By setting up a part of speech ‘adposition’ for a given language, we imply that we are grouping items with shared syntactic behaviour. In the canonical world, members of such a part of speech would have consistent syntactic behaviour (for instance, all would be prepositions or all postpositions); often, however, their behaviour is not consistent. We concentrate on government, and find that we need to split the adpositions according to the case which they govern.5 We may diagram the situation as in Table 3.

This is a relatively familiar situation. Consider these simple phrases:6

While otherwise syntactically similar, these phrases differ in the case governed by the prepositions: genitive in (1) and instrumental in (2). This instantiates our schema (Table 3) as shown in Table 4.

Finding just two adpositions which behave differently is sufficient to demonstrate the split. However, this subsystem fits within a larger system, showing multiple instances of this same type of split. This is clear from Table 5, based on Wade (1992: 416–417), in turn starting from Wade (1983). The table includes just what Wade calls the ‘primary’ prepositions. For each I give the case values governed, with an indication of the meaning for the different uses.

Table 5 demonstrates clearly the need to partition the Russian prepositions. They can govern not only the genitive (1) and instrumental (2) but also the accusative, dative and locative.7 At the most basic level there is clear split, since prepositions govern differently. (See Timberlake 2004: 174–184 for a clear summary of the system.) That is the point of Table 5, for the present. We also see that several prepositions show a split in that they have more than one possibility; for instance, some distinguish location (place) versus motion (path), a common characteristic to which we return in §4.1, where we discuss more complex splits of adpositions.

Governees are not given in Table 5, since in almost all instances they are not relevant; the case value to be used depends on the preposition (there are few exceptions, see §3.3. for discussion of the second locative, while the remarkable role of the governee with po ‘up to, along, after’ is referenced in the discussion of Finnish in §4.3). Other languages, however, give the governee a much greater role, and we now turn to examples of that.

There may also be splits in the other dimension; that is, case requirements may differ according to the governee, as diagrammed in Table 6.

This type of split is shown by data from Mari (chm), formerly known as Cheremis (a Uralic language spoken mainly in the Mari El Republic of the Russian Federation), where the governees of postpositions are split. Personal pronouns stand in the genitive, as in (3).

However, nouns (and extended nominal complements) normally appear in the nominative (which has no overt marker in (4)):

For further examples see Alhoniemi (1993: 50), Kangasmaa-Minn (1998: 236–237) and Burukina (2023); the major source is Riese & Bradley & Yefremova (2022), who focus on Meadow Mari (mhr). These examples show the need for additional partitioning of the governees. While we might expect that governees would behave alike, here they are split, as shown in Table 7.

The account given is true in broad-brush terms. There is a good deal of complicating detail, for which see especially the careful account in Riese & Bradley & Yefremova (2022: 154–191). In particular, a minority of adpositions govern other case values (exclusively or in addition to the usual pattern); this includes the possibility of taking personal pronouns in the nominative in some persons (Riese & Bradley & Yefremova 2022: 157), a point taken up in §3.5. There are complex patterns of combinations of pronouns and adpositions, including instances where adpositions border on being case markers. Hence Table 7 presents an idealization of a part of the system of Mari.

More generally, governees can be split in two different ways. We may find a lexemic split when different governees are governed differently, as in Mari and in Armenian (to be analysed in §4.2). But we may also find a featural split of the governees, where the governee of an adposition stands in a different case according to its own number value; see Corbett (2023a: 126–127) for an example from Belarusian. 8 For present purposes we will focus on lexemic splits of governees.

Our two major types of split (in adpositions and in governees) may interact. That is, some adpositions govern differently in a way that targets just some governees. We can represent this situation as in Table 8.

We see an instance in Russian. Some prepositions govern the locative, and within those, there are two (v ‘in’ and na ‘on’) which in locational use govern the “second locative”. However, only some nouns have a dedicated second locative (other nouns use the normal locative). This situation is worth setting out step by step. First, we see the preposition o ‘about, concerning’ which in this meaning simply governs the locative case, for all types of noun:

In each the sense is ‘about, concerning’ and the locative case is found. Contrast the situation with v ‘in’:

This preposition can govern the second locative, when available. With most nouns, v ‘in’ takes the locative, as in (6a); a listable minority of nouns have a second locative, as illustrated by sad ‘garden’ in (6b); for more examples and analysis see Brown (2007) and Corbett (2012: 206–209); for the semantics see Plungian (2002), and for the possible role of contact with Finno-Ugric languages in the development of the second locative see Breu (1994: 50–52).

We can represent this situation as in Table 9.

Both the adpositions and the governees are split.9 At first glance, this is less of a departure from the canonical situation of Table 1 than either Table 3 or Table 4. But it is more serious, since it requires two splits rather than one. Moreover, this split is embedded in the complex Russian system given above in Table 5.

The sequence from §3.1 to §3.3 is not accidental. We first considered splits in the adposition, where different adpositions govern different case values (§3.1). This may well involve straightforward instances of free combinations, where adpositions take a wide range of governees, and impose their government requirement on them consistently. When we move to the splits in governees (§3.2), we are already leaving the domain of free combinations; the adposition needs to “peek” at the governee to determine which case it will govern. It is significant that the examples here often involve a split between pronouns and nouns, as occurs in other instances of case marking split.10 And the third type, involving splits in both the adpositions and the governees (§3.3), suggests phrases which are at least partially fixed. Thus the Russian second locative (§3.3) is restricted to the prepositions v ‘in’ and na ‘on’; but a given noun typically occurs with only one of these: v sadu ‘in the garden’ but na mostu ‘on the bridge’. The noun determines the preposition, and the preposition governs the special second locative.

Such restrictions on adpositional phrases (from both adposition and governee), which are signs of partially fixed expressions, can be manifested in various ways besides those which are our focus. There can be restrictions on items intervening between adposition and governee. Their meaning can become less than fully compositional. There may also be rather specific rules, as for instance with the Latin preposition cum ‘with’, which is postpositional just with a small inventory of items, including the personal pronoun, as in mecum ‘with me’ (see Fortson 2010: 136–140 and references there).11 And in Russian, the preposition o ‘about, concerning’ (taking the locative, Table 5) has this as its basic form; there is a shape condition such that before vowels it has the form ob (but with some exceptions both ways). In the specific phrases obo mne ‘about me’ and obo vsem ‘about everything’ it has the third form obo (Timberlake 2004: 179); also before other forms of ves′ ‘all, every’, such as obo vsex delax ‘about all matters’.12 At the extremes, there are even situations where it is arguable whether we have adpositions inflected for person and number or personal pronouns with adpositions realizing case; see Stewart & Joseph (2009) and Hutchinson (2025) on Scottish Gaelic, and Lamb (2025: 159–192 for a wealth of examples), and Spencer & Stump (2013) on Hungarian.

I take it as given that the existence of languages like Mari, where familiar rules of government are not fully adequate, are of interest, both for syntax and for typology. And this is true whether such languages are few or numerous. However, a referee suggests that splits in the governee can always be handled by enriching the feature system (introducing an extra case value to be realized differently on different governees), and so the issue is not of general interest. I believe that this reasoning misses the point in two different ways: (i) it would not remove the interest; and (ii) it does not work.

Suppose first that we could find a solution for each language presented, in which by enriching the feature system we could “handle” apparent instances of governee splits, still using a single rule of case government. Even then it would remain true that English and Mari are not equivalent. Somewhere in the grammar we need to acknowledge that there is an interesting difference, which goes beyond the need for an additional case value.

And second, the proposal fails. This can be seen most readily as soon as a system is not fully categorical. We noted in §3.2 that there are complications in Mari, including the possibility of a minority of adpositions governing personal pronouns in the nominative in some persons (Riese & Bradley & Yefremova 2022: 157). This complication is helpful in that it rules out the suggestion of introducing an adpositional case, which would be realized as the nominative for nouns and the genitive for pronouns. This idea would give a straightforward syntactic rule of adpositional government, reserving the complications for the inflectional morphology. However, since there are instances where adpositions can govern pronouns in the nominative, we would be no further on, since adpositions could then govern pronouns in the adpositional (= genitive for pronouns) or the nominative. Thus introducing an adpositional case value for Mari would be an unhelpful complication.

The situation in Russian looks more promising for a simple featural solution. In §3.3 I focused on the clearest instances. With these, since the second locative has a unique form for some nouns, we could treat it as an additional case value, governed by v ‘in’ and na ‘on’ in their locative use. All governed nouns would take this case value, but it would be realized distinctly only for those nouns with a unique second locative form (and otherwise by the normal locative). However, this additional value of the feature case is not all that is needed; in some instances, there is a gradient effect, of the type to be discussed in the next section. Specifically, there are some nouns which have a second locative but do not always use it. Instead, it alternates with the normal locative, even with the prepositions v ‘in’ and na ‘on’ in locational use. This complication is significant. For those nouns which allow but do not require the second locative with v ‘in’ and na ‘on’, we need to recognize an opposition between locative and second locative, and its resolution depends on the specific noun which is the governee. Since gradient effect prove important in the choice of possible analyses, we turn to them next.

We start with the splits triggered by a particular adposition or group of adpositions (shown in Table 10).

Note that a/b in our tables indicates the possibility of two different case values, each of which is justified by independent analysis, and each of which can occur with adpositions. In other words, we are dealing with two morphosyntactic case values, however they are realized, and not with morphological overabundance (Thornton 2019). We do not propose additional case values in such instances (see Corbett 2012: 215–216 for discussion, following Zaliznjak 1973: 62). The slash indicates only that there is an alternation of case values. The conditions on this alternation can be of different types, not just semantic as discussed in this section (see further in §5). The relative frequency of A and B may be balanced or skewed.

Table 10 represents a relatively common situation, which was previewed in the Russian data in Table 4 (there involving locative vs accusative). Here we illustrate from German. We find some prepositions which govern the dative (like mit ‘with’), and others which govern the accusative (like durch ‘through’). Those which interest us can govern both dative and accusative; they include auf ‘on(to)’, hinter ‘behind’, in ‘in(to)’ and neben ‘near’ (Draye 2014). As the glosses suggest, motion is involved in the distinction:13

Broadly speaking, with the adpositions which allow this alternation, the dative (7) indicates location (place), while the accusative as in (8) is used for motion (path). And while many instances are clear-cut following this criterion, there is much more to be said about the less obvious examples (Smith 1995; Draye 2014). The German evidence involving just these case values (omitting nominative and genitive) is summarized in Table 11.

The basic semantic distinction of location (place) versus motion (path) is widely found (see, for example, Lestrade 2006a: 89 on Russian); in many accounts they are assigned different structures (Garzonio & Rossi 2020a: 7–9 and references there).14 Note, however, that the strategy above, where the expression of motion depends just on the case value is common in Indo-European languages, but infrequent beyond these languages (Lestrade & de Schepper & Zwarts 2011: 266). For the related directive-locative alternation see Zaika (2016). Here, and more generally, we need to spell out the conditions on “/”; that is, we need to specify the factors that determine each case value. The condition may be clear-cut, as with examples (7) and (8). However, there are other instances where adpositions govern different case values and possible motivating factors are hard to establish. And at the extreme, the alternatives may be so different as to raise the issue of different lexical entries. Thus Table 5 includes the Russian preposition o; when taking the accusative it means ‘against’ (as in a collision), while when taking the locative it means ‘about, concerning’. It is debatable whether this should be treated as one adposition (with a split in case government) or two adpositions.

More generally, we have a split in government provided that: (i) we have a reasonable assumption that the value will be the same; but (ii) it is different. And indeed, given the numerous canonical situations where there is a single case value, it is reasonable to enquire further about those instances where specifying both the adposition and the governee is not sufficient to give the case value that will be found. We should not ‘hard wire’ our expectations, by assuming that the condition will be of a particular nature, for instance, semantic. This is one type of possible condition, but just one in the list. Besides being of different types, conditions generalize over the different situations we are discussing. Hence we reserve discussion of them for §5.

We now switch to the other dimension, where the alternation depends on the governee, as in Table 12.

In this system, the governee determines the outcome of the case alternation. A possible example might have these characteristics: governee₁ could be the first and second person pronouns, taking one case value; governee₃ could be nouns, taking another; and governee₂ could be the third person pronoun, allowing an alternation between both case values. I have not encountered this system in pure form. However, a comparable situation does occur as a part of more complex systems. Thus in Modern Eastern Armenian, many adpositions take the dative of nouns, as in (9):

Note that for nouns there is no distinct form of the genitive; the forms traditionally called dative function as both genitive and dative. Nouns are included for completeness only; it is the pronouns which are our focus here, since they are the locus of the split. Pronouns have separate forms for the two case values. In written and standard Armenian, we find adpositions governing third person personal pronouns in the genitive:

For the first and second persons, however, in written and standard Armenian, we find the dative:

In colloquial Armenian, the genitive (found with third person pronouns, and demonstratives) is generalized to the first and second persons:

For the sake of completeness we note that in colloquial Armenian the personal pronoun may also be replaced here by the possessive suffix:

The important point is that for a substantial part of the adpositional system, though certainly not for all, there is an alternation of case values which is determined by the governee, specifically the particular pronoun involved, as represented in Table 13 (again nouns are included only for completeness).15

The situation in Modern Eastern Armenian does indeed show partitioning of the governees and an alternation of case values. It is somewhat marginal in that it affects only the pronouns. And this raises the issue which often arises with pronouns, namely whether the split is to be specified in featural terms (according to person) or whether the pronouns should be treated as different lexemes (Corbett 2005).

We now move to the situation, diagrammed in Table 14, where there is a split in the adpositions, and a split in the governees, both contributing to determining the outcome of an alternation of case values.

This might seem unlikely, but we have a carefully described instance of just this situation. It is found in Finnish, as analysed by Lestrade (2010). Finnish has some adpositions which govern the genitive (the more common situation) and some which govern the partitive (see Huumo 2024 for rich data on the partitive). There are four adpositions which govern both genitive and partitive: lähe- ‘near’, keske- ‘in the middle’, ympäri ‘(a)round’, and yli ‘over’,16 as illustrated in (14) and (15):

These examples illustrate two general regularities. First, these adpositions occur as both postpositions and prepositions;17 see Huumo (2013) for discussion. As postpositions they more often assign genitive case as in (14), while as prepositions they normally assign partitive case (15); however, this is only a tendency (Lestrade 2006b: 27). Second, we see a more literal meaning in (14) and a more abstract meaning in (15). There are additional complications; for instance, within the partitive case, combining with a preposition gives the more literal meaning and combining with a postposition yields a more abstract meaning (Lestrade 2010: 605). The main point for us is that the adposition plays a key role in the split; mainly in that it is only some adpositions that induce the choice, and secondly in that the position (preposition or postposition) also has an effect.

What then of the other dimension? Does the governee matter? It turns out that it does. Lestrade (2010) conducted a corpus investigation of five newspapers (approximately 27 million words) and checked lähe- ‘near’, keske- ‘in the middle’ and ympäri ‘(a)round’. Objects appearing fewer than three times were excluded, and this yielded 3671 examples (Lestrade 2010: 618). While the three adpositions should be able to assign either case to a large number of nouns, in fact for any given noun they typically assign the same case. Just 25 nouns are found in both genitive and partitive with the same adposition (Lestrade 2010: 623); information on all 25 is given in Table 15. The axes have been switched compared with Table 14 and similar tables; those combinations of adposition and governee where both case values are found are again indicated with a heavy frame. To be clear, for a noun to be included requires that there is at least one adposition which takes this noun in both genitive and partitive in the corpus.18

To appreciate Table 15, it is important to recall that there are adpositions which govern a single case value: sisä ‘in’ takes the genitive, while päin ‘against’ takes the partitive (Lestrade 2006b: 24). These would form an outer frame to Table 15. This table homes in on those adpositions which in principle take both genitive and partitive (based on Lestrade 2010: 623). In his sizeable corpus, only a small number of nouns appeared in both possible cases, and then not with all combinations of possible adpositions and possible nouns. Lestrade’s example noun kaupunki ‘city’ as in (14) and (15) is in a sense the “best” example. It is found with both cases in the corpus with lähe- ‘near’ and keske- ‘in the middle’; not, however, with ympäri ‘around’, where only the partitive appeared, even though both case values are in principle possible. On the other hand, five other nouns do appear with both possible case values when governed by ympäri ‘around’.

Table 15 demonstrates that the case value depends on both adposition and governee. Moreover, the environments where an alternation is found are quite specific: particular combinations of adposition and noun governee. Only a small number of nouns (25) allow the alternation. And indeed no noun alternates with all three adpositions. Several combinations are not attested in the corpus (indicated ‘–’). Finnish thus provides a fine example of the situation diagrammed in Table 14, and Lestrade’s investigation allows us to appreciate the detail.

The role of frequency is interesting here. The 25 nouns which show the case alternation have a significantly higher frequency than nouns which consistently take one case value.19 Note, however, that there are some highly frequent nouns which do not take both cases (Lestrade 2010: 623). Some of the gaps in Table 14 are inherent (Chuang & Brown & Baayen & Evans 2022: 396); even with a much larger corpus we would not expect to find examples. Others are contingent; given sufficient data we might well come across an example. (See also Nikolaev & Bermel 2023 for discussion of missing forms in corpora.) For our purposes, the important examples are those in Table 14 where we do find alternation. And to identify those (and indeed to investigate further the missing cells) we need to refer both to the governor (the adposition) and to the governee (the noun). Now certain other alternations discussed earlier have been described in terms of an opposition between pronoun and noun as governees. It seemed surprising that the part of speech should matter, since it implies that the adposition “peeks” ahead when determining its case government. But that is what we also find in the partitioning of the noun inventory. The two issues can potentially be brought together, with frequency playing a role, since we expect pronouns to be more frequent than individual nouns.

A further instance of this general type, an alternation involving both the adposition and the governee, can be found in Russian, involving the single preposition po ‘for’ (also ‘on, about, by, according to’); there is a detailed account in Muravenko (2014), discussed in Corbett (2023a: 129–130), together with a range of other sources, and see also Severskaia (2022). The historical development of this alternation shows that both types of governee restriction were involved (§3.2); both the part of speech of the governee (noun vs pronoun) and its feature value (singular vs plural) played a role. For the analogous situation in Belarusian see Corbett (2023a: 126–127).

We have been considering the partitioning of adpositions and governees with alternations of case values. An intriguing comparison is found in Italian (Garzonio & Rossi 2020b). It requires us to recognize a distinction between lexical prepositions (like sotto ‘under’) and ‘functional’ prepositions (a ‘to’ or di ‘of’). In complex prepositional phrases a lexical preposition may take a functional preposition, as in (16):

Whether a functional preposition is found or not depends both on the lexical preposition and on the governee (complement). The lexical prepositions are of three types and the governees can be divided into full nominal governees or pronouns. The possibilities are schematized in Table 16 (based on Garzonio & Rossi 2020b: 88). Note that there is considerable inter-speaker variation.

It is striking that to determine whether or not there will be a functional preposition (and which it will be) we need to partition the lexical prepositions, and of the governees. This double source of determination is analogous to what we see with case government in Finnish.

We have already seen examples where the condition on an alternation is semantic. The widespread distinction between location and motion, seen for German in §4.1, is clearly semantic. We also find syntactic conditions. A notable one is word order, where the relative order of adposition and governee plays a role. For example, Draye (2014: 120–121fn2) states that German entlang ‘along’ has different possibilities according to whether it behaves as a preposition or postposition: as a preposition it takes accusative, genitive and dative, but as a postposition accusative and dative only.20 And in Finnish (§4.3) we noted the strong link between the order of adposition and governee and the case value selected.

A further type of condition is register, which is related to the issue of diachrony. As the case governed by adpositions changes, conservative usage is likely to be perceived as stylistically marked. Thus in Russian, at the beginning of the 19^th century, meždu ‘between, among’ could take genitive or instrumental. The instrumental was the more frequent, and it became increasingly dominant through the 19^th century (Filippova 1964: 79–83). In the 20^th century the genitive was still found, with Graudina, Ickovič & Katlinskaja (1976: 57) suggesting that it tends to occur in poetical usage (see also Kiričenko 2000 on its semantics). In German, there has been movement in adpositional government, for particular subsets of adpositions, both towards the dative and towards the genitive, interacting with register, as documented by Di Meola (2004); and most recently see Vieregge (2025) on speakers’ reactions to the alternation. In Armenian too (§4.2), we noted change in the system leading to differences in register of the possible case outcomes.

I now ask where such information should be located; in particular, does it logically attach to the lexical entries of adpositions (§4.1), is it part of the specification of governees (§4.2) or both (§4.3)? The picture is not always clear-cut, but in most instances the information primarily adds to the lexical entries of adpositions. In the semantic condition location vs motion, this is a generalization over a subset of the adpositions. (Note that we cannot assign the difference directly to the case values, since these typically have varied semantics going well beyond location and motion; furthermore it is only a subset of the potentially relevant adpositions which are involved, and so these have to be specified.) On the syntactic condition involving word order, the evidence again points to locating this with the lexical entry for particular adpositions. And for the examples cited of conditions relating to register, it appears again that the lexical entries of the individual adpositions are the logical place for the additional conditions on alternations. Further research may well establish conditions which are located on the governee, or on both adposition and governee.