILNs (*à le gros intérieur de, *aux intérieurs de, respectively), and their replacement or distribution with pronominal or indexical forms (e.g., *à quel intérieur de). Thus, ILNs lose their status as nouns and become fixed morphological segments of these items.

Coordination of prepositions and semantic opacity represent the fourth and the fifth criteria. Prepositions can head phrases that can be coordinated (e.g., à côté et au-dessous de la voiture ‘next to and underneath the car’). This fact indirectly suggests that they can act as heads of a prepositional phrase, irrespective of their morphological type. Semantic opacity refers to the fact that the precise contribution of ILNs and simple prepositions cannot easily be teased apart. Thus, complex and locutional prepositions are treated as the result of diachronic processes that lexicalize morphologically complex items capturing specific spatial meanings. Hence, French spatial prepositions seem to form a still evolving, partially heterogeneous category.

French prepositions have not been thoroughly addressed in generative work, with the exception of Roy (Reference Roy2006).Footnote ³ Building on Rooryck (Reference Rooryck, Thráinsson, Epstein and Peter1996), who compares French, Dutch, and German prepositions, Roy (Reference Roy2006) observes that “body part” nouns (our ILNs) can have two distinct meaning types that emerge in possessive constructions. If the “whole” noun refers to an animate entity, then these nouns refer to a given body part. If the “whole” noun refers to an inanimate object, then ambiguity can arise. Thus, a sentence such as la tête du lit est encore humide ‘The head of the bed is still damp’ may involve either a “fixed” or a “relative” meaning. In the first case, la tête refers to the damp headboard of a bed; in the second case, it refers to its surrounding region.

Furthermore, Roy (Reference Roy2006) argues that relative meaning types lack certain distinctive properties of “full” nouns, such as the ability to take plural morphology, undergo adjectival modification, and take indefinite articles, among others (e.g., *une tête de lit ‘a headboard’). Therefore, her work proposes that these relative meanings are introduced via a category known as “Axial Part” (AxPart), whereas fixed/referential meanings are introduced via a category known as “Relational Noun” (RelN: see Svenonius Reference Svenonius2006, Reference Svenonius, Cinque and Rizzi2010 for English). Roy's work therefore dovetails with non-generative work in identifying prepositions, including AxPart items as a type of locutional preposition (see Fagard Reference Fagard2006: Ch. 2). However, it leaves open the question of the relevance of these categories beyond their distribution in BLCs.

Overall, previous work on the morpho-syntactic properties of French prepositions identifies at least three sub-types: simple prepositions (e.g., en), complex prepositions (e.g., autour de), and locutional prepositions (e.g., à l'intérieur de). They propose that French prepositions form a sizeable but closed set and that they can act as heads of PPs in BLCs. Crucially, most studies do not explore the interaction of this category with MPs, leaving the matter unexplored. Before we tackle the topic of the interaction of prepositions with MPs, we discuss previous semantic accounts.

3. Old and Novel Data

The goal of this section is to present a broader overview of the distribution of MPs with French prepositions. We present novel data involving MPs once we have shown that all prepositional sub-types (i.e., simple, complex, locutional) display syntactic properties warranting a unified account. We begin by observing that some of the studies reviewed in section 2.1 (e.g., Melis Reference Melis2003, Le Pesant Reference Le Pesant, Neveu, Toke, Durand, Klingler, Prévost and Teston-Bonnard2012) discuss in detail the distribution of PPs in sentences (our BLCs) and their interaction with different types of verbs and ground NPs. Melis (Reference Melis2003: Ch. 2) observes that PPs can be involved in various forms of extraposition, one example being a form of PP fronting known as locative inversion (Cornish Reference Cornish, de Groot and Hengeveld2005; den Dikken Reference den Dikken2006, Reference den Dikken, Cinque and Rizzi2010; Fuchs Reference Fuchs, Sarda, Carter-Thomas, Fagard and Charolles2014). These previous studies offer evidence mostly focusing on simple and complex prepositions; here we focus on locutional prepositions and their role in BLCs.

PPs may also be given as fragment answers to questions involving the spatial wh-word où ‘where’. Spatial prepositions can act as PPs forming congruent answers to où-questions (Jackendoff Reference Jackendoff1972, Merchant Reference Merchant2001: Ch. 2, Boone Reference Boone2014: Ch. 2).Footnote ⁴ For French prepositions, the previous literature has not fully discussed its relevance; hence, we take this occasion to discuss its theoretical import via (11)–(15):

The fragment answers in (11)–(13) confirm that any morphological type (simple à, complex derrière and devant, locutional à l'intérieur de, and à côté de) can head a PP answer to an où-question. Similarly, a fronted or inverted PP can include any type of preposition as its head, as (14)–(15) show. We have complex derrière in (14), and locutional à gauche/à l'intérieur de in (15). Together with the BLC data, these data show that each preposition type can head a PP, irrespective of their morphological sub-type (i.e., simple, complex, or locutional).

Let us now turn to a discussion of NP ellipsis (Melis Reference Melis2003: 112–115, Fagard and De Mulder Reference Fagard and De Mulder2007: 20, Le Pesant Reference Le Pesant, Neveu, Blumenthal and Querler2011: 20–21). This operation targets a ground NP and possibly the head of the PP containing it (Merchant Reference Merchant2001, Svenonius Reference Svenonius, Cinque and Rizzi2010, Boone Reference Boone2014: Ch. 4). The pronounced part or remnant usually involves the segment(s) that can refer to a specific location, at least in English prepositions (e.g., in front in in front (of the car)). An open question that we can easily answer is whether this test applies to simple and complex prepositions. Previous work (e.g., Melis Reference Melis2003) offers ample evidence regarding locutional prepositions. Crucially, the test shows that simple prepositions cannot undergo ellipsis, unlike the other types (16). Complex prepositions involve nuanced licensing patterns, however: chez, entre, parmi, sur, and sous cannot undergo ellipsis, while all other complex prepositions can (as in derrière/devant in (17)).Footnote ⁵ For locutional prepositions, the remnant invariably involves the unit corresponding to the combination of an (un)inflected marker and an ILN (18):

These data also show that remnants can involve two types of internal structures. The first involves an apparently single vocabulary item (e.g., derrière in (17)); the second involves a possibly inflected preposition acting as a “marker” of an ILN (à l'intérieur in (18)). Previous work suggests that prepositions lacking this marker may have undergone a diachronic process of univerbation (e.g., au-tour from au and tour: e.g., Vandeloise Reference Vandeloise1998; Le Pesant Reference Le Pesant, Neveu, Blumenthal and Querler2011; Fagard Reference Fagard2010, Reference Fagard2012). These processes may simply represent the result of two vocabulary items being reinterpreted as a single item (and, possibly, category). We can therefore conclude that PPs may potentially include at least three types, with respect to syntactic tests. The first is a simple preposition that can undergo ellipsis. The second is a ground NP. The third is a preposition/marker combining with an ILN and forming a distinct, possibly complex unit that can stand as a remnant.

Let us move to the MP data, based on an elicitation task. Native speakers of (continental) French (N = 30) were asked to evaluate sentences using a Likert scale (“1” being “unacceptable”, “5” being “perfect”). All examples were designed and verified with the help of a native speaker informant, who also acted as a pilot participant for the test. We opted for this experimental method because corpora data involving MPs turned out to be very rare. We also aimed at establishing a form of triangulation, by having corpora data verified against experimental data (Rothbauer Reference Rothbauer and Given2008). We leave open the possibility that regional varieties might have played a role in judgments, though most speakers offered explicit observations regarding the role of influencing factors (e.g., register).

All examples marked as “#” received average scores of 2;0 or lower; perfectly acceptable examples, scores of 4;0 or higher. Speakers were invited to write comments. For many examples scored as “3” and “4”, most speakers observed that the examples were not perfect but nevertheless quite acceptable (or were not quite deviant, between 3;0 and 4;0). For a number of examples, participants often offered sharply contrasting answers (e.g., most participants offering “5” as answers, but some offering “1” for the same test sentences). For this reason, we report average scores and token answers per value below the examples (following De Clercq and Haegeman Reference de Clercq Karen2018). Using this information, we show that intra-speaker variation may be considerable, but we also underline the fact that general conclusions on acceptability are consistent with this form of variation (Schütze and Sprouse Reference Schütze, Sprouse, Podesva and Sharma2013).Footnote ⁶

As foreshadowed in the introduction via (1)–(4), the distribution of prepositions with MPs acts as a test that establishes a preposition's semantic type. While projective prepositions can combine with MPs (e.g., derrière), geometrical prepositions (e.g., en) cannot do so. Once we consider a broader set of data, the picture becomes considerably more complex. To illustrate this, we introduce the novel data in (19)–(28):Footnote ⁷

In (19), a distance of one kilometre is measured as the distance that Mario walks within Paris. Although the polysemous dans can cover strict and loose ‘inclusion’ (cf. English inside), it covers at least one meaning that is projective in nature (Vandeloise Reference Vandeloise2008). This is not the case for en and à, as (20) shows. These two simple prepositions only denote geometrical relations (inclusion for en and “abstract” geometrical relation for à) between figure and ground, and lack a distance component in their meaning. A similar reasoning extends to par and à travers in (21). Both prepositions cover meanings describing a figure navigating a stretch of space. However, while par seems to imply that the distance involves the whole field and may not involve directed movement, à travers involves a form of directed and hence measurable movement (see also Aurnague and Stosic Reference Aurnague and Stosic2002). Thus, the first preposition resists combination with MPs; the second licenses it.

Similarly, devant and derrière can combine with an MP (22). Furthermore, autour de may combine with the MP un kilometre (23), whereas à côté de cannot do so (24). For the latter preposition, participants observed that à côté de may be highly deviant but not unacceptable with MPs if very small distances are involved (e.g., un centimetre à côté de la voiture ‘one centimetre next to the car’). However, for the most part, the participants rejected its co-occurrence with MPs in sentences. The locutional prepositions à l'intérieur/extérieur de block MPs, as (25) shows, as do sur le sommet de and au milieu de, and locutional prepositions including en and sur as markers ((26)–(27)). On the other hand, à gauche/droit de can occur with MPs, as (28) shows. Two non-exhaustive lists summarising this distribution are given in (29)–(30):

Overall, four key results emerge from our discussion. First, simple prepositions resist distribution with MPs except dans, which can cover meanings related to an ‘internal’ projection. Second, complex and locutional prepositions display a heterogeneous distribution. Whether an item belonging to one of these two morphological classes can be classified as a projective or a region preposition depends only on its meaning. Third, the presence of (un)inflected à or other markers (i.e., en, sur) seems not to play a role in the occurrence of complex and locutional prepositions with MPs. Fourth, the occurrence of PPs with MPs is not a clear-cut matter, for at least two reasons. First, intra-speaker variation reveals points of divergence even when sentences are perfectly acceptable. Second, certain prepositions (e.g., dans, à côté de) may also depend on the presence of certain verbs (and perhaps their selected meaning in the environment of these verbs), for their licensing with MPs. What the lists in (29)–(30) represent, then, is that different simple, complex, and locutional prepositions may (indirectly) combine with MPs: their morphological type does not determine this pattern. Furthermore, each item can be associated with a particular semantic type, but ambiguity is possible.

These results also suggest that extant classifications may not entirely capture these data. We can highlight three reasons for this claim. First, ILNs may combine with definite articles and thus act as relational nouns (e.g., à côté de). Alternatively, they disallow this pattern and occur “bare” (i.e., as AxPart items: à gauche). Hence, Roy (Reference Roy2006)'s morpho-syntactic distinction does not map onto a clear semantic distinction, since some putative AxPart items cannot combine with MPs, and some RelNs can do so. Second, while à côté de and à l'intérieur de denote external internal parts or regions of their grounds, respectively, they both resist occurring with MPs. Conversely, dans and derrière denote internal and external projections, respectively, but can also occur with MPs. Thus, Aurnague and Vieu's (Reference Aurnague, Vieu, Astésano and Jucla2015) distinction between internal and external region prepositions seems to act as a semantic dimension orthogonal to the region/projective dimension. Third, accounts distinguishing between projective and non-projective prepositions conflate simple and locutional prepositions (e.g., Zwarts and Winter Reference Zwarts and Winter2000). Our data, however, suggest that locutional prepositions can resist distribution with MPs because this type of preposition can capture distinct meaning types.Footnote ⁸

Overall, we have reached our first goal: to offer an overview of the novel data that also supports the introduction of region prepositions as a distinct morpho-semantic type. We also have shown that this type is intimately related to projective prepositions, and that geometrical prepositions can be mostly reduced to the directional/locative alternation. That is, simple prepositions (à, en, de) cover “general” spatial relations, which may also involve aspects of motion or stasis (cf. Vandeloise Reference Vandeloise1991). Thus, region and projective meaning types may both be connected to complex prepositions. Furthermore, we have shown that previous accounts require more fine-grained tools of analysis to account for the locutional preposition type. This is the subject of the next section.

4. The Account

Our account of the morpho-syntactic properties of French prepositions follows the “P within P” hypothesis proposed within Lexical Syntax (Hale and Keyser Reference Hale and Keyser2002: Ch. 4, Mateu Reference Mateu2002). The key assumptions in our account are as follows.

First, language-specific categories can project one of four language-general head types. Depending on the valence of a vocabulary item and its syntactic context, an item instantiates a 0-place, 1-place, or 2-place head type, based on how many arguments they combine or merge with. A 0-place head can act as a “bare” argument (i.e., a phrase). A 1-place head can act as an affix or as a marker. A 2-place head type represents a “relational” head merging with a specifier and a complement. The Lexical Syntax framework proposes another type of 2-place head; however, we can ignore this distinction without loss of precision in our analysis (Mateu and Amadas Reference Mateu and Amadas2001, Hale and Keyser Reference Hale and Keyser2002: 13–14).

Second, we take a partially different stance with regards to the structure of prepositions and PPs than is seen in classic and cartographic approaches (e.g., Roy Reference Roy2006). The central assumption here is that ILNs and their markers form a distinct unit that occurs as one of the two arguments of a head preposition. We make this choice for three reasons. First, complex items occurring in complex and locutional prepositions often involve forms of conflation. Examples include conflated items (e.g., au-tour ‘around’) or items having undergone univerbation (e.g., derrière ‘behind’ from Latin de retrum ‘from back’). Second, these items can become remnants in ground NP ellipsis structures, hence acting as a single unit with respect to this operation.

Following these two assumptions we can conclude that these complex units act as arguments of a larger PP phrase. The ground NP data suggest that remnants are a single syntactic unit that can nevertheless contain different morphological structures. If we consider them “stacked” projections of a preposition, as cartographic proposals suggest, complex prepositions would involve apparently heterogeneous structures acting as remnants (roughly, RelN phrases or AxPart phrases). These structures would also involve functional projections assigned to ILNs and their embedding markers. However, our data invariably suggest that ILNs carry lexical content, which then becomes part of a preposition (Aurnague and Vieu Reference Aurnague, Vieu, Astésano and Jucla2015, Matushansky and Zwarts Reference Matushansky, Zwarts, Stockwell, O'Leary, Xu and Zhou2019). Similarly, our data suggest that remnants seem to be arguments of PPs, rather than projections of prepositional heads. Therefore, treating these units as arguments allows us to capture their lexical and syntactic properties in a direct manner.

Third, these complex prepositions inherit their meaning from the ILNs that diachronically act as their “roots”. For instance, autour, derrière, and à l'intérieur inherit their projective or region meanings from their underlying ILNs (e.g., intérieur, arrière, tour). However, each meaning type can block or license the presence of MPs once a full phrase is formed, as part of a larger PP. We thus assume that ILNs may involve a distinction that resembles the AxPart/RelN distinction in cartographic approaches. Yet, this distinction does not determine the category and structure assigned to each vocabulary item, but only the features assigned to a preposition and the PP it projects from this preposition. In other words, ILNs determine the region or projective features assigned to a preposition, but do so as “part” of a preposition being merged as the argument of a second preposition. This cumulative effect is consistent with treatments in previous literature that also focus on the semantic properties of this category (e.g., Vandeloise Reference Vandeloise1988, Melis Reference Melis2003, Aurnague and Vieu Reference Aurnague, Vieu, Astésano and Jucla2015).

Fourth, we assume that simple prepositions qua heads (e.g., de) project 2-place heads, which take ground NPs as their complement and possibly another prepositional phrase as their second argument (a specifier, in standard generative terms). Each projected category can be further enriched with its assigned morphological features. We capture this latter aspect by using formal (morpho-syntactic) features (Chomsky Reference Chomsky and Kenstowicz2001, Adger Reference Adger, Kibort and Corbett2010, Adger and Svenonius Reference Adger, Svenonius and Boeckx2011, Sag et al. Reference Sag, Boas, Kay, Boas and Sag2012). Each head H is represented as projecting a category (e.g., P, V, D, and N).Footnote ⁹ The features associated to each category (e.g., tense for verbs) follow the label and are presented as “attribute:value” pairings (e.g., tense:past). Categories and features with values are represented as ordered sequences of sub-scripts (e.g., H_<P,p>, where this reads: a preposition P is a head carrying a p(rojective) feature). We trade precision for readability, so we only assign features to prepositions below.

We use the following notation. Simple prepositions correspond to 2-place head types, represented with the label P. In their distribution as markers, they are 1-place head types, represented with the label P’ (P prime). The resulting phrase is a P'P, a variant on the “P within P” hypothesis. We then assume that simple Ps are ambiguous: they can carry d(irectional) or l(ocative) “path” feature values. ILNs correspond to NPs (i.e., 0-place heads) and are merged as the complements of members of the P’ category. They carry the feature values p(rojective) or r(egion). Thus, we represent region and projective features as a morpho-semantic dimension orthogonal to the locative/directional dimension.Footnote ¹⁰ Verbs and nouns carry temporal/aspectual features. However, we omit them since they are not relevant to our discussion.

Before we spell out the details of our analysis, a preliminary comparison between this analysis and previous proposals may help readers to better appreciate the novelties we introduce. In cartographic approaches, prepositions involve sequences of functional heads (e.g., Svenonius Reference Svenonius, Cinque and Rizzi2010). In our account, prepositions involve a “pure” functional head mediating between a ground DP and an embedded P'P containing lexical and functional categories. We consider ILNs to be a super-category of RelNs and AxParts, though we treat them as arguments (i.e., NPs) of another preposition. Since we consider p and r to be possible feature values associated with ILNs, we do not differentiate prepositional structures into two slightly different types involving the AxPart- and RelN-based structures. As the use of Lexical Syntax entails, we do not assume that each morpheme in a preposition projects a functional (i.e., 2-place) head. Different elements have different valences, an analysis that finds support in the data discussed in section 3.

We can illustrate the resulting novel proposal by offering the structures corresponding to each preposition type in (31)–(34):

For simple prepositions, we assume that they involve null P'Ps (e.g., au piano in (31b)). We simplify the representation of inflected prepositions by treating definite features of the definite article as an integral part of their structure (i.e., we have < P,±d,def>: Caha Reference Caha2009, Svenonius Reference Svenonius, Siddiqi and Harley2016). Since these prepositions either have a directional or locative meaning, we use a positive or negative value for the feature d to represent this ambiguity. The corresponding PP is also ambiguous: it must merge with a verb to be disambiguated (Tungseth Reference Tungseth2008, among others).

This entails that features percolate from the heads to phrases, in P'Ps and PPs. The proper treatment of percolation involves a formal apparatus that would take us too far afield from the topic of this article if presented in full. Here we present only the core principle: that features belonging to each merged category are unified into a single feature structure. For instance, a P’ with a feature s merging with a spatial noun with a feature r(egion) forms a P'P with the r and s features. From <P’,s> and <NP,r> we have <P'P,s,r> (Shieber Reference Shieber1986: 27, Travis and Lamontagne Reference Travis and Lamontagne1992, Adger Reference Adger, Kibort and Corbett2010: 230–234, Sag et al. Reference Sag, Boas, Kay, Boas and Sag2012: 89–94, Svenonius Reference Svenonius, Siddiqi and Harley2016: 202).Footnote ¹¹ Features can cyclically project along the clausal spine, and may form complex feature clusters associated to “higher” constituents (e.g., “spans” in Svenonius Reference Svenonius, Siddiqi and Harley2016). Thus, features may become accessible to various syntactic operations beyond their local merge domain, e.g., agreement, anaphoric relations, and feature matching, when MPs are involved.

Locutional prepositions involve the merger of an ILN carrying either p or r features. Once this ILN, a 0-place head, merges with a P’, the resulting P'P inherits these features via percolation. In the P'P sur le sommet ‘on top’ in (32b), the P’ sur le marks the noun sommet as part of a preposition, and not as the definite NP le sommet ‘the top’. The same mechanism is at work in (33b), which includes the inflected preposition au milieu ‘in the centre’. Thus, we predict that P’ heads determine that a spatial noun's meaning refers to a location rather than to a part. If prepositions as 2-place heads (i.e., Ps) capture geometrical relations, then in their distribution as 1-place heads (i.e., P's/markers) they capture relations involving region/projective senses.

Complex prepositions including optional heads involve two slightly different structures, as the pairs in (34b–c) and (34d–e) show. While derrière involves a silent head, autour includes de as the pronounced head. “Novel” complex prepositions such as par receive a similar treatment. For instance, par includes a silent “∅” head that can select a deictic ground NP, ici, and thus the feature “+dx” ((34e), based on (9)), or a silent head and a definite article (34d). We can now offer an account of BLCs, as (35) shows:

The structure in (35a) shows that BLCs do not select a specific PP as the complement of a verb. Any of the morphological types of preposition can thus merge in these sentences.

Consider now où-questions and locative inversion sentences. We import two key assumptions from frameworks modelling question-answer data in discourse contexts to account for these data (Jäger Reference Jäger2005, Sag et al. Reference Sag, Boas, Kay, Boas and Sag2012, Ursini and Long Reference Ursini and Long2020). We consider our analysis to be preliminary, though hopefully on the right track. First, we consider answer phrases to be phrases of a type matching the features that a wh-word carries. French où lacks overt spatial features but nevertheless requires answers of a matching type. We model this fact via the structures in (36b–c). The structure in (37b), on the other hand, captures the locative inversion data. A silent head R takes a VP (i.e., a clause) and a PP as its arguments and establishes that the PP acts as an adjunct phrase to the clause (den Dikken Reference den Dikken2006, Reference den Dikken, Cinque and Rizzi2010):

Given the template in (37a), (37b–c) show that derrière la voiture ‘behind the car’ and à l'intérieur de la voiture ‘at the interior of the car’ can be answers to où-questions. They match the features of où and offer a specific feature value (projective for derrière la voiture and region for à l'intérieur de la voiture) that establishes the cohesiveness of the question-answer pairs (see Jäger Reference Jäger2005: Ch. 6). The structure in (37a) also shows that any type of preposition can head a fronted or inverted PP, as per our predictions. Thus, via (36)–(37) we formally capture how the ou-question and the locative inversion tests confirm the full-fledged status of locutional prepositions. Therefore, our account can already systematically capture the subtle differences in the distributional properties of preposition (morphological) types. It does so by offering an alternative view to current cartographic approaches to AxPart and RelN categories (e.g., Svenonius Reference Svenonius, Cinque and Rizzi2010, Franco Reference Franco2016). However, our account shares with these frameworks a view where different preposition types heading PPs can combine with MPs or be answers to où-questions, etc.

Before we fully expand this point, we address the ellipsis data. The account of ground NP ellipsis patterns is based on two key assumptions. First, ellipsis can be licensed when a phrase refers to an entity that is discourse-given (i.e., inferable from the context). Thus, the features that the elided constituent carries establish the content of this reference relation (Merchant Reference Merchant2001: Ch. 2, Jäger Reference Jäger2005: Ch. 4, Adger Reference Adger, Kibort and Corbett2010: 240, Sag et al. Reference Sag, Boas, Kay, Boas and Sag2012). Second, the remnant must be a constituent, even if the elided material can form a fragment (Boone Reference Boone2014: Ch. 2). Consider (38):

The template is given in (38a). The specific structure in (38b) shows that the remnant P'P devant licenses an inference about a (silent) governing head de and a ground NP, la voiture. Hence, the constituent and established identity can be elided. The same analysis applies to à l'intérieur de la voiture, as (38c) shows. P'Ps qua remnants can act as “spatial” complements of a verb. Therefore, only complex and locutional prepositions can undergo ellipsis. Since simple prepositions lack overt P'Ps, ellipsis would need to either elide the whole PP, creating the ungrammatical *Mario va (à la gare), or the ground NP. Our account correctly excludes this possibility (i.e., we also obtain *Mario va à (la gare)). Thus, our account predicts the syntactic distribution of all preposition types in ellipsis contexts.

We can now account for the distribution of MPs with prepositions. Our central assumption is that a silent Deg head can merge with a PP and an MP, its complement and specifier, respectively. This head only licenses a grammatical and interpretable structure when the spatial features of MP and PP match (Svenonius Reference Svenonius, Cinque and Rizzi2010, Morzycki Reference Morzycki2015, Franco Reference Franco2016). We present this analysis and its sentential import in (39)–(42):

As (39a) shows, sommet carries an r feature that percolates at a PP level. The Deg head identifies this feature with the p feature that dix mètres ‘ten meters’ carries; the mismatch causes the structure to be uninterpretable. This is not the case in (40b), because derrière also carries a p feature. The simple preposition à in (41b) involves a silent/empty P'P, and hence it does not specify a feature value corresponding to a region/projection meaning. A feature mismatch arises again: hence, the uninterpretability of the structure.Footnote ¹²

The structure in (42a), then, shows the general structure for BLCs, including MPs. When an MP and a PP match in features, they license an interpretable phrase and sentence (see derrière in (42b)). The opposite holds when a region preposition is merged (see au milieu de in (42c)). A geometrical preposition similarly causes a sentence to be uninterpretable, although this result occurs because no matching feature is merged (as in à in (42d)). Thus, our account can capture how region and geometrical prepositions can resist combination with MPs, although via different principles. For region prepositions, there is a feature mismatch between MP and the full PP embedding a region P'P. For geometrical prepositions, there is no P'P that can establish a feature-matching relation.

Before we move to the discussion, we wish to offer some considerations regarding the status of PPs and MPs as arguments in BLCs. One could argue that in combination with verbs lacking a goal component (e.g., marcher), PPs act like adjunct-like elements. Thus, PPs would denote the extension of the locations in which an event of motion happens. The MP would then be the direct complement of a verb (see Romeu Reference Romeu2014: Ch. 4, Ursini Reference Ursini2015 for recent discussions). Our account excludes this possibility, since it aims to treat all PPs as (syntactic) complements of verbs, given their uniform distribution. We believe that in an adjunct-oriented approach, however, the features that PPs and MPs contribute would enter into a matching relation at a different point in a derivation and would render a sentence uninterpretable if this matching did not take place. In other words, even if our structural analysis of PPs and MPs might turn out not to be accurate, the feature-matching principle would still offer a correct account of the data.