2.1 The lexical versus functional distinction

The notion that there is a distinction between lexical and functional linguistic items has a long history. Meillet’s (1912: 131) early definition of one type of language change, “l’attribution du caractère grammatical”, ‘grammaticalization’, presupposes such a distinction. Grammaticalization is “le passage d’un mot autonome au rôle d’élément grammatical”, ‘the development of an autonomous word to the role of a grammatical element’, with grammatical linguistic items referred to as a distinct class from those items which are autonomous. A distinction between meaningful and grammatical items has found wide support over time, in studies of language change (Kuryłowicz 1965; Hopper & Traugott 2003; Lehmann 2015) and from synchronic perspectives (Cann 2000; Boye & Harder 2012; Bresnan et al. 2015: 93).¹

There is psycholinguistic support for such a distinction amongst linguistic items. For example, research has found that individuals with one type of aphasia, Broca’s aphasia, experience loss of items identified as serving grammatical and linguistic functions, including inflectional morphology, tense-marking auxiliaries and some pronouns (for example Bastiaanse et. al 2011 and Martínez-Ferreiro et al. 2019), but not loss of other items. These findings indicate a cognitive reality to the differentiation between functional items and lexical items, even if there remains debate as to the motivation for these losses with Broca’s aphasia (see Friedmann 2006 and references therein).

The lexical versus functional distinction amongst linguistic items has been defined with reference to various features. A number of properties have been proposed as relevant: membership of an open or closed category (Lehmann 2015: 36); heavier or lighter phonological weight (Hopper & Traugott 2003: 98; Lehmann 2015: 134ff.); variability or invariability in form (Lehmann 2015: 44); real-world reference or procedural, linguistic function (Narrog & Heine 2021: 3); reference to entities or to relations (Hagège 2010: 1; Lehmann 2015: 4); being open or not to positional adjustment (Hopper & Traugott 2003: 149; Lehmann 2015: 167); optionality or obligatoriness (Hopper & Traugott 2003: 57; Lehmann 2015: 148); lower or higher frequency (Sun & Correira Saavedra 2020); available to semantic coercion or unavailable to semantic coercion (Cann 2000: 7); relative distributional freedom or relative distributional restrictedness (Hopper & Traugott 2003: 100; Lehmann 2015: 147–148). A more recent proposal is that of Boye & Harder (2012: 11). Lexical items are those “with the potential for contributing the primary element of utterance meaning”, and functional items are those which must be secondary, a distinction Boye & Harder suggest can be tested by focalizability (2012: 14) and by addressability in subsequent discourse (2012: 14).

The nature of the lexical versus functional distinction differs across these various treatments: sometimes the distinction is envisaged as an absolute split of linguistic items into two sets (Boye & Harder 2012); other times, the distinction has been viewed as a continuum (amongst many others, Hopper & Traugott 2003; Lehmann 2015: 15).

2.2 Adpositions: the lexical versus functional distinction

Applied to adpositions, the properties proposed for distinguishing lexical from functional items are problematic in a way they are not problematic for other linguistic items. This is due to interactions between the properties which define adpositions as a syntactic category and several of the properties used to distinguish lexical from functional linguistic items.² Adpositions denote relations between entities: a relation like location underneath might have real world reference, indicating the real world disposition of two entities, but by virtue of marking a relation between linguistic items also has linguistic reference. A distinction between real-world and linguistic reference cannot easily be made in the case of adpositions: as Froud puts it, adpositions are “neither conceptual nor procedural” (2001: 9). Adpositions might be thought to have a greater selectional restriction than nouns and verbs, but greater selection freedom than items like determiners (see for instance Lehmann 2015: 146ff.), rendering selectional freedom and restriction inapplicable.

Some properties of adpositions lead to the generalization that all adpositions are functional. Adpositions serve a linguistic function by marking relations between entities rather than the entities themselves (Hagège 2010: 1, 5–6). Criteria for the lexical versus functional distinction based on the differences between entity-denoting and relation-denoting items lead to the identification of all adpositions as functional items, as do criteria based on formal variability or invariability (Hagège 2010: 138, 147). Adpositions are relatively light in terms of morphophonological weight (Hagège 2010: 128–129) and tend to have a fixed position in relation to their head: on these grounds too all adpositions are functional linguistic items. Boye & Harder’s (2012: 11) discourse-based definition of lexical and functional items leads to the same conclusion, since adpositions do not have the potential to be primary in discourse, being secondary to the entities whose relations they denote.³

The generalization that all adpositions are functional fails to take account of the evidence from aphasiology that heterogeneity amongst adpositions in terms of the lexical versus functional distinction is a reality which must be accounted for (Freiderici 1982; Froud 2001; Mätzig et al. 2010: 370, 372; Martínez-Ferreiro et al 2019: 507–508). Psycholinguistic research has found that individuals with Broca’s aphasia have variable difficulties with some adpositions but not others (Mätzig et al. 2010: 372), and that individuals with and without Broca’s aphasia make errors with different sets of adpositions (Martínez-Ferreiro et al. 2019: 507). In addition, defining all adpositions as functional ignores the heterogeneity of adpositions in terms of other properties, like lower or higher frequency, lesser or greater restrictions on use, optionality or obligatoriness, which scholars have sought to capture in exactly the terms of a functional versus lexical distinction (van Oosten 1977; Tseng 2001: 20ff.).

One response to the challenges of applying many of the properties proposed for the lexical versus functional distinction to adpositions has been that adpositions stand outside of a binary distinction between lexical and functional, being neither fully lexical nor fully functional (Zwarts 1997: 2; Mardale 2011: 64). Mardale posits a third class of linguistic items, semilexical, to which all adpositions, by virtue of the problematic properties just mentioned, belong. For Zwarts, adpositions are resistant to the lexical versus functional distinction amongst linguistic items: based on properties like morphophonological invariance, all adpositions must be functional, but since there are reasons to reject this conclusion, adpositions are instead identified as being “nonfunctional, nonlexical”. However, designating all adpositions as belonging to a third grouping, neither lexical nor functional does no more than acknowledge the problematic properties of adpositions outlined above. Nor does it seem justified, on the basis of the psycholinguistic evidence, to treat adpositions monolithically, whether as lexical, functional, or semilexical. A rejection of a semilexical definition for all adpositions is vindicated by a careful examination of Mardale’s proposal. Although the claim made initially (Mardale 2011: 64) is that all adpositions belong to a special semilexical group, there is shift to the weaker claim that some individual adpositions are lexical, others are functional, and some sit between these two poles, somewhat lexical and a somewhat functional (2011: 69, 70). Mardale’s initial claim is abandoned in favour of an approach in line with the psycholinguistic evidence, treating adpositions as a heterogeneous group individually available to classification as lexical or functional.

The more common response to the identification of adpositions as lexical or functional has been to propose a lexical versus functional distinction along lines specific to adpositions (van Oosten 1977: 454; Tseng 2001: 96–100). Adpositions like French de and present-day English to are thereby distinguished from adpositions like French pendant and English after (Cann 2000: 9; Tseng 2001: 16–17; Mardale 2011: 67). This distinction has been viewed on the one hand as an absolute split between two types of adposition (Bresnan et al. 2015: 299–300) or between three types of adposition (Abeillé et al. 2006), and on the other as a continuum (Mardale 2011: 69, 70). For Tseng (2001: 19), pre-theoretically at least, “there is a gradient of prepositional uses between Type A and Type B”, where Type A adpositions are lexical and Type B functional.

Since some of the properties proposed for the general lexical versus functional distinction cannot be applied to adpositions, alternative properties have been used, including: whether an adposition has spatial or non-spatial meaning (Abeillé et al. 2006: 6; Svenonius 2010: 130–132; Vincent 2017: 296); is polysyllabic or monosyllabic (Mardale 2011: 65); can be replaced by another adposition or not (Mardale 2011: 67); does not vary with case or does vary with case (Zwart 2005: 689; Cann 2000: 8,9); has lower or higher frequency (Sun & Correia Saavedra 2020); has specific meaning or does not have specific meaning (van Oosten 1977: 454; Tseng 2001: 21; Zwart 2005: 689; Mardale 2011: 65–69; Vincent 2017: 302). In some approaches, combinations of different properties constitute the distinction (Tseng 2001; Mardale 2011: 69); in others, individual properties are privileged to the exclusion of others, notably semantic specificity (van Oosten 1977; Zwart 2005).

However, there are problems with some aspects of these proposals for the lexical versus functional distinction specific to adpositions. Firstly, as Cann (2000: 4) notes with respect to monosyllabicity and polysyllabicity, an individual adposition will potentially be identified as lexical on the basis of some of these properties but functional on the basis of others: the properties proposed in the literature do not align to constitute a lexical versus functional distinction. Secondly, reliance on spatial or non-spatial meanings to differentiate adpositions (Abeillé et al. 2006; Svenonius 2010: 130–132) cannot account for the wide range of literal and metaphorical meanings available to individual adpositions and does not therefore allow classification of a given adposition.⁴ Thirdly, some semantic properties do no more than restate the heterogeneity of adpositions and therefore cannot apply to individual adpositions. Mardale (2011: 69) defines functional adpositions with features including the semantic features [-lexical] and [-theta-assigning]. Whilst the property of theta-role assignment (to the complement of the adposition) might be tested, an adposition being [-lexical] is a re-use of the lexical versus functional distinction, and could not be applied to a particular adposition to determine lexical or functional identity.

Outstanding questions which this article addresses are as follows: what properties constitute the distinction which can be used to identify individual adpositions as being lexical or functional? Is the lexical versus functional distinction amongst adpositions best viewed as an absolute split or a continuum? Is there one single property which defines the distinction or many? Why have individual adpositions proven so resistant to classification as lexical or functional?

2.3 The adposition of

Of is an adposition of Indo-European origin, with cognates in other Germanic and Indo-European languages: for instance, Gothic af, Modern Swedish av, Latin ab. The adposition is reconstructed for proto-Germanic as *aba or *aba (Kroonen 2013: 1). The accepted etymology is that *aba/*aba derives from the ablative case form of an unknown noun not attested in any Indo-European daughter language. In its prehistory, therefore, of would seem to have already undergone a process of grammaticalization from noun to adposition. Since this grammaticalization occurs in prehistory, involves an unattested noun, and does not shed light on the lexical versus functional distinction specific to adpositions, it is set aside, and the focus is only on the historically-attested adposition.

The standard history of of in English, following the sketch given in influential studies like Mustanoja (1960), Mitchell (1985), and Rosenbach (2002) is that of_OE is an infrequently-used spatial adposition approximately equivalent in meaning to present-day English from, before grammaticalizing to take over the functions of the genitive case after the Norman Conquest. Some scholars like Mustanoja (1960: 76–77) have seen a causal connection between changes in the use of of and the influence of the French possessive-marking adposition de. Others, like Allen (2008: 72) argue that contact with de coincided with and sped up an existing process of grammaticalization of of within English.

A diachronic development whereby an adposition used to mark spatial relations becomes a marker of possessive relations, seen in the history of of is paralleled in other languages (Harris & Campbell 1995; Heine & Kuteva 2002). Heine and Kuteva (2002: 33), in their discussion of the development of de and of identify a grammaticalization path ablative > possessive > partitive, that proposed by Harris & Campbell (1995: 339) is similar. The data considered in this article on of_OE gives reason to doubt the ordering of the development from possessive to partitive. For the time being, it is sufficient to note that the history of of has been taken as one example of a grammaticalization path observed cross-linguistically.

Of_OE has received far less scholarly attention than its present-day counterpart. Ceolin’s (2021: 3) assessment is representative: “although OE already had the preposition of, this was limited to partitive readings, and therefore could not be used to express possession or to introduce arguments”. Of_OE is generally characterized as different from of_PDE on two counts (Mitchell 1985: §1199): having spatial semantics, and not being an available variant of the genitive case in expressing possession in the noun phrase. Curme (1913) is a notable exception, dating the use of of_OE as a marker of possession to the early Old English period. Curme’s argument relies heavily on a small number of textually problematic passages from the Bible, some of which might be interpreted as having a more spatial meaning than Curme allows although the insight that of_OE has a range of uses not usually recognised is a valid one supported by this study.

On the other hand of_PDE has received a great deal of attention for its part in the so-called genitive alternation of present-day English (amongst others, Rosenbach 2002; Feist 2012; Grafmiller 2014; Szmrecsanyi et al. 2016). These investigations have given a detailed picture of the contexts in which of_PDE tends to appear.

4.1 Comparing syntax

Both of_OE and of_PDE meet the syntactic characteristics—such as they are— for the category adposition. Both take at least nominal complements (Huddleston & Pullum 2002: 598; Hagège 2010: 58) but are not governed by any one category of head (Hagège 2010: 191,232,236,246). There is no change over time in the categories on which of-phrases depend, only a change in terms of the frequencies with which of-phrases depend on items of these various categories. In terms of internal syntax, of_PDE has an expanded range of non-nominal complements compared to of_OE, which selects only nominal complements.

4.2 Comparing morphophonological form

Of is formally stable in the history of English, with two exceptions. The etymon of of, *aba/*aba, is also the etymon of the Old English verbal prefix a(f)-, seen in verbs like a-flieman, ‘banish’. The prefix adds a meaning of motion away from a source (Campbell 1959:§73; see also Elenbaas 2007: 135). In the prehistoric period, the adposition of and the prefix a(f)- have become phonologically distinct. A second change is seen in the formal split of off, from of, an adposition, in the post-Conquest history of English (Elenbaas 2007). The adposition, adverb, and particle off_PDE is morphophonologically and historically related to of_OE, and the difference in form between the two has traditionally been ascribed to stress (OED online, 2022).

Of_PDE can be reduced phonologically, represented in writing as the -a in forms like lotsa. Also observed in BNC is the representation of phonologically reduced of_PDE as -uv- in the form helluva (Pierce 2017). These forms also represent loss of word boundaries between heads and of_PDE, another important indicator of formal reduction.

In the Old English data there is no orthographic evidence to indicate loss of word boundaries before or after of_OE or phonological reduction. The view found in handbooks and grammars, based partly on assumptions about stress and partly on metrical evidence from poetry (Campbell 1959: §96–97), is that of_OE, like other Old English adpositions, is unstressed in most contexts, in other words, showing relative phonological reduction compared to non-adpositional items in its immediate context, but has full stress more commonly applicable to nouns, verbs, and adjectives, when it appears in the syntactic context of preposition stranding. However, verifying these claims about stress and reduction in Old English prose is not possible. Of_PDE has undergone some loss of form and phonological integrity, but there is no evidence in the corpus data to allow comment on whether or not of_OE was subject to similar processes.

Of_PDE is also phonologically reduced in the context of the hedges kind of and sort of, which are grammaticalizations from (a) kind of and (a) sort of constructions (for the development of the hedges, see for instance Brems & Davidse 2010, Margerie 2010). In kind of and sort of hedges, the element of_PDE no longer marks a relation between entities.

The reduced forms kinda, sorta are sometimes observed in noun phrases where there is a part-whole or classification relation between the nouns sort and kind and the nominal complement of of_PDE, whilst the unreduced forms sort of, kind of are less frequently observed as hedges, as shown in Table 10. Nevertheless, in the BNC, there is a statistically significant correlation between the phonological reduction of of_PDE to -a and the hedge function of kind of and sort of (χ² = 949.501, df =1, p< 0.00001). This is a distribution also observed in Joseph (2011: 195) and used there to criticize associations between morphophonological reduction and grammaticalization.

Morphophonological reduction is one of the few instances of difference and discontinuity between of_OE and of_PDE, evidence for the relatively small differences between this lexical and functional adposition.

Already at this stage in the comparison between of_OE and of_PDE it is clear that the distinction between the adposition cannot be simplified to a single difference. There are differences in the dimensions considered thus far, syntax and morphophonology, and as Section 4.3 will show, differences in semantics too. On this basis, it is suggested that the lexical versus functional distinction amongst adpositions generally consists of multiple properties. This finding is not unexpected given previous assessments, and is in the line with the arguments of Zwarts (1997: 1), Tseng (2001: 34), and Mardale (2011: 69).

4.3 Comparing semantics

Of_OE marks the following relations: cause (19), repeated from (4); the point from which something is removed or separated (20), repeated from (1); parts of wholes (21), repeated from (3); the origin point of movement (22); the material from which something is made (23); agents of passives and in the eventive noun phrase (24); stimulus or theme arguments in the clause and eventive noun phrase (25); the place or space over which someone has control (26); the social group to which someone belongs (27). Several of the relations which have been taken to support the semantic non-specificity of of_PDE are in fact already in evidence in the Old English period. These include possessive relations like part-whole marking, which are taken as evidence of non-specificity by van Oosten (1977: 456) and Zwart (2005: 689), and argument marking in the eventive noun phrase, which has been taken as evidence of the semantic non-specificity by Ceolin (2021: 3) and in some studies of the genitive alternation (Feist 2012: 266).

What is coveted is the source of the covetousness; again a sense of origin is relevant.

Philip is the king who rules over France, he is himself of French birth, providing an association between landed control and social origin.

The relations of kinship (28) and alienable possession (29) are observed marked by of_OE, in 1 example each.

Of_PDE can be used to mark all these relations, although for the origin point of movement (22), of_PDE is only observed as a complement to out (8) or off (9) (Huddleston & Pullum 2002: 639). Of_PDE marking removal and the stimulus of perception is exemplified above (11), (12). The following examples show of_PDE marking subsets (30), material (31), agents (32), cause (33), control (34), and social origin (35).

Of_PDE marks relations not observed marked by of_OE, or in the case of alienable possession and kinship, marked by of_OE in two examples: alienable possession (36); all and any kinship relations (37); characteristics of various sorts (38); topics of discourse (39); body parts (40); what is represented in an image or artistic depiction (41); the author or creator of a work (42); place (43).

As further evidence that of_OE has a smaller semantic range than of_PDE, the specific case of part-whole relations between entities is now considered. The semantic notion of part-whole relations can be divided up into various types of semantic relations, when viewed in more granular detail. In the remainder of this discussion of semantic relations, “part-whole” is used as an umbrella term to refer to a group of different but interrelated semantic relations, following the distinctions between part-whole relations and the labels applied to the different relations used by Payne et al. (2013) to describe of_PDE. The finding is that of_OE marks fewer of these part-whole relations than does of_PDE. Of_OE is observed marking subsets (3), membership (44), and the contents of collections (45). Table 11 gives the frequencies with which of_OE is observed marking these relations.

The set of part-whole relations marked by of_PDE likewise includes subsets (30), membership, (46), and the contents of collections (47).

Of_PDE is also observed marking part-whole relations not marked by of_OE: body parts of animate beings (40), non-bodily inherent parts (48), sub-periods of time (49), and indications of measures and quantities (50).

Part-whole relations are an instance of semantic continuity uniting the two adpositions, but there is also difference insofar as the functional adposition of_PDE marks a greater number of relations covering a greater semantic range than does the lexical adposition of_OE.

Some semantically-oriented characterizations of the lexical versus functional distinction amongst adpositions have stated the distinction as being between adpositions which have a limited number of specific meanings, and adpositions which lack specific meaning. For van Oosten (1977: 456), for instance, the distinction is between meaningful and “colourless” adpositions; for Tseng (2001: 21) it constitutes a continuum from “minimal to maximal” meaningfulness; for Mardale (2011: 69), the distinction is captured by feature [±theta assigning]. It has been established that of_PDE marks more relations than does of_OE, and might be tempting to take this difference in available relations as equivalent to characterizations of the lexical versus function distinction as being between specific and non-specific meaning. The distinction would therefore be between of_PDE which has no specific meaning, marking multiple dependency relations, and of_OE which has a specific meaning, origin. The relations marked by of_OE, such as like materials, causes, and departure points, all relate to literal and metaphorical origins. However, interpreting the fewer and greater relations marked by of_OE and of_PDE in this way ignores the possibility of identifying specific points of semantic difference and similarity. For example, if a collection relation is between a container or group noun and the contents of the collection, of_OE can be used, but not if the relation is between a unit of measurement and the contexts. On the other hand, of_PDE can be used in both circumstances. Likewise, of_OE marks the material from which something is made, but is not observed marking other characteristics of objects like size; of_PDE is observed marking both relations. If of_PDE as a functional adposition is described as having minimal meaningfulness or being semantically colourless these insights into the semantic connections between the relations available to of_PDE are missed. Rather than characterize semantic differences between lexical and functional adpositions in terms of limited meaning versus no particular meaning, it is therefore proposed that a characterization more open to semantic similarities and only small semantic differences between lexical and functional adpositions is adopted instead. This entails a language-specific relative distinction between smaller and greater numbers of discrete, identifiable semantic relations. The semantic distinction between of_OE and of_PDE, and perhaps the lexical versus functional distinction amongst adpositions more widely, can be characterized as a difference between an adposition able to mark n relations and an adposition able to mark n-plus relations, allowing for a situation like that of of_PDE, in which a functional adposition retains the uses of its earlier lexical counterpart.

4.4 Comparing of_OE and of_PDE with respect to variation

One of the most commented-upon properties of_PDE is its participation in variation with the s-possessive, the so-called “genitive alternation” of PDE illustrated in (51); by contrast a comparable variation between of_OE-phrases and genitive-case-marked noun phrases in OE has been routinely dismissed in the literature (Allen 2008: 73–74; Ceolin 2021: 8). Any examination of of_PDE must acknowledge the genitive alternation as an important aspect of the use of the adposition, and this investigation has also found that there is variation between genitive-case-marked noun phrases and of_OE-phrases.

6.1 Characterizing the lexical versus functional distinction

One of the open questions in the literature concerns whether the lexical versus functional distinction amongst adpositions can best be characterized as an absolute split between lexical adpositions on the one hand and functional adpositions on the other or as a continuum between lexical and functional poles. Firstly, the degree of continuity between of_OE and of_PDE suggests that the distinction between the adpositions is not an absolute split. Continuity is evident with respect to variation with case (4.4.2), the predominance of nominal complements, and some semantic relations (4.3). Secondly, there are only a few dimensions in which of_PDE has some property of_OE lacks. These are morphophonological reduction, and taking non-nominal complements, yet in these areas the differences between of_OE and of_PDE have been shown to be contextually limited. Other differences between of_OE and of_PDE are differences in which of_OE has a property to an extent, and of_PDE has the same property to a different, greater extent. It is possible that the properties which constitute the lexical versus functional distinction amongst adpositions are continual whilst the underlying distinction itself is absolute; the nature of the properties and the nature of the distinction need not be identical. However, the predominance of continual properties is suggestive of a continuum distinction, especially combined with the evidence that of_OE and of_PDE have properties in common. A continual distinction between lexical and functional adpositions would not only mean that the category of adpositions can contain items of various lexial or functional status, but also that any individual adposition, like of_eME might be identified as more or less lexical or functional relative to other adpositions.

The findings of continuity between of_OE and of_PDE seem to support the view of the relationship between the lexical versus functional distinction and the category of adpositions arrived at in Mardale’s final assessment (2011: 69–70), also voiced by Tseng (2001: 19). Some adpositions are lexical, others are functional, and some, perhaps like of_eME are more or less lexical or functional. The fact that of_eME shows some differences from of_OE but is similar to of_OE in ways that of_PDE is not supports this assessment of adpositions as a mixed set of lexical, functional, and partly functional, partly lexical items.

However, the differences between these different sorts of adposition are minimal. The differences between of_OE and of_PDE are found to be slight, with the similarities between them sometimes more striking than the differences. This is especially the case in terms of variations, where there is no difference between the adpositions. Those differences that exist are mostly questions of degree (the number of semantic relations marked, the proportion of adverbial and adnominal uses, the number of distinct heads selecting for an of-phrase). It is not unreasonable to suppose that the lexical versus functional distinction amongst adpositions more widely is also a relatively small or close distinction. Items belonging to the category adposition emerge as more functional than items of other categories like nouns and verbs, by virtue of some of the few available definitional properties of adpositions like morphological invariance, a role marking relations, and the inability to have discourse primacy (see Section 2.2). A close distinction between lexical and functional adpositions follows from the fact that even lexical adpositions are more functional than lexical items belonging to other categories. The lexical or functional status either of adpositions generally or the category adposition would therefore seem to be best resolved as functional: the sometimes quite limited differences between a lexical and a functional adposition evidenced by of_OE and of_PDE speak to this conclusion. Some of the features of of_OE, like variation with case and non-spatial semantics, which have sometimes been assumed to indicate the functional status of an adposition, lead to the same conclusion that even lexical adpositions are in fact rather functional, and in this sense adpositions generally might be termed “semilexical” (Mardale 2011). The apparent smallness of these distinctions also offers a general explanation why adpositions generally and some individual adpositions in particular have proven resistant to definitions of lexical and functional status.

Following on from the characterization of the lexical versus distinction amongst adpositions as continual and small, the category adposition can be said to consist of some members which are functional items and some members which are lexical items, and some items which are more or less functional or lexical. In this respect the category adposition would seem to be relatively similar to some other major categories, like the category verb in containing items with differing statuses relative to the lexical versus functional distinction. Nevertheless, the category adposition may well be distinctive insofar as the differences between lexical and functional items are not prominent. On the evidence of of_OE and of_PDE, the extent of semantic relations marked is a major difference between a lexical and a functional adposition (4.3). The importance of semantic relations in the lexical versus functional distinction might also be specific to the category adposition in contrast to other major categories. All adpositions are more functional than lexical in respect to some non-semantic properties, like morphological invariance and light weight, such that semantics is one of a relatively small number of dimensions in which individual adpositions might differ from one another. One of the definitional properties of an adposition is marking relations between entities, and as such it is unsurprising that the number and nature of the relations marked should be important in classifying individual adpositions with respect to the lexical versus functional distinction. The property, marking relations between entities, which is one of the small number of properties which might unite the otherwise hetergeneous category adposition can also be used to draw distinctions amongst adpositions (Asbury et al. 2006; Hagège 2010: 259–261; Deacon 2014: 3).

Of_OE and of_PDE have been found to differ in multiple dimensions, such as the semantic relations marked, types of heads, types of complements. Although semantics is the dimension in which there are the most obvious and greatest differences between Of_OE and of_PDE, there is no single property by which this lexical and functional adposition can be distinguished. Neither variation with case nor spatial semantics, properties previously proposed as distinguishing a lexical from a functional adposition can be used to successfully differentitate textitof_OE and of_PDE, indicating that these properties have no special status for the lexical versus functional distinction amongst adpositions. The lexical versus functional distinction established here comprises multiple properties, in line with the positions taken by Zwarts (1997), Tseng (2001), and Mardale (2011). What the comparison of of_OE and of_PDE also shows is that the multifaceted nature of the distinction is neither an epiphenomenon of making multiple comparisons between different adpositions (as in Tseng 2001) nor a result of drawing together findings from multiple previous assessments (evident in Mardale 2011).

6.2 Grammaticalization

The comparison of of_OE and _PDE has implications too for some descriptions of grammaticalization. The following discussion assumes that of_PDE is the outcome of a grammaticalization process which involves some morphophonological reduction as well as expansion in semantics and to an extent in syntax.

Semantic bleaching, defined in terms of the lowered prominence but retention of core meanings of a lexical item (Hopper & Traugott 2003: 98; Narrog & Heine 2021: 84ff.) aligns with the semantic differences identified between of_OE and Of_PDE (see Section 4.3). Of_PDE can mark relations associated with space and origins but these are less prominent than the corresponding relations in the Old English period. The term semantic bleaching has sometimes been defined in terms of the loss of earlier meaning, not just as the lowered prominence of earlier meaning, as has the term generalization (Lehmann 2015: 34, 88). However, of does not show evidence for this sort of loss of meaning as part of grammaticalization. The relations marked by of_OE are a subset of those marked by of_PDE, such that any relation of_OE marks, of_PDE also marks.

Following the notion that semantic expansion and syntactic context extension occur in parallel (Lehmann 2015: 35; Bybee 2015: 133, 239), it is anticipated that an expansion in the number of available semantic relations will be accompanied by a commensurate expansion in terms of the syntactic contexts in which of is observed. This expectation is not met in the case of of: there is semantic expansion and syntactic differences which seem to be motivated by semantic expansion, but the syntactic differences are not themselves expansions and are not motivated by the general fact of semantic expansion itself but rather by specific semantic relations. What syntactic distinction there is between of_OE and of_PDE consists of a shift in which syntactic context is most common, although the relevant contexts are common to both adpositions: of_OE-phrases are predominantly selected for by verbs, but of_PDE-phrases are predominantly adnominal (see Tables 3 and 4). This distinction would seem to be related to the semantic expansion of of to mark relations associated more with the noun phrase than the verb phrase or clause, like alienable possession, kinship, and body-part relations (36), (37), (40), (48). Nouns (in English, at least) are the core means of referring to body parts, property, and family members. The syntactic distinction between adverbial of_OE and adnominal of_PDE would therefore seem to be a consequence of the identity of a number of particular semantic relations involved in the semantic expansion of the adposition. The evidence that of_eME can mark more possessive and part-whole relations (64), (65), (66) than can of_OE and is also more frequent as an adnominal adposition than of_OE supports this position (see Tables 3 and 13). Such associations mean that it is possible to link syntactic distinctions between of_OE and of_PDE to individual semantic relations available to of_PDE. However, there is no requirement to link syntactic distinctions between the adpositions to the general fact that of_PDE marks a greater number of relations than does of_OE.

The finding that syntax is not a major differentiating dimension between of_OE and of_PDE as compared with semantics contradicts some proposals concerning grammaticalization in constructional contexts. Both within and beyond the framework of Construction Grammar, it has been suggested that linguistic items do not undergo grammaticalization in isolation, but grammaticalize within constructions (Traugott 2003: 643ff.; Noël 2007; Lehmann 2015: 194; Bybee 2015: 132). Grammaticalization is said to apply to an item in “highly constrained pragmatic and morphosyntactic contexts” (Traugott 2003: 645). The reduction and semantic changes to of in the context of the hedges kinda (18) and sorta are an illustrative example of the adposition undergoing change as part of a grammaticalizing construction (see Section 4.2). There is no constructional context, however, for other changes to of. The syntactic contexts in which of is observed remain relatively stable over time, an initial indicator that the mostly semantic grammaticalization of of does not occur in any particular construction (see Section 4.1). The noun phrase is the environment in which the most obvious semantic changes impact of (see Section 4.3). Therefore, identifying adnominal dependency as a construction in which of grammaticalizes might be a possibility. However, no “highly constrained” (Traugott 2003: 645) noun phrase or set of noun phrases in which of grammaticalizes is identifiable. For instance, no individual lexemes or noun phrases fulfilling particular roles are the context for changes to of. Moreover, given that of_OE-phrases are observed as adnominal dependents with several of the same semantic relations as are observed with of_PDE-phrases in the noun phrase, describing the changes to of in terms of an adnominal construction offers no analytical advantage. An understanding of grammaticalization as occurring within a construction misses the fact that it is the possible range of relations which can be marked by of which differs between the Old English noun phrase and its present-day English counterpart. The noun phrase has not grammaticalized and the set of possible relations holding between head and dependent (however these are marked) has not altered. Of represents an interesting example of grammaticalization seeming to affect an item not limited to a particular context.

The change from of_OE to of_PDE does not require two distinct categories, one for the lexical adposition and one for the functional. The grammaticalized status of of_PDE does not seem to have any consequences for the categorial status of the adposition. Of_PDE remains an adposition in the majority of these uses. This is attested, for example, by the fact that nominal complements of of remain in the majority over time (see Tables 8 and 9), nominal complemention being a common but not consistent property of adpositions (Huddleston & Pullum 2002: 598–599). The view of the category adposition proposed in light of the evidence from of is that the category contains functional, lexical, and more or less functional and lexical items allows for of_OE and of_PDE to belong to the same category. A grammaticalizing adposition in this case has not undergone decategorialization presumably because the category adposition is already more functional than lexical.

Determining whether lexical of_OE or functional of_PDE is the more prototypical is challenging given the well-known difficulties in identifying a prototypical adposition at all (Huddleston & Pullum 2002; Hagège 2010). Of_PDE more frequently marks a relation between two NPs and might therefore be considered the more protoypically relation marking (see Tables 4 and 9), yet of_OE has a stronger association with spatial relations (22) (27), and might therefore be considered the more prototypical. It has been concluded here that adpositions are more functional than lexical, in which case of_PDE would be considered the more prototypical. Such a conclusion would entail positing a change from a less to a more prototypical adposition, contrary to some previous characterizations of grammaticalization (Hopper & Traugott 2003:106).

In relation to the categorial status of of_PDE, the evidence that of_OE varies with the genitive case (see Section 4.4.2), comparable to how of_PDE varies with the s-possessive, combined with the evidence that of_PDE varies with other adpositions, including lexical adpositions (see Section 4.4.1), suggest that analysing of_PDE as a case marker, heading a Kase Phrase (as per Svenonious 2010) is not appropriate. The contexts in which the form of is not an adposition in present-day English are adverbial qualifiers: hedges and, the phrase of course. There is therefore an association between category change and grammaticalization of of, but in a small number of contexts, and where there is some evidence of loss of word boundaries.

Finally, the consideration of of in the history of English also has consequences for the specific grammaticalization path proposed for adpositions like English of and Latin and French de. Harris & Campbell (1995: 339) and Heine & Kuteva (2002: 33) propose grammaticalization paths in which partitive uses arise from possessive uses (more accurately, for Harris & Campbell 1995: 339 both possessive and ablatival relations feed into partitive relations). Of_OE is far more frequently observed marking part-whole relations (45), (46) than marking possessive relations (26). The same is true of of_eME, although the set of possessive relations marked by of_eME is larger. In addition the part-whole uses of of_OE chronologically precede the expansion into greater numbers of possessive uses of the adposition evident from the early Middle English (65), (66) and present-day English data (36), (38), (42). There is sufficient evidence to challenge at least the consensus that marking part-whole relations is a development further along in grammaticalization than the marking of possessive relations. Of_OE can mark some part-whole and some possessive relations, of_eME can mark more part-whole and some possessive relations, and of_PDE can mark a greater number of part-whole and possessive relations. It therefore seems that the interaction between partitive and possessive meanings are more complex than existing accounts allow for.