2.1 Physical attachment between the possessor and possessee

While, in some cases, both external and attributive possessives can be uttered to denote the same eventuality, the two constructions encode different kinds of possessive relationships. Unlike attributive possessives, which can presuppose various kinds of relationships depending on the semantics of the possessee, accusative external possessives always presuppose the relationship of physical attachment.

The evidence for this comes from the external possessives with non-body-part entities. These entities, usually elements of clothing or gear, can occupy the position of the possessee (cf. (7)). Yet, authors do not discuss differences between body parts and non-body parts occurring in external possessives. But it is precisely these differences that clearly show that the meaning of the possessive relationship in external possessives is physical attachment, while for attributive possessives, this is not so.

Let us take a look at the following two examples.

Both attributive possessives in the sentences in (8) presuppose possessive relationships. In (8a), Pauls Arm presupposes a body-part relationship ‘arm of’, denoted by the relational noun Arm. The possessee in (8b) is a sortal (or non-relational) noun Rucksack ‘backpack’, which on its own does not provide any information about the nature of the free possessive relationship between the backpack and its possessor Paul. The meaning of the relationship, in this case, has to be supplied by the context: it can be a backpack Paul possesses, a backpack Paul has gifted, a backpack Paul has sewn, etc. In any case, as both examples show, the relationship of physical attachment is not presupposed by either of the two attributive possessives.

Unlike attributive possessives, external possessives do presuppose physical attachment (e.g., (9) and (10)). In fact, they do not presuppose a body-part relationship. So the backpack in example (10a), despite not being a body part, is perfectly acceptable in the position of the possessee.

We can conclude that, unlike body-part relations, physical attachment constitutes a part of the meaning of external possessives. Body-part possessees in these constructions are understood as physically attached to the possessor too, and thus are by default interpreted as natural body parts of the possessor. So the most reasonable interpretation of Arm in (9a) is the one on which it is a naturally attached body part of Paul and thus as an integral part of his body. This has implications for the analysis of relational and sortal nouns, which are discussed in more detail in Section 4.3.

Let us now turn to another meaning component usually attributed to accusative external possessives: physical contact.

2.2 External possessives entail (potential) physical impact on the undergoer

According to a standard view, verbs appearing in accusative external possessives must entail physical contact (cf. Levin 1993; Dowty 2001 for English). The purpose of this section is to weaken this claim. As we will see below, external possessives allow verbs that do not entail physical contact as long as the (potential) result they entail is physical. For instance, the verb verletzen ‘to injure’, which does not entail physical contact (cf. (11)), but entails a physical impact on the undergoer, appears in accusative external possessives (cf. (12)). The action that leads to the physical impact remains unspecified.

Other German verbs that confirm the assumption of physical impact without physical contact requirement are attackieren ‘to attack’ and zurichten ‘heavily damage or injure’. They do not entail physical contact, but the (potential) impact of the action they denote is physical. The requirement of (potential) physical impact allows us, therefore, to predict the behavior of these verbs in accusative external possessives.

The potentiality of physical impact is assumed to account for such verbs as berühren ‘to touch’ and küssen ‘to kiss’. These verbs entail physical contact, but they are neutral towards physical impact on the undergoer, i.e., they neither entail nor exclude its physical impingement. Following Beavers (2011), I will treat these verbs as the ones entailing a potential change in the undergoer. We will come back to this topic in Section 3 of the paper.

As for break, which is commonly discussed in this connection (cf. Levin 1993; Dowty 2001), this verb entails a change in inherent physical properties of the undergoer, such that the undergoer can no longer perform its original function (Dowty 2001: 184). That is, the requirement of a (potential) physical impact is not enough to account for the behavior of this verb in the accusative external possessive frame.

Further challenging verbs are those like zerkratzen ‘to scratch all over, to lacerate’. While kratzen ‘to scratch’ and zerkratzen ‘to scratch all over, to lacerate’ both entail physical impact on the undergoer, only the former can occur in the external possessive frame (cf. (13)).

The behavior of the verbs mentioned, and other challenging verbs will be accounted for in the following section.

2.3 The double entailment constraint

So far, we have considered two meaning components of accusative external possessives: the physical attachment of the possessee to the possessor and the (potential) physical impact entailed by the verb. However, as examples in (13) show, this is not enough to capture the complete picture. In this section, I argue that the central meaning component of accusative external possessives is what I call here double entailment. That is, the (potential) physical impact entailed by verbs in external possessives holds equally for both the possessor and the possessee (cf. (14)). The compatibility constraint on verbs that can occur in this construction can, therefore, be formulated accordingly: to appear in accusative external possessive constructions, verbs must have the double entailment property.

Let us consider an example first. For the already familiar cases repeated below, the same (potential) result requirement holds for both the possessor and the possessee. In (15a) with the verb kratzen ‘to scratch’, the entailed potential result, the scratch, is necessarily on both the hand and the person: (15a) entails both (15b) and (15c). In contrast, double entailment does not hold for attributive possessives: (15d) entails (15c), but it does not entail (15b) (for instance, in the case when Anna’s hand is detached from Anna’s body).

The same is true for the verbs alternating between the accusative external possessive frame and prepositional attributive possessive frame, like the verb beißen ‘to bite’ in (16). Also here, the potential result of a bite, for example, a bite mark, holds for both participants, the possessor and its body part.

The effects described are directly related to the interaction of two factors: thematic roles assigned to both undergoers (they appear to bear one and the same thematic role) and mereological relationships between them. Both factors will be discussed in more detail in Sections 4.1 and 4.2 of the paper.

Evidence for what I propose as the meaning of the external possessive frame is twofold: the underlying existence of an affected possessee, and semantic restrictions on possessees occurring in external possessives. Both pieces of evidence are considered in more detail below.

In accusative external possessives, prepositional phrases containing possessees can always be omitted without loss of grammaticality (cf. (17) and (18)). The omitted possessees are, however, always implied. That is, when a verb occurring in accusative external possessives appears in a transitive clause with an animate undergoer in accusative object position, there is a second implicit thematic participant, a body part or other attached entity, that undergoes the action in the same way as the participant expressed by the accusative object, its possessor. Therefore, the physical impact of this action holds for both the possessor and the possessee, even if the latter is not expressed overtly. That is, whenever a person is kissed, scratched, or bitten, there is always a body part of that person that is kissed, scratched, or bitten in one and the same event. Implicit possessees can be accounted for as dummy body parts whose existence is presupposed, along the lines of the analysis proposed by Postma (1997) for reflexive pronouns.

The second piece of evidence in support of the double entailment constraint on verbs in external possessives is the restrictions on attached possessees. The attached possessees must be such that the double entailment can hold. Consider the examples with verbs packen ‘to grab’ and berühren ‘to touch’ below. The possessee Arm ‘arm’ is acceptable in both (19a) and (19b), while Rucksack ‘backpack’ is acceptable in (19a) but not in (19b), and Hut ‘hat’ and Fahrrad ‘bicycle’ are not acceptable in either case. It is, for instance, possible to grab somebody by grabbing their backpack if it is attached to them, but it is impossible to touch somebody by touching their backpack. That is, the way the possessee is attached to the possessor must be such that the action expressed by the verb can have the same physical impact on both participants of this relationship, the possessor and the possessee. For the verb packen ‘to grab’ it would be control over the possessor and the attached possessee which would hold only if they are attached to each other firmly.

In summary, in this section I have argued that whether a verb can occur in the external possessive construction depends on the satisfaction of the double entailment constraint. This is true equally for verbs that differ in other semantic properties that have been thought to be relevant for their acceptability in this construction.

3.1 ‘Break’ verbs

As has been stated for English (cf. Massam 1989; Levin 1993; Dowty 2001), but is also true for German and many other European languages, ‘break’ verbs do not allow possessor raising, i.e., they do not occur in accusative external possessives (cf. (23)). Given the proposed account, the explanation for their behavior is straightforward.

On their physical readings (the only readings that might lead to any physical impact), ‘break’ verbs do not take animate complements.¹⁵ Therefore, these verbs cannot fulfill the double entailment condition: the physical results they entail cannot hold for animate possessors.

3.2 ‘Torture’ verbs

Much the same is true for ‘torture’ verbs. Just like ‘break’ verbs, they impose a restriction on the kind of arguments they take. These verbs subcategorize for animates exclusively, and the result state they entail can only hold for animates. Verbs like kränken ‘to hurt’ or peinigen ‘to tantalize’ on their physical readings (cf. (26a)) entail that the necessarily animate undergoer physically perceives some unspecified action and suffers from it; the verb wachküssen ‘to kiss awake’, behaving in the same way, entails a result state of being awake, which is only possible for animate undergoers. Consequently, verbs of this class can neither occur in the external possessive (cf. (26c)) nor in the attributive possessive frame (cf. (26b)) where the complement position is filled by an inanimate body part or another attached entity.

3.3 ‘Shred’ verbs

The last and the most puzzling class of verbs not appearing in the accusative external possessive frame are ‘shred’ verbs. As shown in the examples repeated below, verbs of this class occur in the attributive possessive frame and also take animates as direct objects (cf. (27a) and (27b)). That is, the result state they entail can hold likewise for animate and inanimate entities. Nonetheless, they do not occur in accusative external possessives (cf. (27c)). In what follows, I argue that the reason for that is the kind of result state they entail: it is quantized and, therefore, it is not universally the same for two entities standing in the mereological relationship to each other, like the one presupposed by accusative external possessives. This makes ‘shred’ verbs fail the double entailment constraint and explains why they cannot occur in accusative external possessives.

There are two striking characteristics of ‘shred’ verbs: first, many of them are marked by prefixes zer- ‘apart, completely’ or durch- ‘through’; and, second, many of them have a non-prefixed counterpart compatible with the external possessive frame, e.g., zerkratzen ‘to scratch all over, to lacerate’ vs. kratzen ‘to scratch’.

German prefixes zer- and durch- are related to the result state entailed by the verb. The prefix zer- ‘apart, be in pieces’ adds to the meanings of the verbs with which it combines that the thematic argument of an event described by the verb is divided or separated into parts, seriously damaged or annihilated. Zer- thereby specifies that the events expressed by verbs have culminated (Stiebels & Wunderlich 1994; Stiebels 1996).¹⁶ (These effects are visible in the English translations of the zer- verbs provided in footnote 13.) Likewise, when the prefix durch- ‘through’ combines with verbs, it specifies the result and entails a culmination (e.g., beißen ‘to bite’ vs. durchbeißen ‘to bite through’).

Unlike verbs that occur in accusative external possessives, durch- and zer- verbs, as well as other verbs of this class, entail a quantized result (Hay et al. 1999; Beavers 2011), also known as a definite state (cf. Dowty 1979). Such a result state is quantized with respect to some physical property of the undergoer of the action (e.g., volume, surface, thickness, etc.). Consequently, it does not hold simultaneously for two entities being in a relationship of physical attachment with each other (which, in most instances of accusative external possessives, is a part-whole relationship).¹⁷ In contrast, verbs that entail a non-quantized result state, like some of the verbs in ‘touch’ and ‘hit’ classes (e.g., kratzen ‘scratch’), can appear in accusative external possessives.

Let us consider as an example the pair of verbs kratzen ‘to scratch’ and zerkratzen ‘to scratch all over, to lacerate’, the first of which appears in external possessives, while the second does not. The potential result state entailed by the alternating verb kratzen is the particular kind of damage to the undergoer’s material integrity (at least one scratch). This result can be taken in two ways, as damage to the part and as damage to the whole: the damage to the part is the damage to the whole. It is in this sense that double entailment should be understood.

The verb zerkratzen ‘to scratch all over, to lacerate’ entails a different kind of result. Not only does it entail the presence of scratches, but it also entails that they are distributed over the surface of the undergoer, specifically, that the undergoer is covered by scratches. To say that the body part is fully covered with scratches is a different claim from the one that the body as a whole is fully covered with scratches. The two claims are logically independent: not only can the hand be fully covered with scratches without this being true for the body as a whole, but it is also possible for the body to be conceived as fully covered, while yet one would not necessarily want to say this about the hand. While one can imagine scenarios on which the result state of being scratched all over is true for both, the possessor and its body part (e.g., the person is scratched all over and the person’s hand too), this does not hold universally, that is, it is not a part of the verb meaning. Similar logic applies to all other members of the ‘shred’ verb class, such as durchbeißen ‘to bite through’, zerstückeln ‘to break/cut into pieces’, waschen ‘to wash’ and others.

The explanation for the behavior of ‘shred’ verbs in accusative external possessives proposed here is also supported by the evidence from resultative constructions. As we have seen, ‘touch’ and ‘hit’ verbs, which do not entail a quantized physical change in the undergoer, occur in accusative external possessives (cf. (28a) with a ‘touch’ verb lecken ‘to lick’). While their attributive counterparts can be enriched by a specific result state, as in (28b), this is not possible in case of the external possessives (28c), owing to the incompatibility of quantized results with the double entailment.

3.4 ‘Touch’ and ‘hit’ verbs and what else they tell us about the meaning of accusative external possessives

As we have seen, ‘touch’ and ‘hit’ verbs can occur in accusative external possessives. However, the two verb classes differ in their behavior in attributive possessives. In this section, I discuss these differences and argue that, against the common claim that the use of external possessives is associated with the expression of affectedness on its most general understanding (cf. Chappell & McGregor 1996, Heine 1997, Wierzbicka 1988, Payne & Barshi 1999, Podlesskaya & Rakhilina 1999), these constructions do not encode it.

The term affectedness used in connection with external possessives is usually understood very broadly and is not limited to the kinds of physical impact on the undergoer. It captures the difference between sentences (29a) and (29b) as alternative descriptions of the same scenario as follows. While the physical impingement of both the possessor and the possessee is the same in both cases, the possessor’s surfacing in the more prominent position of the accusative object emphasizes its affectedness by the event expressed in the sentence. The use of attributive possessives instead evokes an inference of the distant perception of the event by pure virtue of the availability of the alternative encoding with external possessives connected to the expression of the affectedness (cf. Lehmann 2021).

While most native speakers of German share the above intuitions about differences between accusative external and attributive possessives, it is not entirely clear whether the information about the possessor’s affectedness is directly encoded in the external possessive frame or derived by inference. In what follows, I will argue for the latter.

In the discussion of the double entailment carried out in the previous sections, two alternating verbal classes, ‘hit’ and ‘touch’, were considered in opposition to the other three classes which do not occur in the accusative external possessive frame. However, as we have seen in Section 1, verb classes ‘hit’ and ‘touch’ behave differently when it comes to the formation of the attributive possessive frame. Unlike ‘touch’ verbs, inanimate undergoers of ‘hit’ verbs are encoded not as accusative objects, but as obliques (cf. (30a) and (30b)). With other words, only animates can occupy the accusative object position of ‘hit’ verbs.

Verbs in the ‘touch’ verb class are so heterogeneous that it is hard to make any generalizations about them besides that all of them, as discussed previously, entail a potential or non-quantized physical impact on their animate or inanimate undergoer. The class of ‘hit’ verbs is, in contrast, well-defined. Verbs in this class have been studied for a variety of languages and found to have three basic meaning components: physical contact, movement, and force (cf. Levin 1993; Gao & Cheng 2003; Goldschmidt et al. 2015). Indeed, in German these verbs take oblique complements in accusative case encoding goals (as opposed to ‘touch’ verbs mostly appearing with obliques in dative expressing location).¹⁸ Due to the mentioned peculiarity in their surfacing patterns with inanimate undergoers (also observed in Dutch and exemplified once again in (31)), these verbs have been discussed in the context of paradigmatic differential object marking (cf. Fleischhauer 2018 for German and de Swart 2014; de Swart & de Hoop 2018 for Dutch).

Based on the animacy restrictions exhibited by ‘hit’ verbs, they have been accounted for in terms of sentience (de Swart 2014; cf. Dowty 1991 for the term) or experiential affectedness (Fleischhauer 2018; cf. Lundquist & Ramchand 2012: 229 for the term). According to these accounts, ‘hit’ verbs entail experiential affectedness of the undergoers appearing in the accusative object position. Therefore, this position can only filled by animates (inanimates are not sentient and, thus, cannot be affected experientially). That is, the accusative external possessives with ‘hit’ verbs expresses affectedness.

Yet in German, ‘touch’ verbs do not impose sentience restrictions on their undergoer arguments: not only do they take likewise animates or inanimates as accusative objects, but most of them also occur in external possessives, with inanimates standing in part-whole relationships to each other (cf. (32) and (33)). Consequently, we can state that in German, affectedness meaning, as understood by the last mentioned authors, is not directly associated with the external possessive frame.

Nonetheless, intuitively not only for ‘hit’ verbs but also for ‘touch’ verbs, animate possessors in external possessives still seem to be more affected than in their attributive counterparts (cf. (29)). There are several tentative non-competing explanations for this, which require further research not pursued here.

Firstly, affectedness inferences might be drawn from differences in the information structure. In particular, the appearance of the possessor in the more prominent object position may lead to the inference of its higher importance in the event denoted by the sentence. Secondly, canonical objects, as opposed to the deeper embedded elements like attributive possessors, have a natural affinity towards affectedness. That is, from all the participants, the ones encoded by accusative objects are most likely to be affected (cf. Dowty 1991; Beavers 2010).

Furthermore, the following observations might also be relevant to the matter. In cases with an animate possessor, when both external and attributive possessives can be formed, the former seems to be preferred over the latter. This fact might reflect generalizations captured in several prominence hierarchies in all of which animates are ranked higher than inanimates and, thus, are expected to occur in more prominent positions in the sentence (cf. empathy hierarchy by DeLancey (1981), personal hierarchy by Siewierska (2008) and animacy hierarchy by Comrie (1989) among others). In addition, attributive possessives tend to occur in particular text genres that lack emotional coloring, such as medical handbooks, legal protocols, or descriptions of paintings. As for inanimate possessors, external possessives with such possessors are somewhat rare, even if grammatically acceptable. For inanimate entities standing in a part-whole relationship with each other, the choice the speaker is facing seems to be between the attributive frame and the frame in which the whole occupies the accusative object position without further specification of the subpart (e.g., He holds the stalk of the plant vs. He holds the plant). This is possibly related to the fact that the specification of the subpart (or the secondary undergoer) has more relevance in the discourse when it comes to animates.

In sum, the observations mentioned above might explain the inferences about the possessor’s affectedness associated with the use of the external possessive frame. However, as shown by the lack of animacy restrictions on participants in the accusative object position with ‘touch’ verbs, the external possessive frame does not entail the possessor’s experiential affectedness.

4.1 Thematic roles

One of the basic assumptions about thematic roles is uniqueness. According to the uniqueness principle, one event can have only one thematic role of a particular kind (Carlson 1984). For instance, in one and the same event of eating, there will be one and the same eater and the same food to be eaten. Events described by external possessives are different in that there are two argument phrases with the same thematic role.

While there are semantic differences between accusative external possessives and attributive possessives (cf. Section 2), there are cases when both structures can be used to denote one and the same event. So, in the pair of examples repeated below, whenever (34b) is true, (34a) is also true. That is, the thematic roles of the possessor and the possessee are semantically equivalent.

According to Carlson (1984), the constraint on thematic uniqueness arises from the natural properties of eventualities; thematic uniqueness is not a linguistic property of verbs. Accusative external possessives with ‘touch’ verbs are an exception. The existence of verbs violating uniqueness has been pointed out by Krifka (1998). He briefly discusses two verbs of this kind: see and touch. Krifka proposes to capture uniqueness as a general principle that can be overridden by more specific rules, for instance, by the rule below formulated for the verb touch (cf. (35)). Krifka points out that this kind of rule serves as a conceptual basis for possessor raising.

Indeed, as we have seen, accusative external possessives reveal a whole class of verbs that denote events in which the part and the whole play the same thematic role. However, to formulate a more general rule that would apply to all of them does not seem possible. One reason for that is that the relationships of physical attachment between possessors and possessees are not uniform, they cannot be uniformly defined for all verbs in this class. For instance, the relationships suitable for touching and grabbing are not the same: one can grab somebody by the backpack, but is not possible to touch somebody on the backpack. Eventualities involve entities that come with their mereological properties. If, due to these properties, the part and the whole are involved in the event in the same way, they can be assigned equal thematic roles.

Unlike those described by ‘touch’ verbs, events described by ‘hit’ verbs do not assign the same thematic roles to the possessor and possessee. As discussed in Section 3.4, they exhibit differential object marking: their animate undergoers surface as accusative objects, while inanimate ones surface as obliques. Although they undergo the same physical action and the entailed potential physical change holds for both of them, the thematic roles assigned to them are different. Besides the potential physical result, the thematic role assigned to the animate possessor includes sentience, while the one assigned to the part lacks it. In other words, ‘hit’ verbs do not violate the uniqueness principle.

According to another common assumption, no argument is assigned more than one thematic role (Bresnan 1982; Carlson & Tanenhaus 1988). Also this is challenged by accusative external possessives. As we have just seen, along with the possessor, the possessee plays the role of an undergoer. However, it also seems to have the role of location, direction, or manner (depending on the verb and the analysis). So in the already familiar example with‘touch’ verb below, questions targeting the possessee in sentence (36b) are questions about location or manner (less common) (cf. (36a)). The same is true for ‘hit’ verbs: the possessee in external possessives appears to be the theme and the location, direction or manner at the same time (cf. (37)).

4.2 Verb meaning and nominal mereology

Aspectual properties of predicates correlate with the meaning of their nominal arguments (cf. Garey 1957; Verkuyl 1972; Platzack 1979). It is well established that the (a)telicity of eventive predicates interacts with the mereological composition of their undergoers. So atelic eventive predicates, like drink or push, yield atelic readings due to the scalar characteristic of the non-quantized change they entail in combination with the quantizedness of the undergoer: while the predicate in (38a) is atelic, the one in (38b) is telic. The culmination of an event, like the one denoted by (38b), is partly induced by the physical finiteness of the undergoer (the event reaches its logical culmination when there is no more beer left) (cf. Krifka 1989; 1992; 1998; Beavers 2011).

Quantized and non-quantized result states are kinds of event culminations that, for the verbs discussed in this paper, are characterized by a physical change in the undergoer. Verbs like kratzen ‘to scratch’ and zerkratzen ‘to scratch all over, to lacerate’ discussed previously are good examples for this distinction that, as we have seen, determines the verb behavior in accusative external possessives. In German, English, and Spanish, verbs entailing quantized change cannot occur in external possessives because a quantized result is determined in relation to some physical property of the undergoer and, therefore, cannot hold for a whole and a part at the same time (cf. Section 3.3). Thus, the mereological constitution of the undergoer correlates not only with the culmination of the event denoted by the predicate but also with the kind of culmination, i.e, with the quantization of the entailed result state.

It is interesting in this connection how the mereological structure of the undergoer is conceived. In accusative external possessives, the possessor is affected through the possessee. That is, the action denoted by the verb is carried out on the possessee and, as a consequence of the physical attachment, thereby also on the possessor. The possessor and the attached possessee are, therefore, conceptually considered as two constituents of a single totality. However, unlike animates, which are typically perceived as wholes whose body parts are their natural components, animates with attached elements of clothing or gear are generally not conceptualized as integral wholes. That is, accusative external possessives allow for the creation of, so to speak, new, ad hoc wholes that do not correspond to any established mereological concept (cf. accidental integrated wholes by Moltmann (1997)). The common denominator between the two conceptually very different kinds of part-whole relationships (natural and ad hoc ones) encoded by accusative external possessives, is the physical attachment at the event time that allows the (potential) physical result to be transferred (or to spread, cf. Dowty 2001) from the possessee to the possessor.

4.3 Possessive relationships

There are two commonly distinguished kinds of nouns: relational and sortal (cf. Löbner 2011). While sortal nouns, like table or backpack, denote entities, relational nouns, like father or arm, denote entities and relationships in which they stand to another entity, their possessor. Despite this semantic distinction between the two kinds of nouns, both of them can appear either accompanied by a possessor (cf. (39a) and (40a)) or without it, on their own (cf. (39b) and (40b)). There is a not yet settled discussion of how to analyze formally equal but semantically different attributive possessives with relational and sortal nouns (cf. (39a) and (40a)) (cf. Hellan 1980, Partee 1983/97, Jensen & Vikner 1994; Barker 1995; Vikner & Jensen 2002; Partee & Borschev 2003). Specifically, there is no consensus about whether possessors occurring with both kinds of nouns should be considered their arguments or modifiers. Nonetheless, there is a general agreement that in cases like (39b), where relational nouns appear without an NP-internal possessor, their meaning is shifted from its original relational function to the sortal function.

Although in accusative external possessives, relational nouns appear without NP-internal possessors, their meaning does not seem to shift. Not only can relational nouns be interpreted as such, but their sortal interpretations appear to be disfavored.

As discussed in Section 2.1, accusative external possessives presuppose physical attachment. They allow non-relational nouns, like backpack, in the possessee position. However, relational possessees, like nouns for body parts, are still interpreted as such. So, in example (41) repeated below, der Arm ‘the arm’ not accompanied by an NP-internal possessor is interpreted as a body part of Paul and, along with this, as an arm attached to Paul’s body. That is, it stands in two relationships to a single referent (viz., the possessor). In addition, the body-part relationship can be made explicit by an optional NP-internal possessor, as in example (42). In this case, both relationships, the body-part relation and the physical attachment relation, are directly encoded in the sentence.

Similar in this sense are German dative external possessives (cf. (43) repeated below). Hole (2014) argues that nominal phrases containing possessees include an implicit anaphor picking up the possessor antecedent appearing in the position of the dative argument. In this regard, possessees resemble associative definites (also known as binding definites or inferables) (cf. Clark 1977; Prince 1981; Löbner 1998 among others). However, unlike instances of associative anaphora, like the one given in (44), in dative, but also in accusative external possessives, the assumed implicit anaphora must be bound clause-internally. That is, the possessor must occur in the same clause as the possessee.

While dative and accusative external possessives are similar in that possessees occurring in these constructions do not have to bear an overt NP-internal possessor marking, they also show some crucial differences, so that the analysis developed by Hole (2014) for dative external possessives is not applicable to the accusative ones without further ado. In dative external possessives, the possessee appears in the accusative object position and cannot be omitted without lost of grammaticality. According to Hole (2014), free datives, not licensed by the verb, always must bind a clause-internal anaphor, which, like in our case, can be implicit. This bound anaphor is always contained in the nominal phrase encoding the possessee.

Applying Hole’s (2014) analysis to accusative external possessives discussed here, we can state that the anaphor contained in the nominal phrase encoding possessee is likewise always bound by the possessor. However, the latter does not have to bind per se: as mentioned in Section 2.3, the prepositional phrase encoding the possessee in accusative external possessives can always be omitted. That is, unlike dative possessors, accusative possessors do not have to bind, but the possessees in these constructions, when they are overtly expressed, must be bound clause-internally.

The presence of a zero-anaphor would explain the interpretability of the possessive relationship (physical attachment) for all kinds of possessees occurring in accusative external possessives, including sortal non-body-parts. In addition to that, further piece of evidence suggests that sortal interpretations of relational nouns appearing in accusative external possessives are generally dispreferred. Consider the example (45) below. If, according to the standard assumption, a shift from relational to sortal meaning occurred whenever a relational noun appears without an NP-internal possessor DP, we would expect the dependent clause to be felicitous. This is, however, not the case. The relational interpretation of Arm collides with the sortal interpretation required for the dependent clause to be felicitous.

Similarly, English speakers confronted with the parallel example (46) have pointed out that, to make the sentence acceptable, the possessee phrase must include adjectives fake or false. These adjectives necessarily shift the meaning of the body-part noun from relational to sortal. Without them, the arm would receive a relational interpretation which, again, would collide with the sortal interpretation required for the felicity of the dependent clause.

Consequently, we can state that relational possessees accompanied by contracted definite articles in accusative external possessives do not shift from relational meanings to sortal. This, however, does not mean that relational nouns cannot shift at all. As shown by example (47) below, such a shift is possible when a full, or non-contracted, definite article is used (compare (47) with (45)). Unlike their contracted counterparts, which in German, mark dependent definite noun phrases characterized by the reduced referential potential in discourse (cf. Schwarz 2009, 2014; Krifka 2021), regular definite noun phrases can introduce discourse referents that not only can be picked up more easily in the subsequent discourse but can apparently also be reevaluated regarding their relationality. So Arm in (47) can receive a sortal interpretation.

In a nutshell, accusative external possessives show that type-shifting is not a necessary condition for a relational noun to appear without an overt NP-internal possessor. Furthermore, accusative external possessives offer an insight into the meaning of body-part relationships. They show that these relations are conceptual bonds between two entities as participants in a given eventuality e, which go beyond a mere physical relationship which holds between them independently of e.