Persian follows an accusative alignment. The subject controls agreement which is realized as agreement suffixes on the verb (3).

Psychological constructions, decomposed into a non-verbal and a verbal element, show an anomalous agreement pattern.3 With these constructions, specified in brackets (4–7), we do not see the expected subject agreement suffixes. Instead, oblique clitics, that are co-indexed with the sentence-initial DP, obligatorily appear on the nominal component of these constructions and the verb obligatorily inflects as [3sg] (the asterisk indicates the obligatory presence of these clitics).

The pattern in (4–7) raises the question of why we do not observe the expected agreement suffixes with these predicates. These patterns further raise a question regarding the nature of the oblique clitics. In some previous accounts for Persian psych constructions, the similarity of these constructions to possessives (8) has been taken as the base for considering a possessive nature for all psych constructions, taking the oblique clitics as a possessor-related element. (8a) illustrates a regular genitive structure, where the possessor that is expressed as a full DP is linked to the possessum by a nominal linker (Ezafe)4. (8b) shows that the possessor, realized as a clitic pronoun within the possessive phrase, is doubled with a sentence-initial possessor serving as the topic of the clause, identical to the pattern obtained with psych constructions in (4–7). I will return to this discussion in the following section and argue that a possessive-based analysis for all these predicates faces problems (for different possessive-based analyses of these constructions see S. Karimi 2005; Haig 2008; Y. Karimi 2013; Kazeminejad 2014).5

Some other authors have not taken a possessive base for these constructions. For instance, Ghomeshi (1996) takes the clitics as pronominal elements encoding an experiencer argument. Under this account, the sentence-initial DP is taken as a topic that is base-generated in the Spec of a Topic Phrase. Barjasteh (1983) posits that the subject of these constructions is underlyingly accusative, and the clitics are the result of accusative subject cliticization via which the subject cliticizes to the non-verbal element. Sedighi (2001) takes these clitics as the realization of clitic left dislocation that are co-indexed with the sentence-initial DP, merged as an experiencer topic. Sedighi (2005; 2009) analyzes these clitics as pronominal elements obligatorily encoding the experiencer argument that is introduced as an applied argument by what she calls a “Super High Applicative” phrase. Under some historical accounts, the oblique clitics have been taken as agreement markers that have evolved from pronominal enclitics in the language (see Rasekh-Mahand 2010, Jügel and Samvelian 2020).

The previous analyses all offer valuable insights which contribute to a better understanding of psych constructions in Persian. However, there are issues in these analyses that need to be addressed. One issue that arises regarding the possessive analyses is that while some of these constructions seem to have a possessive base, the majority of them do not. Furthermore, under the majority of possessive-based accounts, obtaining a possessor-related interpretation in these constructions has remained unexplained. Finally, previous accounts have treated these psych constructions as a unified phenomenon, an assumption which I will show to be misguided (see also Samiian and Moradi 2023 for observations regarding some mixed properties of these structures particularly regarding the status of the sentence-initial DP).

In languages with possessive datives, a possessor often carries an added meaning. Crucially, it has been shown that in some languages possessive datives can have a genitive alternation with a different thematic interpretation. In such languages, in the dative counterpart, the possessor is affected by the action denoted by the verb, and thus, receives a different thematic interpretation, compared to its genitive counterpart. For example, in the dative counterpart of the French example in (9a), the possessor is interpreted as a benefactive argument of the verb. By contrast, in the genitive counterpart (9b), the possessor does not convey any such added meaning besides being the possessor of the object. These possessive dative constructions have been attested in several unrelated languages. The thematic interpretation of these constructions, however, is subject to variation both within and across languages. For example, in the Spanish possessive dative construction illustrated in (10a), no affected interpretation is involved compared to its genitive counterpart (10b). However, crucially, there is still a thematic difference between the two. In the dative counterpart, the object of admiration is Valeria, as opposed to the genitive counterpart in which the object of admiration is Valeria’s patience.6 Here too, in the dative construction there seems to be an added meaning (other than an affected interpretation) that is absent in the genitive counterpart.

Under some previous accounts, possessive datives have been characterized as instances of external possession where the added meaning is encoded in a thematic position external to the possessive phrase. It has been proposed that the added meaning is obtained either via raising of the possessor to a new thematic position (Landau 1999; Lee-Schoenfeld 2006; Kato and Ordóñez 2019; Rodrigues 2023), or via a c-commanding/control relation established between an argument that is coreferential with a possessor (Guéron 1985; Borer and Grodzinsky 1986; Hole 2005; see Pylkkänen 2002 for further discussion on these viewpoints). For some other authors possessive datives are taken to be merged in the specifier of a thematic phrase––i.e., Applicative Phrase (Cuervo 2003; Torres Morais and Lima-Salles 2016).

In terms of word order and the placement of clitics, Persian psych constructions show striking similarities to dative possessive structures in Dominican Spanish (Kato and Ordóñez 2019). It has been proposed that the possessor (‘these teams) in (11) is merged inside a possessive DP and raises to an external position where it receives dative case from a dative head and obtains an affected interpretation in that new position (see Kato and Ordóñez 2019 (following Landau 1999); see also Rodrigues 2023 for an analysis along similar lines for dative possessives in Brazilian Portuguese). The possessor ultimately raises to the left-periphery of the clause and is doubled by a clitic through clitic left dislocation.

All Persian psych constructions that we saw above carry an affected interpretation involving some sort of psychological (and in some places a physical) relation with the event denoted in the clause. At surface, this interpretation could be best captured by an experiencer role under which an individual undergoes a mental or psychological experience as a result of the event denoted by the predicate (for similar affected experiencer interpretations see Cuervo 2003; Bosse and Bruening 2011; Bosse et al. 2012; Jiménez-Fernández 2020). Given the similarity of psych constructions to possessives in Persian, and along the lines of some previous accounts for dative possessives across languages, one might posit that the sentence-initial DP in psych constructions is a possessor that is base-generated within a genitive structure and obtains the experiencer interpretation via raising to a new thematic position, while being doubled by a clitic in the merge position (along the lines of e.g., Landau 1999; Kato and Ordóñez 2019; Rodrigues 2023). Based on the cross-linguistic observation made above, if psych constructions have a possessive base, one of our expectations is that these constructions should have a regular genitive alternative form as well (see French and Spanish patterns in (9–10)). Crucially, this expectation is borne out only with a small subset of the psych constructions in Persian. Therefore, it would be problematic to assume a possessive base for all of these constructions. As shown below, in the genitive pattern (12b–14b), the possessor DP immediately follows the possessum and is linked to it by a nominal linker (aka Ezafe). This alternation leads to the absence of the clitic on the nominal. Note that the possessor cannot be expressed as a full DP possessor, simultaneously on the nominal element and in the clause-initial position (12c–14c).

The question that we need to address is whether this possessive subset of psych constructions could be analyzed as an instance of a possessive dative structure found across languages (see Y. Karimi 2013 for the dative possessive analysis of psych constructions in Persian). In what follows, I will argue that these psych constructions are not instances of possessive datives.

As discussed above, across languages, there is always a thematic difference between dative possessives and their simple genitive structures. Crucially, in Persian, the genitive counterparts (12b–14b) convey the same interpretation as that of their clitic-bearing counterparts co-occurring with a sentence-initial DP (12a–14a). In other words, the genitive alternation, in which the possessor remains within the possessive phrase, is not correlated with the absence of an affected interpretation for the possessor, or any added meaning beyond affectedness which was shown to be the case in a language like French and Spanish. As such, the presence of a sentence-initial possessor in the (a) counterparts could not be correlated with the assignment of a new theta-role to the external possessor.

Furthermore, a part-whole relation, often formed through an inalienable possession relation is a common property of possessive datives across languages (reported in Cuervo 2003; Boneh and Nash 2013; Torres Morais and Lima-Salles 2016; Jiménez Fernández 2020; Rodrigues 2023).7 Considering this typological observation, it is important to note that the part-whole relation between the possessum DP and its possessor in Persian examples does not necessarily express an affected interpretation for the possessor. In (15), we see that the part-whole relation is found in what seems to be just a run-of-the-mill possessive structure, with the topicalized possessor lacking any affected interpretation (15c). Crucially, the part-whole relation in such possessive constructions involves no psychological import. As a result, the possessor does not obtain an affected/experiencer meaning, yet the same structure persists.8 In other words, (15) expresses nothing beyond its literal meaning: ‘my leg got cramped’. By contrast, the possessive psych constructions discussed earlier express some form of emotional or psychological experience. Thus, they convey an affected interpretation for the possessor. For instance, the event of ‘one’s heart being broken’ suggests that the possessor of ‘heart’ is emotionally affected by feeling heartbroken.

Additionally, some possessive psych constructions in Persian do not even contain the typical body-part possession relation. As observed in the examples given above, the possessum DP can be a nominal such as mood, etc. Finally, it should be noted that the external possessor structure is also found in genitive constructions that do not even involve a part-whole relation between the possessor and the possessum.9 I will return to this discussion in section 4.2 and argue that the possessive-type psych constructions are derived from a simple genitive structure. The sentence-initial DP in these possessive constructions is taken as an external possessor moving to the CP domain, hence serving as the topic of the clause.

In light of these observations, I posit that the possessive class of psych constructions is formed through a regular genitive construction and does not involve a dative structure. I further propose that the affected interpretation is not obtained structurally (as is the case with possessive datives in some languages), rather, it is derived from the combination of particular lexical items that are involved in this class, giving an idiomatic flavor to these constructions (e.g., the breaking of one’s heart ‘feeling heartbroken’, burning of one’s heart ‘feeling sympathetic’, one’s mood being ruined ‘feeling upset’, etc.). In other words, the combination of the lexical items in possessive psych constructions conveys a psychological experience leading to their characterization as a psych construction. Meanwhile, such psychological imports cannot be involved when the combination of lexical items in an expression lacks a reference to a psychological, mental or an emotional experience, even under a similar part-whole inalienable possession relation (e.g., ‘my leg got cramped’).

As idiomatic expressions, these psych constructions are expected to disallow the manipulation of their components. As expected, crucially, unlike regular possessive phrases, the pluralization of the nominal element in possessive psych constructions leads to an infelicitous result (16b), as opposed to (16a) which has a singular nominal.10 I will return to the discussion on this class of constructions in the following sections and propose a detailed account for their syntax.

Above we saw that a subset of psych constructions allows a genitive alternation. Crucially, unlike what we observed above in (12–14), we see that another class of psych constructions, shown in (17–19), does not allow a genitive alternation. In this class, the sentence-initial DP carries an experiencer role under which an individual undergoes an emotional or psychological experience.11 I take the failure of the genitive alternation test to indicate that this class––henceforth the experiencer class–– is not derived from a possessive base. Therefore, the clitics could not be taken as possessor pronouns in this class. The distinction between the two classes raises an immediate question regarding the source of these clitics in this new class of psych constructions. I will return to this question in section 4 and provide an account for the clitics in the experiencer class.

In this section, I showed that Persian psych constructions fall into two main categories. The ones that passed the genitive alternation test were argued to involve a simple possessive base. The ones that did not pass the genitive alternation test were categorized as the experiencer type, lacking a possessive structure. In the next section, we look at these constructions more closely and examine their syntactic derivations.

Complex predicates in Persian are formed by a non-verbal element (e.g., noun, adjective, PP) and a light verb. This is a common compositional property of complex predicates among Iranian languages (see Vahedi-Langrudi 1996; S. Karimi 1997; Karimi-Doostan 1997; Megerdoomian 2002; 2012; Folli et al. 2005; Gündoğdu 2015). One of the properties of complex predicates is their resistance against syntactic operations like the displacement of their components. This type of resistance could be attributed to the incorporated nature of the components of these predicates.12

One case of displacement occurs in cleft structures. As shown in the following examples, a regular complex predicate (20a) does not allow the clefting of its non-verbal element (20b). By contrast, the combination of an unaccusative verb and an internal argument (21a) allows a cleft structure (21b) via which the internal argument of an unaccusative verb undergoes clefting.

Furthermore, complex predicates do not typically allow the stranding of their non-verbal element. In (22), we see that the non-verbal element of the complex predicate raqs kardan ‘to dance’ cannot be stranded. This stands in contrast with the behaviour of an unaccusative simplex predicate selecting an internal argument. In (23), we see that the internal argument of an unaccusative verb allows stranding.

Finally, adverbials, even the ones known as low or vP-internal adverbials such as manner adverbs, often cannot intervene between the components of the complex predicate (the (b) counterparts of (24–25); the predicate is specified in brackets and the adverb is highlighted) (see Sedighi 2009). Following much of the literature on complex predicate constructions across languages (Megerdoomian 2002; 2012; Folli et al. 2005; Lomashvili 2010; Folli and Harley 2013; Lyutikova and Tatevosov 2013; Gündoğdu 2015; 2019), I take complex predicates as vP projections. I further take low adverbials as adjunction to vP (Figure 1).13 As such, the resistance of the components of a complex predicate against the separation by a manner adverb could be attributed to the adverb’s adjunction position, which typically occupies a higher structural position than where the predicate is formed.14

Unlike the observation above regarding complex predicates, a manner adverb can intervene between a simplex predicate and its theme argument. This pattern could be the result of the internal argument raising from the vP internal position to [Spec,TP].15 This movement leads to the placement of the adverb between the simplex unaccusative verb and its complement (the subject).

In light of these observations, let us now have a look at the behaviour of psych constructions with respect to these tests. The application of these tests to the two types of psych constructions, identified above, reveals a contrast between them. As shown below in (28–30), the possessive class allows the cleft structure. In the (a) counterparts, we see the possessive psych construction as the base structure. In the (b) counterparts, we see that the non-verbal element is clefted. Note that the bare nominal cannot be clefted on its own, without the presence of the possessor following it. As a result of this restriction, the possessor has to obligatorily appear following the non-verbal element. This restriction is indeed another indication of the possessive base for this class of constructions.

It is worth pointing out that in cleft structures (the (b) counterparts), we seem to need the full pronoun only. Therefore, the placement of a clitic on the non-verbal element yields an ungrammatical result (or at best a marginally acceptable result; compare the clitic bearing counterpart (30b) with (31)). The ungrammaticality/markedness of (31) could receive an independent explanation. It seems that where clefting involves a possessive structure, with a contrastive focus on the possessor, which is the case in (28–30), a full possessor DP is always favored to its pronominal counterpart. In (32), we see a regular possessive/genitive construction involving clefting. Here, likewise, the full DP possessor is favored over its clitic pronominal form. This restriction could be due to the fact that cleft structures are often associated with an inherent contrastive/focused interpretation (Kiss 1998; Drubig 2003 and references therein). As such, they generally require a full form of a pronoun so that a contrast is highlighted.16

Note that the cleft structures presented above sound even more natural under ellipsis where the non-verbal element is stranded with the possessor DP following it (33). This pattern again clearly suggests that firstly, these constructions have a possessive base, and secondly, they do not form a complex predicate, given that the non-verbal element can be stranded after ellipsis (while carrying a possessor). These two properties (i.e., stranding of the non-verbal element and its modification by a possessor) cannot be extended to complex predicates.

Above, we observed that the application of clefting and ellipsis yields grammatical results for the possessive class, providing an argument against a complex predicate source for these constructions. With these observations in mind, let us now have a look at the behavior of the experiencer class regarding clefting and ellipsis. As shown in (34–36), the predicates in the experiencer class do not allow clefting (i.e. (b) counterparts). They do not allow ellipsis either ((c) and (d) counterparts; (c) counterparts involve both clefting and ellipsis and (d) counterparts involve only ellipsis with the stranded non-verbal element). Note that since this class does not allow a full DP on its nominal (see section 2.2), we only see the clitic-bearing form.

Finally, if psych constructions form a complex predicate, similar to regular complex predicates, they are expected to resist the separation of their components by an adverbial. By contrast, if they are formed through the combination of an unaccusative verb and a theme argument, they should allow the separation by an adverbial. As we see below, the possessive class (37) allows the separation of its components by an adverb, while the experiencer class (38) resists it. Note that for establishing the contrasts more clearly, I am applying this test on the same constructions examined above. As such, while the constructions in (37) allow a regular genitive alternation, the ones in (38) do not.

Based on the diagnostic tests provided in this section, we saw that the predicates in the experiencer class form a complex predicate, while the ones in the possessive class are formed through the combination of a simplex unaccusative verb and a possessive internal argument. Crucially, the lack of a possessive base for the experiencer class correlates with their complex predicate nature since the nominal element of a complex predicate typically needs to be bare and non-referential, and thus, does not allow modification by a possessive pronoun.17 By contrast, in the possessive class, formed by a simplex verb selecting an internal argument, the nominal element is not involved in the formation of a complex predicate. Therefore, as expected, the nominal element can be stranded, while hosting a possessor as a type of nominal modification (either as a strong pronoun or an oblique pronominal clitic).

Finally, a question may arise regarding the unaccusative nature of the constructions in the possessive class. In what follows, I will apply two diagnostic tests showing that we are dealing with unaccusative predicates. The use of agent-oriented adverbs (Roeper 1987; Pylkkänen 1999) as well as control constructions could provide two suitable test cases in this regard. Agent-oriented adverbs modify agentive predicates and control constructions often require an agentive predicate in the embedded clause. As such, unaccusative predicates are predicted to be infelicitous in these contexts.18 (39) illustrates the infelicitous use of agent-oriented adverbs as well as control constructions with the predicates in (37a) (predicates in this class consistently exhibit the same restriction). The embedded CP in the control construction is specified in brackets. Compare (39) with (40) illustrating the felicitous use of an agentive-oriented adverb and a control construction with an agentive predicate (i.e., unergative dance).

The diagnostic tests presented above showed that the possessive constructions are infelicitous in agent-denoting contexts, providing evidence for their unaccusative nature. Overall, based on the discussion in this section, we can conclude that despite the striking surface similarities of psych constructions, they are in fact of two distinct types diverging from each other in two main aspects: their thematic nature and their compositionality. In the following section, I address the syntactic derivation of these constructions more closely and provide an account for the nature of the sentence-initial DP in the two classes as well as their agreement patterns. We will see that their agreement does not follow the same system either.

In the previous section, I argued that the psych predicates in the experiencer class form a complex predicate. I take the dative argument in this class as an affected argument that is involved in some psychological or emotional (or physical) experience, as a result of the event denoted by the predicate (Cuervo 2003; Bosse and Bruening 2011; Bosse et al. 2012; Jiménez-Fernández 2020). I further take the derivation of this class to involve an applicative phrase, where the experiencer is merged in the specifier of a high Applicative Phrase (along the lines of Pylkkänen 2002; Cuervo 2003; 2011; McGinnis 2008; Fábregas et al. 2017; Jiménez-Fernández 2020). As such, the structure of these predicates differs from the regular complex predicates in that these predicates obligatorily involve an ApplP on top of their vP, as part of their structure. This additional functional projection derives the dative semantics of these predicates (Figure 2).

I take the Appl head a φ-probe and the oblique clitics in this type of constructions as the result of a Spec-Head agreement between the Appl head and its specifier upon merge (Figure 3), taken as an instance of ‘inherent agreement’ (Wiltschko 2006; Coon 2017; Forbes 2018; Coon et al. 2021). This agreement is realized as oblique clitics occupying the second position within the verb phrase. This distributional behavior of clitics is reminiscent of the distribution of ergative agreement clitics in some Iranian languages (see Haig 2008; Y. Karimi 2010; 2017; Akkuş et al. 2024; Taghipour 2024 for different Kurdish dialects; see Moghaddam 2016 for Davani (Northwestern Iranian)). As such, these clitics obligatorily appear on the nominal component of the predicate that is the first element within the verb phrase.19

The current analysis of these oblique clitics as agreement markers raises the question of why these markers are not taken as clitic doubling on par with object clitic doubling found in the language. As shown in (41), the object (‘you’) is doubled by a pronominal clitic on the verb.

Importantly, while the doubling of the object is a consistently optional process, the presence of clitics in the experiencer psych constructions is obligatory (see Samvelian and Tseng 2010 for a discussion on object clitic doubling in Persian; for the optionality of clitic doubling or constraints on its occurrence in other languages see Dobrovie-Sorin 1990; Alexiadou and Anagnostopoulou 2000; Anagnostopoulou 2003; Yuan 2020). It is worth pointing out that the agreement analysis of the clitics in the experiencer class is along the lines of earlier accounts in the literature on ergative subject clitics in Iranian languages analyzing them as agreement elements (Roberts 2000; 2001; Moghaddam 2016; Akkuş et al. 2024; Taghipour 2024).20

We are still left with a question regarding the precise nature of the sentence-initial DP in this class of predicates. Datives have been shown to move to the left-periphery of the clause, serving as the topic in some instances (for this discussion in the context of Romance languages see Fábregas et al. 2017; Jiménez-Fernández and Rozwadowska 2017; Kato and Ordóñez 2019; Jiménez-Fernández 2020; Rodrigues 2023). If dative experiencers in Persian move into the TP domain, they are expected to exhibit subject properties, and hence could be taken as subjects. In Persian, one of the key properties of subject DPs is their ability to control agreement (see Bošković 2016 for taking the lack of regular agreement morphology with wh-subjects as an indication of their movement to [Spec,CP], bypassing [Spec,TP]; see also Kato and Ordóñez 2019 for a similar effect). We have already seen that the experiencer DP does not control subject agreement in Persian dative psych constructions. Below, I argue that such DPs are in fact topicalized elements moving into the CP domain, while bypassing [Spec,TP]. This analysis is extended also to the possessive class (see section 4.2).

The epistemic modal tavānestan ‘can’ has been taken as a raising predicate in Persian (Béjar and Kahnemuyipour 2023).21 In (42), the DP in the matrix clause (‘students’) triggers agreement on the modal, indicating the raising of the DP to a subject position––a canonical property of subjects in Persian.22

Crucially, the sentence-initial DP in dative psych constructions does not trigger agreement on the modal (43–44). The contrast with (42) indicates that the dative experiencer could not have moved to a subject position, otherwise the agreement should have been controlled by the experiencer DP (see Boeckx 2000; Tubino 2009; Jiménez-Fernández and Rozwadowska 2017 for using seem/appear type of raising constructions as a diagnostic test for subject properties of datives).23

Additionally, in clauses formed by the progressive auxiliary dāštan ‘have’, agreement is triggered both on the main verb and on the auxiliary (45).24 Crucially, in clauses formed by dative psych predicates (46), while we see the expected oblique clitics tracking the dative experiencer, the progressive auxiliary consistently shows the [3sg] inflection, indicating that the sentence-initial experiencer is not occupying subject position (i.e., [Spec,TP]).

In light of these observations, I propose that the experiencer DP in the dative experiencer class moves to the CP domain and serves as the topic of the clause.25 I further take the [3sg] verbal agreement in the experiencer dative constructions as the default realization of the φ-features on T, which have not been valued in syntax. I assume that an unvalued feature in a well-formed derivation can survive LF (along the lines of Béjar 2003; Preminger 2011; 2014; Kornfilt and Preminger 2015; Levin 2015). As such, the presence of unvalued φ-features on T does not induce an ungrammatical result (see Y. Karimi 2010; 2017; Parker 2020; Taghipour and Kahnemuyipour 2021; Taghipour 2024 for a similar viewpoint about the [3sg] inflection of the verb in the ergative alignment of some Iranian languages).26

I will close this section by a brief discussion on the resistance of the dative experiencer predicates against nominalization. In section 3, it was argued that the predicates in the dative experiencer class form a complex predicate since their behaviour in a number of diagnostics (clefting, ellipsis and adverbial placement) conforms to the behaviour of regular complex predicates. This pattern stands in contrast with the behaviour of the possessive class showing diverging properties in these respects. The predicates in the dative experiencer class, however, illustrate an important distinction compared to regular complex predicates: unlike regular complex predicates, these types of predicates resist nominalization. Before proceeding any further, let us have a look at the patterns of nominalization in Persian.

In Persian, nominalization (also known as masdar) is formed by attaching the nominalizing suffix -an to the past verbal stem (47). In (48), we see the nominalization of regular complex predicates by the attachment of the nominalizer suffix to the light verb.

In contrast to regular complex predicates, in (49) we see that the dative experiencer predicates, categorized as complex predicates, do not allow nominalization. The failure of nominalization with this class of predicates raises a question regarding the complex predicate nature of such predicates (see Sedighi 2009).

It has been proposed that vP is the structural height of nominalization in Persian (Kahnemuyipour 2022). Non-specific objects (50a) and manner adverbs (50b) have been shown to be within this domain, as opposed to specific objects (in non-generic contexts) (50c) and speaker-oriented adverbs (50d) that are excluded from nominalization (see Kahnemuyipour 2022 for further discussion on this topic).27

As noted earlier, vP is a level that includes also both components of the complex predicate. In other words, a regular complex predicate is formed as a predicate within this domain. Crucially, the situation differs with dative experiencer psych predicates. These predicates involve an applicative phrase (ApplP) above vP in their derivation, a structural property that sets them apart from regular complex predicates in the language. I propose that nominalization fails to produce a grammatical result with the experiencer dative psych predicates because the predicate is not fully formed when nominalization occurs (cf. Y. Karimi 2013). In other words, this failure arises due to the presence of an ApplP above vP, leaving the predicate incomplete at the vP level. As such, I do not take the failure of nominalization as a point against a non-complex predicate source of these constructions. Furthermore, other tests presented in section 3 provided support for taking the dative constructions as complex predicates. In the next section, I will provide my proposal regarding the agreement patterns in the possessive psych constructions.

In section 2, it was argued that the possessive psych constructions, exemplified in (51a), could not be taken as an instance of dative possessives attested in other languages. Rather, I posited that this class of psych constructions is no different from a regular genitive structure in Persian. On par with regular possessive structures, I propose that the external possessor in this subclass is introduced within the possessive phrase and raises to the CP domain, serving as the topic of the clause. This movement further leads to clitic left dislocation in the base position within the possessive phrase which is reminiscent of the pattern found in the external possessor constructions in Romance languages (see e.g., Kato and Ordóñez 2019), schematized in (52). As such, in the presence of the external possessor in the sentence-initial position, the full DP possessor on the possessum is predicted to be ill-formed ((51b) versus (51c)). Furthermore, in contexts where the possessor raising does not occur, the possessor can optionally be realized either as a full DP, or in its pronominal clitic form following the possessum (51d).

The lack of subject agreement with the external possessor could indicate that the possessor’s landing site lies outside the TP domain.28 The same agreement obtains in regular possessive constructions that contain a topicalized possessor. In (53), the [3pl] agreement tracks the possessum DP, rather than the topicalized external possessor.

Additionally, the [3sg] agreement obtained in constructions formed by the epistemic modal tavānestan, taken as a raising predicate (54) and the progressive auxiliary dāštan (55) provides further support for taking the external possessor in the CP domain, bypassing [Spec,TP]. In section 4.1, we saw that subjects always control agreement in these contexts (also see section 4.1 for the application of these tests on the experiencer dative class).

It is worth pointing out that taking the possessor to raise to a position above TP aligns with the account proposed by Kato and Ordóñez (2019) for external possessors in Dominican Spanish. However, Persian possessor raising presents a notable contrast. In Dominican Spanish, the possessor is argued to first move to a dative thematic position, acquiring an affected interpretation, before raising to a topic position above TP (see Kato and Ordóñez 2019 for further discussion; see also section 2.2). As mentioned in section 2, the external possessor in psych constructions and regular possessives in Persian is not associated to an affected interpretation or any added meaning. As such, in the current analysis, I do not consider the landing of the possessor in an intermediate thematic position. Rather, I take the possessor directly raise to the CP domain. Additionally, I analyze the whole possessive phrase as the subject of the clause. In Section 3.1, we observed that in cases of vP ellipsis involving nominal negation, the entire possessive phrase is stranded. Below, we further see that the whole possessive phrase, consisting of the possessor, also survives the too (ham) replacement targeting the vP (56). This indicates that similar to regular subject DPs, the whole possessive phrase is outside of the vP domain.

The behaviour of the possessive psych constructions in nominalization provides further support for taking the whole possessive phrase as the subject of the clause. If the whole possessive phrase is the subject of the clause, similar to a regular subject of an unaccusative verb (57), the whole possessive phrase (consisting of the possessum and the possessor), should be excluded from nominalization. As we see in (58), this prediction is borne out. This indicates that the whole possessive phrase is the subject of the unaccusative verb, and thus similar to the pattern in (57), this DP cannot be part of a nominalized structure.29 Finally, due to the [3sg] property of the possessum DP, these clauses always show [3sg] agreement (cf. Sedighi 2005; 2009). Figure 4 illustrates the structure of possessive-type constructions (dotted line indicates agreement).

In light of the observations made in this section, I further conclude that the Persian external possessor examined above cannot be analyzed similar to previously reported patterns of external possession, where the possessor raises to an athematic A-position, as reported in Nez Perce and Tagalog (see Deal 2013; Nie 2019). Firstly, it has been proposed that in such instances, the possessor raising is a case-driven movement, and thus, it is always obligatory. Crucially, possessor raising in Persian is not an obligatory operation. Secondly, by examining various constructions in this section, I showed that the possessor could not have raised to a subject/argument position. Rather, I argued that the landing site of the external possessor is the CP domain. Thus, it was taken as an instance of A-bar movement.

In this section, I examined the argument structure and the agreement patterns of psych constructions. I provided an account for their differences at thematic and syntactic levels. Overall, the facts examined in this paper showed that how two seemingly identical constructions can be derived from distinct underlying systems.