Non-syntactic factors and accessibility to relativization: evidence from Armenian

2.2.1 Material between ‘filler’ and ‘gap’

In order to shed light on these apparent AH violations, it is necessary to investigate the mechanisms that have been proposed for the AH effects: why do we find these effects, and what is causing them? Many proposals focus on complexity in the sense of the amount of material intervening between filler (the relativized noun, or, in some proposals, the relative pronoun if there is one) and gap (argument position of the relativized element in RC). The simplest version of this type of theory, as seen in Tarallo and Myhill (1983), simply counts the number of intervening words, so an English subject relative clause (SRC), such as (4a), will be easier than an object relative clause (ORC), such as (1b), because in (4a), there is only one word (‘that’) between filler (relativized element ‘the man’) and gap, but in (1b) there are four words (‘that the woman likes’):

However, this would not have universal application, due to the variety of word order patterns in the languages of the world. For example, in Malagasy, with clause-final subjects and postnominal RCs, the subject would be expected to have the lowest accessibility (N[VOS], with the whole clause intervening between the head noun and RC subject position), but in fact, this language can only relativize subjects. These facts have led to the proposal that what is important in determining the complexity of different constructions or dependencies is not linear distance or number of intervening referents, but rather structural distance, in terms of the depth of syntactic structure between the elements linked by the dependency.

Hawkins (2004), within the framework of a wider theory of processing efficiency and complexity in grammars, explains the AH as an instance of his proposed universal principle ‘Minimize Domains’, which states that one way that efficiency is increased is by “minimizing the domains (i.e., the sequences of linguistic forms and their conventionally associated properties) within which certain properties are assigned” (Hawkins 2004: 9). Subject relative clauses (SRCs) will always be more difficult to process than object relative clauses (ORCs), regardless of word order, because the intervening VP node in the latter means that there is an extra layer of structure between filler and gap, increasing the complexity of the construction. Hawkins’s (2004) theory claims to explain not only the reason why SRCs are easier to process than ORCs, but is extended to the whole length of the AH, with obliques conceived of as containing an extra layer of structure such as a PP, and possessive constructions containing a possessor phrase, so that their relative accessibility will depend on the role of the possessee in the clause (possessor of the subject will be more accessible than the possessor of DO, etc.), giving the following hierarchy (Figure 2).

However, there is evidence that it is not simply the amount of material between filler and gap that is relevant, but that the properties of the intervening elements can affect accessibility to relativization. Notably, ORCs have been shown to be easier to process when the subject of RC is a pronoun, i.e., denotes a referent that is already present in the discourse. This leads Gibson (1998 etc.) to propose that the crucial factor is the number of new discourse referents introduced between filler and gap. However, Gordon et al.’s (2001) results show that processing of ORCs is also eased in cases when the subject is a proper name,^[2] even though this constitutes a new discourse referent, casting doubt on this aspect of Gibson’s theory. Gordon et al. (2001) propose that the crucial issue is not new discourse referents, but the similarity between subject and DO. When the subject and DO have similar properties, this causes problems for ORCs. When the subject and DO have different properties, e.g., when the object is a lexical noun and the subject is a pronoun, these problems do not arise. Further support for this proposal comes from the fact, reported in many studies (e.g., Mak et al. 2002), that ORCs are less problematic when the object is inanimate and the subject is animate, i.e., the referents differ in terms of this semantic feature.

Similarity between subject and object referents has been proposed to be the key factor behind these data in both generativist (e.g., Belletti et al. 2012; Hamann and Tuller 2010) and psycholinguistic (e.g., Humphreys et al. 2016) studies, the former interpreting it as an instance of Rizzi’s (1991) principle of Relativized Minimality,^[3] the latter pointing to the fact that referents with similar properties inhibit each other’s cognitive accessibility, and that competition may arise when two nouns are good candidates for the same role with regard to the verb, leading to processing difficulties. However, if the similarity between subject and object referents were the only factor behind the problems caused by ORCs, we would expect that ORCs in which the subject is inanimate and the object is animate should show a high level of accessibility, analogous to that of ORCs with an animate subject and inanimate object. Various experimental studies (Traxler et al. 2002, 2005; Wu et al. 2011) show that not only is this not the case, but that in fact there is evidence that this type of ORC shows a lower level of accessibility than those in which subject and object have the same value for animacy (Traxler et al. 2005: 218).

2.2.2 Role-reference association

These results suggest that the crucial factor behind the difference in accessibility between ORCs in which the referents have different semantic characteristics is not in fact similarity between the subject and object referents in itself, but rather role-reference association. According to Haspelmath’s (2021) Role-Reference Association Universal, constructions are easier to process when each referent plays a role that is frequently associated with elements having its referential properties. This means that constructions in which the subject referent has typical subject properties, being high in cognitive salience (as represented here by the Silverstein hierarchy of prototypical subject referents as shown in Figure 3), will be easier to process than ones in which it is low on the hierarchy, and especially ones where it is lower than the object.

This explains why the accessibility of ORCs is improved when the subject is a pronoun or a proper noun, as it is higher in cognitive salience than the object, a lexical noun. In the same way, an ORC with an inanimate object and an animate subject shows improved accessibility, while one with an inanimate subject and an animate object is actually worse than one in which both referents are animate, which would be completely unexpected if similarity were the main issue, but exactly what we would expect from the point of view of role-reference association patterns.

Role-reference association can also explain the general preference for SRCs over ORCs. As pointed out by Kuno (1976), a relative clause is inherently a statement ‘about’ the relativized element, thus the relativized element always has topic status within RC. Topic status is one of the key properties frequently associated with syntactic subject status, thus constructions in which the relativized element (topic) plays subject role will always be easier to process than ones in which it does not. This is why, when the subject and object referents occupy the same position on the Silverstein hierarchy, SRCs are preferred to ORCs, as the fact of being the relativized element itself adds to the cognitive salience of the referent in question, effectively raising its position on the hierarchy. The subject preference may be overridden if the semantic properties of the relativized element are particularly atypical for subject referents, e.g., if it is inanimate, and/or if there is another referent in the context with a much greater degree of salience, e.g., a pronoun or a personal name.

That role-reference association is a key factor behind the subject preference in relativization is confirmed if we look in more detail at an important source of evidence that has often been cited in favor of the subject preference, namely errors in elicited production experiments. It is very common for subjects to produce subject RCs ((5) and (6a)) in circumstances designed to elicit RCs in which the relativized element plays the role of object or some other grammatical relation ((5) and (6b)). These include both cases where the semantic roles are reversed, with the relativized element being expressed as agent rather than patient (5), and cases where the semantic roles remain the same, but the relativized patient is expressed as syntactic subject of a passive construction (6).

These are recorded as errors in the production of ORCs (or other non-subject RCs), and taken as evidence that these structures present more difficulties than SRCs. Diessel and Tomasello (2005), in an elicited imitation task carried out by English- and German-speaking children, note that most of the errors involve ‘promotion’ of the relativized element to subject, e.g., Example (5). However, they did find some examples of ‘demotion’ of subject to DO, for Example (7):

What we actually have here is a promotion to subject of the proper name object, which has a higher level of cognitive salience than the lexical NP subject. Parallel errors are found in German. If we were dealing with a mechanical reflex of syntactic structure universally favoring SRCs over ORCs, we would never expect to find such errors. However, if they are viewed from the perspective of role-reference association patterns, they make perfect sense. What all these errors have in common is that the referent with the highest position on the Silverstein Hierarchy is expressed as the subject of the relative clause.

2.2.3 Topicality and AH

As we have seen, one probable factor behind the subject preference that is the most thoroughly studied part of the AH is the fact that every relativized element intrinsically has topic status in RC (the relative clause is always a statement ‘about’ the relativized element, Kuno 1976), and elements with topic status are most frequently articulated as the syntactic subject. In this context, we would expect that languages where topics are particularly frequently or obligatorily articulated as syntactic subjects, for example using mechanisms such as passive if they are not thematic subjects, would show a stronger subject preference in relativization than those in which the link between the topic and syntactic subject is weaker or non-existent. There is some evidence that this is the case. For example, in some Malayo-Polynesian languages such as Malagasy, there is a wide range of voices that are used to articulate topical elements with almost any thematic role as ‘subject’ (agreement trigger) (see Pearson 2005, etc.). Thus in these languages, ‘subject’ status is assigned based on topicality more than thematic role. We would expect that relativized elements, as topics, would generally be articulated as a subject. In fact, only subjects can be relativized in Malagasy and some other languages with similar systems, such as Tagalog.

In contrast, some languages, such as Chinese, do not have a strong association between topic and syntactic subject status. This concerns Li and Thompson’s (1976) ‘topic-prominent’ languages, where topic has a much higher functional load than the thematic subject, typically receiving some surface coding (most often initial position, sometimes morphological marking, rarely verb agreement), while the subject may not. As the notion of topic plays a more fundamental role than that of subject in sentence construction in these languages, there is no need for a construction promoting topical non-subjects to subject status, and the link between topic and subject status is weak. Non-subject relativized elements (topics) will not be promoted to syntactic subject in the RC as they frequently are in languages that make regular use of passive-type constructions for this purpose, thus subject RCs will be less frequent in texts, and we will not expect to find passive SRCs produced instead of ORCs in elicitation experiments.^[4] The experimental data on relativization in Chinese has given very mixed results, with some studies (notably Hsiao and Gibson 2003, which contains references to other studies giving similar results) finding a preference for ORCs over SRCs, while others (Hsu et al. 2009; Lau 2016; Lin and Bever 2006, etc.) find a preference for SRCs over ORCs. It appears that, although there is no strong association of topichood with syntactic subject status, there is still a tendency to produce utterances in which the topic (relativized element in RC) plays the most prominent semantic role, e.g., (8).

This is the main type of evidence that has been proposed for a subject preference in Chinese. Thus there is good evidence that the link between topichood and subject status is a major factor in the subject preference in relativization, as shown by the fact that the subject preference varies in its strength and type of manifestations depending on the nature of the link between topic and subject status in the language in question.

Direct promotion of the relativized element to syntactic subject status is not the only way in which topicality affects accessibility to relativization. A related type of phenomenon involves the use of specialized relativization strategies typically associated with subjects for relativized elements with other roles if they are the main topic of the clause or show a particularly high degree of cognitive salience either intrinsically or compared to other referents. For example, possessors are generally animate and often function as topics in the ‘aboutness’ sense (Keenan and Comrie 1977: 91 discuss cases where they are preferentially articulated as subjects for this reason), and in languages with a subject participle, this is not infrequently extended to relativize possessor of the subject as well as subject (see Shagal 2017: 87–88). This type of process has different implications for the AH as compared to direct promotion to subject. Direct promotion to subject would particularly reduce the accessibility of non-subject roles typically associated with animate referents, such as possessor or IO (beneficiary of ditransitive), as they will be the most affected by promotion to subject. However, for the same reason, they are also the type of non-subject roles most likely to be associated with subject-specialized strategies, so, from this point of view, they will be considered high in accessibility.

DO, occupying second position in most versions of the AH, is generally associated with topicality to a higher degree than other object-like referents (with the possible exception of beneficiary in ditransitives). This role has been claimed to be typically associated with secondary topics (Dalrymple and Nikolaeva 2011), and some languages, notably many Bantu languages, have syntactic resources (applicative constructions) for promoting topical indirect objects and adjuncts to DO status, in a way that is parallel to the use of passive for promoting topical referents to subject status. Keenan and Comrie (1977) observe that in many languages possessing this type of construction, relativization is restricted to subject and DO. As in the case of Malagasy, we find that in languages where syntactic prominence has a strong association with pragmatic prominence, rather than depending on purely semantic factors, there are strict syntactic restrictions on relativization. Another syntactic phenomenon associated with secondary topic status is object agreement on the verb (see Dalrymple and Nikolaeva 2011, etc.). This gives rise to another way in which topicality can affect accessibility to relativization, as in some languages, such as Basque, only agreement triggers (subject and object) can be relativized. In ditransitive constructions, languages vary as to which object (DO or IO) can or must be the agreement trigger (see, e.g., Dryer 1986), as each may be considered prominent for different reasons, the former as semantically more closely linked with the verb, the latter as typically animate and thus more intrinsically topical than the former. Cases where IO may trigger object agreement, and relativization is restricted to agreement triggers, are key to the relatively high position of IO on AH. This can be linked to the association of IO with secondary topic status.

2.2.4 Affectedness and AH

As observed by Dryer (1986), another type of prominence, distinct from the pragmatic property of topicality, is the semantic property of affectedness, associated with patient or theme semantic roles. While topicality can be identified as the key property uniting the most prominent roles (A and S) in nominative-accusative systems and constructions, affectedness is the property linking the most prominent roles (S and P) in ergative-absolutive systems and constructions. Patient/theme can be envisaged as the semantic role that has the closest link to the activity denoted by the verb, often being physically affected by it, and in some cases, its existence depends on it. Patient/theme status has been identified as a factor that is relevant in accessibility to relativization. Fox (1987) proposes that it is in fact patient status (involving stative intransitive subjects and transitive DO), rather than syntactic subject status, that is key to the highest position on the AH, at least in terms of one of its manifestations, namely textual frequency in spoken discourse (she finds a 1:1 subject-object ratio for RCs). She finds that RCs in spoken discourse are most frequently used to characterize new referents, which is most effectively achieved using stative verbs, or transitive verbs that result in a state, which generally affects the patient object, rather than the agent subject. Transitive patient RCs are also favored as a means to anchor a new referent (typically the patient) by linking it with a referent that is already present in the discourse (typically the agent). Since this study mainly involves elicited data in which the RCs do not perform the function of characterizing new referents, it cannot evaluate the effect of this factor on accessibility to relativization.

However, it is likely that the concept of affectedness will be favorable to accessibility to relativization in another way, which can be measured by this study. Many languages, including Armenian, make use of so-called resultative participles in relativization (see Haspelmath 1994; Shagal 2017), which are typically associated with affected patients (stative intransitive subjects and transitive objects). These have often been said to be restricted to relativizing subject and DO referents (see e.g., Authier 2012: 230 for Persian, Jahukyan 1974: 552 for Armenian), and the restricted distribution of participial relativization in many languages that possess it is well-known as a manifestation of AH (see e.g., Lehmann 1986). However, it is in fact the semantic property of affectedness (having a property that is a state denoted by or resulting from a verb), rather than syntactic grammatical relation in itself, that is key to the operation of this type of participle.

The issue of affectedness, and its link to the characterization function of RCs, is also likely to be relevant to the low position of genitive on some versions of the hierarchy. Diessel and Tomasello (2005) find that this is the role showing the highest number of errors in their production experiments. However, they use contexts (e.g., ‘The man whose cat caught a mouse’) in which the possessor is totally unaffected by and in fact plays no role at all in the state of affairs denoted by the relative clause, thus the use of this clause to characterize the referent is not felicitous. Genitives that can be conceived of as affected by the state of affairs expressed in RC, such as possessor of an affected body part, may be expected to show a relatively high level of accessibility.

4.2.1 General overview

The second measure of accessibility to relativization is the percentage of non-finite (participial) as opposed to finite RC responses (Figure 5 and Table 2). In order to test the significance of this result, we used the same procedure as in the analysis of target responses (Section 4.1). The dependent variable is the choice of finite/non-finite form, and the fixed factor (factor of interest) is the factor Role with 11 levels (the 11 roles in Figures 4 and 5). The different Speakers were added as a random factor to the model. This model was compared to a model without the fixed factor Role. The result shows that the factor Role has a highly significant impact on the choice of finite/non-finite forms (χ² = 470, p < 0.001); see coefficients of the model in Appendix 3(D).

Figure 3:

The Silverstein hierarchy (Silverstein 1976).

Figure 4:

Percentage of target responses.

Figure 5:

Finite versus non-finite (participial) RCs.

Participial RCs, unlike their finite counterparts, lack a means for expressing the role of the relativized element within RC, and thus may be considered simpler and less explicit constructions. It is therefore to be expected that they are generally associated with roles at the top of the Relativization Accessibility Hierarchy: Jahukyan (1974: 552) states that they may only be used when the relativized element has the role of subject or DO in RC. There are two main participles used in relativization in Armenian, the resultative participle in -ats (-uk in some dialects), originally an adjective denoting a property resulting from an action, most frequently used to relativize elements with the patient role, and the subject participle in -oɣ, originally an agent noun, but now used to relativize subjects of all kinds. However, although these participles are most frequently associated with patient and with subject roles respectively, in colloquial speech, both participles may, under the right circumstances, be used to relativize essentially any grammatical relation.

4.2.2 Subject participle

The subject participle was originally an agent noun, and, in the literary language, is strictly confined to relativizing subjects, though not necessarily agentive ones, as shown by (36):

However, in colloquial spoken language, it may under some circumstances be used to relativize non-subjects. In the database, there are 56 instances where the subject participle is used to relativize non-subjects out of 483 instances of SPT (12% of SPT RCs relativize non-subjects). Note that context is crucial for the use of such forms; in isolation, the examples given below would naturally be interpreted as SRCs: (39) ‘the one who bit a mosquito’, (40) ‘your country that works’, (41) ‘the one who gave water’. However, the context provided by the questions makes it clear to which participant the participle refers, thus the speaker is able to use these forms without the risk of ambiguity.

The characteristics of these non-subjects relativized with SPT are exactly what we would expect of prime candidates for the assignment of subject properties (‘promotion to subject’): SPT is used for non-subjects when they are the most referentially prominent element in the clause, which generally occurs in cases when the syntactic subject is lacking in prominence. There are no examples in the corpus with overt, specific, human subject. The subjects of the SPT non-subject RCs attested in the corpus have the following properties:

Very often possessee, often inalienable (37), but not necessarily (38):

Referentially non-prominent, e.g., inanimate or hardly animate (stone, mosquito):

Non-specific/generic: there is even one example of a pronoun:

Referentially prominent but omitted because not salient in the context (lacking discourse prominence):

In addition to the non-salience of the subject, the relativized element itself must be either inherently referentially prominent (animate) and/or thematically prominent (agentive); the only examples of the use of the subject participle for inanimate non-subjects involve instrumentals, which could be understood as having some agentive characteristics (note the relatively high position of instrument in Bresnan and Kanerva’s (1989) thematic hierarchy^[7]):

Note that there is no prohibition on the use of SPT to relativize elements with patient role, as long as these are animate, and the subject is non-salient (see Example (39) and numerous others). Neither is it prohibited from relativizing inanimate elements, as long as they are subjects (see (36)).

The use of the subject participle for non-subjects is crosslinguistically most common when the relativized element is possessor of subject (see Shagal 2017: 87–88). This is part of a more general tendency, observed by Givón (2001: 193), for the assignment of subject properties to possessor of subject, especially when the possessor is human. However, in principle, in colloquial spoken Armenian it seems to be possible to use the subject participle to relativize practically any grammatical relation (with the possible exception of an adpositional object) given the above conditions, as shown by evidence from the corpus (Table 3).

Thus it is possible to interpret the use of the Armenian subject participle to relativize non-subjects as an instance of ‘promotion to subject’, i.e., of the assignment of morphological/syntactic subject properties (here, the use of a verb form generally associated with subjects) to the cognitively most prominent element in the clause, even if it is not the syntactic subject.

4.2.3 Resultative participle

The resultative (also referred to as perfective) participle is always formed on the perfective stem of the verb if there is one (with the ending -ats in most dialects, -uk in some), and in the responses to the questionnaires is strongly associated with perfective contexts: all contexts with 25–100% resultative participle responses were perfective. In the few examples where it is used in imperfective contexts, it is always associated with the patient role:

Perfective participles, and perfective constructions in general, often accord prominence to elements with a patient/theme role. This is the result of semantic properties: in perfective expressions, which describe completed events, what is crucial is often the result of the event, while for incomplete (imperfective) events there would be no result yet, and the focus of attention is more often on the action and/or the agent. The result of an event is most saliently associated with those elements that have undergone a change as a result of the event, i.e., those with an affected patient/theme role. The key properties associated with the use of the resultative participle in relativization are perfectivity and patient/theme role. In syntactic terms, this means that it is most commonly associated with direct object position, as this is the syntactic position most characteristic of the patient/theme role; all but two of the questionnaire contexts that had at least 50% RPT responses were direct objects, but it seems to be semantic role rather than grammatical relation that is crucial, as the two exceptions were patient/theme subjects:

Note that these, the only subject referents with 50+% RPT responses, are both inanimate. It is true that on average there is a higher percentage of RPT responses for inanimate than animate referents (total: 37 vs. 10%, subjects: 50 vs. 6%, DO: 76 vs. 67%). However, a closer investigation of the data reveals that semantic role is a more relevant factor than animacy. All but one of the contexts with 50–100% RPT responses (100% of contexts with 75–100%) have affected patient/theme (undergoing change of state, inc. position or possession) role. Animacy is clearly not a decisive factor, as the referents in this category (contexts with 50–100% RPT responses) are almost evenly divided according to animacy (14 inanimate vs. 12 animate).

As we have seen, RPT is not confined to DO use, but may also be used for other grammatical relations, with some patient/theme inanimate subjects receiving high percentages of RPT responses (up to 86%, average 50%). No animate subjects have more than 18% RPT responses; SPT is preferred to RPT for all animate subject referents. The two with the highest percentage of RPT responses are subjects of the verb meaning ‘go’, so although they are in some sense agentive, they also have the role of theme, undergoing a change of state (position):

However, there are a few examples where RPT is used for unambiguously agentive subjects:

Note that this seems to be possible only in perfective contexts; there are no examples of RPT use in contexts that are both imperfective and agentive. The use of RPT instead of tense/aspect-neutral SPT here serves to indicate that the action is completed and/or has past time reference, rather than that the referent has patient/theme status. These uses are at least in some cases associated with dialects that make use of RPT in the formation of regular past perfective tense forms so that the association with time reference and/or aspect has to some extent eclipsed that of semantic role.

More frequent than agentive subject uses, though still all with <50% RPT responses, are cases where relative clauses with RPT refer to places (48) (fixed location (48a), motion away from (48b) or motion towards (48c)), times (49), or possessors (50). These cases differ crucially from DO/subject uses in that the patient/theme referred to by the participle is not the relativized element, but the subject, as seen in the following examples:

(Examples (48)–(50) from Sakayan 1993: 363, transliteration, glosses and translations adapted)

For time and place (most commonly location, but also ablative and destination), it is noticeable that certain contexts have a relatively high percentage of RPT responses, while others have few or none. The following are the time (51) and place (52) contexts with the highest percentage of RPT responses:

Time and place contexts with a high percentage of RPT use share the following properties: perfective event, referent of participle undergoes salient change of state, referent of participle is highly salient (1st person and/or MC topic). In fact, there is a more general association of non-subject RPT uses with contexts where the subject is a pronoun, i.e., highly salient (Table 4).

In order to understand the patterns of usage of these constructions, it is important to remember that a restrictive RC functions to characterize the relativized element, to define its reference, often in order to make it in some sense identifiable to the addressee. An RC is used when the referent is defined in terms of something that is best expressed by a clausal constituent, i.e., an event or state of affairs, rather than in terms of a simple property that could be expressed by a non-verbal adjective. There are two main types of cases: in one case, the referent is directly involved in the event, typically either as agent or as having undergone a change of state as a result of the event; this mainly concerns subject or object RCs. In the other, less frequent case, the referent is identified in terms of its relation to an event involving another entity. This is more likely to be an effective way of identifying a referent if that other entity is highly salient (e.g., discourse participant and/or topic), and affected in a salient way by the event.

It is also important that the relationship of the relativized element with the state of affairs denoted by the clause is transparent. Thus these constructions are most frequent when the role of the element is obvious either from its own semantics (e.g., if it refers to a time or place) or those of the verb: the only examples of instrumental and postpositional object uses are with verbs that take instrumental and postpositional objects (which have stimulus role, causing a change of state in the subject referent of RPT).

INSTRUMENTAL

POSTPOSITIONAL OBJECT

However, participial RCs seem to be impossible for cases where the case or adposition makes an independent contribution to the meaning that is not recoverable from the context, which would be lost in a participial RC where these elements cannot be morphologically expressed: for example, a construction such as [cat sat] table could be interpreted as ‘the table that the cat sat on’, but equally as ‘under’ or ‘at’ (Shagal 2017: 74).

The only grammatical function for which there are no examples of RPT use is indirect object, i.e., the recipient/beneficiary argument of a ditransitive. It is not the case that the extra complexity of ditransitive constructions is in itself a decisive factor disfavoring participle use, as ditransitive DO referents appear among those with the highest (75+) percentage of RPT responses, for Example (55).

Thus the number of arguments in itself cannot be the reason why ditransitive IO is rarely relativized with participial constructions. There are in fact a few examples where SPT is used:

In this case, as discussed in Section 4.2.2, the subject is not salient in the context, and the IO is the primary topic. In the absence of context, this would naturally be interpreted as a subject relative: ‘the one who gives/gave water’. The same would apply to a parallel construction jur tvatsə with RPT, which would be preferentially interpreted as SRC with past time reference (there is a general tendency to interpret animate RC heads, as intrinsically topical elements, as subject, see Mak et al. 2002). Thus such a construction referring to IO would be dispreferred except when the context makes the reference totally clear. In addition, ditransitive IO generally has the role of recipient/beneficiary, and thus does not undergo a change of state as a result of the event, nor does the subject, the most probable primary topic, therefore RPT would be unlikely to be used.